Stem cell toxicology: a powerful tool to assess pollution effects on human health

Environmental pollution is a global problem; the lack of comprehensive toxicological assessments may lead to increased health risks. To fully understand the health effects of pollution, it is paramount to implement fast, efficient and specific toxicity screening that relies on human models rather than on time-consuming, expensive and often inaccurate tests involving live animals. Human stem cell toxicology represents a valid alternative to traditional toxicity assays because it takes advantage of the ability of stem cells to differentiate into multiple cell types and tissues of the human body. Thus, this branch of toxicology provides a possibility to assess cellular, embryonic, developmental, reproductive and functional toxicity in vitro within a single system highly relevant to human physiology. In this review, we describe the development, performance and future perspectives of stem cell toxicology, with an emphasis on how it can meet the increasing challenges posed by environmental pollution in the modern world.


INTRODUCTION
Environmental pollutants, including chemical and biological contaminants in air, water, food, soil, and radiation can have negative effects on human health. Economic globalization, which increases the interdependence of national economies across the borders through the exchange of resources and production, makes environmental pollution a common problem not restricted to a single country or city. Thus, because of globalization, the environmental problems of China are spread to other countries and vice versa [1,2]. However, China faces greater and more complex environmental challenges than other countries owing to its rapid economic development over the past few decades [1]. In addition, China is now experiencing an important period of socioeconomic transition characterized by an increasing demand for the assessment of health risks caused by industrial chemical waste, drugs, pesticides, food additives, cosmetics and atmospheric fine particles. To cope with this problem on a global scale, we urgently need to implement novel high-throughput and high-sensitivity systems that can provide rapid evaluation of the toxicity of environmental pollutants.
Currently, we still heavily rely on using live animals for toxicity testing, which is time-consuming, resource-intensive and raises ethical problems [3]. More importantly, the theory of 'high fidelity fallacy' postulated by Russell and Burch in 1959 has already warned us about inter-specific variations that make toxicity assays based on experimental animals not always translatable to human health [4]. In fact, there are numerous examples of drugs that passed animal testing but failed during clinical trials. Therefore, the 3Rs (Replacement, Reduction and Refinement) as the principles of alternative toxicology mostly based on in vitro experiments [4] may be more important today than ever before [3,5]. The accepted view is that, by implementing comprehensive in vitro tests based on human biology to identify relevant toxicity mechanisms at the cellular and molecular levels, we would eliminate the necessity for whole-animal testing and still provide adequate environmental-and health-friendly decision-making in the future [3].
The in vitro toxicological assays using human models are primarily based on cell cultures and have several limitations. For example, immortalized or cancer cell lines widely used for toxicity screening because of fast growth and expansion may no longer be representative of original cells because of accumulating mutations or altered cell functions. On the other hand, primary cells directly derived from human tissues are either impossible or difficult to obtain and the procedures are invasive; in addition, primary cells always need standardization prior to use and have limited ability to grow and proliferate in culture. Collectively, these issues can significantly limit the generation of the data, or affect their interpretation [6]. Most importantly, these in vitro assays are generally based on a single cell type, and cannot provide the information on toxicological responses at the tissue or whole-organism levels [3].
The emergence and development of stem cell biology have enlightened the enthusiasm of toxicologists. It is thought that contemporary stem cell technologies applied to the analysis of potential hazardous impacts of pollutants on human health can revolutionize the in vitro toxicology. Our group was the first to clearly define the term 'stem cell toxicology' to characterize a new promising trend in in vitro toxicology [7]. In this review, we describe the application and emphasize the vast prospects of stem cell toxicology, building upon and complementing other recent and excellent reviews, a few of which are listed here [8][9][10][11]. In addition, although embryonal carcinoma and cancer stem cells have been extensively studied, here we only focus on normal stem cells because they are more representative of microenvironments in vivo.

HISTORY OF STEM CELL BIOLOGY
The precise definition of stem cells remains a matter of debate. A commonly accepted one is offered by Douglas Melton [12]: 'A working definition of a stem cell line is a clonal, self-renewing cell population that is multipotent and thus can generate several differentiated cell types.' Taking into consideration the histological origin and differentiation potential of stem cells, we can roughly divide them into pluripotent stem cells (PSCs) and multipotent somatic stem cells (SSCs).

PSCs
Here, we describe the development of PSC biology focusing on the achievements that facilitated the emergence of stem cell toxicology (Fig. 1, top). The first evidence that PSCs have the capacity to differ-entiate into specialized cell types came from studying mouse embryonic carcinoma cells (malignant counterparts of embryonic stem cells) isolated from teratocarcinomas: tumors comprising cells from several differentiated tissues (reviewed in [13]). An important milestone in stem cell biology was set in 1981 by Martin Evans (Nobel Prize in Physiology or Medicine, 2007) et al., who were the first to establish an in vitro culture of non-malignant pluripotent cells from mouse embryos [14], which were later designated mouse embryonic stem cells (mESCs) by Gail Martin [15]. Later, the origin of mESCs was clearly proved to be the inner cell mass (ICM) of the pre-implantation blastocyst [16]. Interestingly, when ESCs were maintained in suspension, they formed small aggregates called embryoid bodies. During this step, cell types of all three germ layers (ectoderm, mesoderm and endoderm) differentiated and interacted with each other to produce different tissue-like structures. Pluripotent mESCs were also demonstrated to produce chimeric mice with germline transmission after blastocyst injection [17]. At that time, the maintenance of mESCs in vitro strictly required the presence of a feeder layer formed by fibroblasts; however, the introduction of leukemia inhibitory factor (LIF) in 1988 was a major contribution to the success of mESC culture because LIF addition to the medium prevented mESC differentiation and promoted their growth under feeder-free conditions, while LIF withdrawal resulted in EB formation [18]. After that, studies have focused on the identification of molecular mechanisms regulating mESC pluripotency, discovering Oct3/4 (POU5F1) [19,20] and other transcription factors such as Nanog, Sox2, Sall4, Esrrb, Tbx3, Klf4, Stella and Rex1 [21,22], which are essential for self-renewal and developmental potential of mESCs. Moreover, efforts have been directed to the establishment of protocols on ESC differentiation to different types of cells and tissues.
In 2006, Shinya Yamanaka and colleagues made a groundbreaking discovery in stem cell biology by reprogramming adult mouse fibroblasts into induced PSCs (iPSCs) through retroviral transduction of four transcription factors (Oct4, Sox2, c-Myc and Klf4) [23]. The generated iPSCs were virtually identical to ESCs but were of adult rather than embryonic origin. In that study, the authors combined the milestone achievements of John Gurdon on nuclear reprogramming (transplantation of somatic amphibian nuclei into oocytes) [24], and the important study demonstrated the possibility to convert one type of differentiated cells (fibroblasts) into another (muscle cells) through overexpression of the transcription factor MyoD [25]. In the followup investigation, Yamanaka's and other groups demonstrated that the gene expression, DNA methylation and chromatin status of murine iPSCs were similar to those of mESCs, and that iPSCs could produce competent germline chimeras by blastocyst injection [26,27]. In 2009, the laboratories of Qi Zhou and Shaorong Gao demonstrated that iPSCs could generate full-term mice via tetraploid blastocyst complementation, which indisputably established iPSCs as fully pluripotent cells [28,29] with a potential to substitute ESCs in all applications. For their breakthrough achievements in nuclear reprogramming, Shinya Yamanaka and John Gurdon were awarded the Nobel Prize in 2012.
The generation of human ESCs (hESCs) did not rapidly follow that of mESCs, and it took 17 years before James Thomson and his group [30] first obtained hESCs in 1998. In contrast to mESCs, hESCs do not respond to LIF, but require fibroblast growth factor 2 (FGF2) and transforming growth factor β1 (TGF-β1) or Activin A for their self-renewal [31,32], and are also more difficult to culture in vitro. Still, hESCs and mESCs are similar in terms of virtual immortality, marker gene expression and the ability to differentiate into all three primary germ layers. However, it is not possible to assess hESC pluripotency by blastocyst injection because of ethical and legal issues. In 2007, a year after the publication of their seminal article on mouse iPSCs, Yamanaka and his group reported that human somatic cells such as adult dermal fibroblasts could also be efficiently reprogrammed into human iPSCs (hiP-SCs) by retroviral transduction of Oct4, Sox2, Klf4 and c-Myc [33]. Later the same year, James Thomson and colleagues showed that lentiviral transduction of a different set of transcription factors, including Oct4, Sox2, Nanog and Lin28, was sufficient to allow iPSC generation from human fetal and adult fibroblasts [34]. These findings were truly remarkable, as they opened a way to patient-specific regenerative medicine and, thus, can be qualified as one of the most significant events in the development of stem cell biology [35].
Also in 2007, a new cell line derived from the mouse epiblast (at a later developmental stage as compared to mESCs) and designated postimplantation epiblast-derived stem cells (EpiSCs) was demonstrated to share defining features with hESCs [36,37]. In fact, until that year, the differences between human and mouse ESCs/iPSCs were simply attributed to unknown species-specific genetic traits. EpiSCs are capable of differentiating into cells of all three germ layers in vitro or in teratoma assays, demonstrating pluripotency; however, they are inefficient in yielding chimeric animals once injected into pre-implantation epiblasts. For ethical reasons, no attempts have been made to obtain human equivalents of EpiSCs. In 2009, the two kinds of pluripotency were defined as naïve (mESCs) and primed (mEpiSCs and hESCs) [38]. Ground state naïve pluripotency is achieved in the pre-implantation epiblast and refers to a completed unrestricted cell population able to generate all the embryonic lineages. Upon implantation, the epiblast becomes 'primed' for lineage specification and commitment driven by signals from the extraembryonic tissues. Nevertheless, it is important to keep in mind that hESCs are not equivalent to mEpiSCs, which can be considered to be relatively less primed (reviewed in [39]). In 2010, hESCs and hiPSCs were proved to be maintained in a unique mESC-like pluripotent state by ectopic expression of Oct4, Klf4 and Klf2, and cultured in the presence of LIF and inhibitors of glycogen synthase kinase 3β (GSK3β) and mitogen-activated protein kinase (ERK1/2) (LIF/2i conditions) [40]. Since then, many studies have demonstrated that hESCs and hiPSCs could achieve naïve pluripotency under transgene-dependent or -independent conditions, although not as efficiently as mESCs or human ICM [39]. However, because of the limitations in performing chimeric analysis in humans, the accurate classification of the pluripotent state of isolated naïve hPSCs remains debatable.
Gametogenesis is also an import feature of the differentiation ability of PSCs. In 2003, two studies provided the first evidence that mESCs could potentially differentiate in vitro into germ cells, including oocytes and spermatocytes [41,42]. Yao et al. 433 Primordial germ cells (PGCs) are another important type of pluripotent cells, which can generate germ cells, eggs and sperm. PGC-like cells can also be induced in vitro from mESCs or iPSCs through epiblast-like cells that are highly similar to pregastrulating epiblasts but distinct from EpiSCs [43].

REVIEW
Another important study demonstrated that naïve hESCs and hiPSCs could be converted to PGC-like cells [44]. These findings are very useful for stem cell toxicology because PGC-like cells and germ cells could be employed in reproductive toxicity assays in vitro described below.
In conclusion, the basic characteristics of PSCs include: indefinite self-renewal under well-defined culture conditions, differentiation into cells of the three germ layers in vitro and in vivo, clonogenicity, a normal karyotype, and the ability to withstand freezing and thawing. Since a clear definition of pluripotency in vivo is still under debate, we determine it according to the species of origin and the level of pluripotency demonstrated in vitro. Thus, in animals, naïve PSCs can be rigorously identified by observing whether blastocyst-injected cells can contribute to all somatic lineages and show germline chimerism. In addition, both naïve and primed PSCs should be able to form teratomas in vivo, which contain differentiated cells from the three germ layers. However, in the case of hPSCs, their in vivo pluripotency can only be assessed by teratoma assays because of ethical problems.

Multipotent SSCs
During the gastrulation stage of embryogenesis, pluripotent cells in the ICM are reorganized into the germ layers that eventually produce all the tissues of the body and, thus, become restricted in their potential to differentiate into all lineages. These new cells are less plastic and have the ability to generate a limited range of cells mostly within one or a few specific tissue types, so they are multipotent; some of them persist in adults and are called SSCs [45]. The existence of SSCs in adult tissues was first demonstrated in the hematopoietic system in 1960-1961 by a seminal work of McCulloch and Till, showing that cells from murine bone marrow could give rise to multilineage descendants while retaining the ability to self-renew (Fig. 1, bottom) [46,47]. Therefore, SSCs are also commonly referred to as adult stem cells; however, they can also be found in fetal tissues such as the umbilical cord and placenta, and are designated fetal stem cells. Moreover, SSCs are classified, according to their histological origin, into mesenchymal stem cells (MSCs), neural stem cells (NSCs), hematopoietic stem cells (HSCs), skin stem cells, etc. Here, although we recognize the im-portance of other SSCs, due to space limitation, we will only focus on MSCs (Fig. 1, bottom) as an example of SSCs.
Alexander Friedenstein is considered the first to put forward the concept of prototype MSCs. In 1968, he and his colleagues isolated, for the first time, adherent, fibroblast-like, colony-forming cells from mouse bone marrow with a high replication capacity in vitro, and demonstrated that those cells were able to differentiate into osteoblasts and reconstitute the hematopoietic microenvironment after subcutaneous transplantation [48]. When similar fibroblast colony-forming cells were detected in human bone marrow aspirates in 1980 [49,50] and showed a potential for differentiating into osteogenic, adipogenic, chondrogenic and myogenic mesenchymal lineages in vitro (reviewed in [51]), the term 'mesenchymal stem cells' coined by Caplan in 1991 [52] has been applied. By now, MSCs have been isolated from almost all fetal and post-natal tissues, including fat, dental pulp, periodontal ligament, tendon, umbilical cord, skin, placenta, amniotic fluid, muscle, liver and brain (reviewed in [53]). Multipotency has been demonstrated in some adult MSC lines generated from bone marrow [54,55] and adipose tissues [56]. However, because of their heterogeneity, not all of these plastic-adherent cells have comparable self-renewal and differentiation potential in vivo. Therefore, in 2005, the International Society for Cellular Therapy recommended the term Multipotent Mesenchymal Stromal Cells for the fibroblast-like plastic-adherent cells [57] and, in 2006, issued the minimal criteria to define human MSCs: adherence to plastic under standard culture conditions; expression of CD105, CD73 and CD90; no expression of CD45, CD34, CD14 or CD11b, CD79a or CD19, and HLA-DR; and ability to differentiate into osteoblasts, adipocytes and chondroblasts in vitro [58]. Although there is still no agreement on the use of the terms 'stem' and 'stromal' [51,59], it did not affect the increasing application of MSCs in both basic research and clinical trials. In fact, MSCs can not only function as multipotent cell progenitors, but also modulate immune reactivity, support hematopoiesis and regulate the release of trophic factors in response to injury [51,53,59]. As of May 2016, a search of the website clinicaltrials.gov using 'mesenchymal stem cells' would yield a list of over 600 clinical trials covering a surprisingly vast array of pathological conditions.
Other commonly studied types of SSCs are also very important in the history of stem cell biology. For example, NSCs successfully isolated from both developing and adult brains can self-renew in vitro [60] and differentiate into the three major central nervous system lineages: neurons [61], astrocytes and REVIEW oligodendrocytes [62]. Another example is HSCs, which are derived from bone marrow and undergo the process of hematopoiesis to generate all types of blood cells. However, they are not discussed here because of limited space.

PSCs in toxicology
In 1991, 10 years after mESC cultures were established in vitro, Horst Spielmann et al. [63] (all the references from now on, in the development of stem cell toxicology section, are also included in Table 1), were the first to apply PSCs in toxicology by measuring cytotoxic effects of potential teratogens in vitro. The same group also developed an in vitro embryotoxicity test using differentiating mESCs [64,65]. Similarly, another assay using proliferating and spontaneously differentiating mESCs (induced by LIF removal) was reported in 1994 and 1996, for testing the teratogenic activity of 25 compounds; teratogens were identified if their IC 50 for differentiating cells was less than that for proliferating ESCs [66,67]. Although the assay was not very accurate and had poor prediction rate, at that time, it could potentially replace the traditional micromass test used to assess developmental toxicity. However, it was not developed any further.
In 1995-2004, the European Centre for the Validation of Alternative Methods (ECVAM) proposed, defined and validated the Embryonic Stem Cell Test (EST) [68][69][70][71][72][73] pioneered by Horst Spielmann as an in vitro assay for embryotoxicity [74,75]. In the EST, the effects of tested chemicals were assessed according to a prediction model based on three endpoints: inhibition of ES differentiation into the contracting myocardium analysed by EB formation (hanging drop technique) and cytotoxicity for ESCs and mouse 3T3 fibroblasts (control differentiated cells). The correlation between the EST and in vivo data (78%) was similar to that observed for other in vitro tests such as the micromass assay (70%) and the post-implantation rat whole-embryo culture assay (80%). However, the advantage of the EST was that it did not use embryos or primary embryonic tissues isolated from pregnant animals [74]; therefore, the establishment of the EST is considered a milestone in the history of stem cell toxicology. The EST has been evaluated for drug screening [76,77], developmental toxicity testing of panels of related compounds [78] and for embryotoxicity assessment of nanomaterials [79]. Nevertheless, it had several weaknesses such as limited differentiation assessment based solely on the detection of sponta-neous myocardial differentiation (which in turn was subjectively evaluated by microscopic observation of beating areas), and the absence of metabolic analysis [80]. Moreover, in another EST evaluation study sponsored by the ECVAM and ReProTect, only 2 out of 13 previously untested compounds were classified correctly, highlighting the limitations of the EST prediction model [81].
To improve the EST performance, several modifications were proposed. Thus, instead of the visual microscopic evaluation of beating cardiomyocytes, more quantitative molecular endpoints were introduced, such as analysis of gene and protein expression of myocardial markers by reverse transcription quantitative PCR (RT-qPCR) and fluorescenceactivated cell sorting (FACS) [82][83][84][85][86][87]. In addition, toxicity was assessed based on the differentiation to other cell types, including neuronal, skeletal muscle, pancreatic and osteogenic lineages [88][89][90][91]; for this, genetically engineered ESCs were introduced, which expressed green fluorescent protein (GFP)-labeled lineage-specific markers [92][93][94] or cardiac/neuralspecific luciferase reporters [95]. Based on these improvements, a molecular multiple-endpoint EST was developed in 2004, which incorporated quantitative analysis by RT-qPCR and assessment of the differentiation to osteogenic, chondrogenic and neural cells in addition to the traditional cardiomyocyte differentiation [96]. Other improvements of the typical EST are mentioned in Table 1 [97][98][99]. An official ECVAM validation of the EST was published in 2011 in the journal Nature Protocol [100]. Moreover, ReProTect Inc. conducted a feasibility study with a modified EST method and showed correct prediction for 9 of the 10 blinded chemicals [101]. In another major global drug company, Roche, the EST has been routinely employed to detect potential teratogenic/embryotoxic liabilities during optimization of early pharmaceutical drug candidates [102].
Starting from the introduction of the EST, most of the developmental ESC-based toxicity tests analysed cell differentiation by EB formation. However, several later studies reported the assessment of ESC  64, 65 1993, 1994 They utilized differentiating mESCs and demonstrated the inhibitory influence of the teratogen retinoic acid on blood and myocardial cell development in contrast to a stimulating effect on the nerve and skeletal muscle cell development Newall et al. 66, 67 1994, 1996 They used mESCs to design their stem cell tests. 25  They compared the neuronal differentiation propensity of hESCs and hiPSCs and demonstrated that a CREB (cAMP-responsive element-binding protein) pathway inhibition could be involved in cellular and molecular neurotoxic effects, and qualified the use of hiPSC-derived neuronal model for studying chemical-induced neurotoxicity resulting from pathway perturbations Waldmann et al. 152 2014 In an attempt to define universal rules for neurotoxicity tests, they tested the dose-dependent transcriptome deviations in an assay that recapitulated the development of hESCs into neuroectoderm, and suggested the use of the highest non-cytotoxic drug concentrations for gene array toxicogenomics studies. They argued that higher concentrations would yield wrong information on the mode of action of each chemical, and lower drug levels would result in low gene expression changes which might be difficult to detect Schwartz et al. 153 2015 They cultured 3D neural constructs with hESC-derived neural progenitor cells, endothelial cells, MSCs and microglia/macrophage precursors on chemically defined polyethylene glycol hydrogels in serum-free medium to model cellular interactions within the developing brain. They also used linear support vector machines to construct a reliable predictive model from RNA-Seq data acquired from 240 neural constructs treated with 34 toxic and 26 nontoxic chemicals Senut et al. 154 2016 They assessed the effects, of gold nanoparticles (AuNPs) on the viability, pluripotency, neuronal differentiation ability and DNA methylation status of hESCs. They identified a type of AuNPs highly toxic to hESCs and demonstrated the potential of hESCs in predicting nanotoxicity and characterizing nanoparticle ability to alter DNA methylation and hydroxymethylation patterns in the cells differentiation in cell monolayers, which could complement or replace that based on EBs. In fact, in 2011, using a simplified EST, the effects of chemicals were analysed by observing spontaneous differentiation of D3 ESCs cells in monolayers over 5 days and which provided faster, technically simplified analysis, while maintaining the same prediction rate [123]. Later that year, Barrier and colleagues developed the adherent cell differentiation and cytotoxicity (ACDC) assay using In-Cell Western analysis for mESC differentiation into cardiomyocyte lineage [124], which also did not rely on EB formation. The efficiency of the ACDC assay was confirmed by evaluating 309 environmental pollutants [125]. Further toxicity studies employing mESC monolayers [126][127][128][129][130][131][132][133] are described in Table 1.
An important limitation of toxicological analysis performed using mouse stem cells is that the results may not be directly applicable to humans because of inter-species variations. Therefore, toxicity testing based on hESCs is likely to generate more clinically relevant data. The first application of hESCs in toxicology was described in 2007 by Cezar et al. [134], who demonstrated that mass spectrometry-based small-molecule metabolite profiling of hESCs and hESC-derived neural precursors subjected to chemical stress could help elucidate molecular mechanisms of toxicity. Later, a more reliable developmental toxicity model based on the hESC system was developed [135,136]. By performing metabolomic profiling, they identified biomarkers of developmental toxicity and demonstrated that their model could correctly predict the teratogenicity of 7 out of 8 drugs (88%) and of 9 out of 11 environmental toxicants (83%). Further examples of metabolomics application in hESC-based toxicological assays [137,138] are described in Table 1. Another interesting study, published in 2008 [139], utilized hESC-differentiated into fibroblastic progenies for in vitro toxicology screening and showed that hESC-derived fibroblasts exhibited a more sensitive dose-response to mitomycin C compared to human lung fibroblast L929 cells. The same year, research from the ECVAM followed up on the successfully validated mouse EST and established an equivalent assay based on hESCs and human fibroblasts. By assessing cytotoxicity (IC 50 values) and lineage marker expression (RT-qPCR) during EB differentiation, they demonstrated a predictability similar to that of the mouse EST [140]. The ability of hESCs to differentiate via EB formation was also shown by other research [141][142][143][144] (Table 1).
In several studies, hESCs were investigated with the aim to specifically assess toxic effects on neural differentiation. In 2009, a large-scale five-year European Union project titled Embryonic Stem cell-based Novel Alternative Testing Strategies (ESNAT) was launched to establish novel prenatal developmental toxicity tests based on hESCs (http://www.esnats.eu). Within this project, in 2011, a protocol was designed based on the generation of neural rosettes from hESCs, with the emphasis on early neural development, for the detection of neural toxicity [145]. In 2013, several groups within the ESNAT project employed transcriptomics for the analysis of developmental neurotoxicity (DNT) and reproductive toxicity (RT) using hESCs; as a result, they were able to classify human DNT/RT toxicants on the basis of hESC transcriptome profiles [146]. Several other studies mentioned in Table 1 [147][148][149][150][151][152] also used hESCs to assess environmental neurotoxicity, and one of them employed epigenetic profiling to dissect the underlying molecular mechanisms [149]. In 2015, the group of James Thomson who first generated hESCs also published a toxicological study in which cellular interactions within the developing brain were modeled by culturing hESC-derived neural progenitor cells, endothelial cells, MSCs and microglia/macrophage precursors on chemically defined polyethylene glycol hydrogels in serum-free medium. As a result, they constructed, using linear support vector machines, a reliable predictive model based on RNA-Seq data acquired from 240 neural constructs treated with 34 toxic and 26 nontoxic chemicals [153]. In 2016, at the time of writing this review, four additional reports on using hPSCs for neurotoxicity testing were published [154][155][156][157] (Table 1).
Another popular direction in the development of human stem cell toxicology was the design of assays on cardiotoxicity [158][159][160][161][162][163][164][165][166] and hepatotoxicity [167][168][169][170][171][172] (Table 1). Thus, in 2013, it was reported that the approach based on hESCand hiPSC-derived hepatocytes could predict toxicity in a way comparable to that of standard hepatotoxicity assays, representing a major advancement in the field [168]. Two other studies addressed toxic effects of bisphenol A (BPA), a chemical commonly used to harden plastics, on the reproductive systems using hESCs differentiated into mammary epithelial cells in 3D conditions [173] and human prostate organoids [174]. Another study published in 2014 used three-day monolayers of mesendoderm-differentiated hESCs to assess teratogenicity by the reduction in nuclear translocation of the transcription factor SOX17; the method named the hPSC test was validated by high-throughput screening of 71 drug-like compounds, 15 environmental toxicants and 300 kinase inhibitors [175].

SSCs in toxicology
The application of SSCs in toxicology can be first traced to 1982, when Robert M. Pratt and his colleagues prescreened environmental teratogens using cultured mesenchymal cells from the human embryonic palate [176]. After that, numerous studies were conducted on SSC-based toxicity assays; here, we will focus on those studies that used MSCs and NSCs. However, we recognize that there are plenty of toxicological studies with other SSCs, such as HSCs and so on. All these works are important in stem cell toxicology. Nevertheless, due to space constrictions, we cannot address them thoroughly in this review.
In 2011, Cao and coworkers were the first to evaluate the applicability of hMSCs for in vitro cytotoxicity testing to correctly assess LD 50 values and predict the hazard category of the tested chemicals according to the globally harmonized system of classification (GHS) [177]. Their findings indicated that hMSCs provided a more accurate modeling of in vivo conditions compared to the validated 3T3 cell test and Normal Human Keratinocyte/Neutral Red Uptake methods. In 2012, Akhavan et al. [178] demonstrated that low concentrations (0.1 mg/ml) of reduced graphene oxide nanoplatelets exerted genotoxic effects on hMSCs due to DNA fragmentation and chromosomal aberrations. Moreover, in 2015, Strong and colleagues exposed hMSCs to the endocrine-disrupting chemical dichlorodiphenyltrichloroethane (DDT) and revealed profound alterations in self-renewal, proliferation, differentiation (adipogenesis and osteogenesis) and gene expression, which could partially explain homeostatic imbalance and increased cancer incidence among the affected individuals [179].
In 2006, Tamm with coworkers demonstrated that neural stem cell line C17.2 and primary embryonic cortical stem cells were highly sensitive to methylmercury (MeHg) as evidenced by the effects on survival and differentiation, offering new perspectives for evaluating the biological consequences of MeHg exposure at low levels [180]. In 2009, Buzanska et al. [181] established a human neural stem cell line from umbilical cord blood (HUCB-NSC) and used it to test developmental neurotoxicity by analysing such parameters as cell proliferation, apoptosis, and neuronal and glial differentiation.

MAJOR FEATURES OF STEM CELL TOXICOLOGY
The studies described above mentioned many applications of stem cells in toxicology. However, in none of them was the term 'stem cell toxicology' used. Our group was the first to clearly define, in 2015, 'stem cell toxicology' as a new branch of toxicology [7]. In this section, we will then describe the major features of stem cell toxicology in more detail.

Pluripotent stem cell toxicology
Pluripotent ESCs and iPSCs are capable of differentiating in vitro into virtually all the cell types of the adult organism, including germ cells (Fig. 2). This property defines the very core of stem cell toxicology and explains why only stem cells offer such a great potential in toxicity testing compared not only to other cell types used in vitro, but also to experimental animals. Even a simple cytotoxicity assay can be more informative if performed with stem cells because of their pluripotency, which provides not only higher sensitivity compared to somatic cells, but also enables the assessment of harmful developmental effects. PSCs can mimic the early stages of embryogenesis in vitro by forming EBs which, under differentiating conditions, give rise to the three primary germ layers and cell lineages and, thus, can be used to evaluate embryotoxicity or teratogenicity of environmental pollutants (Fig. 3). This is particular relevant for testing deleterious effects on human embryogenesis that cannot be reliably investigated using other experimental models.
PSC-formed EBs spontaneously differentiate into multiple cell lineages at the same time, which limits the sensitivity and specificity of toxicological assays, especially when subtle toxic effects on a specific tissue are masked by stronger responses of another tissue. To overcome these problems, protocols have been developed to promote preferential differentiation of EBs to a single germ layer and then to particular progenitor or somatic cells, which enables assessing tissues-specific toxic effects, such as neurotoxicity, cardiotoxicity, hepatotoxicity, etc. (Fig. 2). Many PSC differentiation protocols are based on monolayer conditions that facilitate the performance of developmental toxicity tests and provide faster data collection. Therefore, although differentiation assays conducted in cell monolayers do not reproduce in vivo 3D conditions, they are very useful for quick preliminary screening of toxic compounds.
In addition to the analysis of developmental toxicity, stem cell toxicology provides functional assessment of PSC-differentiated cells in such cases when pollutants do not influence embryonic development and lineage commitment, but rather affect subsequent functional performance of differentiated tissues. In this respect, stem cell toxicology presents an advantage of analysing specific cell types  without resorting to complicated and often invasive procedures of isolation from live tissues, which could be even practically impossible. Further applications of PSC toxicology include RT assessments because PSCs can generate in vitro PGCs as well as potentially terminally differentiated gametes and, thus, can be used to assess the impact of environmental contaminants on reproduction-related parameters (Fig. 2).

Somatic stem cell toxicology
Contrary to ESCs, SSCs cannot be used for teratogenic and embryotoxicity assays. However, SSCs can still self-renew and differentiate into somatic cells during infant and adolescent periods and, therefore, can be applied to the evaluation of environmental effects on the post-natal development into the adult organism. In adult tissues, SSCs are retained in a quiescent state until triggered to regenerate damaged cells/tissues through cycles of self-renewal and differentiation [182,183]. With physiological aging, tissue homeostasis is progressively disrupted and the ability of SSCs to repair injured terminally differentiated cells gradually declines [183]. Therefore, environmental pollutants can induce irreversible tissue damage that cannot be adequately repaired by SSC differentiation or directly target SSCs, causing their exhaustion and eventual premature aging and/or pathological conditions, including cancer [184]. Thus, primary tissue-derived or PSC-derived SSCs can be used for the in vitro assessment of harmful environmental effects on the development of infants and adolescents into adults (Fig. 4). SSC-based toxicology can also include the assays specifically designed to determine toxic effects of pollutants during tissue regeneration after injury or degenerative diseases, and assess the effects on stem cell exhaustion and aging (Fig. 4).

FUTURE PERSPECTIVES OF STEM CELL TOXICOLOGY
Toxicology is a field that takes advantage of the development and technical innovations in many disciplines, including biology, chemistry, bioinformatics and engineering [185]. This is particularly true for stem cell toxicology because it has just recently emerged. In fact, for instance, stem cell toxicology relies on the constantly developing field of stem cell biology, especially regarding the further investigation of the molecular mechanisms underlying differentiation potential of stem cells to any cell type of the body, including neuronal cells, cardiomyocytes, adipocytes, osteoblasts and hepatocytes. Although  Figure 4. Special features of somatic stem cell toxicology. Multipotent somatic stem cells (SSCs) present in many adult tissues are generally quiescent in vivo but, under certain conditions, such as during tissue regeneration after injury, can start proliferating to maintain the number of stem cells and differentiating to replace damaged cells in the tissue. Pollutants accumulating in the body can negatively affect these processes, causing stem cell aging and exhaustion, which ultimately weaken tissue repair and may cause degenerative diseases. SSC-based toxicology enables testing the effects of toxicants on SSC self-renewal and differentiation in vitro. a number of organ toxicity assays have been developed based on PSC differentiation protocols, further research should concentrate not only on the design of additional differentiation methods, but also on the improvement of established procedures to make them more reproducible, efficient, and less costly and labor-intensive. Those differentiation protocols should also include information about which stages we can freeze cells, to make stocks and subsequently thaw them. This would save time if toxicity assays need to be performed only at late stages of differentiation and there is no requirement to start from undifferentiated stem cells. An urgent need is the development of human stem cell toxicology, which is less advanced compared to the murine system, for several reasons. First, in vitro cultures of hESCs were established much later than those of mPSCs and, because of species-specific differences, the knowledge about the molecular mechanisms functioning in mESCs could not be directly applied to hESCs. Therefore, up to now, there are fewer standardized differentiation procedures available for hESCs than for mESCs. Second, it is more technically challenging to culture and differentiate hESCs, which is also a factor delaying the development of human stem cell toxicology.
A significant aspect of stem cell toxicology is the analysis of RT. ESCs can be first differentiated to PGCs and then to eggs and sperm, thus enabling the in vitro evaluation of toxic effects on the final commitment and function of germ cells. However, the development of differentiation procedures that would yield functional germ cells has been challenging and, up to now, a reliable protocol for the generation of mature eggs and sperm from ESCs in vitro has not been established, although mouse and human ESCs have been successfully differentiated into primordial germ-like cells in vitro. Consequently, stem cell-based RT assays can assess the effects on the development of PGCs, but not on that of functional gametes. Nevertheless, recent technological advancements indicate that the in vitro production of terminally differentiated eggs and sperm from stem cells is a perspective of the nearest future.
In addition, iPSCs could also be used in toxicology. This would avoid the ethical issues associated with human ESCs; however, iPSCs are not identical to ESCs in that they retain some epigenetic memory of the cell type of origin. Nevertheless, they may suffice for stem cell toxicology applications, provided more than one iPSC line, preferably originating from different somatic cell types and reprogrammed with different techniques, is used. Another advantage of iPSCs compared to ESCs is that a variety of iPSC lines representing different ethnic, clinical and environmental backgrounds are available, indicating a possibility for diversified toxicity assessment depending on the genetic and/or pathological conditions of the population. To accomplish this task, iPSC banks that would provide a panel of standardized iPSC lines corresponding to a specific toxicology assay or risk assessment need to be established.
For comprehensive representation of different in vivo microenvironments, 3D cell cultures are more suitable than monolayers, which is particularly relevant when differentiation of stem cells into tissues and/or organs is attempted. Current stem cell differentiation procedures are generally performed in monolayer settings, except when EBs are generated. Even in EB-based protocols, 3D conditions are only used in the early stages of differentiation when EBs are formed. The application of 3D cultures is especially beneficial in stem cell toxicology, as mentioned above, as it facilitates obtaining reliable data on developmental and functional effects of pollutants on early embryogenesis. However, the application of 3D stem cell-derived culture systems, although very promising, is just at the initial stage [186][187][188]. Successful reconstruction of organoids/organs ex vivo depends on careful selection of supporting matrices, either synthetic or derived from decellularized organs, and requires reproducibility in concerted cell assembly on scaffolds, which can only be achieved with automated bio-printing systems. Advances in 3D scaffold design and manufacturing, as well as bioprinting techniques, are paramount for the replacement of in vivo assays in stem cell toxicology. An ideal REVIEW Yao et al. 445 situation would be simultaneous reconstruction of several human organs to mimic the whole organism during toxicity tests. This concept has been brought to life in the so-called 'organ-on-a-chip' technology that uses different miniature organs put together on the same chip and connected by an artificial vascular system providing nutrients and conducting metabolic signals (reviewed in [189]). This technology would allow more comprehensive toxicity evaluation by examining adverse effects on several organs at the same time. Nevertheless, microorgans described above cannot reliably represent real-sized organic systems; therefore, future efforts should be invested into organ-assembling technology based on the differentiation and maturation of whole-organ scaffolds.
An important technological revolution in toxicology was promoted by the completion of the Human Genome Project and recent advances in genome sequencing, transcriptomics, proteomics, metabolomics and global epigenetics, which allowed accumulation of a tremendous amount of relevant biological information carrying enormous potential for toxicity analyses, predictions and risk assessments. Genetic variations could explain why certain individuals and/or populations are more sensitive to a particular pollutant, while global transcriptomics, epigenetic signatures, protein expression analysis and metabolic profiling would help to identify toxicity mechanisms, screen potential toxicants and monitor human exposure to pollutants [34].
The technological revolution described above was accompanied by the development of cuttingedge informatics tools for comprehensive data analyses. Stem cell toxicology would greatly benefit from the accumulation and analyses of the 'omics' data, which should further improve our understanding of global molecular changes in stem cell self-renewal and differentiation elicited by drugs and/or environmental pollutants, especially in the human system. Recent technological developments include chemoproteomics [190,191] and chemical ChIP-SEQ [192] aimed at dissecting the interactions of host genes and proteins with small-molecular-weight compounds such as drugs, metabolites and environmental pollutants. The targeted host molecules would be then identified by mass spectrometry and chromatin immunoprecipitation followed by deep sequencing, providing the data on the molecular mechanisms affected by toxic substances and predicting, with a high degree of accuracy, potential toxicity of untested chemicals for further analyses. Moreover, once the interacting proteins and/or genes targeted by potential toxicants are identified, the information can be complemented using genome-editing techniques such as CRISPR/CAS9mediated gene knock-out [193] to confirm the biological significance of the gene or signal transduction pathway for cell survival or sensitivity to a particular toxicant. The CRISPR/CAS9 system has been proved particularly beneficial for the evaluation of genome editing in hESCs that are not suitable for genetic analysis by other techniques [194].

CONCLUSIONS
Stem cell toxicology may become the gold standard in toxicity testing if fully validated physiologically relevant tests that are reproducible, relatively inexpensive, and not time-and labor-consuming are implemented on a global scale. In this case, stem cell toxicology would eliminate the need for wholeorganism tests not allowed in humans, while providing the platform to evaluate a wide variety of untested chemicals to which we are continuously exposed. In addition, stem cells can differentiate into 3D organoid structures more closely recreating the in vivo microenvironments. Also, multiple SSCs and hiPSCs have been established to match clinical individual differences, with great potential applications in personalized toxicology. Consequently, stem cell toxicology would allow shifting from experimental animal systems that may not generate results fully applicable to human health because of species-specific differences, and may solve the problem of traditional in vitro toxicology that cannot reliably evaluate potential effects on the whole organism, bringing us a step closer to an ideal analytical system implemented exclusively in vitro.