We showed that a modified epiGBS protocol originally proposed by Van Gurp et al. [88] was applicable to further analyze patterns of cytosine methylation at a genome-wide scale in D. labrax. This is the first use of epiGBS in fish and the second in an animal species (Canadian lynx [102]). The addition of a second restriction enzyme illustrates the flexibility of the epiGBS and more generally of reduced-representation bisulfite sequencing (RRBS) protocols for data acquisition and impact [103]. This is not the first RRBS protocol dedicated to the addition of a second restriction enzyme [104], but this is hereby proposed in a context of improved multiplexing of samples offering more opportunities to explore important biological processes, such as the stress response. We showed that RBC’s DNA methylation may respond to stressors in the European sea bass, but remains relatively stable over time and is mostly influenced by genetic background. In spite of confounding factors, results suggest that some epigenetic marks could complement both the traditional evaluation of the stress response based on analyses of blood serum samples [85,105], or on gene expression variation, often evaluated using invasive brain samples [87,93,94,106-111].
Mining the epigenome
The information provided in this study is based on the analysis of 47,983 distinct methylated sites distributed over all sea bass LGs. The mapping efficiency was high (74.5%) when compared to early values retrieved in human (~65%) [112], or in fish studies screening for genome-wide methylation (e.g. 55-60% in [59]; 40% in [41]). Other studies reported similar mapping efficiencies, but reported percentages of mapping for unique best hits that were generally lower. For example, in Kryptolebias marmoratus, Berbel-Filho et al. [43] reported a mean mapping efficiency of 74.2% but 61.1% unique best hits while, in this study, this latter percentage reached 73.0%. This reflects a more robust mapping of the DMCs we detected and significantly enlarge the breadth of the sites that can confidently exploit to retrieve functional information. Taking advantage of the epiGBS protocol that allow to process more samples [88], the number of individuals considered in this study is rather high (n = 70 distinct individuals), when most epigenomic studies in fish dealt with less than 30 individuals (range: n = 3 in [60]; n = 106 in [47] for a population study). In sea bass, Anastasiadi and Piferrer [83] previously reported a study that used 27 samples and as many libraries to be sequenced while our data were obtained from a unique library preparation. Our modified epiGBS protocol provides a considerable amount of information, certainly at a reasonable cost, to decipher methylation landscapes of sea bass or other species.
The operational and statistical thresholds used in the successive steps of this study are conservative, resulting in the discovery of a rather low total number of methylated sites, but certainly limiting the report of false positives. For example a threshold of 30X and nominal cut-off value of 0.001 are quite conservative, when some studies might consider a threshold of 5X or 10X for a CpG to be analysed and associated cut-off values of 0.05 or 0.01 (e.g. [46,59,83]). Relaxing thresholds would enable to retrieve more DMCs, but elevated thresholds should normally ensure that access to relevant information is reached. Thus, only 57 DMCs have been found in RBCs of pre- and post-stress European sea bass, and only few were defined as DMRs. Considering DMRs rather than DMCs is common in genome-wide methylation studies, including sea bass [83]. Methylation changes grouped in clusters (i.e. DMRs) are more likely to influence transcriptional activity at nearby loci than single cytosines [113]. However, other experimental issues associated to the flexibility of RRBS protocols (e.g. single- vs paired-end sequencing, likelihood of annotation, selection and size of fragments) can made the advantages of considering DMRs over DMCs perhaps more disputable [103]. Furthermore,and DMRs are not always accessible [114]. In this study, DMCs were found mostly hypermethylated in post-stress individuals compared to the pre-stress individuals, and mostly located in gene bodies (i.e. the transcriptionally active portion of the genome) of fifty-one different genes. It has been shown that such location of differential methylation may regulate splicing and/or act as alternative promoters to reshape gene expression [115-117].
In addition to DMCs located in gene bodies, a dozen of DMCs were found in intergenic regions (21.0%). Intergenic cytosine methylation has been frequently described, including in stress studies [118], but its role remains poorly understood [119,120]. While numbers of genic vs intergenic DMCs may greatly vary, a ratio of ~80% of DMCs located in gene bodies and ~20% located in other genomic regions has been reported in other fish studies (e.g. [46]). While under-represented in our data, the role for differentially methylated transposable elements is also poorly understood [121, 122], while crucial in the stress response [123].
But mining the unknown
Studies looking at the epigenomic landscape of RBCs in fish are scarce. Gavery et al. [41] reported 85 DMRs between hatchery and wild fish in O. mykiss. This study suggested that blood DNA methylation patterns are indicative of the stressors experienced in the environment, but we are still currently mining the unknown to gain supporting information.
We nevertheless demonstrate that fish RBCs are relevant candidates to investigate the stress response at the epigenomic level in the European sea bass. Numerous DMC-related genes reported here have been previously shown to respond to stress or implicated in the stress response in animal models. Overall, our comparative surveys showed that the DMC-related genes we detected in sea bass using RBCs have more connections with results found in the brain than in other tissues. This relationship is however weak at the epigenomic level and about one third of our DMC-related genes have also been detected in liver of trout [45]. This relationship is far more pronounced when genome-wide gene expression difference is considered. Indeed, genome-wide neurotranscriptomics studies dealing with personality, stress coping styles [93,94], or social regulation [95] collectively reported 38 of our 51 DMC-related genes as expressed and potentially associated to the stress response in zebrafish (one more when including a study on B. splendens [96]). As differential methylation is classically interpreted as leading to differential gene expression (but see, e.g., [124, 125] for a more nuanced consideration), this observation is one interesting signal indicating that RBCs might monitor events occurring in the brain. However, it remains unclear why most of them are related to coping and personality, and to studies where fish were under stress for few minutes [93,94], hours or days [95]. How distinct stress contexts and challenges may apparently rely on similar stress circuitry has to be investigated further.
A brain-derived neutrophic factor (BDNF) network
The roles of the DMC-related genes have to be explored exposed more extensively to illustrate their possible involvement in the stress response of sea bass. This is not place to fully expand on all DMC-related genes (see Additional File 3 for details), but DMCs/DMRs with a potential impact in fear memory and anxiety response as expected in fish after an acute stress challenge are present in our study. This especially includes DMC-related genes from the BDNF network whose role in the stress response is well-known. Indeed, BDNF is a protein synthesized in the brain that offers resistance to neurodegenerative diseases and favors stress adaptation and resilience [126,127]. It also favors energy homeostasis [128]. BDNF has emerged as one of the most important molecule in memory [129, 130]. It consolidates both the within- and between-generation fear memory owing to epigenetic regulation [131,132]. Its activity is strongly linked to glucocorticoid stress to imprint neurogenesis [133] and it acts as both a regulator and a target of stress hormone signaling [127]. In fish, cortisol binds the glucocorticoid receptor (gr) and controls BDNF expression in the brain. With gr and few other genes, BDNF is one of the target genes in fish stress studies, including zebrafish [134], sea bream [135] and sea bass [8687,136-138]. BDNF is also involved in other aspects of the stress response in fish (e.g. regulation of the microbiota-gut-brain axis [139]). Variation in BDNF methylation status could not be investigated in this study as no SbfI restriction site is present within or close to this gene. Nevertheless, a methylation signature of stress associated to BDNF is likely present in our data and illustrates how blood methylation may provide a snapshot of events studied in the brain.
Indeed, several DMC-related genes are involved in the maturation of proBDNF to mature BDNF or the regulation of its activities (ABLIM2, ADCY1b, CRTC2, FURIN, NPAS3, PLG, and possibly SLC22A2). Most of these DMC-related genes were already found differentially expressed in the zebrafish brain [93-95], but an outline of their interaction was not presented. The adenyl(ate) cyclase (AC, ADCY1b gene) is a brain-specific signaling enzyme that synthesizes the cyclic AMP [140]. In the brain, AC is activated by PACAP (Protein Adenylate Cyclase Activating Protein) and stimulates memory and long term potentiation [141]. This inducible signaling pathway participates to the synthesis of the active form of BDNF (proBDNF to mature BDNF) [142]. ProBDNF is processed by furin and the plasminogen system [128], including processing steps that necessitate actions of actin-binding LIM kinases (ABLIM) [143]. Furin - encoded by FURIN - is a subtilisin-like protein proconvertase ubiquitously expressed in fish [144]. It presents an intracellular cleaving activity of numerous biologically relevant molecules in the trans-Golgi network. This includes BDNF in the brain, notably in pre-synaptic axons. In mice, furin has been shown to participate to dendrite morphogenesis and modulate learning abilities and memory [145]. Stress imprinting at FURIN is likely and it has recently been shown that transgenerational epigenetic effects of furin activity were active in brain of mice [146]. ProBDNF cleavage by furin depends on brain AC and CREB (cAMP response element-binding protein) signaling [147,148], but also plasminogen (PLG) [149]. This activity is modulated by stress hormones (corticosteroids) and is essential to brain hippocampal plasticity in mammals [22,150]. One interesting supplementary observation is that CREB signaling necessary to furin is associated to CRTC2 - a CREB co-activator – found as one of the four DMRs reported in this study. In mice, CRTC2 is known to act as a switch for BDNF and glucocorticoids to direct the expression of corticotropin-releasing hormone (CRH) in the hypothalamus [151].
Additionally, in the brain, plasminogen encoded by PLG is converted to plasmin that cleaves BDNF in the extracellular synaptic domain [128,152]. PLG has also been shown to regulate pro-opiomelanocortin (POMC) in the hypothalamic-pituitary axis, then the production of peptides hormones such as the adrenocorticotropic hormone (ACTH) [153].
In relation to BDNF, two other DMC-related genes should be mentioned: NPAS3 (neuronal PAS domain containing protein 3) and SLC22A2. SLC22A2 - also known as OCT2 (organic cation transporter 2) – associated to the unique 3’UTR DMC found in this study is implied in numerous transmembrane transports [154], including at the blood-brain barrier [155]. It was found involved in memory in mice [156,157] or Drosophila [158], and associated itself to neurodegenerative diseases [159]. Substrates of SLC22A2 include neurotransmitters such as norepinephrine, but also dopamine, serotonin and a wide-variety of internal or exogenous compounds [160,161]. Functional links of SLC22A2 with BDNF remain however poorly documented [162], while they have been found co-expressed and co-regulated in the brain during drug administration experiments (e.g. methamphetamine, mimicking the action of catecholamines) [163]. In this study, PLG and SLC22A2 are associated to the same DMC; the functional significance of this situation should be investigated further. Finally, NPAS3 has a well-established action in memory [164,165]. It participates to neurogenesis in the hippocampus within a network that also includes BDNF [166]. NPAS3 is also associated to the glial cell line-derived neurotrophic factor (GDNF) receptor-alpha2 gene (GFRA2) detected in this study and related to stress and anxiety [167]. The DLG1 (Disk-large homolog 1) gene could be also associated to BDNF, but this will be not develop further (see Additional File 3).
Overall, results suggest that even in absence of BDNF sequencing reads, some DMCs/DMRs did not occur only by chance in this study. These DMCs/DMRs are also related to processes that regulate the HPI axis and hormones. The measurement of methylation difference now used to monitor the BDNF activity and its relationship to stress [132,168] could thus potentially depend on the methylation status observed at or close to ADCY1b, ABLIM2, FURIN, CRTC2, PLG and other genes such as SLC22A2 and NPAS3. This should be investigated further to improve our understanding of changes in the directionality of multigenic epigenetic modifications and in inter-individual variation in stress coping. Candidate genes coming from other fish or more generally vertebrate studies with role in memory consolidation could be added (e.g. [112,169] for proposals).
As at least ADCY1b, FURIN, CRTC2, PLG and SLC22A2 have other well-recognized relationships with stress and the stress response (especially the glucose metabolism and the immune function; see Additional File 3), this also illustrates the complex non-linear network modulating the stress response in fish with a pivotal role in the brain associated to BDNF and secondary additional outcomes in other tissues and organs.
Family effect, development and few individuals: a need to improve findings
The uneven representation of families in our sampling protocols did not aim to properly investigate any family effect in sea bass, and prioritized the analysis of methylation changes occurring between pre- vs. post-stress fish. Indeed, the PCA showed that pre- and post-stress individuals have in average distinct methylation profiles that may represent stress imprinting during the challenge test. However, as in other fish species [51,54,63], a strong family effect was also detected in this study, suggesting that sea bass families certainly do not have the same sensitivity to stress. This may arise because of (i) their genomic backgrounds alter how family will develop an epigenetic stress response during the challenge test, (ii) the developmental period during which families were distributed over different tanks was a “critical time window” [170] sufficient to imprint individuals and lead them to develop a familial response to stress during the challenge, or that (iii) some degree of transgenerational inheritance in methylation profiles is present. These proposals are not mutually exclusive, but cannot be tested with data at hand and/or experimental design retained. Transgenerational inheritance has gained interest over the last years [171], notably in marine species [50,172,173], including sea bass [83]. Parents and/or the germline were unfortunately not sampled in this study for further analysis. As Gavery et al. [41] reported that BTR30, DLG1, MMRN2, and NOL4B - also found in this study - were differentially methylated in the sperm of O. mykiss, these differentially methylated genes could be interesting candidates for transgenerational inheritance study of the stress response. Other candidates are possible. For example, Ghalambor et al. [108] reported rapid evolution in brain gene expression of the Trinidadian guppy (Poecilia reticulata) for the TNKS and NCBP2 genes, found as DMC-related genes in this study. It would thus be interesting to investigate how gene expression and cytosine methylation at these genes may co-vary within and between species over generations.
Anastasiadi and Piferrer [83] showed that the domestication process in D. labrax resulted in genome-wide methylation differences for genes involved in the nervous system and neural crest cell differentiation. In our study, GLI2 [174], PBX1 [175], DLG1 [176] and FOXJ3 [177] are DMC-related genes involved in neural crest cell proliferation and differentiation. ROBO3 is involved in the immune response, but also implied in early neurogenesis [178], as well as NPAS3 [179]. With BMP3 [180], FURIN [181]; NOL4B [182], and Myf5 [183,184], most of these genes are engaged in the development of the anterior region and/or the craniofacial skeleton which is modified during sea bass farming [83]. This suggests that methylation patterns observed in sea bass could also potentially rely on the ontogenetic regulation of a particular phenotype between pre- and post-stress fish, rather than being directly related to the stress challenge. Variation in gene expression or methylation accompanying ontogeny are is reported in fish [45, 185]. and, As for other organisms, cytosine methylation has been shown to vary with age in sea bass [84]. Overall, collateral effects of stress might thus be present in our data.
On another hand, the epigenetic profiles of four individuals were analyzed in the pre- and post-stress situations and clustered very closely from each other in each case. While based on few observations of randomly sampled individuals, it may suggests that neither the stress challenge itself, nor development have a strong impact on the cytosine methylation landscape in sea bass. How transgenerational and within-generation stress-imprinting events influenced by ontogeny may interact to shape both the plastic and the heritable component of the stress response in relation to environmental stimuli require in depth evaluation [186,187]. Nature and strength of family-based epigenomic variation are of considerable importance attention to engage future selection breeding improvements in a cultured fish like sea bass [188], including issues about health and welfare [189]. However, this suggests adopting a far more complex and rigorous experimental design that the one followed in the present study, especially to include the study of unchallenged fish to control for developmental aspects.
Blood and RBCs: extending the perspective
Despite limits to this study, as fish engaged in aquaculture programs as sea bass are costly and based on breeders that cannot be easily sacrificed, epiGBS using blood samples may offer a genome-wide assessment of stress-induced epigenetic marks over a significant number of individuals and at reasonable cost. Blood, a tissue subjected to systematic hormonal fluctuations by a centrally produced stress response, is widely used in human epigenomics and has been recently used in farmed chicken [190], but not in fish despite the presence of nucleated cells and minimally invasive sampling that allows for multiple measurements. Using RNA-sequencing, authors [38,39] already suggested that fish erythrocytes would be useful to provide insights on the innate immunity or response to pathogens. Blood is Our study suggests that blood could effectively offer an in depth insight for the evaluation of the stress response from a novel perspective, accompanied by classical physiological measurements (e.g. cortisol, glucose, lactate). This may provide a more complete picture of the stress response in fish than currently performed and DMCs/DMRs be used as markers for stress diagnosis in order to monitor good practices in animal production setups [190].
The link between brain and blood epigenomics remains however to be explored more deeply in fish and requires careful evaluation and validation to correct for tissue specificity, as requested in human [191]. Our attempts to survey fish studies undoubtedly showed that considerably more knowledge has to be accumulated before to reach this issue. In human, the lack of access to brain tissues have been shown to impede some epigenomic studies and naturally raised ethical issues certainly also valid for fish [192,193]. In situ hybridization techniques in brain have been shown to be especially important to link neurobiological activities and stress coping in fish [134,194]. Their use and thus sacrifice of individuals cannot be ruled out. Actually, methylation differences among full brain or brain areas have been rarely studied in fish, mostly in zebrafish [195], but recently in a tilapia [48], a goby [56], and a killifish [43].