Highlighting the gaps in hazard and risk assessment of unregulated Endocrine Active Substances in surface waters: retinoids as a European case study

Regulatory hazard and risk assessment of endocrine-active substances currently specifies four modes of action: interference with sex hormone (oestrogen, androgen) pathways, steroidogenesis, and thyroid hormone signalling. This does not encompass the full complexity of the endocrine system and its extended interfaces with environmental pollutants that can potentially disrupt the carefully maintained balance. Here we take the retinoid signalling pathway as a European case study for both, under- and unregulated endocrine pathways and outline the different levels of interference, discuss their adversity, and indicate crosstalk to other signalling pathways. Retinoid compounds already exist in drinking water sources, occur naturally in cyanobacterial blooms and/or enter surface waters via wastewater discharge, where they pose a potential hazard to the environment and human health - a situation that can be expected to worsen due to water shortages induced by climate-change and population growth. We briefly review relevant aspects of current endocrine disruptor (ED) testing for regulatory purposes and then expand upon the needs for inclusion of disruption of retinoid signalling in (ED) regulatory safety assessment contributing to adverse health outcomes that include cognitive function and neurological disease. An overview of developmental effects of retinoid signalling disruption across species highlights critical processes and potential crosstalk with other signalling pathways. A focused weight of evidence-based evaluation of the biologically plausible associations between neurological disorders and altered retinoid signalling highlights the evidence gaps. We show that monitoring only a limited number of anthropogenic priority chemicals in water is insufficient to address the environmental risks of retinoid signalling disruption. To comprehensively assess impacts on the endpoints, processes, and pathways of the endocrine system that are most vulnerable to chemical interference we need further investigation of the true mixture composition in environmental matrices. On a weight of evidence-basis this information can then be integrated into a reliable, inclusive, quantitative approach that ultimately accommodates all the critical pathways. By focusing on the retinoid signalling pathway, we intend to improve the scope and relevance of an integrated approach for the risk assessment of endocrine disruptors.


Introduction
Internationally, chemical substances are currently screened for endocrine activity in regulatory risk assessments (as for example in the European Union's Biocides regulation [1, 2]), utilizing standard test methods that refer to chemical substances as endocrine active when interfering with sex hormone (oestrogen, androgen) receptors, steroidogenesis, or thyroid hormone signalling (EATS; the available tests are introduced in the "infobox" below) [3]. It is recognized that the endocrine system, however, is a complex interplay of different, often evolutionary highly conserved, mechanisms that by far exceed the above-mentioned four modes of action. It includes all hormone signalling pathways, interlinking and regulating an extensive set of functions, including development, growth, reproduction and metabolism [4][5][6][7][8], and this is being actively examined at inter-governmental levels [8][9][10][11][12][13][14][15][16]. The endocrine system is highly sensitive and circulating hormone levels are in the pM-µM range, making it highly susceptible to interfering compounds [17,18]. Interference of exogenous chemicals with the tightly regulated endocrine system may result in adverse health effects, that, especially when encountered during development, may have sustained and life-long [12 and references therein, 19] or even transgenerational impacts on individuals or contribute to non-communicable diseases like metabolic disorders and cancer [4,18,[20][21][22][23][24][25][26][27][28][29].
To allow an assessment of risks related to chemicals in the environment, information on the ecological or human health hazard of these chemicals is needed [2,3] together with information on exposure to these chemicals and/or mixtures, i.e. their levels and fate in the environment, to conclude as to whether there is a risk of adverse outcomes or not [30,31].
Despite the progress in the development of test methods screening for endocrine disrupting activity, endocrine pathways other than EATS remain under-investigated. Whilst currently there are no specific test methods available with respect to other endocrine mechanisms, these are being actively explored at the European level (https:// eurion-cluster.eu/), and internationally, for retinoids [13, reviewed in 14]. Such comprehensive reviews together with identification of relevant assays with reference and test chemicals are needed to address regulatory needs, prior to the development of the test method tools that can be included into legislative mechanisms.
In surface waters, endocrine disruption gained public attention when altered sex ratios, genital malformations, and reproductive impairment were discovered in aquatic vertebrates [6,[33][34][35][36][37][38]. Most strikingly, feminization of male fish occurred at oestrogen levels below the limit of detection by analytical methods available at that time and also led to the collapse of a fish population in a Canadian experimental lake [39]. The oestrogen levels in the respective water bodies have been frequently attributed to poor treatment of communal wastewaters, containing high levels of human contraceptives [40]. More recently progesterone has been detected in UK shores in molluscs at concentrations equivalent to those used in contraceptives and hormone replacement therapy [41]. The striking impact of compounds interfering with the oestrogen hormone system has expanded the field of environmental endocrine disruption and enabled investigation of other endocrine pathways sensitive to environmental interference [42].
It is intended that this review provides a useful contribution to the discussion of under-and unregulated endocrine pathways, particularly in relation to the gap in hazard and risk assessment approaches to address anthropogenic and naturally occurring toxic retinoid substances for water quality. We focus on two key aspects of environmental chemicals' potential to interfere with retinoid signalling: (1) with respect to the presence of chemicals that elicit retinoid-like activity via retinoid receptors, and (2) the potential of the endogenous retinoid system to be a target for an expanded range of chemicals which could disrupt this system. In addition to retinoid signalling pathway-related developmental and reproductive endpoints [reviewed in 13, 14], here we facilitate the addition of the less well studied endpoints of cognitive function and neurological disease. We thereby intend to contribute to the evidence base needed for the development of the tools and approaches to address endocrine adverse outcomes related to disruption of retinoid signalling pathway.
developmental processes in which retinoids are involved are further discussed in "Phenotypic patterns of interference with retinoid signalling during development" section. Besides the tissue distribution and metabolization of retinoid isomers, also the expression pattern of retinoic acid receptor (RAR) and retinoid X receptor (RXR) isoforms and variants in tissues determines retinoid effects [78,82].
Binding of retinoids, primarily atRA, to RAR results in their heterodimerization with RXR and subsequent transcriptional activation of retinoic acid-responsive elements (RAREs), which govern a number of crucial cellular processes, including inflammation, proliferation, differentiation and carcinogenesis [28, 98,99]. RAR-RXR heterodimers furthermore can recruit co-repressor complexes and, depending on the presence of natural or synthetic ligands, modulate or suppress gene expression [100][101][102].
While RARs show a higher specificity towards retinoid compounds binding and are the main driver in retinoid-mediated patterning and teratogenicity [e.g. recently reviewed by 103], the role of RXRs is broader.
One of the reasons is the molecular promiscuity of RXR. Type II nuclear receptors, characterized by forming heterodimers with RXR, govern the transcription of a large variety of target genes [104]. They are involved in the biological responses to many endogenous ligands, anthropogenic and natural chemicals and therapeutic drugs. The affected functions include lipid metabolism (peroxisome-proliferator activated receptor, PPAR), steroidogenesis, xenobiotic response (pregnane X receptor, PXR; constitutive androstane receptor, CAR), vitamin D receptor (VDR), liver functions (FXR, LXR), orphan nuclear receptors (Nurs), and thyroid hormone signalling (thyroid hormone receptor, TR [67]; Fig. 2) [7, 85,88,[105][106][107]. Whilst the receptors TR, VDR, and RAR form non-permissive heterodimers, the others (Fig. 2) form permissive heterodimers with RXR, where the transcriptional activity is regulated by a ligand binding to one of the dimerization partners [85,104,106]. Dimerization is achieved via the asymmetrical so-called identity box -a small region within the ligand binding domain, which, in the case of RXRα, consists of 40 amino acids [108,109]. This subdomain shows a very high degree of conservation. Especially, the two amino acids A416 and R421 have been shown crucial for dimerization of RXRα with RAR [108,109]. The high conservation of the RXR identity box even across animal phyla underlines the evolutionary importance of RXR [110,111].

Molecular crosstalk in the RXR signalling pathway
The fact that nuclear receptors share the common heterodimerization partner, RXR, indicates the potential for molecular crosstalk between signalling pathways dependent on RXR heterodimerization. The sequestration of ligand-bound RXR from the pool of active RXR monomers with downstream modulating activities is indicated [7], and also direct ligand activation of, e.g. the PPAR family by retinoic acids [112,113] has been reported in addition to activation of retinoid receptors. Additionally, there is evidence of crosstalk to the thyroid hormone signalling pathway by heterodimers of TR with RXR in vitro [114] and augmentation of thyroid hormone-related effects by RXR activation in vivo [67]. Most often, the ubiquitous RXRα isoform is involved in heterodimerization and it is essential for xenobiotic metabolism [7,13,115]. Competitive decrease of effect due to RXRα sequestration by retinoic acid/RAR has been reported for CAR [116], LXR, FXR, PPARα [106], and PPARγ [117] and may be implied in the metabolization and detoxification capacity mediated via activation of, e.g. PPARs. The capacity of RXRs to form heterodimers with several dimerization partners allows integration of signals from simultaneous and independent signalling pathways that can be further modulated by transcription co-factors [13, reviewed in 87]. The importance of allosteric modulators has been also stressed in a recent study on nuclear receptor binding to DNA target sequences (direct repeats and half-sites), where in vitro binding was predictive of in vivo binding, but not of in vivo function [104].
The importance of co-evolution of nuclear receptors and overlapping cis-regulatory elements also becomes apparent at the intersection of RAR/RXR and ERα signalling pathways. RAR/RXR signalling has been demonstrated several times to antagonize ER binding to respective DNA target sequences [118][119][120][121]. Besides the therapeutic use of this observation particularly in ERresponsive breast cancer [121], ERs play a critical role in organogenesis and maturation processes that, thus, can be affected by dietary and environmental factors.
Steroidogenesis critically influences the production and subsequently the circulating amount of the prototype sex steroids oestrogen and testosterone [13]. RAR/RXR play a pivotal role at the beginning of the steroidogenesis pathway, but RXR also as the essential dimerization partner for adjacent and subsequent steps interlinked with lipid metabolism (PPARs, LXR) and xenobiotic response (CAR, PXR) [13,106]. At this interface, delivery of retinoic acid to the various nuclear receptors (RAR, PPAR or VDR) can have different consequences with respect to adiposity, such that, for example VDR activation in fibroblasts induces non-adipogenic gene transcription, whilst PPARγ/RXR heterodimers contribute to adipogenic processes [7,122].
Whilst the therapeutic potential of the interdependency amongst many pathways and retinoid signalling via RXR has already been discovered and drugs specific to RXR, so-called "rexinoids" (e.g. bexarotene) are available to treat certain types of cancers [123], the implications of unintentional deregulation of retinoid signalling remain to be elucidated.

Phenotypic patterns of interference with retinoid signalling during development
Interference with retinoic acid signalling has the highest impact on humans during development and was first observed in vitamin A (retinol) deficiency. This has been understood since the early twentieth century from studies that investigated the teratogenic effects of both excess and a lack of retinoid activity [124][125][126][127][128][129][130]. To date, vitamin A deficiency is still a concern, especially in developing countries where one-third (33.3%) of pre-school age children and 15.3% of pregnant women have serum retinol levels below 0.7 µM [131], resulting in severe risk of vitamin A-preventable blindness that has a fatality rate in children of 50% within one year [131][132][133]. Without nutritional supplementation within the first year of life, this can be considered as irreversible retinoid disruption and interpreted as an example of an (irreversible) adverse outcome in humans justifying classification as an endocrine disruption pathway.
The most critical elements for interference in the retinoid signalling pathways, beyond RARs and RXRs, are the metabolizing enzymes of the cytochrome P450 subfamily 26 (CYP26) CYP26A1, CYP26B1, and CYP26C1 [94,161,162], the retinol-converting alcohol dehydrogenase (ADH), and retinal-oxidizing dehydrogenases (RALDHs) [90,94,163,164]. Efforts to identify key players and switches in the hierarchical signalling network and sort them into adverse outcome pathways (AOPs) continue to be undertaken with respect to vertebrate axial and neural tube development [14, 165,166], as well as mammalian reproduction [reviewed in 13].
Besides retinoids themselves, other environmental contaminants, such as pharmaceuticals (e.g. valproic acid), flame retardants (e.g. polybrominated diphenyl ethers), plasticizers (phthalate esters), and pesticides (triazole fungicides) have been reported to alter retinoid signalling biomarkers and induce retinoid-like teratogenic effects [111,[167][168][169][170] (see also Table 1). Exposure assessment of pharmaceuticals with the retinoid mode of action may require different regulatory approaches, depending upon the route of exposure, i.e. whether there is oral intake/ dermal application, as opposed to their occurrence and potency in (waste)waters.

Contribution of retinoids to chronic neurological disorders
The contribution of endocrine disruptors to neurological disorders is receiving more attention, and increasing resources are being put into funding such research [e.g. 32, 171]. In addition to known adverse teratogenic effects during development, particularly, brain and neurodegenerative conditions such as Alzheimer's and Parkinson's disease, or depression may be linked to altered retinoid signalling [172,173].
While retinoid compounds are well described as early morphogens of the central nervous system (CNS) during development, their role in postnatal development of the brain is less investigated. Retinoid signalling is implied in neural plasticity, required for the formation of new memories and for learning [151,[174][175][176], in affective disorders [177], and in ageing -namely in Alzheimer's disease and dementia [178].
To date, neurodegenerative diseases such as dementia, Parkinson's, Alzheimer's, and Huntington's disease, are the 6 th leading cause of deaths in adults in the US [179,180]. Unlike mortality due to heart disease, stroke, or HIV, deaths linked to Alzheimer's disease have more than doubled between 2000 and 2018 [179]. Even more severe is the situation regarding neurological disorders in general, which are the second leading cause of death after heart disease and the leading cause of disability worldwide [181,182]. To date, there is no cure for dementia and treatment focuses on ameliorating the symptoms of  [73, 151-153, 173, 188, 192, 193, 203, 374, 381, 391, 393-398] Rat Hippocampal volume is reduced after 3 weeks of 13cRA treatment [190]  Interestingly, the involvement of RA signalling in learning processes in vertebrates is not limited to mammals and may even play a role in invertebrate learning [197]. It has been shown that RA is critical for learning and song maturation in songbirds (zebra finches) [198,199]. Unlike vocalization in mice, song maturation involves a learning aspect similar to that in human speech [200]. Convergent signalling mechanisms of forkhead box protein P2 and RA have been hypothesized to play a role in learned vocalization in both, birds and humans [200].
Involvement of RA signalling in these key cognitive functions raises the questions of its role in neurodegenerative diseases. Age-related cognitive decline with impaired spatial learning and memory is associated with decreased RA signalling in elderly vertebrates (human and mice) [201,202]. In elderly mice with impaired cognitive function, the administration of 13cRA re-established RA signalling and hippocampal RARβ and RXRβ/γ expression and rescued the cognitive impairment [153,201,203].

Therapeutic applications
As a specific form of dementia, Alzheimer's disease (AD) is characterized by formation of amyloid-β plaques in the CNS, which leads to inflammation and subsequent neurodegeneration [179,204]. Due to a reported decrease in RA signalling in AD patients' brains, particularly in the hippocampus region responsible for the formation of memories [201,202], retinoids, especially 13cRA, are proposed for AD treatment as neuroprotectants [123,[205][206][207][208][209]. Furthermore, specific synthetic agonists of RARα/β, such as tamibarotene, that are in use for cancer treatment, have been explored for AD treatment [209], but the clinical trials seem not to have progressed since [210]. For AD, disruption of RA signalling was linked to increased amyloid-β deposition in rats [211], and RAR-agonism was effective to act both preventively and therapeutically to decrease amyloid-β-induced damage to human cell cultures in vitro and to mice in vivo [212][213][214][215][216][217][218]. Besides RARs, also RXRs are being explored for their pharmacological potential in neurodegenerative and inflammatory disease treatment, though the results are not conclusive so far [123,219].
In contrast to the recovery of learning and memory abilities in elderly mice, longer-term (6 week) administration of 13cRA at a therapeutic dose (1 mg/kg/day, i.p.) during young adulthood decreased cell proliferation in the murine hippocampus and was associated with impaired learning and memory formation [192]. The authors consider this result to be due to an insufficient growth factor supply to maintain a large differentiating neuron population, leading to premature neuronal death and longer-term decreased performance in RA-sensitive cognitive decline and increasing or maintaining quality of life [179].
While risk factors associated with dementia are mostly age, genetic predisposition (family history of dementia), or life-style related (high BMI, non-healthy diet, lack of physical and cognitive exercise) [179,182], the contribution of environmental exposure has also been considered [183][184][185].
Most observations regarding retinoid signalling and adverse health outcomes in adolescents and adults have been derived from human intervention and clinical studies together with animal in vivo modelling. Of particular relevance are studies with pharmacological application of 13cRA for the treatment of acne or cancer [177,186,187].
Retinoic acid signalling is necessary for the differentiation and speciation of cell types, particularly in neurons [82]. While the differentiation of neurons is often perceived as restricted to early developmental stages, in fact many postnatal processes, including memory and learning, are dependent on neural differentiation and speciation (also known as neural plasticity) throughout life [174,[188][189][190].

Cognitive function, memory, learning, and dementia
Postnatal disturbance of RAR/RXR signalling impairs cognitive functions, especially in the forebrain and hippocampus [191]. Indeed the hippocampus region retains high postnatal RAR expression, and so is most susceptible to RA signalling interference [192][193][194]. One potential role of RA signalling in the adult brain is the modulation of synaptic plasticity, that is required for learning and the formation of memories [173,174,188]. In a mouse model, functional expression of retinoid receptors has been shown to be critical for long-term potentiation (RARβ) and long-term depression (RARβ and RXRγ) [151]. The same study linked the decreased synaptic plasticity with a substantial performance loss in spatial learning and memory tasks in mice. While Chiang et al. [151] did not study the role of RARα, Aoto et al. [195] observed a rapid increase in synaptic strength upon treatment of primary rat neurons in vitro with 1 µM atRA or increased endogenous atRA synthesis due to decreased neuronal activity. The homeostatic modulation of synaptic strength was mediated by atRA via dendritic, i.e. non-nuclear, RARα and upregulated postsynaptic glutamate receptor 1 (GluR1) expression in a transcriptionindependent fashion [195]. Membrane-bound RARα was also shown to be involved in the differentiation of spine neurons from the hippocampus [196], suggesting a nontranscriptional role of RARs and perhaps of RA as a paracrine signalling molecule.
tasks. However, the impaired cognitive functions could also be linked to affective depression disorders, as reported in humans after extended periods of RA treatment [177].
Affective disorders -altered mood, depression, and suicide RA was first linked to altered behaviour in rats in 1986 [220]. Later a link between affective disorders and RA in humans was proposed after 13cRA was approved for medical use as a treatment of severe cystic and recalcitrant acne in 1982, leading to the inclusion of a warning on the label [221,222]. A systematic review conducted by Marqueling and Zane [223] and an almost parallel review of studies by Strahan and Raimer [224] concluded that the current data available neither confirm nor disprove the association. The latter however noted that changes in mood can be accounted to 13cRA [224]. Further case studies and reports on the involvement of retinoid exposure in affective disorders have been summarized by Bremner et al. [225].
Here, we briefly introduce mechanistic data generated from animal models, with respect to the involvement of RA signalling in the dopaminergic system, as this is of particular interest with respect to the development and manifestation of affective disorders, as well as schizophrenia.
In the late 1990s, mutation and knockout of RARβ, RXRβ, and RXRγ in mice were observed to be linked to impaired locomotion and decreased signalling via dopamine receptors 1 and 2 (D1R and D2R) [226,227]. Also, the involvement of the orphan nuclear retinoid receptor 1 (Nurr1) in the differentiation and/ or maturation of dopaminergic neurons was hypothesized [226,228]. Consecutively, the involvement of RA signalling in the development of the dopaminergic system, particularly in the expression of D2R, was confirmed in mice [229] and rats [227,230,231]. Of interest are also studies of chronic 13cRA administration in mice that better reflect an extended exposure to retinoids. Administration of therapeutic doses (1 mg/kg/day) to young adult mice over 6 weeks did not alter general locomotor activity, but increased depression-like behaviour in the forced swim test and tail suspension test [232]. A follow-up literature review to this study proposed alterations to the serotonin neurotransmitter system rather than dopamine [233]. This is in line with an extended 13cRA exposure study in rats, which affected the serotonin rather than the dopamine neurotransmitter system [187]. Interestingly, a parallel study with chronic exposure to 13cRA or atRA in rats did not confirm the observed behavioural despair (forced swim test) observed by O'Reilly et al. [232], indicating species differences in sensitivity to RA [234].
Another parallel between depression in (elderly) human and in mice was drawn rather recently, when Qi et al. [235] observed post-mortem a decrease in mRNA levels of brain-derived neurotropic factor (BDNF) and RA signalling pathway elements in the brain of depressed patients and were able to confirm this observation in mice (BDNF is a biomarker also considered for inclusion in AOPs for developmental neurotoxicity and learning impairment [236][237][238]). Additionally, they identified a RA-responsive element in the tropomyosin receptor kinase B (TrkB; receptor for BDNF) promoter region specifically targeted by RARα and thereby confirmed crosstalk between the RA and BDNF signalling pathways [235].

Schizophrenia
While the multifactorial aetiology of schizophrenia includes genetic and environmental risk factors, critical areas are early neurodevelopment, social behaviour and cognitive ability [239,240]. In fact, genetic predisposition by itself is not necessary nor sufficient for the development of schizophrenia and the developmental cascade leading to the disease should include interactions with the environment [241]. Whilst no discrete substance has been proven to cause schizophrenia, it has been hypothesized that the neurochemical processes affected by some recreational drugs play a role in the development of schizophrenia and psychoses [summarized in 241]. These processes are signal transduction via dopaminergic [242] and glutamergic synapses [243], the endocannabinoid system [244], and (neuro-)inflammation [245].
Twenty years ago retinoid signalling was postulated to be involved in the development of schizophrenia [246], and whilst further mechanistic evidence is present, it is not sufficient for confirmation. The link has been established based on a predisposition for schizophrenia in children with congenital anomalies similar to RA signalling disturbance, convergent gene loci of schizophrenia risk factors and the RA signalling cascade (esp. CYP26B1) [240,247], and the already mentioned sensitivity of the dopamine neurotransmitter system, particularly D2R, to RA interference (see "Affective disorders -altered mood, depression, and suicide" section) [outlined in 246]. Besides the dopamine system, also γ-amino butyric acid (GABA)-ergic interneurons in the prefrontal cortex, whose aberrant development is associated with neurological disorders including schizophrenia, have been shown to be sensitive to RA [248].
Although a considerable number of studies address the biologically plausible link between retinoid signalling and neurological diseases, on balance, the evidence supporting the link is currently insufficient to attribute causation (see Table 2).
Whilst the causal link between aberrant retinoid signalling and neurological disease is currently weak, the biological plausibility of the association is high: RA is a morphogen during early development and is strongly involved in shaping the CNS, including differentiation and maturation of neurons. Despite the significant role during development, the role of RA signalling in the adult or postnatal brain is less clear. Still, the conserved mechanisms of RA signalling are most likely to act also in the adult brain, though the effects may be less evident due to the multitude of parallel processes and potential influencing factors. Also, it is difficult to simulate and assess the many hues of neurological disease in animal models that are distinctly different from humans and it is not possible to assess behavioural changes in in vitro systems.
The "Cognitive function, memory, learning, and dementia"-"Schizophrenia" sections introduced several neurological conditions that share common affected personality traits due to changed connections in the CNS and altered neural plasticity. While modified RA signalling is not the single cause of adverse psychological and neurological outcomes, it is a strong candidate for connecting environmental exposure to neurological and/or neurodegenerative disease by modulating neurotransmitter systems (i.e. the dopaminergic system) and altering the base-line population of (non-) differentiated cells in the CNS. The role of the retinoid signalling pathways is especially pronounced, because interference of environmental chemicals does not have to be mediated via the molecular initiating events of the nuclear receptors (RARs and RXRs) directly, but could interfere with the endogenous retinoid homeostasis, e.g. by altering RA degradation (via CYP26 enzymes) or its biosynthesis (via ALDH and/or RALDH).

Sources of retinoids in surface water -exposure
Whilst retinoids are an intrinsic part of the diet for terrestrial animals and humans, aquatic animals in particular may be susceptible to involuntary exposure to excess retinoids at critically sensitive early-life stages [79] due to the prevalence of retinoid sources, both natural -cyanobacteria (blue-green algae) in eutrophic (fresh)water ecosystems [249][250][251][252] -and/or anthropogenic -wastewater discharge [253].
Retinoid-like activities in environmental matrices mediated via RAR or RXR can be measured by in vitro receptor transactivation assays, similar to ER and AR (see "Infobox"). In fact, ligand binding to specific nuclear receptors that leads to the transcription of target genes which, in the case of reporter assays, govern the expression of an easily detectable (e.g. luciferase) product, has become the method-of-choice for recent screening programmes targeted at uncovering endocrine activities of chemicals in a high-throughput manner [254][255][256]. The signal reflecting the extent of receptor transactivation can be quantified relative to the reference ligand (atRA for RAR, 9cRA for RXR) [257][258][259]. Detected retinoid-like activity for the different types of samples is expressed as equivalent concentration of the reference ligand that would cause the same response. These retinoic acid equivalent concentrations integrate the potential of a given mixture to activate the transcriptional response of the receptor and are more informative than targeted analyses for a limited set of compounds.

Cyanobacteria
An important source of retinoids to surface waters is cyanobacterial blooms in eutrophic freshwater ecosystems. Anthropogenic eutrophication of water bodies is driven by agricultural activities and insufficient removal of nutrients (mainly nitrogen and phosphorous) from communal wastewaters [260,261]. The ability of cyanobacteria to fix dissolved carbon dioxide (HCO 3 − ) by photosynthesis makes their occurrence independent of bioavailable carbon [262]. Together with global climate change, these are the biggest factors enhancing cyanobacterial blooms in (fresh-)water environments [260,263]. As a result, the main limiting nutrients are bioavailable inorganic nitrogen (nitrate, NO 3 − ) and phosphorous (phosphate, PO 4 3− ) [261]. These nutrients are further concentrated in long, dry warm periods in summer, that are increasing with, and exacerbated by, global climate change. Evaporation and increased abstraction from surface water bodies leads to increasing water temperatures especially in shallow surface waters, further fuelling the development of cyanobacterial blooms [264][265][266]. Greater abstraction will also be expected with the growth in human population. In addition, climate change increases the frequency and size of flooding events which, in turn, (a) increase sediment loss to surface water (which is a key mechanism via which phosphorus enters water [267]) and (b) promote resuspension of nutrientladen benthic sediment, both of which further exacerbate cyanobacterial blooms [268]. Besides being an integral part of the aquatic ecosystem, cyanobacteria produce a large variety of secondary metabolites, many of which show bioactive or even toxic properties [269, reviewed in 270]. Amongst others, cyanobacterial bloom biomass and affected waters were shown to contain retinoids, elicit retinoid-like activity in vitro, and to cause in vivo teratogenic effects in Xenopus laevis tadpoles and Danio rerio embryos, which implies relevance towards wildlife populations [79,249,251,252,[271][272][273][274][275]. Although algae contain retinoids at comparable levels in their biomass to cyanobacteria [251,271,274], it is the latter that are major contributors to retinoids in surface waters due to their proliferation. While the occurrence of cyanobacteria themselves is natural, their hazardous massive blooming events are strongly driven by human actions making it an "anthropo-natural" phenomenon.

European and Asian environmental case examples
The chemical assessment of environmentally occurring retinoids or a quantification of retinoid-like activity is a monitoring data gap. However, the few studies systematically analysing water samples reveal highly concerning levels of retinoids or their activity.
Besides occurring in cyanobacterial blooms, retinoids also enter the environment via wastewater effluents [reviewed in 253]. Humans, as well as animals, excrete retinoids most often as 4-oxo derivatives [277][278][279]. Even though retinoids are sensitive to oxidation and isomerization processes that significantly alter their bioactivity, in municipal wastewater treatment plants the treatment efficiency may not be sufficient for their complete removal and, consequently, retinoids can be released to the receiving water bodies at concentrations of up to 11.5 ng REQ × L −1 [253,271,[280][281][282][283]. Amongst the detected isomers, oxidized (i.e. 4-oxo-) derivatives of retinoic acids dominate over the parent compounds. Besides dietary excreted retinoids, pharmaceutical retinoids (tretinoin (atRA), alitretinoin (9cRA), isotretinoin (13cRA), bexarotene, and others; pharmaceutical use: cancer, acne (≥ 0.5 mg isotretinoin/kg/days), eczema treatment [222,284,285]) and cosmetically used retinoids (retinol, retinyl palmitate, retinyl acetate; cosmetic use: body lotion (≤ 0.05% retinol equivalent), hand/face cream or rinse-off products (≤ 0.3% retinol equivalent) [279,286]) can be excreted into wastewaters or washedoff after topical application. The use of retinoic acid in cosmetics is restricted within the EU and Norway [70,286,287], recommendations are provided by the European Medical Agency, with respect to oral use of the retinoid containing medicinal products in pregnancy, and also for those suffering from neuropsychiatric disorders [222]. The potential contribution routes to retinoid compounds in surface waters are also summarized in Fig. 3.

Anthropogenic and natural endocrine-active substances
The presence of unregulated endocrine-active substances (EAS), like retinoids, in waters raises the questions: to what extent do they contribute to mixture effects? Do we need to update the assessment of water safety for human consumption? To date, EAS in the environment are managed by regulating their commercial release: the manufacturer is responsible for correct labelling of the product and has to assure an acceptable risk of the active ingredient, or the preparation, to the environment. While this management scheme covers anthropogenic releases of chemicals, it cannot capture natural sources that contribute to the cumulative effects observed in the environment in situ. Hence, it does not reflect the need to tackle mixture effects in the environment directly or just before release of complex mixtures like waste water effluents into the environment [288][289][290][291].
The European Union's visionary concept of water legislation, encompassed in the Water Framework Directive (WFD, [291]), the Urban Waste Water Treatment Directive [288,292], the Nitrates Directive [293], and the Drinking Water Directive [DWD,294], aims to secure safe drinking water now and for future generations. However, the demand for water, for nutritional, recreational and agricultural purposes, increasingly challenges water supply managers and may require switching to lessfavourable water sources to provide the consumers with the desired supply [295,296].
To meet the high expectations for surface and drinking water safety, a holistic assessment at the point of abstraction for drinking water, including sub-acute and/ or longer-term effects of EAS and putative endocrine disruptors, rather than managing only anthropogenic inputs of active substances (e.g. pesticides, fertilizers, pharmaceuticals) to the environment would be beneficial for risk assessment and more reflective of the true burden of EAS exposure via drinking water [44,290,297,298].     Epidemiological data on the association of PD with vitamin A/carotenoid blood levels are insufficient to draw conclusions; recommended reporting schemes for epidemiological data need to be followed Age-related memory deficit in mice (C57BL6 Jico) due to decreased RAS associated with decreased RAR expression [153,203] Observations are clearly linked to RAS and The impact of substances such as pesticides, fertilizers, and pharmaceuticals upon the growth of cyanobacterial blooms causing increased "anthropo-natural" retinoid production is also a critical factor to include in the hazard and risk assessment process.

Discussion -implications for risk assessment
We have presented evidence here indicating that altered endogenous retinoid signalling is plausibly implicated in a variety of major public health areas of concern. In particular these include brain and neurodegenerative conditions such as Parkinson's disease, dementia, schizophrenia, and depression (see sections "Contribution of retinoids to chronic neurological disorders"-"Schizophrenia", Tables 1 and 2), as well as developmental effects. However, the available evidence causally linking these diseases with aberrant retinoid signalling is currently weak -particularly with respect to environmental exposure. Retinoids in the form of vitamin A and its precursors are essential nutritional requirements, and it is well established that too little or too much can lead to adverse health outcomes. Whilst vitamin A dietary reference values are clearly specified [70,72,97], these are advisory and on the whole are not controlled by regulatory bodies, with the exception of fortified functional/novel foods, where fortification can impact upon vulnerable populations, as for example, with infant formula [71]. Retinoid-based orally administered pharmaceuticals can also contribute to the daily exposure, while not being accounted for by nutritional reference values [222]. Critical limitations in attributing altered retinoid signalling to environmental exposure are the lack of monitoring of retinoid compounds and virtual absence of effect-based screenings for retinoid-like activity, even though monitoring reports from the Czech Republic and Asia indicate significant (anthropo-)natural sources of retinoids [249][250][251][252]271]. This data gap is aggravated by the retinoid signalling pathway not (yet) being included as a contributory pathway that can be adequately assessed by standardized test methods, as part of endocrine disruption hazard assessment and by the unknown hazards related to environmental mixtures of anthropogenic contaminants and (anthropo-)natural compounds adversely interfering with this pathway. Retinoid signalling is directly involved (via RAR) in a multitude of developmental, neurological and repair processes as well as indirectly via the universal heterodimerization partner RXR. This contributes to the non-linear intercommunication web of cause-effect relationships that are observed upon disturbed retinoid signalling and additionally allows pleiotropic effects through crosstalk with other pathways such as TR [114], or PPARβ/δ signalling [113] and steroidogenesis [7]. In addition, the substantial evidence for the teratogenicity of retinoids is usually addressed under developmental toxicity hazard assessment, rather than endocrine disruption per se (see Table 1), whilst the proposed adverse outcomes in relation to spermatogenesis and male reproduction [summarized in 13], fall under reproductive toxicity, and the biologically plausible hypothesis of involvement in the development and manifestation of neurological disease, under (developmental) neurotoxicity (see Table 2). Furthermore, the elucidation of the link between neurological disease and altered RA signalling also requires more basic investigation into the role of retinoid signalling in the brain (using experimental models and clinical investigations) together with populationbased studies -as has been done, e.g. for polychlorinated biphenyls [299,300]. This would need a characterization of the exposure to compounds with potential retinoid signalling disruptive effects -as depicted in Table 2.
In addition to hazard assessment, the characterization and quantification of environmental levels of retinoid compounds and retinoid-like activity is key for the exposure evidence base needed to assess whether they are likely to pose a risk to the environment and human health. This needs to include a consideration of exceedance of vitamin A nutritional requirements and dietary sources of exposure. (Anthropo-)natural sources of EAS such as cyanobacterial blooms often exhibit pronounced cyclic recurrent (i.e. seasonal, non-continuous) patterns [270,301]. Consequently, derived exposure limits should consider intermediate longer-term values in addition to lifetime-daily exposure, such as seasonal, monthly, or weekly exposure. This has been recently proposed and conducted by the WHO for a few selected cyanobacterial toxins [270]. In surface waters, wastewater treatment effluents are a constant contributor of anthropogenic bioactive compounds including retinoids [253]. To address transformation of pharmaceutically and cosmetically used retinoids, it is suggested that work being initiated at the OECD to develop a test guideline on anaerobic transformation of veterinary pharmaceuticals, organic compounds and biocides in liquid farmyard manure [302] could include retinoid-like monitoring endpoints. This would contribute to filling the assessment data gap of retinoid-like activity originating from farmyard manure and its fate in the environment. This could also be a tool of potential value in addressing subsequent monitoring of retinoid removal in urban wastewater effluents.
Besides recognizing the occurrence of endocrine disruptors in the environment, it is also critical to develop more accurate tools to assess their potential impact and hence any associated risk. In the case of interference with retinoid signalling, this means mainly to direct research efforts into the augmentation of already existing test guidelines and the validation of (non-animal alternative) methods for regulatory testing of retinoid signalling pathway disruption, which is already initiated at international intergovernmental levels (see "Introduction" section).
In addition to distinct test methods, AOPs are being developed with the intention of regulatory applications, to convey biologically plausible hierarchical structures of causes, effects, and outcomes from basic research to regulatory actions. It is important to refine and strengthen AOPs under development for retinoid signalling disturbance [13,303]. Besides only linking the sequential "event-train", recent efforts to define tipping points for transition between key events could make AOPs become quantitative, thus more useful for computational predictive approaches [21, 303,304]. An analogous approach to AOPs has also been taken in exposure science with aggregate exposure pathways (AEPs). They aim to summarize exposure from different sources, and integrate target site exposure, e.g. at a receptor in the tissue [305]. AEPs take into account potential environmental or metabolic transformation of a substance or cumulative effects of structurally similar substances in mixtures and are inclusive to substances of natural origin that may contribute to target site effects [305,306]. The integration of exposure and effect assessment is also called for by European partnerships to achieve the ambitious goals laid out in the WFD [307].
Substances of emerging concern often show endocrine activity and/or are candidate endocrine disruptors [308,309]. Although not intended, these substances often find their way into the environment, and, most importantly their environmental occurrence is augmented by human actions [309]. For the sustainable development of society, we need to recognize our environmental impact and try to retain or re-establish the delicate balance of maintaining and protecting landscapes and ecosystems (see Fig. 3). Only then will we be able to achieve the visionary milestones identified and articulated, e.g. by the United Nations Organization as the "Sustainable Development Goals" [310, Goal 6: "Clean water and sanitation"], by the European Commission in water-related directives (e.g. WFD [291], DWD [294]) and, most recently, by the European Green Deal, which aims at ensuring a "toxicfree environment", including a zero pollution approach and the development of an action plan regarding endocrine disruptors in the environment and circular economies [311]. With respect to mixtures in the environment, it was recently proposed to combine all EU chemicalrelated legislation, independent of the use scenario, in order to allow an inclusive mixture impact assessment [312]. A further proposal is to formulate "human health protection goals", similar to the protection goals defined in the WFD for aquatic environments, with respect to involuntary and cumulative exposure to chemicals [312].

Conclusion
Here, we have presented the (anthropo-)natural occurrence of retinoids in freshwater environments as a case study example to highlight the importance of regulatory recognition of non-EATS endocrine disruption pathways, Fig. 3 Exposure scenario to environmentally occurring retinoids, including challenges for water catchment management. Retinoid-like activity in the environment is not only associated with anthropogenic contaminants, but also from human-driven naturally occurring ("anthropo-natural") cyanobacterial blooms. N & P: nitrogen and phosphorous specifically the retinoid signalling pathway. Elaborating on diffuse and especially (anthropo-)natural sources of these teratogenic EAS, we highlight the necessity of including exposure to mixtures from different environmental media and evaluating environmental and human health impacts of compounds, irrespective of and independent to their initial use, e.g. biocide/plant protection product; environmental matrices like water or soil are indifferent to the use-case of a product.
The (anthropo-)natural occurrence and production of retinoids in water bodies in addition to anthropogenic sources suggests a human health hazard. However, due to insufficient data on environmental levels of retinoids, especially spatio-temporal screening data, an adequate risk assessment cannot be conducted to date. Future monitoring studies need to take into account both point sources such as wastewater treatment plants and diffuse (anthropo-)natural sources of EAS that include retinoids.
The retinoid signalling pathway is conserved at least across vertebrates and plays a pivotal role during prenatal development, such that its disturbance can cause teratogenic effects that range from mild malformations to lethality. Phenotypically similar developmental defects were observed in aquatic vertebrates exposed to environmental cyanobacterial bloom extracts with retinoid-like activity.
Postnatal roles of retinoids include epithelial integrity and spermatogenesis, and retinoid signalling disruption may play a role in the epidemic of neurological and neurodegenerative disease. A preliminary weight-ofevidence matrix for the association of disturbed retinoid signalling with neurological disease was presented to flag uncertainties in the experimental design or the biological link, however despite biological plausibility, the weight of evidence to date is insufficient to support the causality of retinoid signalling disturbance in neurological diseases.
Also, agonistic/antagonistic and additive actions that are not covered by the current assessment methods may occur due to the high degree of molecular crosstalk between different endocrine signalling pathways, as depicted for example for RXR.
To strengthen the retinoid relevant AOPs for regulatory applications, future toxicological studies need to further address and elucidate the toxicological tipping points from one key event to the next. Understanding the adaptive stress response in a concentration and timedependent manner is crucial to derive not only acute and chronic (i.e. life-time daily) exposure limits, but also more realistic prolonged-short time exposure limits that, for example, reflect seasonal variations in exposure scenarios as recently conducted by WHO [270], although retinoids are not currently included in this proposed approach. It may also lead to a better understanding of life-stage and gender differences in toxic effects. The development of high-throughput methods and an increasing number of validated non-animal methods will enable more rapid and efficient understanding of these differences that could ultimately contribute to safer waters in the futurefor humans and ecosystems.

Acknowledgements
Not applicable.

Disclaimer
The opinions are those of the authors' and not necessarily the organisations' involved. The European Union's Research Executive Agency cannot be held responsible for any use that may be made of the information it contains.

Funding
The research has received funding from the Czech Science Foundation, Czechia, project No. 18-15199S, the RECETOX Research Infrastructure (Projects LM2018121 and the CETOCOEN EXCELLENCE Teaming 2 project supported by European Union's Horizon 2020 (857560) and the Czech Ministry of Education, Youth and Sports (02.1.01/0.0/0.0/18_046/0015975)), the European Union's Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No. 722493 NaToxAq (BK), and grant agreement No. 825,753 ERGO (KH). MNJ was supported by Public Health England, CR did not receive specific funding for this study. The European Union's Research Executive Agency was not involved in any step of the study design.

Availability of data and materials
Not applicable.

Ethics approval and consent to participate
Not applicable.

Consent for publication
Not applicable.

Competing interests
The authors declare no conflict of interest. MNJ is on the Scientific Advisory Board of ERGO.