DNA methylation is one of the most studied and characterized epigenetic marks. It corresponds to the addition, via an enzymatic process, of a methyl (CH₃) group covalently linked to a cytosine residue in position 5 (5-methylcytosine, 5mC). DNA methylation that occurs mainly on palindromic CpG sequences, which can form CpG islands, takes place during early development and does not increase over time (Jang et al., 2017). However, recent studies indicate that DNA methylation can also occur on non-CpG sequences (called CpH, where H is A, C or T), especially in the brain of vertebrates (de Mendoza et al., 2021). Methylation of CpH, unlike CpG, is established de novo during the post-natal period, which corresponds to the maturation processes of neurons, and is increased later in life (Guo et al., 2014; Stroud et al., 2017). DNA methylation on CpG usually leads to inhibition of gene expression when it occurs at the promoter region of genes, while it increases transcriptional activity when it occurs in the transcribed regions. Although many studies suggest that CpH methylation is inhibitory, this is still debated (Jang et al., 2017). The addition of the CH₃ group is catalysed by DNA methyltransferases (DNMTs). There are 3 DNMTs, DNMT3A and DNMT3B being involved in de novo methylation, while DNMT1 catalyses the maintenance of methylation during DNA replication. DNA methylation, although quite stable, is a reversible mechanism mediated by the action of Ten-eleven translocation (TET) enzymes. This family of 3 isoforms catalyses the oxidation of 5mC to hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC), the last two products can be excised from DNA through base excision repair, thus resulting in the reversion to demethylated cytosine (for review Lacal and Ventura, 2018). Interestingly, the activity of DNMTs and TETs is particularly high during neurodevelopment and the balance between these two families of enzymes is thought to play a crucial role in the dynamic regulation of brain genes involved in development and in cell differentiation (Tao et al., 2018; Cisternas et al., 2019). A recent study reported the methylation patterns of the Presenilin-1 gene (PSEN1), which encodes the catalytic peptide of the gamma-secretase complex (a key enzyme that cleaves the AβPP), in the brain of AD animals and humans. Interestingly, a temporal correlation between dynamic modifications in the PSEN1 CpG and non-CpG methylation patterns and mRNA expression during neurodevelopment and AD neurodegeneration has been reported (Monti et al., 2020). Moreover, several studies have shown that a decrease in TET and DNMT activity is associated with AD and cognitive impairment (Christian et al., 2020; Zhang et al., 2020; Antunes et al., 2021). Although it is likely that 5hmC represents the first step of active DNA demethylation, several data suggest that the 5hmC mark, which is enriched in adult CNS, is an active and independent epigenetic mark promoting the recruitment of factors allowing or not gene expression (Mellén et al., 2012; Shi et al., 2017; Stoyanova et al., 2021). In addition to DNA, methylation can also occur on mRNAs and lncRNAs. In this case, the addition of the methyl group takes place on adenine at position 6 (m6A) and is catalysed by a protein complex including METTL3 and METTL4 that carry the catalytic domain. This post-transcriptional modification of RNA promotes binding to other proteins, allowing pre-miR formation, translation or mRNA degradation. As with DNA methylation, this mechanism is reversible via the action of FTO and ALKBH5 (for review Zhou et al., 2020).

In mammalian cells, histones interact with DNA to form chromatin. There are four types of histones: H2A, H2B, H3 and H4, which are present in duplicate to form an octamer, around which DNA is bound, forming a nucleosome. In the same way as DNA, histones can undergo several post-translational modifications (acetylation, methylation, phosphorylation, ubiquitination … ), altering the structure of chromatin. Histone methylation, via histone methyltransferase (HMT) activity, on the N-terminal tail changes the degree of chromatin compaction, making the DNA accessible or inaccessible to transcription factors, regulating gene expression. Histone methylation can enable or repress gene expression depending on the histones methylated and the amount of methylation. The mechanism is dynamic as histones can be demethylated by histone demethylases (HDMs). Histones can also be acetylated on the lysine residues of the N-terminal tail. The addition of an acetyl group is catalysed by histone acetyltransferase (HAT), while histone deacetylases (HDACs) remove acetyl groups. Acetylation of lysines opens up chromatin, making the chromatin structure transcriptionally active with a greater access of transcriptional activators. In contrast, deacetylation compacts chromatin, preventing the binding of transcription factors, thus inhibiting gene expression (for review Bannister and Kouzarides, 2011).

There are two types of non-coding RNAs (ncRNAs) classified according to their size, ncRNAs below 200 nt are defined as “short” (sRNAs) and those above this length as “long” (lncRNAs). NcRNAs are single-stranded and the result of post-transcriptional maturation. Among sRNAs, microRNAs (22–25 nt), the most studied, can assemble to a protein complex and associate by complementarity with RNA segments usually on the 3’ untranslated region, leading to translational repression or mRNA degradation depending on the level of complementarity. The lncRNAs, whose functions are less studied, have genes similar to the genes that code for proteins and have different modes of action by linking to DNA, RNA and proteins. Non-exhaustively, they can inhibit transcription by preventing the recruitment of the pre-initiation complex, regulate other epigenetic modifications by guiding enzymes (HDACs, HATs, HMTs, HATs, DNMTs … ) to their site of action, activate the expression of neighbouring genes, bind to mRNA leading to mRNA degradation, bind to miRNAs limiting the action of the latter, regulate the splicing (for review Hombach and Kretz, 2016).

Although the emergence of Omics technologies has allowed to better characterize epigenetic modifications in the context of AD, these studies, that have been the subject of recent reviews (Monti et al., 2020; Coppedè, 2021; Nikolac Perkovic et al., 2021), are outside the scope of the present manuscript which is focused on the putative influence of the early environment on AD programming.

Exposure to lead (Pb) during early-life has been particularly studied in the context of AD. Indeed, lead pollution is a major public health issue in developing cities. A meta-analysis using data on Pb exposure in Mexico City residents showed high blood Pb levels associated with a 5-point reduction in Intellectual quotient (IQ), and intellectual disability in children aged 0–4 years-old (Caravanos et al., 2014). Although no epidemiological study reports the effects of long-term Pb exposure, experimental studies in rodents (Bihaqi et al., 2014; Gąssowska et al., 2016) and monkeys (Bihaqi and Zawia, 2013) indicate that early exposure to Pb augments the phosphorylation of Tau, and is associated with an increase in the level of the cyclin-dependant kinase 5 (Cdk5), a major Tau kinase that is involved in abnormal phosphorylation of Tau in AD brain (Cruz and Tsai, 2004). In addition to these effects on Tau, studies also showed an increase in AβPP expression and plaque formation in the aging primate brain induced by early Pb exposure (from birth until 400 days of age). That was associated with a decrease in DNMT1 and DNMT3a activity and with a reduction in the acetylation and methylation of certain histones, resulting in a reprogramming of gene expression associated to neurotransmitter, growth factors and signal transduction pathways, suggesting an epigenetic effect (Wu et al., 2008; Bihaqi et al., 2011).

The action of bisphenol A (BPA), an endocrine disruptor used to manufacture polycarbonate plastics, is also well documented. Interestingly BPA, due to its lipophilic nature, is able to penetrate the blood brain barrier and has been found in the placenta, amniotic fluid, blood breastmilk, suggesting that it could act on foetuses/neonates during the perinatal life (Hines et al., 2015). Indeed, epidemiological studies have associated BPA exposure during early life stages with low birth weight and an increased risk of developing metabolic diseases in adulthood (Chevalier and Fénichel, 2015). In mice, maternal BPA exposure disrupts brain function and induces cognitive deficits (Tian et al., 2010) and learning-memory impairment in offspring (Xu et al., 2010). It also augments AβPP levels and Tau phosphorylation via increased activities of GSK3β and CDK5—two major Tau kinases (Sergeant et al., 2008)—but decreases insulin signalling and alters synaptic plasticity in adulthood (Fang et al., 2016; Xue et al., 2020). Prenatal BPA exposure deregulates the offspring hippocampal transcriptome with genes associated with AD, oxidative stress and inflammation (Sukjamnong et al., 2020). In particular, it induces elevation of NF-κB protein and its AD-related target BACE1, a key enzyme involved in AβPP processing, suggesting that prenatal BPA exposure triggers neuroinflammation in the hippocampus and increases the susceptibility of AD through NF-κB. These modifications may explain the cognitive impairments observed in other studies (Negishi et al., 2004; Ryan and Vandenbergh, 2006; Wang et al., 2014), even with a single dose of BPA at PND10 (Viberg et al., 2011). Indeed, in utero and neonatal exposure to low dose of BPA may cause deregulation of miRNAs expression, DNA methylation levels and histone modifications (Singh and Li, 2012). A genome-wide analysis of the foetal mouse forebrain epigenome shows that exposure to low doses of BPA modifies the methylation in the promoter-associated CpG islands at several loci, which can alter the brain development (Yaoi et al., 2008). In rat, the exposure to the Di-(2-thyhexyl)-phtalate (DEHP), another endocrine disruptor, during gestation and lactation also increases the phosphorylation of Tau, without modification of AβPP level in hippocampus, and is associated with cognitive impairments in adulthood (Sun, 2014). However, so far no studies have investigated the long-term consequences of early-life BPA on epigenetic changes in the Microtubule Associated Protein Tau (MAPT, the gene encoding Tau) and AβPP genes.

Exposure to xenobiotics, man-made chemicals found in the environment but not endogenously produced, is increasingly common, and their presence in the brain during the perinatal period can alter CNS development. Polybrominated diphenyl ethers 209 (PBDE 209), a combustion inhibitor that spreads in the environment, has been found in breast milk and in consumer products. Studies in animals show that a single exposure to PBDE 209 at PND10, at the time of brain growth spurt in rodent, does not modify the level of Tau but affects CaMKII and synaptophysin levels in the hippocampus (Viberg, 2009), two proteins playing an essential role in neurodevelopment, whose alterations may likely explain impaired spontaneous behaviour seen in adulthood (Johansson et al., 2008). Similarly, exposure to another xenobiotic, the non-dioxin-like polychlorinated biphenyls (NDL-PCB), during lactation in Swiss albino mice led to a more pronounced decrease of the synaptic proteins synaptophysin and PSD95 and impairment of long-term memory when the brain is challenged by the injection of Aβ oligomers (Elnar et al., 2016).

Maternal obesity has considerably increased among women of reproductive age in the last decades. Current estimates suggest that by 2025 more than 21% of women in the world will be obese, and in 2014, the estimated percentage of overweight and obesity among pregnant women was 33% in the United States (Poston et al., 2016). Cumulative data suggest that maternal obesity leads to metabolic and cognitive disorders in the offspring (Hasebe et al., 2021), which makes maternal obesity a major public health issue. However, although cognitive and metabolic impairments are two components of AD, the consequences of maternal perinatal high-fat diet on Tau and Aβ have been poorly investigated. A recent study by Di Meco et al. reports that a maternal high-fat diet (mHFD with 42% calories from fat) during gestation decreases total Tau and pathological Tau conformation level and increases the synaptic integrity, leading to improved cognitive performance in adult mice (Di Meco and Praticò, 2019). However, it has also been reported that a mHFD (45% fat) during gestation and lactation increases the level of Aβ peptides at the vascular level, and is associated with a modification of the neurovascular environment in the brain of adult mice, which leads to the alteration of the perivascular clearance of Aβ peptides (Hawkes et al., 2015). Although many works have looked at the epigenetic changes induced by mHFD in the programming of peripheral metabolic diseases, very few studies have looked at the consequences in the CNS of the offspring (Gawlińska et al., 2021a). For instance, it has been reported that consumption of a high-fat diet during gestation and lactation in the Wistar rat decreases the global level of histone H4 acetylation in the hippocampus at P50, leading to transcriptional repression (Gonçalves et al., 2017). Interestingly, a diet rich in grape juice has an opposite effect on the level of this epigenetic mark at PND21 and is associated with a beneficial effect on the brain with an antioxidant effect (Gonçalves et al., 2017; Schaffer et al., 2019) which could limit deleterious consequences of mHFD by reducing oxidative stress and acetylcholinesterase activity (Proença et al., 2021). However, to our knowledge, the putative consequences of mHFD on epigenetic changes in MAPT and AβPP genes have never been reported.

According to the WHO, more than 1.3 billion people worldwide smoke, about 40% of them being women. In low- and middle-income countries, where the proportion of people who smoke is highest, the prevalence of pregnant women who smoke is 2.6%, but can be as high as 30.8% in Russia (Caleyachetty et al., 2014). The deleterious effects of smoking on offspring are widely accepted and can lead to IUGR and neurodevelopmental changes (Perera et al., 2005; El Marroun et al., 2016). An epidemiological study on 471 individuals in Finland showed that male offspring of mothers who were exposed to cigarette smoke prenatally had more cognitive problems, although these effects were minor or absent in females (Ramsay et al., 2016). In mice, it has been shown that exposure to cigarette smoke during lactation increases the phosphorylation of Tau, and the levels of the 3R form of brain Tau, leading to a deregulation of the 3R/4R ratio of Tau during a critical period of brain development in 4-week-old offspring (La Maestra et al., 2011). Indeed, Tau phosphorylation but also the proportion of each isoform plays an important role in brain development, and particularly during lactation, during which synaptogenesis and axonal growth take place. Although not reported to date, these early changes could lead to neuronal dysfunction and exert long-term consequences.

Humans are routinely exposed to ionizing radiation (IR) from natural sources, and from man-made sources such as nuclear energy. IR is also increasingly used in the medical field in the diagnosis or treatment of diseases, for example in X-ray imaging. For example, in children, for the treatment of brain tumours, exposure at a very young age can also influence the development of CNS. An epidemiological study on 3,094 Swedish males indicated that low-dose IR exposure during early human development can have a negative impact on cognitive abilities during childhood (Hall et al., 2004). An experimental study in mice reported that a single low-dose exposure of IR to PND10 animals deregulates the hippocampal and cortical proteome and transcriptome, impairs synaptic plasticity, reduces neurogenesis and leads to neuroinflammation later in life. Interestingly, miRNAs analysis showed an increased expression of miR-132 and miR-134 repressing the translation of proteins of the Rac1-Cofilin signalling pathway involved in synaptic plasticity and neurogenesis, highlighting a potential epigenetic mechanism (Kempf et al., 2014). Another team using a similar protocol demonstrated an increase in cortical Tau levels at PND11 (i.e. 1 day after exposure) and at 6 months. That was associated with a decline of long-term spatial memory when IR is combined to ketamine exposure, an anaesthesia commonly used to facilitate radiotherapy in children. Although it has not been directly assessed, these long-term consequences on Tau protein levels and cognitive deficits suggested the involvement of epigenetic mechanisms (Buratovic et al., 2014; Buratovic et al., 2018).

Perinatal asphyxia is a defect in blood flow or gas exchange from the mother to the foetus or vice versa, occurring immediately before, during or after birth. It accounts for up to 20 cases per 1,000 births in developing countries, with mortality during the neonatal period of 15–20%, while 25% of survivors develop short- and long-term neurological disorders, associated with neonatal encephalopathy in most cases (Gillam-Krakauer and Gowen Jr, 2021). The most common diagnosis of perinatal asphyxia is addressed by chest X-ray, which, as mentioned in the previous paragraph, can be deleterious, especially in the young individual. Thus, it is important to uncover biomarkers released into the cerebrospinal fluid (CSF) during perinatal asphyxia to avoid X-ray exposure. With this idea in mind, one team evaluated the levels of Tau, pTau and Aβ42 in the CSF of newborn pigs that had undergone a perinatal hypoxia event a few hours after birth. They showed a decrease in the level of Aβ42, with no change in the level of Tau and pTau. They also observed an increase in S100B (Benterud et al., 2015), a glial marker of neuroinflammation which is also increased in CSF of infants with perinatal asphyxia (Massaro et al., 2014). In contrast, an epidemiological study reported that delivery by caesarean, known to limit hypoxia in cases of perinatal asphyxia, was associated with a reduction in cord blood Tau protein levels, indicating that Tau is modulated by blood oxygen concentration (Tunc et al., 2010) although it could not be excluded that Tau variation results from other factors associated with caesarean delivery. However, it is clear that perinatal asphyxia modified Tau, pTau and Aβ42 in the fetus/neonate, it remains to be determined if these early-life changes influence (or not) the development of late-onset AD.

As previously mentioned, mHFD can have long-term deleterious effects on the levels of Tau and Aβ peptides. The consequences of mHFD have also been studied in the 3xTgAD mouse, a model of AD that develops early amyloid pathology and later Tau pathology. A first team reported that the application of mHFD (60% fat) during gestation and lactation impaired short- and long-term spatial memory and increased the number of Tau-positive neurons in the hippocampus without altering Aβ peptides level, suggesting that mHFD exacerbated Tau pathology and may sensitize to the development of AD (Martin et al., 2014). Surprisingly, using the same animal model it has also been reported that mHFD (42% fat) only applied during gestation protects from synaptic dysfunctions and associated cognitive disorders and decreases Tau and amyloid pathology by reducing the amount Aβ40 and Aβ42. This putative protective effect of the maternal diet could result from an increased gene expression of the transcriptional repressor FOXP2, which could be responsible for the decreased gene expression of Tau, CDK5 (a Tau kinase) and BACE-1 (involved in the amyloidogenic pathway) possibly via epigenetic mechanisms (Di Meco et al., 2019). It is important to note that an important limitation of this model is that the mice are homozygous for all 3 mutant alleles, and thus the observed phenotype is the result of the diet given to transgenic dams. Therefore, it is not entirely clear if changes result from the diet effect only or from the convergent impact of the diet and the transgenic phenotype of dams. Using the Tg2576 mouse, another model of AD expressing the human Swedish AβPP mutation and developing only amyloid pathology, it has been reported that a mHFD (45% fat) starting before crossing until weaning increases the level of soluble Aβ and the number of amyloid plaques. It has been suggested that these alterations could modify the extracellular matrix leading to increase Aβ plaques retention within the parenchyma (Nizari et al., 2016). These seemingly contradictory results may be explained by the different amount or types of fat used in the diets and/or by the timing of exposure of mHFD, as well as by the model used regarding amyloid pathology. Although the biological mechanisms that explain the consequences of mHFD on the development of AD are still poorly understood, recent studies suggest an important role for the gut microbiota. Indeed changes in maternal diet during the perinatal period modify both the mother’s and the offspring’s microbiota with maternal transmission both at delivery and through milk intake (Mueller et al., 2015). In mice, perinatal mHFD alters the short-term memory and is associated with a deregulation of the gut microbiota in offspring, which could partly explain the observed phenotype (Sanguinetti et al., 2019). Interestingly, a high-fibre diet in mothers or offspring can restore cognitive deficits in offspring by regulating the composition of bacteria and SCFA metabolites in the long-term (Liu et al., 2021). It has been suggested that early deregulation of the microbiota may participate to cognitive impairment and to the onset of later lesions in the 3xTgAD mouse model (Bello-Medina et al., 2021). Interestingly, addition of bioactive food to the diet restores these effects and modulates the composition of the gut microbiota (Syeda et al., 2018). Together, these studies indicate that the modification of the diets or the addition of probiotics, both of which being able to modulate epigenetic mechanisms, offer the opportunity for identification of new diagnostic biomarkers and treatments (Ji and Shen, 2021).

Tea and coffee are two commonly substances consumed in the world, whose main component is caffeine, a psychoactive substance. Numerous epidemiological studies in humans and experimental studies in AD mouse models have shown the beneficial and long-term protective effects of moderate caffeine consumption in adulthood in reducing the development of lesions and related to cognitive impairment, decreasing the risk of developing AD by 65% in humans (Eskelinen et al., 2009; Flaten et al., 2014; Polito et al., 2018). Sustaining epidemiological studies, caffeine treatment of adult mice developing amyloid or Tau demonstrate positive outcomes at the pathological and behavioural levels (Arendash et al., 2006; Arendash et al., 2009; Laurent et al., 2014). In sharp contrast perinatal caffeine, by blocking adenosine A2A receptors, not only exert a detrimental impact on neuronal migration but caffeine receptors themselves strongly impact GABAergic synapse stabilization during synaptogenesis (Gomez-Castro et al., 2021). Further, perinatal caffeine strongly impacts hippocampal excitability and memory at the adult stage. Whether perinatal caffeine modulates the later development of AD pathology or has cognitive consequences remains largely unclear. However, recent data showed that perinatal caffeine consumption starting before gestation until PND15 in a transgenic mouse model of tauopathy, even not modifying the pathological load, leads to earlier long-term memory impairments together with excitatory/inhibitory synaptic drive in adulthood. These results indicate that early-life caffeine may influence brain development and exert long-lasting consequences in a Tau background (Zappettini et al., 2019).

Evidence indicates that chronic stress can participate to the progression of AD and other neurodegenerative disorders. In 3xTgAD mice, chronic social stress in adulthood, by regularly changing the composition of the mice in the cage, decreases BDNF levels and increases amyloid pathology and anxiety, with no change in control animals indicating a greater sensitivity to stress in mutant mice (Rothman et al., 2012). During the perinatal period, although few studies have been reported, Sierksma et al. have shown that chronic restraint stress performed during gestation (E1-E7) in APPswe/PS1dE9 mice model of AD induced sexual-dimorphic results in the adult offspring with males exhibiting altered spatial memory, while females have improved spatial memory associated with decreased amyloid load in the dorsal hippocampus (Sierksma et al., 2012). Although these modifications were not associated with changes in DNMT expression nor 5mC and 5hmC levels, other epigenetic mechanisms targeting histones or non-coding RNAs remain to be studied. Using the same AD animal model, it has been shown that repeated maternal separation during lactation, which is also used to induce stress during early-life, increases amyloid deposits in the hippocampus and cortex, and impairs learning and spatial memory in adulthood (Hui et al., 2017). The same experimental protocol increases Tau and amyloid pathology, decreases synaptophysin and PSD95 (two synaptic markers), impairs memory and is associated with hyperactivity of the hypothalamo-pituitary-adrenal (HPA) axis in adult male rats (Martisova et al., 2013). These observations are in agreement with the elevated glucocorticoid levels in adulthood found associated with a higher risk of developing AD, by altering glial function, Tau and amyloid lesions, and synaptic plasticity (Vyas et al., 2016). The first evidence that early-life stress could induce epigenetic programming in the rat brain came from the pioneer studies of Weaver et al. who showed that poor maternal care leads to increased methylation in the glucocorticoid receptor (GR) promoter which was accompanied by aberrant behaviours and altered HPA responses in adult offspring (Weaver et al., 2004). Using the 3xTgAD mice, it has been reported that handling offspring during lactation (PND2-PND9), by separating the dam from her pups for 15 min, increases maternal care of the dam towards her pups upon reunion. Increased maternal care leads to a decrease both in short-term memory decline and in amyloid load in the hippocampus, suggesting that prenatal stress could be mitigated or to have antagonistic effect by postnatal maternal care.

Recently, it has been reported in a subclinical AD-like rat model, that adult animals which were prenatally-stressed (PS) exhibited increased dysfunctions of HPA axis activity and working and long-term spatial memory as well as modified cellular organisation in CA1 hippocampal region. Interestingly, these alterations were alleviated by increased postnatal maternal care during lactation. Unfortunately, the authors did not report the putative consequences of cross-fostering in adult PS rats on Aβ plaques and NFT nor on phosphorylation and protein levels of Tau (Rostami et al., 2020).

In the present review, we provided strong evidence that early-life stress may constitute another risk factor contributing to the development of late-onset AD possibly via long-term influence of epigenetic mechanisms (Figure 1). However, important questions, which are still unanswered in the context of AD, such as the putative contribution of transgenerational epigenetic inheritance, paternal contribution and sexual dimorphism on late-onset AD development, have been overlooked. It should be kept in mind that although epigenetic marks may be transmitted transgenerationally (Burton and Greer, 2021), the existence of multigenerational epigenetic inheritance in humans is still a matter of debate (Cavalli and Heard, 2019). Epigenetic mechanisms are indeed plastic, modifiable and reversible processes providing thus opportunities to try to reverse deleterious effects of environment via early-life interventional strategies (Lesuis et al., 2018). Recently, the consequences of mHFD (60% fat) have been examined in F1 (intergenerational inheritance) and F2 (transgenerational inheritance) generations of SAMP8 mice, a model of accelerated ageing leading to increased cognitive decline associated with the development of neuropathological AD hallmarks (Liu et al., 2020). The HFD applied to female mice of the F0 generation was supplemented or not with resveratrol, a natural phytoalexin known to have beneficial effects in pathological conditions, notably via its anti-inflammatory effects, and by reducing amyloid pathology and playing a role in epigenetic mechanisms (Rahman et al., 2020). Interestingly, the impairment of both short- and long-term memory and metabolic parameters (increased plasma levels of triglycerides and leptin) induced by the HFD was attenuated by resveratrol supplementation in the F0, F1 and F2 generations supporting a multigenerational effect. In addition, the level of DNA and RNA methylation (on 5mC and m6A, respectively) is increased in F1 offspring and associated with a change in the expression of enzymes involved in these epigenetic mechanisms, strongly suggesting epigenetic inheritance (Izquierdo et al., 2021). Future studies will be clearly required to understand if and how early nutritional interventions in mothers may exert beneficial effects in offspring and mitigate the long-term deleterious effects of adverse environmental exposure on late-onset AD.

Although most of the studies focused on the consequences of maternal environmental stress in the offspring, a growing body of evidence indicates that paternal preconception stress may also have long-term effects on the offspring. Studies in rodents indicate that paternal obesity induced by HFD alters the development of the 2-cell embryo. These early events have long-lasting consequences and decrease neurogenesis and memory, and are associated with epigenetic marks changes in adulthood, including hypermethylation of BDNF promoter, a neurotrophic factor that plays a key role in brain development and function (McPherson et al., 2015; Zhou et al., 2018). It has also been reported that repeated paternal stress (Harker et al., 2015, 2018), as well as paternal BPA exposure (Luo et al., 2017), lead to behavioural alterations, associated with changes in dendritic spines and brain organisation in the offspring. In addition, paternal stress during spermatogenesis leads to altered behaviour and DNA methylation patterns with increased level in the hippocampus of both male and female PND21 offspring (Mychasiuk et al., 2013). In mice, paternal cognitive spatial training improves cognitive performance in adult offspring and is associated with an increase in the hippocampal expression of synaptotagmin, a pre-synaptic protein. This was associated with elevation of histone acetylation at the promoter of synaptotagmin in the hippocampi of both fathers and offspring as well as in fathers sperm (Zhang et al., 2017). Thus, it would be interesting to delineate the role of paternal stress in animal models of AD and to determine whether both maternal and paternal exposure to stress would have a cumulative effect. Although it would more closely mimic what happens in humans, to our knowledge, this has never been reported in the context of AD.

As previously mentioned, a large majority of works have examined the consequences of maternal stress on male offspring. To date and to the best of our knowledge, only three studies look at the consequences of early-life environment on females, even though Alzheimer’s disease affects women preferentially. In the US, nearly 70% of AD patients are women, which can be explained by a longer life expectancy, but also by intrinsic biological differences, including brain organisation and function, age-associated loss of protection exerted by steroid hormones, higher inflammatory response, but also different susceptibility to genetic variations (for review Fisher et al., 2018). In addition, during early-development brain volume is higher in males, while the hippocampus is larger in females, pointing out to sexual dimorphism during a critical time-window (Kaczkurkin et al., 2019). Finally, although few studies have been reported in the context of AD, the use of mouse models led to discordant results in males and females showing the importance of studying the effects in the two sexes independently.

In conclusion, although the way by which perinatal epigenetic mechanisms act to program long-term memory and cognition remains to be clarified, it seems clear that the development of multi-omics strategies will allow molecular dissection of this fascinating new field of research. In particular, it will offer novel opportunities to identify early markers of AD and to develop novel preventive strategies to attenuate age-associated epigenetic alterations.