Introduction
Among the biological variables characterizing a population of wild animals, the age of individuals is one of the most relevant to acquire (Le Clercq et al. 2023; Piferrer and Anastasiadi 2023). Indeed, age assessments are carried out for a wide variety of purposes; for example, assessments of age-class structures over multiple years are becoming crucial for the management of exploited populations and the conservation of threatened species (Piferrer and Anastasiadi 2023). Age-based information, such as the duration of life stages (e.g., dispersal phase and reproductive maturity) or growth and mortality rates, is fundamental to parametrize both demographic (Iannelli and Milner 2017) and dispersal models (Swearer, Treml, and Shima 2019). These models allow conservation efforts to be prioritized toward the most vulnerable zones (e.g., major “source” areas, nursery zones) and stages (e.g., early juveniles and most fecund adults; Beger et al. 2022). Various complementary methods have been developed to assess the age of wild individuals based on size, mark-recapture approaches (Amstrup, McDonald, and Manly 2005), or predictably evolving biochemical/morphological features (e.g., sclerochronology, pigments, and hormones; Zhang et al. 2024). However, they all have drawbacks in terms of resolution, universality (most are group-specific), feasibility, efficiency, or cost (Le Clercq et al. 2023). To address these challenges, several genetic-based methods have been proposed recently, such as telomere length (Haussmann and Vleck 2002). Cooke and Smith (1986) initially identified telomere length as a genetic feature that evolves with age in humans, and it has been used since 2002 to age more than 100 species (Le Clercq et al. 2023). However, a relationship between epigenetics and age was discovered more recently by Horvath (2013). Such a relationship has already been characterized for nearly 100 species, and epigenetics now appears to be more promising than telomere length for estimating age, in terms of resolution (Le Clercq et al. 2023).
Among the epigenetic mechanisms that control gene expression without actual changes to DNA sequences (e.g., histone/chromatin modification and non-coding RNA), the potential of DNA methylation for constructing epigenetic clocks has been the most extensively studied to date (Booth and Brunet 2016; Trautner et al. 2017). DNA methylation is a reversible dynamic process that involves binding a methyl group or its oxidative derivatives (e.g., hydroxymethyl, formyl, or carboxyl group) to all four types of nucleotides (Carell et al. 2018; O'Brown and Greer 2016). Unlike epigenetic drift and environmental regulation, which are sources of variability between individuals' epigenomes, DNA methylation consistently evolves with age across individuals and even tissues (Tangili et al. 2023), although the underlying mechanisms are not yet well understood (Piferrer and Anastasiadi 2023). Certain differentially methylated sites across ages (referred to as “aging sites” hereafter) even share common characteristics among species (Klughammer et al. 2023), which has made it possible to fit universal epigenetic clocks for 185 mammal species (Lu et al. 2023; Robeck et al. 2021) and four fish species separated by up to 433 million years (Mayne et al. 2020, 2021). Methylated nucleotides were first detected through DNA modifications using antibodies, restriction enzymes, and bisulfite treatment (restricted to cytosine methylation; Anastasiadi and Piferrer 2023). The direct detection of methylated bases without altering DNA is now possible using third-generation sequencing, as methyl groups produce distinct electrical patterns (Nanopore; Wescoe, Schreiber, and Akeson 2014) or speed patterns (PacBio; Flusberg et al. 2010). This approach also offers the potential to reliably discriminate among various methylation types: 4mC, 5mC, 5hmC, and 6 mA currently (Liu et al. 2021).
In parallel to the epigenetic and third-generation sequencing breakthroughs in recent years, environmental DNA (eDNA) detection has revolutionized the assessment of biodiversity (Beng and Corlett 2020; Díaz-Ferguson and Moyer 2014; Polanco et al. 2021). Detecting transient DNA traces of animals in water, sediments, or digestive systems is now an affordable and non-destructive method that has been applied to swiftly and simply monitor communities' biodiversity (Beng and Corlett 2020; Rishan, Kline, and Rahman 2023). Notably, eDNA has various advantages for population genetic structure assessments, which are as crucial to demographic/dispersal models as age (Adams et al. 2019; Rishan, Kline, and Rahman 2023; Yao et al. 2022). While eDNA sampling is becoming more efficient and standardized, a major limitation is that it is not capable of obtaining individual-scale information, such as sex, age, and condition (e.g., hormones and isotopes), or individual counts (Adams et al. 2019; Beng and Corlett 2020). eDNA epigenetics assessed from full-length sequences present in the environment, using third-generation sequencing instead of short metabarcoding regions, has the potential to overcome some of these challenges (Yao et al. 2022). Its ability to discriminate among the four life stages of the freshwater snail
Age is particularly important for stock assessments of bony fishes (Actinopterygii; Punt, Allen Akselrud, and Cronin-Fine 2017), as over 50% of fish meat still comes from wild stocks (> 2200 species; Boyd, McNevin, and Davis 2022). This sets these fishes apart from most other heavily exploited taxa, emphasizing the need for proper management to prevent collapse in the face of the growing demand (Boyd, McNevin, and Davis 2022). Counting daily (in early life stages) or seasonal increments of otoliths (ear stones) is the standard age assessment method for fishes, and it has been estimated that several millions of otoliths are aged each year (Campana and Thorrold 2001). However, this method involves killing the fish, does not work for all fish species, suffers from various methodological biases, and is highly time-consuming (Piferrer and Anastasiadi 2023). These drawbacks have prompted several attempts to improve this process through automated annotation and/or spectroscopy (Benson et al. 2023), as well as the use of radiocarbon/radiometry aging for a maximum resolution of about 1 year in adults (Piddocke et al. 2015). Despite the importance of developing new aging methods for fish species, only a few studies (Anastasiadi et al. 2018) on epigenetic clocks have been conducted with fish (e.g., Anastasiadi and Piferrer 2020), in contrast to 100 studies for wild mammals (Le Clercq et al. 2023). Considering wild fish species in particular, two studies showed that epigenetic clocks had uncertainty ranges comparable to those observed with otolith and radiocarbon aging, which are commonly used as calibration methods (Mayne et al. 2023; Weber et al. 2022). Epigenetic clocks might even be more accurate than other methods for early life stages, as they are characterized by a higher rate of methylation change (Bertucci et al. 2021). This is suggested by the higher accuracy of epigenetic clocks fitted to juveniles compared with those for adult fish (Anastasiadi and Piferrer 2020; Mayne et al. 2023).
In this study, we used eDNA to fit an epigenetic clock, taking advantage of two promising features of nanopore sequencing that have not yet been applied for the detection of animals through eDNA and aging: amplification-free sequencing and detection of multiple methylation types at the single-nucleotide scale. We focused on the larvae of seabass (
Materials and Methods
We carried out our experiments on a pool of seabass larvae (
[IMAGE OMITTED. SEE PDF]
To limit human contamination of the samples, we conducted each experiment in a dedicated aquaculture zone, where it was mandatory to wear latex gloves, surgical masks, laboratory coats, and hair nets. The first phase of the experimental procedure involved decontaminating all sampling materials and tanks with bleach (10%), then rinsing them with distilled water. We then used a 2 L beaker to transfer 16 L of seawater containing seabass larvae from the aquaculture tank to a bleached tank with a tap fitted near its base, which was topped with a suction strainer. We released about 13 L of seawater through the tap before refilling the tank with filtered seawater, which had been pumped at sea and filtered through a 1 μM mechanical filter. We repeated this operation three times to ensure that most seawater from the aquaculture tank had been eliminated. Then, using a 1 L beaker, we transferred 4 L of filtered seawater containing larvae into three 5 L canisters (replicates), whose screwed caps were sealed using parafilm. As a control measure to monitor potential DNA contamination of the samples, we filled another canister closed similarly and filled only with 1 μM filtered seawater from the aquaculture station. We left it in the same conditions than the replicates (i.e., positioned next to them; Figure 1).
After 30 min, we filtered a volume of 2–3 L from each canister through a Sylphium eDNA 0.45 μM dual filter capsule using an Athena peristaltic pump (Proactive Environmental Products LLC, Bradenton, Florida, USA; nominal flow of 1.1 L min−1) and disposable sterile tubing for each filter capsule. The filtered volume was only 1.3 L for one replicate at 7 DPH, as the filter clogged too quickly, and it was increased from 2 to 3 L after 12 DPH, due to lower larval densities. The filtering was performed sequentially by a single person, which means that larvae had 0.5–1.5 h to shed eDNA (duration noted for each filter). At the end of each filtration, we emptied the water inside the filter capsules and then filled the capsules with 5 mL of Longmire lysis buffer solution.
We extracted DNA from all filter capsules within 2 weeks after the end of the experiment. One replicate at 7, 10, and 26 DPH could not be processed because the filter tip was broken during transport and all liquid leaked out. We carried out the eDNA extractions in facilities dedicated to this purpose, following a protocol adapted from Pont et al. (2018). Modifications were as follows: We retrieved 2 mL of buffer for the extraction; when removing the supernatant, we left 0.6 mL of liquid at the bottom and then mixed it with 1.32 mL of absolute ethanol and 60 μL of 3 M sodium acetate; and we performed the final elution step with 2 × 50 μL SE buffer. We measured DNA concentrations with the Qubit high sensitivity dsDNA kit (Thermo Fisher Scientific, Bremen, Germany) following the manufacturer's instructions. We detected no contamination during extraction, as blanks for each round of extraction presented a DNA concentration below the Qubit detection limit (< 0.0005 ng μL−1). Finally, we stored the filter extracts at −20°C until further processing.
We pooled the extracts from the three sample replicate filters in 1.5 mL Eppendorf tubes to increase the amount of total DNA for sequencing. In cases where one sample was unavailable due to leaked filters, only two were pooled, resulting in a total of nine pooled samples (P1–P9) and nine negative controls (C1–C9) ordered by age (7–28 DPH). Following the pooling procedure, the total DNA in most pooled samples remained below the required 400 ng per sample for Oxford Nanopore Technologies (ONT) sequencing (Table S1 in Data S1). However, three pooled samples met the sequencing criteria (Table S2 in Data S1).
Following advice provided by Oxford Nanopore Technologies in the case of low DNA input, we fragmented the DNA molecules to ensure that a sufficient number of molecules would pass through the nanopores and that none would be blocked. We concentrated each sample to 50 μL by heated evaporation (2 h, 30°C) using a Concentrator 5301 (Vaudaux-Eppendorf, Schönenbuch, Switzerland). We added 15 zirconium oxide beads (1.4 mm; Precellys P000927-LYSK0-A, Bertin-Instruments, Montigny-le-Bretonneux, France) to each tube, placed the tubes in a MM 301 homogenizer (Retsch, Haan, Germany), and agitated them twice for 10 s at 30 Hz, inverting tube positions between the two runs to ensure symmetrical treatment of all samples. We recovered as much supernatant as possible (typically ~45 μL, i.e., 90%) and centrifuged it for 1 min at 21,300 × g to pellet proteins. Again, we kept only the supernatant for the next steps, discarding the protein pellets. DNA contents measured using the Qubit high sensitivity dsDNA kit (Thermo Fisher Scientific) revealed that only two pooled samples contained the amount of total DNA required for ONT library preparation (Table S2 in Data S1).
We then concentrated the samples to 11 μL, using the same heated evaporation method as above (45 min, 30°C). Subsequently, we prepared the library using the Native Barcoding Kit 24V14 (SQK-NBD114.24, Oxford Nanopore Technologies, Oxford, United Kingdom) according to the ligation sequencing gDNA native barcoding v14 protocol from ONT, using a short fragment buffer during adapter ligation and cleanup and adding BSA (Bovine Serum Albumine) during flow cell flushing. P8 and P9 were inadvertently mixed during library preparation, prohibiting differentiation between the last two experimental stages (26–28 DPH). We analyzed the final library on an TapeStation 4150 (Agilent, Santa Clara, United States) using Genomic DNA ScreenTape, which showed a peak fragment length of 7 kb (94.19% of fragments were between 3058 and 23,454 bp). We loaded and ran the library on an R10.4.1 flow cell (FLO-MIN114, ONT) on a MinION Mk1B (MIN-101B, ONT) sequencer (Figure 1).
Basecalling and Methylation Calling
We basecalled the sequencing files not filtered by their quality score using the latest and most accurate basecalling model of ONT ([email protected]) available to date in the open-source basecaller for Oxford Nanopore reads, Dorado. We demultiplexed the file generated using the guppy_barcoder program and subsequently aggregated all generated outputs per barcode and converted the aggregate to a single FASTA file, keeping track of read names and barcodes.
We detected seabass reads among all the other sequenced reads using a 90% threshold of minimum similarity in VSEARCH (Rognes et al. 2016) to a reference seabass whole genome. We downloaded this genome from NCBI (RefSeq: GCF_905237075.1, Tine et al. 2014), which was chunked in 1 kb fragments to speed up the assignation on a CPU server and obtain the most assigned reads (Figure S14 in Data S7). We selected this similarity threshold because it corresponds to the optimal gap to delineate genera for many mitochondrial genes (Ruiz et al., in prep), enabling the identification of all Dicentrarchus reads while accounting for potential basecalling errors. We verified the absence of the only other species in this genus (
After these cleaning steps, we performed the methylation calling on seabass reads using the submodels 5mC, 5mCG_5hmCG (CG meaning in the CpG context only), and 6 mA integrated in the Dorado program. We mapped the output of the methylation calling both on the reference mitogenome (RefSeq: NC_026074) and on the full reference genome (containing this mitogenome) using the aligner program from Dorado, as well as the sort and index programs from SAMtools (Danecek et al. 2021). We then used the modbam2bed program from ONT to generate methylation summaries based on the mappings across the whole reference genome for each type of modification (i.e., modC, 5mC, 5hmC, modA, and 6 mA) detected separately by each submodel (eight combinations).
From these analyses, we obtained the methylation summary for each mapped site (e.g., modification type, reliability score, coverage, and methylation level), and we summarized this information for each barcode using R version 4.1.1 (Data S5). As samples collected at 7 DPH (i.e., too few reads in barcode 1) and 26/28 DPH (i.e., mixed Barcodes 8 and 9) were not suitable for further analyses, we used R to identify aging sites as differentially methylated sites (i.e., at least one unequal methylation level) across the Barcodes 2–7 (10–24 DPH).
Mapping Analysis
For each type of mapping (i.e., mitogenome or full genome), we summarized the correspondence between the reference genome and seabass reads, potentially reflecting natural intraspecific variations or basecalling errors, using the “read accuracy” and “read identity” concepts introduced with third-generation sequencing. We applied a modified version (i.e., no implementation of read identity and a different formula for read accuracy) of an R function (import_bam_file) provided by Gleeson et al. (2022) to convert the file from the methylation calling into the concise alignment format CIGAR, from which we computed both metrics. First, we calculated the read accuracy, which corresponds to the ratio of the number of matches to the alignment length. Second, we computed the read identity as the ratio of the number of matches to the number of bases aligned (i.e., not considering gaps). This definition was the most useful in our case because we did not expect any nucleotide insertion or deletion caused by basecalling errors (Data S4).
Epigenetic Clocks
We fitted epigenetic clocks using penalized regularizations (lasso, ridge, or elastic-net regressions) that individually shrank the weight of each aging site since their number was much greater than the number of ages in this study. To determine if factors other than methylation levels had to be taken into account when using grouped penalized regularizations, we preliminarily used a permutational multivariate analysis of variance (PERMANOVA) and a Mantel test to check for any significant links between methylation levels and variables characterizing aging sites (Data S6). Based on these tests, we decided to fit epigenetic clocks taking into account all four variables characterizing the tested aging sites: (1) the gene obtained using a custom R function from the GenBank file of the reference mitogenome (NC indicating “non-coding” for non-annotated sections), (2) the modification type (5mC, 5hmC, Other modC, and 6 mA), (3) the mean coverage across experimental stages, and (4) the reliability score, taking into account the chances of confusion with another modification type for each read, which would ultimately affect the modification frequency.
First, we objectively chose the best values of the penalty factor α by testing each value between 0 and 1 by increments of 0.01 to fit both ridge regression (α = 0) and lasso regression (α = 1), as well as elastic-net regression for all values of α in between 0 and 1. Second, we objectively determined the optimal number of folds for the cross-validation, although in our case with only six different ages we could only use a number of three (also not split into training/testing datasets). Third, due to large variability between models fitted on the same values, and to ensure convergence toward a stable optimal model (Figure S4 in Data S3), we employed bootstrapping (10 parallelized iterations) to fit epigenetic clocks for all combinations of penalty factors and numbers of folds. Fourth, along with modification frequency, we considered all possible combinations of the four factors mentioned above (Figure 5), by pasting the considered features so that sites with the same label were grouped together for regularization. Finally, we selected the best models, defined as those with the lowest median absolute error (MAE) during the cross-validation, as well as the lowest MAE during the final prediction, using the full training dataset as input. We standardized these two metrics by giving them the same weight and then summed them to select the best model, that is, the one with the lowest overall score (Data S6). We created two R functions based on the “glmnet” (Friedman, Hastie, and Tibshirani 2010) and “grpnet” (Helwig 2023) R packages to change fitting/selection criteria and make profit of a trained model to predict new ages from a testing matrix (see commented functions on Github: ).
We fitted epigenetic clocks both on the full dataset and on datasets containing only methylated adenosines (6 mA) or only methylated cytosines, which could be 5mC, 5hmC, or modC that were not characterized as 5mC or 5hmC by both models (named “Other modC”). We removed sites characterized as 5mC by the 5mC submodel and as 5hmC by the 5mCG_5hmCG submodel (six in total).
Results
Nanopore Sequencing of Seabass Larvae
The amount of eDNA extracted averaged 303 ng per sample, but the mean coefficient of variation between experimental stages (87%), and even between replicates (69%) was very high. A robust linear regression revealed that the estimated biomass of larvae per canister was the best predictor of the extraction yield, compared with the duration of eDNA shedding, the volume filtered, or the larval age (t(19) = −4.754, p < 0.001, rs = 0.28; Figure S1 in Data S1). The DNA input was lower than the recommended 400 ng for ONT library preparation, both for control samples (μ = 99, σ = 89 ng) and for most pooled testing triplicates (μ = 260, σ = 333 ng), notably due to a great loss during shearing (μ = 77%). Nevertheless, we were able to obtain a total of 1.8 million reads, with an average length of 1.35 (σ = 1.85 kb). Neither the total number of reads (t(16) = 0.230, p = 0.82) nor the number of reads (t(16) = −0.192, p = 0.85) presenting a similarity ≥ 90% to the 1 kb chunked reference seabass genome (“seabass reads”) were significantly linked to the DNA input weight (Figure S2 and Table S3 in Data S1).
Seabass reads were found in all control and test samples (Table S3 in Data S1), but their mean proportion of the total number of reads per sample was much higher for test samples (μ = 6, σ = 5%) than for control samples (μ = 1, σ = 2%). Complementary blasts of all fast basecalled seabass reads (Q-scores > 8) to all
The two approaches using the whole NCBI nucleotide database or reference complete mitogenomes (i.e., RefSeq database from NCBI) as targets during the assignation of fast and superior basecalled reads, respectively, were also adopted to identify non-seabass reads. Human DNA contamination was successfully minimized during sampling (e.g., 0.1% of total eukaryotes' mitogenomic reads). However, our protocol—carried out in a non-sterile environment using 1 μM filtered seawater—was hampered by airborne and waterborne contamination from microorganisms (i.e., > 30% of bacteria and > 23% of fungi), as 94% of the reads obtained in test samples with amplification-free nanopore sequencing were not assigned to
Although a correct quantification of the read accuracy was not possible since we did not know the genomes of our specific seabass population, the read identity showed that there was at least < 0.7% (median) of basecalling errors both for all fragments (Figure 2) and for mitogenomic fragments only (Figure S6 in Data S4). Longer reads had a significantly lower read identity, and the other significantly correlated variables in a robust regression were those characterizing the alignment quality with the reference genome (Figure S7 in Data S4).
[IMAGE OMITTED. SEE PDF]
Detected Aging Site Types
Taking into account the reliability score, we detected a methyl group on 0.8% of the mapped cytosines or adenosines, which yielded a mean per test sample of interest (10–24 DPH) of 99,187 modA sites and 154,772 modC sites (Figure S8 in Data S5). By comparing outputs from the 5mC and 5mCG_5hmCG submodels, we estimated that about 84% of modC sites were in a CpG context, and 5.6% of these CpG sites were hydroxymethylated (5hmC could only be detected in a CpG context at the time of analysis). Another fraction (not quantified) of modC sites in CpG contexts only (5mCG_5hmCG) was categorized as neither 5mC nor 5hmC.
Unlike methylated sites, there were more modA than modC aging sites (ratio from 0.64 to 1.55) for a total of 493 aging sites, and the proportion of 5mC over all modC sites also decreased from 97.7% to 89.6% compare to all methylated sites (Figures 3B and S8 in Data S5). We only detected aging sites on the mitogenome which represents less than 0.06% of the mapped cytosines and adenosines (Figure 3A). Indeed, mitogenomic mapped sites and aging sites had a mean coverage of 3.00× instead of 1.01× if all mapped sites were considered, which had a large significant effect in a non-parametric analysis of variance (ANOVA; F(1,36) = 44.55, p < 0.001, ) compared with the modification type and the type of sites (i.e., candidate or aging sites), which were not significant (Data S5). Aging sites also seemed to have a non-random distribution in the mitogenome because they were found only on certain genes, and even certain gene portions, despite having a similar coverage across ages for most other genes (Figure 3A). They mainly occurred on codon-organized genes (i.e., 73% of aging sites; Figure 3C), such as NADH dehydrogenase subunits 1 and 5 (ND1 and ND5) genes and the cytochrome c oxidase subunit I (COX1) gene (ND2 to a lesser extent). Nevertheless, the average density of aging sites per gene (7.5% vs. 14.7% of gene length) was two times higher on the nine much smaller tRNA genes and even on two supposedly non-coding sections (Figures S10 and S11 in Data S5). Aging sites were generally methylated for a single experimental stage (no sites were methylated throughout the entire experiment), and most methylation levels of 100% occurred at 10 DPH, without clear methylation patterns per gene or modification type (Figure 4). However, the methylation level was significantly linked to both qualitative (gene and modification type) and quantitative factors (coverage and reliability score; Figure S12 in Data S6), either per experimental stage (Mantel Test, p < 0.001 for all variables) or averaged across the experiment (PERMANOVA, p < 0.05 for all single variables except the type only significant in interaction with genes: p = 0.02), even though standardized effect sizes () were categorized as small for all variables except the gene (Data S6).
[IMAGE OMITTED. SEE PDF]
[IMAGE OMITTED. SEE PDF]
Efficient Prediction of Larvae Ages
To take into account all these factors, along with methylation levels, for seabass age prediction, we developed a method to fit an objective epigenetic clock based on grouped penalized regularizations (i.e., lasso, ridge, or elastic-net regressions) systematically tested for minimum error rate across penalty factors (α) and bootstrap iterations. Indeed, we detected complex relationships between α values and the resulting model performance, which greatly varied between bootstrap iterations (Supplement 6). Using 10 bootstrap iterations, we were able to reach stable optimal performances of epigenetic clocks fitted across all possible combinations of additional factors and across the three datasets (Figure 5). Overall, most optimal α values were close to 0, yielding a large number of non-neutralized “selected” sites, even though the link between α and the number of selected sites was not straightforward because the sites' coefficients were set to be very close to, but not at, 0 (Figure 5). The minimum MAE during cross-validation was very similar (around 4 days) among groupings for clocks fitted on the modC dataset, while it varied greatly (between 2.6 and 5 days) for the two other datasets (Figure 5).
[IMAGE OMITTED. SEE PDF]
MAE was much lower after training when no groups were used (around 0.1 days), while this metric was between 0.5 and 1 day for most other clocks. The best epigenetic clocks selected for the modC and modA datasets were therefore fitted without groupings (Figure 5). Conversely, epigenetic clocks fitted on the full dataset that accounted for the gene, the mean reliability score, and to a lesser extent the mean coverage had the best results in terms of cross-validated MAE (Figure 5). Indeed, we obtained a cross-validated MAE reaching 2.61 days in this case, compared with 3.94 and 3.43 days for the smallest modC and modA datasets, respectively. Despite such a result, the MAE after training was slightly lower for the full dataset (0.43 days), even if the overall correspondence between known and predicted values was high (R2 = 0.99, Pearson correlation coefficient = 1; Figure 6A). To characterize the importance of each gene in the age prediction, we summed the absolute coefficients of their respective sites instead of just accounting for the number of selected sites per gene, which was more linked to the penalty factor chosen (Figure 6B). Generally, the most important genes for each clock were those with the largest number of aging sites with high methylation levels per dataset (especially for the modC best clock with α = 0), even though the tRNA-Trp gene had a very high importance—despite having only nine sites—compared with codon-organized genes (especially the ND2 gene; Figure 6B). Indeed, seven aging sites on the tRNA-Trp gene were 6 mA, and most had a methylation level of 0% at 10–14 DPH, followed by an average increase of 8.5% at 17 DPH, before reaching 100% at 19 DPH and 24 DPH for three and two sites, respectively (Figure 4B). Overall, mitogenome methylation seems sufficient for the accurate age prediction of the seabass larvae studied here.
[IMAGE OMITTED. SEE PDF]
Discussion
Our study shows that eDNA can be used to fit epigenetic clocks, but also that their accuracy might be equivalent to, or even better than, most other clocks previously established. We obtained the highest correlation (Figure 6) of all methylation-based (μ = 0.92, 95% CI [0.89; 0.94]) and telomere-based (μ = 0.31, 95% CI [0.25; 0.37]) clocks listed in the review by Le Clercq et al. (2023). The cross-validated median absolute error (MAE) obtained in our analysis with the best clock (2.61 days) considering all aging sites exceeded previously obtained cross-validated MAE values for non-model animals (μ = 1.28 years; Tangili et al. 2023). However, the MAE scaled to the age range was twice as high (i.e., 0.19 vs. μ = 0.08) due to the very short age range considered here. The improvement in cross-validated MAE between the clock previously established for
The higher accuracy of this new epigenetic clock for larval
Compared with previous approaches employed to detect methylated nucleotides, which involved microarrays or bisulfite sequencing in epigenetic clock studies (Tangili et al. 2023), nanopore sequencing has various advantages that might explain the low error rate of our clock. First, with this method it is possible to search for aging sites in a non-targeted way, as it does not involve amplifying/enriching a specific region of DNA, and the method performs better at detecting CpG sites than the equivalent bisulfite-based technique (i.e., WGBS; Liu et al. 2023). Second, it does not require any DNA modification, which can generate errors, and it enables the retrieval of long reads that are easier to map to a reference genome (Laine et al. 2023; Schatz 2017). Third, it allows detection of cytosine methylation in all contexts, while simultaneously detecting 5hmC and 6 mA modifications, unlike previous methods. In our case, this increased the number of methylated nucleotides (all: 254 K; CpG only = 123 K) and the number of aging sites detected by more than half (Figures 3B and S8D in Data S5). These two less-studied modification types might even be informative since they are both thought to be involved in the aging process (Shi et al. 2017; Xie et al. 2023). They made up a larger proportion of aging sites versus 5mC sites compared with the equivalent proportion for methylated sites (35% vs. 52%; Figure 3B), and the adenosine-based clock had a lower MAE than the cytosine-based clock (3.43 vs. 3.94 days; Figure 6). The main criticisms of nanopore sequencing focus on its sequencing accuracy, but this has improved over the past few years, and we achieved < 0.7% in our study using the latest materials available (R10.4.1 + V14).
The high accuracy of our clock is due to our newly developed methodology, which fits optimal epigenetic clocks based on objective selection criteria, taking advantage of various qualitative and quantitative variables characterizing aging sites obtained with nanopore sequencing (Figure 1). Most epigenetic clocks utilize penalized regression to prevent overfitting, as the number of aging sites is often greater than the number of samples (Anastasiadi and Piferrer 2023), which requires manually tuning (e.g., α = 0.5; in fishes: Bertucci et al. 2021; Mayne et al. 2020) the parameter determining the amount of shrinkage (α). The use of a systematic and bootstrapped approach to objectively determines parameters reduced the stochasticity of performances for equal model parameters (Figure S13 in Data S6), but our results question the repeatability of performances between previous studies that have not yet been tested (Piferrer and Anastasiadi 2023). In our study, grouping aging sites that shared similar features before fitting an epigenetic clock made it possible to reduce error rates through the objective choice of the best groupings in the global model (i.e., gene and uncertainty-associated variables; Figure 5). However, this approach is not necessarily the best for smaller subsets sharing similar modification types (Figure 6). Aside from nanopore sequencing, this approach could be generalized to account for any qualitative (e.g., tissue and CpG context) or quantitative (e.g., replicates and alignment/mapping scores) features, depending on the methylation assessment methodology, and could be applied to further explore if neighboring methylated sites influence each other in terms of characteristics (Laine et al. 2023). The main limitation of our method is that it only predicts age from evolving methylated patterns as an animal ages, without any insight into the underlying biological processes; informed deep neural networks could be a solution to this problem (Prosz et al. 2024). Another limitation specific to this exploratory study is the absence of an independent validation dataset, which forced us to predict new values from the same training dataset (Figure 6A). This likely led to a significant underestimation of the model's true performance (i.e., R2 and final MAE). As solely using cross-validation is common in machine learning studies and considered as safe (King, Orhobor, and Taylor 2021; Levman et al. 2023), we advocate, in line with Tangili et al. (2023), that only cross-validated MAE obtained after testing on ⅓ of the initial dataset the model trained on the other ⅔ should be considered as reliable.
The exclusive presence of aging sites on mitochondrial DNA (mtDNA) is in contrast to the many methylated sites detected on nuclear DNA (nuDNA), and mtDNA generally makes up 0.1%–1% of extracted tDNA (Ramón-Laca, Gallego, and Nichols 2023). This pattern might be explained by the observed coverage bias (Figure S9 in Data S5), in that it is more difficult to obtain sufficient coverage (nuDNA: 1×; mtDNA: 3×) to detect differentially methylated sites for all ages considered in nuDNA, possibly because mtDNA and methylated groups are more stable inside the mitochondria due to a “double-protection” as well as a short size (Deiner et al. 2017; Jo, Takao, and Minamoto 2022; Jensen et al. 2021). So far, aging sites detected on the whole genome (WGBS) have either been found exclusively within nuDNA (Qiu et al. 2022; Raddatz et al. 2021; Sun et al. 2021) or have been removed during processing (Meer et al. 2018). Mitochondrial methylation is controversial (Chatterjee, Das, and Chakrabarti 2022), but some evidence indicates a predominance of 6 mA methylation over 5mC methylation (the inverse of the pattern in nuDNA; Hao et al. 2020). Cytosine modifications have typically been used to fit epigenetic clocks, but they might be scarce or even absent from the mtDNA of some taxa (Shao, Han, and Zhou 2023; Sharma, Pasala, and Prakash 2019; Sturm et al. 2023). Decreased 5hmC levels but not 5mC levels (Dzitoyeva, Chen, and Manev 2012) have also been observed for nuclear 5mC sites (Suarez-Bregua et al. 2021), while higher 6 mA levels within mtDNA have been linked to aging (Sturm et al. 2023). Both 5mC and 6 mA are distributed throughout most of the mtDNA, but we observed apparent non-random aging site distribution patterns (Figure 3A) that do not necessarily match zones with the highest density and methylation levels for 5mC in the fish
Limitations and Perspectives
The main limitations of this study are its low number of successful time points (i.e., only 6 due to technical issues), the short age range considered (10–24 DPH) and the absence of an independent validation dataset. Considering this, we do not aim to provide a reference epigenetic clock, but we instead propose a set of new tools to explore a new promising research question efficiently. Additionally, an advantage of using eDNA is that this approach resulted in pooling DNA from hundreds of individuals in each vial, meaning that the 376 mtDNA seabass reads used to fit epigenetic clocks likely originated from different individuals. Such number of individuals exceeds the minimum optimal sample size of 134 individuals advised by Mayne, Berry, and Jarman (2021) for tDNA. To more thoroughly validate the potential of eDNA epigenetics as a new age assessment method, further work should focus on various species/stages using larger filtration volumes (e.g., 60 L). Researchers should also aim to predict the age of individuals from different populations/species/environments as a way to assess the universality of their clock. A major challenge of amplification-free eDNA sequencing will be to obtain sufficient eDNA for detecting aging sites (i.e., sufficiently high coverage; Laine et al. 2023) from marine waters, which typically contain much lower densities of organisms than aquaculture tanks, even though abundances might be very high for many commercial species (e.g., pelagic schoolers). To that end, it might be better to filter larger volumes of water and/or use enrichment methods instead of pooling replicates and shearing samples, notably since nanopore sequencing provides satisfying results even at 6.25% of their recommended input (Heavens et al. 2021). Targeted methylation-sensitive polymerase chain reaction (PCR) toward previously identified aging regions could be used for very small DNA inputs, but the method still does not offer a single-nucleotide resolution (Mayne et al. 2020; Qi et al. 2021; Sturm et al. 2023). Enrichment appears more attractive in other cases, as it would make it possible to enrich circular DNA (exonuclease; Ramón-Laca, Gallego, and Nichols 2023) or organelles (differential centrifugation; Jo et al. 2019) for targeting mtDNA, if their integrity is preserved. It would additionally enable depletion of bacterial DNA (Feehery et al. 2013), which is less methylated than animal DNA and was the main source of contamination in our case (Figure S3 in Data S2). Recently, ONT introduced a new method called adaptive sampling to enrich target DNA during sequencing, which already showed promising results enriching environmental mtDNA using sets a full reference mitogenomes from all mammalians (Frank et al. 2024) or all expected species of parasites and hosts (Kipp et al. 2023).
Another major challenge will be to test if reference epigenetic clocks used later for eDNA epigenetics can be obtained from tDNA of wild individuals, as many species are complicated to maintain in captivity during all life stages, and as captivity might cause methylation changes even if there is no evidence that they affect epigenetic clocks (Tangili et al. 2023). This might not be straightforward, as Zhao, van Bodegom, and Trimbos (2022) observed different methylation patterns between eDNA and tDNA for the same ages. However, at least the tissue, sex, and potential error in training age (if estimated) do not seem to have a large effect on epigenetic clocks (Le Clercq et al. 2023; Mayne, Berry, and Jarman 2023; Tangili et al. 2023). Finally, epigenetic markers other than DNA methylation that affect the expression of genes, such as non-coding RNA or various chromatin features now detectable with nanopore sequencing (Yue et al. 2022), could also be used for age assessments. In particular, as for eDNA, environmental RNA (eRNA) could be useful for identifying species, age, sex ratio, condition, and stress (Stevens and Parsley 2023). eRNA's ability to discriminate life stages of various amphibian species—both in captivity and in the wild—was recently demonstrated (Parsley and Goldberg 2023). This suggests that, when used in conjunction, eRNA and eDNA epigenetics could have an even greater potential for population management and species conservation. After overcoming technical challenges, eDNA/eRNA surveys might become very efficient methods for rapid, synoptic, and non-invasive screening of a whole ecosystem's health and biodiversity on the basis of multiple variables, since eDNA sampling is generally much easier than the sampling of individual organisms.
Author Contributions
Eliot Ruiz: conceptualization (lead), formal analysis (lead), methodology (lead), resources (lead), software (lead), visualization (lead), writing – original draft (lead), writing – review and editing (lead). Fabien Leprieur: supervision (equal), writing – review and editing (equal). Gérard Sposito: resources (equal), supervision (equal). Martina Lüthi: methodology (equal), resources (equal), writing – review and editing (equal). Michel Schmidlin: methodology (equal), resources (equal), writing – review and editing (equal). Jacques Panfili: supervision (equal), writing – review and editing (equal). Loïc Pellissier: conceptualization (equal), funding acquisition (equal), writing – review and editing (equal). Camille Albouy: conceptualization (equal), funding acquisition (equal), project administration (equal), supervision (equal), writing – review and editing (equal).
Acknowledgements
We thank the OREME (Mediterranean Coastal Environment Station, Sète, France) observatory for hosting our experiments and providing seabass larvae. We also thank members of the GDC (Genetic Diversity Center, ETH Zürich, Switzerland), in particular Silvia Kobel and Aria Minder, for valuable help conducting library preparation and sequencing.
Conflicts of Interest
The authors declare no conflicts of interest.
Data Availability Statement
All data and scripts generated during this study are publicly accessible. Codes and raw figures were deposited on Github: . Data were deposited on FigShare: .
Adams, C. I. M., M. Knapp, N. J. Gemmell, et al. 2019. “Beyond Biodiversity: Can Environmental DNA (eDNA) cut It as a Population Genetics Tool?” Genes 10: 192. https://doi.org/10.3390/genes10030192.
Amstrup, S. C., T. L. McDonald, and B. F. J. Manly. 2005. Handbook of Capture‐Recapture Analysis. Princeton: Princeton University Press.
Anastasiadi, D., N. Díaz, and F. Piferrer. 2017. “Small Ocean Temperature Increases Elicit Stage‐Dependent Changes in DNA Methylation and Gene Expression in a Fish, the European Sea Bass.” Scientific Reports 7: 12401. https://doi.org/10.1038/s41598‐017‐10861‐6.
Anastasiadi, D., and F. Piferrer. 2020. “A Clockwork Fish: Age Prediction Using DNA Methylation‐Based Biomarkers in the European Seabass.” Molecular Ecology Resources 20: 387–397. https://doi.org/10.1111/1755‐0998.13111.
Anastasiadi, D., and F. Piferrer. 2023. “Bioinformatic Analysis for Age Prediction Using Epigenetic Clocks: Application to Fisheries Management and Conservation Biology.” Frontiers in Marine Science 10: 1096909. https://doi.org/10.3389/fmars.2023.1096909.
Anastasiadi, D., M. Vandeputte, N. Sánchez‐Baizán, F. Allal, and F. Piferrer. 2018. “Dynamic Epimarks in Sex‐Related Genes Predict Gonad Phenotype in the European Sea Bass, a Fish With Mixed Genetic and Environmental Sex Determination.” Epigenetics 13: 988–1011. https://doi.org/10.1080/15592294.2018.1529504.
Beger, M., A. Metaxas, A. C. Balbar, et al. 2022. “Demystifying Ecological Connectivity for Actionable Spatial Conservation Planning.” Trends in Ecology and Evolution 37: 1079–1091. https://doi.org/10.1016/j.tree.2022.09.002.
Bellizzi, D. 2017. “Mitochondrial Genome and Epigenome Two Sides of the Same Coin.” Frontiers in Bioscience 22: 888–908. https://doi.org/10.2741/4523.
Beng, K. C., and R. Corlett. 2020. “Applications of Environmental DNA (eDNA) in Ecology and Conservation: Opportunities, Challenges and Prospects.” Biodiversity and Conservation 29: 2089–2121. https://doi.org/10.1007/s10531‐020‐01980‐0.
Benson, I. M., T. E. Helser, G. Marchetti, and B. K. Barnett. 2023. “The Future of Fish Age Estimation: Deep Machine Learning Coupled With Fourier Transform Near‐Infrared Spectroscopy of Otoliths.” Canadian Journal of Fisheries and Aquatic Sciences 80: 1482–1494. https://doi.org/10.1139/cjfas‐2023‐0045.
Bertucci, E. M., M. W. Mason, O. E. Rhodes, and B. B. Parrott. 2021. “The Aging DNA Methylome Reveals Environment‐By‐Aging Interactions in a Model Teleost.” https://doi.org/10.1101/2021.03.01.433371.
Blondeau‐Bidet, E., G. Banousse, T. L'Honoré, E. Farcy, C. Cosseau, and C. Lorin‐Nebel. 2023. “The Role of Salinity on Genome‐Wide DNA Methylation Dynamics in European Sea Bass Gills.” Molecular Ecology 32: 5089–5109. https://doi.org/10.1111/mec.17089.
Booth, L. N., and A. Brunet. 2016. “The Aging Epigenome.” Molecular Cell 62: 728–744. https://doi.org/10.1016/j.molcel.2016.05.013.
Boyd, C. E., A. A. McNevin, and R. P. Davis. 2022. “The Contribution of Fisheries and Aquaculture to the Global Protein Supply.” Food security 14: 805–827. https://doi.org/10.1007/s12571‐021‐01246‐9.
Burgerhout, E., M. Mommens, H. Johnsen, A. Aunsmo, N. Santi, and Ø. Andersen. 2017. “Genetic Background and Embryonic Temperature Affect DNA Methylation and Expression of Myogenin and Muscle Development in Atlantic Salmon (Salmo salar).” PLoS One 12, no. 6: e0179918. https://doi.org/10.1371/journal.pone.0179918.
Campana, S. E., and S. R. Thorrold. 2001. “Otoliths, increments, and elements: Keys to a comprehensive understanding of fish populations?.” Canadian Journal of Fisheries and Aquatic Sciences 58, no. 1: 30–38. https://doi.org/10.1139/f00‐177.
Campos, C., L. M. P. Valente, L. E. C. Conceição, S. Engrola, and J. M. O. Fernandes. 2013. “Temperature Affects Methylation of the Myogenin Putative Promoter, Its Expression and Muscle Cellularity in Senegalese Sole Larvae.” Epigenetics 8: 389–397. https://doi.org/10.4161/epi.24178.
Carell, T., M. Q. Kurz, M. Müller, M. Rossa, and F. Spada. 2018. “Non‐canonical Bases in the Genome: The Regulatory Information Layer in DNA.” Angewandte Chemie 57: 4296–4312. https://doi.org/10.1002/anie.201708228.
Chatterjee, D., P. Das, and O. Chakrabarti. 2022. “Mitochondrial Epigenetics Regulating Inflammation in Cancer and Aging.” Frontiers in Cell and Development Biology 10: 929708. https://doi.org/10.3389/fcell.2022.929708.
Cooke, H. J., and B. A. Smith. 1986. “Variability at the Telomeres of the Human X/Y Pseudoautosomal Region.” Cold Spring Harbor Symposia on Quantitative Biology 51: 213–219. https://doi.org/10.1101/SQB.1986.051.01.026.
Covelo‐Soto, L., M. Saura, and P. Morán. 2015. “Does DNA Methylation Regulate Metamorphosis? The Case of the Sea Lamprey (Petromyzon marinus) as an Example.” Comparative Biochemistry and Physiology Part B: Biochemistry and Molecular Biology 185: 42–46. https://doi.org/10.1016/j.cbpb.2015.03.007.
Danecek, P., J. K. Bonfield, J. Liddle, et al. 2021. “Twelve Years of SAMtools and BCFtools.” GigaScience 10, no. 2: giab008. https://doi.org/10.1093/gigascience/giab008.
de Pontual, H., K. Heerah, J. Goossens, et al. 2023. “Seasonal Migration, Site Fidelity, and Population Structure of European Seabass (Dicentrarchus labrax).” ICES Journal of Marine Science 80: 1606–1618. https://doi.org/10.1093/icesjms/fsad087.
Deiner, K., M. A. Renshaw, Y. Li, B. P. Olds, D. M. Lodge, and M. E. Pfrender. 2017. “Long‐Range PCR Allows Sequencing of Mitochondrial Genomes From Environmental DNA.” Methods in Ecology and Evolution 8, no. 12: 1888–1898. https://doi.org/10.1111/2041‐210X.12836.
Díaz‐Ferguson, E. E., and G. R. Moyer. 2014. “History, Applications, Methodological Issues and Perspectives for the Use Environmental DNA (eDNA) in Marine and Freshwater Environments.” Revista de Biología Tropical 62: 1273–1284. https://doi.org/10.15517/rbt.v62i4.13231.
Ding, B., X. Zhang, Z. Wan, et al. 2023. “Characterization of Mitochondrial DNA Methylation of Alzheimer's Disease in Plasma Cell‐Free DNA.” Diagnostics 13: 2351. https://doi.org/10.3390/diagnostics13142351.
Dzitoyeva, S., H. Chen, and H. Manev. 2012. “Effect of Aging on 5‐Hydroxymethylcytosine in Brain Mitochondria.” Neurobiology of Aging 33: 2881–2891. https://doi.org/10.1016/j.neurobiolaging.2012.02.006.
FAO. 2022. The State of World Fisheries and Aquaculture 2022: Towards Blue Transformation, the State of World Fisheries and Aquaculture (SOFIA). Rome, Italy: FAO. https://doi.org/10.4060/cc0461en.
Feehery, G. R., E. Yigit, S. O. Oyola, et al. 2013. “A Method for Selectively Enriching Microbial DNA From Contaminating Vertebrate Host DNA.” PLoS One 8: e76096. https://doi.org/10.1371/journal.pone.0076096.
Flusberg, B. A., D. R. Webster, J. H. Lee, et al. 2010. “Direct Detection of DNA Methylation During Single‐Molecule, Real‐Time Sequencing.” Nature Methods 7: 461–465. https://doi.org/10.1038/nmeth.1459.
Fontoura, L., S. D'Agata, M. Gamoyo, et al. 2022. “Protecting Connectivity Promotes Successful Biodiversity and Fisheries Conservation.” Science 375: 336–340. https://doi.org/10.1126/science.abg4351.
Franco‐Sierra, N. D., and J. F. Díaz‐Nieto. 2020. “Rapid Mitochondrial Genome Sequencing Based on Oxford Nanopore Sequencing and a Proxy for Vertebrate Species Identification.” Ecology and Evolution 10: 3544–3560. https://doi.org/10.1002/ece3.6151.
Frank, L. E., L. L. Lindsey, E. J. Kipp, et al. 2024. “Rapid Molecular Species Identification of Mammalian Scat Samples Using Nanopore Adaptive Sampling.” Journal of Mammalogy 105, no. 5: 965–975. https://doi.org/10.1093/jmammal/gyae044.
Friedman, J. H., T. Hastie, and R. Tibshirani. 2010. “Regularization Paths for Generalized Linear Models via Coordinate Descent.” Journal of Statistical Software 33: 1–22. https://doi.org/10.18637/jss.v033.i01.
Gleeson, J., A. Leger, Y. D. J. Prawer, et al. 2022. “Accurate Expression Quantification From Nanopore Direct RNA Sequencing With NanoCount.” Nucleic Acids Research 50: e19. https://doi.org/10.1093/nar/gkab1129.
Hao, Z., T. Wu, X. Cui, et al. 2020. “N6‐Deoxyadenosine Methylation in Mammalian Mitochondrial DNA.” Molecular Cell 78: 382–395.e8. https://doi.org/10.1016/j.molcel.2020.02.018.
Haussmann, M., and C. Vleck. 2002. “Telomere Length Provides a New Technique for Aging Animals.” Oecologia 130: 325–328. https://doi.org/10.1007/s00442‐001‐0827‐y.
Heavens, D., D. Chooneea, M. Giolai, et al. 2021. “How Low Can You Go? Driving Down the DNA Input Requirements for Nanopore Sequencing.” https://doi.org/10.1101/2021.10.15.464554.
Helwig, N. E. 2023. “grpnet: Group Elastic Net Regularized GLM.”
Horvath, S. 2013. “DNA Methylation Age of Human Tissues and Cell Types.” Genome Biology 14: 3156. https://doi.org/10.1186/gb‐2013‐14‐10‐r115.
Iannelli, M., and F. Milner. 2017. “The Basic Approach to Age‐Structured Population Dynamics: Models, Methods and Numerics.” In Lecture Notes on Mathematical Modelling in the Life Sciences. Netherlands, Dordrecht: Springer. https://doi.org/10.1007/978‐94‐024‐1146‐1.
Jensen, M. R., E. E. Sigsgaard, S. Liu, et al. 2021. “Genome‐Scale Target Capture of Mitochondrial and Nuclear Environmental DNA From Water Samples.” Molecular Ecology Resources 21, no. 3: 690–702. https://doi.org/10.1111/1755‐0998.13293.
Jo, J., H.‐G. Lee, K. Y. Kim, et al. 2019. “SoEM: A Novel PCR‐Free Biodiversity Assessment Method Based on Small‐Organelles Enriched Metagenomics.” Algae 34: 57–70. https://doi.org/10.4490/algae.2019.34.2.26.
Jo, T., K. Takao, and T. Minamoto. 2022. “Linking the State of Environmental DNA to Its Application for Biomonitoring and Stock Assessment: Targeting Mitochondrial/Nuclear Genes, and Different DNA Fragment Lengths and Particle Sizes.” Environmental DNA 4, no. 2: 271–283. https://doi.org/10.1002/edn3.253.
King, R. D., O. I. Orhobor, and C. C. Taylor. 2021. “Cross‐Validation Is Safe to Use.” Nature Machine Intelligence 3, no. 4: 276. https://doi.org/10.1038/s42256‐021‐00332‐z.
Kipp, E. J., L. L. Lindsey, M. S. Milstein, et al. 2023. “Nanopore Adaptive Sampling for Targeted Mitochondrial Genome Sequencing and Bloodmeal Identification in Hematophagous Insects.” Parasites and Vectors 16, no. 1: 68. https://doi.org/10.1186/s13071‐023‐05679‐3.
Klughammer, J., D. Romanovskaia, A. Nemc, et al. 2023. “Comparative Analysis of Genome‐Scale, Base‐Resolution DNA Methylation Profiles Across 580 Animal Species.” Nature Communications 14: 232. https://doi.org/10.1038/s41467‐022‐34828‐y.
Laine, V. N., B. Sepers, M. Lindner, F. Gawehns, S. Ruuskanen, and K. van Oers. 2023. “An ecologist's Guide for Studying DNA Methylation Variation in Wild Vertebrates.” Molecular Ecology Resources 23: 1488–1508. https://doi.org/10.1111/1755‐0998.13624.
Le Clercq, L.‐S., A. Kotzé, J. P. Grobler, and D. L. Dalton. 2023. “Biological Clocks as Age Estimation Markers in Animals: A Systematic Review and meta‐Analysis.” Biological Reviews 98: 1972–2011. https://doi.org/10.1111/brv.12992.
Levman, J., B. Ewenson, J. Apaloo, D. Berger, and P. N. Tyrrell. 2023. “Error Consistency for Machine Learning Evaluation and Validation With Application to Biomedical Diagnostics.” Diagnostics 13, no. 7: Article 7. https://doi.org/10.3390/diagnostics13071315.
Liu, S., A. N. B. Tengstedt, M. W. Jacobsen, et al. 2022. “Genome‐Wide Methylation in the Panmictic European Eel (Anguilla anguilla).” Molecular Ecology 31: 4286–4306. https://doi.org/10.1111/mec.16586.
Liu, X., Y. Ni, D. Wang, et al. 2023. “Unraveling the Whole Genome DNA Methylation Profile of Zebrafish Kidney Marrow by Oxford Nanopore Sequencing.” Scientific Data 10: 532. https://doi.org/10.1038/s41597‐023‐02431‐5.
Liu, Y., W. Rosikiewicz, Z. Pan, et al. 2021. “DNA Methylation‐Calling Tools for Oxford Nanopore Sequencing: A Survey and Human Epigenome‐Wide Evaluation.” Genome Biology 22: 295. https://doi.org/10.1186/s13059‐021‐02510‐z.
Lu, A. T., Z. Fei, A. Haghani, et al. 2023. “Universal DNA Methylation Age Across Mammalian Tissues.” Nature Aging 3: 1144–1166. https://doi.org/10.1038/s43587‐023‐00462‐6.
Mayne, B., O. Berry, and S. Jarman. 2021. “Optimal Sample Size for Calibrating DNA Methylation Age Estimators.” Molecular Ecology Resources 21: 2316–2323. https://doi.org/10.1111/1755‐0998.13437.
Mayne, B., O. Berry, and S. Jarman. 2023. “Calibrating Epigenetic Clocks With Training Data Error.” Evolutionary Applications 16: 1496–1502. https://doi.org/10.1111/eva.13582.
Mayne, B., T. Espinoza, D. A. Crook, et al. 2023. “Accurate, Non‐destructive, and High‐Throughput Age Estimation for Golden Perch (Macquaria ambigua Spp.) Using DNA Methylation.” Scientific Reports 13: 9547. https://doi.org/10.1038/s41598‐023‐36773‐2.
Mayne, B., T. Espinoza, D. Roberts, et al. 2021. “Nonlethal Age Estimation of Three Threatened Fish Species Using DNA Methylation: Australian Lungfish, Murray Cod and Mary River Cod.” Molecular Ecology Resources 21: 2324–2332. https://doi.org/10.1111/1755‐0998.13440.
Mayne, B., D. Korbie, L. Kenchington, B. Ezzy, O. Berry, and S. Jarman. 2020. “A DNA Methylation Age Predictor for Zebrafish.” Aging 12: 24817–24835. https://doi.org/10.18632/aging.202400.
Meer, M. V., D. I. Podolskiy, A. Tyshkovskiy, and V. N. Gladyshev. 2018. “A Whole Lifespan Mouse Multi‐Tissue DNA Methylation Clock.” eLife 7: e40675. https://doi.org/10.7554/eLife.40675.
Nedoluzhko, A., R. Mjelle, M. Renström, K. H. Skjærven, F. Piferrer, and J. M. O. Fernandes. 2021. “The First Mitochondrial 5‐Methylcytosine Map in a Non‐model Teleost (Oreochromis niloticus) Reveals Extensive Strand‐Specific and Non‐CpG Methylation.” Genomics 113: 3050–3057. https://doi.org/10.1016/j.ygeno.2021.07.007.
O'Brown, Z. K., and E. L. Greer. 2016. “N6‐Methyladenine: A Conserved and Dynamic DNA Mark.” Advances in Experimental Medicine and Biology 945: 213–246. https://doi.org/10.1007/978‐3‐319‐43624‐1_10.
Parsley, M. B., and C. S. Goldberg. 2023. “Environmental RNA Can Distinguish Life Stages in Amphibian Populations.” Molecular Ecology Resources 24: e13857. https://doi.org/10.1111/1755‐0998.13857.
Piddocke, T. P., G. L. Butler, P. A. Butcher, S. W. Purcell, D. J. Bucher, and L. Christidis. 2015. “Age Validation in the Lutjanidae: A Review.” Fisheries Research 167: 48–63. https://doi.org/10.1016/j.fishres.2015.01.016.
Piferrer, F., and D. Anastasiadi. 2023. “Age Estimation in Fishes Using Epigenetic Clocks: Applications to Fisheries Management and Conservation Biology.” Frontiers in Marine Science 10: 1062151. https://doi.org/10.3389/fmars.2023.1062151.
Polanco, A., E. Richards, B. Flück, et al. 2021. “Comparing the Performance of 12S Mitochondrial Primers for Fish Environmental DNA Across Ecosystems.” Environmental DNA 3: 1113–1127. https://doi.org/10.1002/edn3.232.
Pont, D., M. Rocle, A. Valentini, et al. 2018. “Environmental DNA Reveals Quantitative Patterns of Fish Biodiversity in Large Rivers Despite Its Downstream Transportation.” Scientific Reports 8: 10361. https://doi.org/10.1038/s41598‐018‐28424‐8.
Prosz, A., O. Pipek, J. Börcsök, et al. 2024. “Biologically Informed Deep Learning for Explainable Epigenetic Clocks.” Scientific Reports 14: 1306. https://doi.org/10.1038/s41598‐023‐50495‐5.
Punt, A. E., C. Allen Akselrud, and L. Cronin‐Fine. 2017. “The Effects of Applying Mis‐Specified Age‐ and Size‐Structured Models.” Fisheries Research 188: 58–73. https://doi.org/10.1016/j.fishres.2016.11.017.
Qi, H., K. Kinoshita, T. Mori, K. Matsumoto, Y. Matsui, and M. Inoue‐Murayama. 2021. “Age Estimation Using Methylation‐Sensitive High‐Resolution Melting (MS‐HRM) in Both Healthy Felines and Those With Chronic Kidney Disease.” Scientific Reports 11: 19963. https://doi.org/10.1038/s41598‐021‐99424‐4.
Qiu, H., H. Li, R. Fan, et al. 2022. “Genome‐Wide DNA Methylation Profile Indicates Potential Epigenetic Regulation of Aging in the Rhesus Macaque Thymus.” International Journal of Molecular Sciences 23: 14984. https://doi.org/10.3390/ijms232314984.
Raddatz, G., R. J. Arsenault, B. Aylward, R. Whelan, F. Böhl, and F. Lyko. 2021. “A Chicken DNA Methylation Clock for the Prediction of Broiler Health.” Communications Biology 4: 76. https://doi.org/10.1038/s42003‐020‐01608‐7.
Ramón‐Laca, A., R. Gallego, and K. M. Nichols. 2023. “Affordable de Novo Generation of Fish Mitogenomes Using Amplification‐Free Enrichment of Mitochondrial DNA and Deep Sequencing of Long Fragments.” Molecular Ecology Resources 23: 818–832. https://doi.org/10.1111/1755‐0998.13758.
Richards, W. J., and K. C. Lindeman. 1987. “Recruitment Dynamics of Reef Fishes: Planktonic Processes, Settlement and Demersal Ecologies, and Fishery Analysis.” Bulletin of Marine Science 41: 392–410.
Rishan, S. T., R. J. Kline, and M. S. Rahman. 2023. “Applications of Environmental DNA (eDNA) to Detect Subterranean and Aquatic Invasive Species: A Critical Review on the Challenges and Limitations of eDNA Metabarcoding.” Environmental Advances 12: 100370. https://doi.org/10.1016/j.envadv.2023.100370.
Robeck, T. R., Z. Fei, A. T. Lu, et al. 2021. “Multi‐Species and Multi‐Tissue Methylation Clocks for Age Estimation in Toothed Whales and Dolphins.” Communications Biology 4: 642. https://doi.org/10.1038/s42003‐021‐02179‐x.
Rognes, T., T. Flouri, B. Nichols, C. Quince, and F. Mahé. 2016. “VSEARCH: A Versatile Open Source Tool for Metagenomics.” PeerJ 4: e2584. https://doi.org/10.7717/peerj.2584.
Schatz, M. C. 2017. “Nanopore Sequencing Meets Epigenetics.” Nature Methods 14: 347–348. https://doi.org/10.1038/nmeth.4240.
Shao, Z., Y. Han, and D. Zhou. 2023. “Optimized Bisulfite Sequencing Analysis Reveals the Lack of 5‐Methylcytosine in Mammalian Mitochondrial DNA.” BMC Genomics 24: 439. https://doi.org/10.1186/s12864‐023‐09541‐9.
Sharma, N., M. S. Pasala, and A. Prakash. 2019. “Mitochondrial DNA: Epigenetics and Environment.” Environmental and Molecular Mutagenesis 60: 668–682. https://doi.org/10.1002/em.22319.
Shi, D.‐Q., I. Ali, J. Tang, and W.‐C. Yang. 2017. “New Insights Into 5hmC DNA Modification: Generation, Distribution and Function.” Frontiers in Genetics 8: 100. https://doi.org/10.3389/fgene.2017.00100.
Stevens, J. D., and M. B. Parsley. 2023. “Environmental RNA Applications and Their Associated Gene Targets for Management and Conservation.” Environmental DNA 5: 227–239. https://doi.org/10.1002/edn3.386.
Sturm, Á., H. Sharma, F. Bodnár, et al. 2023. “N6‐Methyladenine Progressively Accumulates in Mitochondrial DNA During Aging.” International Journal of Molecular Sciences 24: 14858. https://doi.org/10.3390/ijms241914858.
Suarez‐Bregua, P., A. Pérez‐Figueroa, J. Hernández‐Urcera, P. Morán, and J. Rotllant. 2020. “Temperature‐Independent Genome‐Wide DNA Methylation Profile in Turbot Post‐Embryonic Development.” Journal of Thermal Biology 88: 102483. https://doi.org/10.1016/j.jtherbio.2019.102483.
Suarez‐Bregua, P., S. Rosendo, P. Comesaña, et al. 2021. “Dynamic Changes in DNA Methylation During Seahorse (Hippocampus reidi) Postnatal Development and Settlement.” Frontiers in Zoology 18: 52. https://doi.org/10.1186/s12983‐021‐00436‐7.
Sun, D., T. S. Layman, H. Jeong, et al. 2021. “Genome‐Wide Variation in DNA Methylation Linked to Developmental Stage and Chromosomal Suppression of Recombination in White‐Throated Sparrows.” Molecular Ecology 30: 3453–3467. https://doi.org/10.1111/mec.15793.
Swearer, S. E., E. A. Treml, and J. S. Shima. 2019. “A Review of Biophysical Models of Marine Larval Dispersal.” In Oceanography and Marine Biology, 325–356. Boca Raton: CRC Press.
Tangili, M., A. J. Slettenhaar, J. Sudyka, et al. 2023. “DNA Methylation Markers of Age(Ing) in Non‐model Animals.” Molecular Ecology 32: 4725–4741. https://doi.org/10.1111/mec.17065.
Tine, M., H. Kuhl, P.‐A. Gagnaire, et al. 2014. “European Sea Bass Genome and Its Variation Provide Insights Into Adaptation to Euryhalinity and Speciation.” Nature Communications 5: 5770. https://doi.org/10.1038/ncomms6770.
Trautner, J. H., S. Reiser, T. Blancke, K. Unger, and K. Wysujack. 2017. “Metamorphosis and Transition Between Developmental Stages in European Eel (Anguilla anguilla, L.) Involve Epigenetic Changes in DNA Methylation Patterns.” Comparative Biochemistry and Physiology Part D: Genomics and Proteomics 22: 139–145. https://doi.org/10.1016/j.cbd.2017.04.002.
Valdivieso, A., D. Anastasiadi, L. Ribas, and F. Piferrer. 2023. “Development of Epigenetic Biomarkers for the Identification of Sex and Thermal Stress in Fish Using DNA Methylation Analysis and Machine Learning Procedures.” Molecular Ecology Resources 23: 453–470. https://doi.org/10.1111/1755‐0998.13725.
Valdivieso, A., N. Sánchez‐Baizán, N. Mitrizakis, N. Papandroulakis, and F. Piferrer. 2023. “Development of Epigenetic Biomarkers With Diagnostic and Prognostic Value to Assess the Lasting Effects of Early Temperature Changes in Farmed Fish.” Aquaculture 563: 738918. https://doi.org/10.1016/j.aquaculture.2022.738918.
van der Lingen, C., and J. Huggett. 2003. “The Role of Ichthyoplankton Surveys in Recruitment Research and Management of South African Anchovy and Sardine.” In Proceedings of the 26th Annual Larval Fish Conference, edited by H. Browman and A. Skiftesvik, 303–343. Bergen: Institute of Marine Research.
Weber, D. N., A. T. Fields, W. F. Patterson, B. K. Barnett, C. M. Hollenbeck, and D. S. Portnoy. 2022. “Novel Epigenetic Age Estimation in Wild‐Caught Gulf of Mexico Reef Fishes.” Canadian Journal of Fisheries and Aquatic Sciences 79: 1–5. https://doi.org/10.1139/cjfas‐2021‐0240.
Wescoe, Z. L., J. Schreiber, and M. Akeson. 2014. “Nanopores Discriminate Among Five C5‐Cytosine Variants in DNA.” Journal of the American Chemical Society 136: 16582–16587. https://doi.org/10.1021/ja508527b.
Wu, S., Y. Huang, S. Li, et al. 2018. “DNA Methylation Levels and Expression Patterns of Smyd1a and Smyd1b Genes During Metamorphosis of the Japanese Flounder (Paralichthys olivaceus).” Comparative Biochemistry and Physiology Part B: Biochemistry and Molecular Biology 223: 16–22. https://doi.org/10.1016/j.cbpb.2018.05.002.
Xie, L., X. Zhang, J. Xie, Y. Xu, X.‐J. Li, and L. Lin. 2023. “Emerging Roles for DNA 6mA and RNA m6A Methylation in Mammalian Genome.” Int J Mol Sci 24: 13897. https://doi.org/10.3390/ijms241813897.
Yao, M., S. Zhang, Q. Lu, et al. 2022. “Fishing for Fish Environmental DNA: Ecological Applications, Methodological Considerations, Surveying Designs, and Ways Forward.” Molecular Ecology 31: 5132–5164. https://doi.org/10.1111/mec.16659.
Yue, X., Z. Xie, M. Li, et al. 2022. “Simultaneous Profiling of Histone Modifications and DNA Methylation via Nanopore Sequencing.” Nature Communications 13: 7939. https://doi.org/10.1038/s41467‐022‐35650‐2.
Zhang, Y., J. Bi, Y. Ning, and J. Feng. 2024. “Methodology Advances in Vertebrate Age Estimation.” Animals 14: 343. https://doi.org/10.3390/ani14020343.
Zhao, B., P. M. van Bodegom, and K. B. Trimbos. 2022. “Environmental DNA Methylation of Lymnaea stagnalis Varies With Age and Is Hypermethylated Compared to Tissue DNA.” Molecular Ecology Resources 23: 81–91. https://doi.org/10.1111/1755‐0998.13691.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2025. This work is published under http://creativecommons.org/licenses/by/4.0/ (the "License"). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
ABSTRACT
While acquiring age information is crucial for efficient stock management and biodiversity conservation, traditional aging methods fail to offer a universal, non‐invasive, and precise way of estimating a wild animal's age. DNA methylation from tissue DNA (tDNA) was recently proposed as a method to overcome these issues and showed more accurate results than telomere‐length‐based age assessments. Here, we used environmental DNA (eDNA) for the first time as a template for age estimation, focusing on the larval phase (10–24 days post‐hatch) of cultured
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details


1 MARBEC, Univ Montpellier, IRD, IFREMER, CNRS, Montpellier, France
2 Mediterranean Coastal Environment Station, University of Montpellier, Sète, France
3 Department of Environmental Systems Science, Ecosystems and Landscape Evolution, Institute of Terrestrial Ecosystems, ETH Zürich, Zürich, Switzerland, Land Change Science Research Unit, Swiss Federal Research Institute WSL, Birmensdorf, Switzerland