Human stem cell models to study placode development, function and pathology

The fields of developmental biology and stem cell biology are inherently linked, as the process leading pluripotent stem cells (PSCs) to the acquisition of a definitive cell state in vitro is a close recapitulation of the developmental steps that drive such specification in vivo. Accordingly, knowledge coming from developmental biology is crucial in establishing the necessary conditions for in vitro differentiation, and, inversely, differentiating PSCs that recapitulate the complex molecular changes underlying cellular differentiation and tissue development are an invaluable tool to model development in vitro (Camp et al., 2015; Frum and Spence, 2021; Sridhar et al., 2020; Tao and Zhang, 2016). This tool is particularly relevant when it comes to human development, the study of which is hindered by the limited availability of fetal tissue samples and by ethical concerns regarding embryo research (Wamaitha and Niakan, 2018). For example, exceedingly early steps of development, particularly difficult to study in vivo, can be studied in vitro starting from human PSCs (hPSCs) (Box 1): germ layer specification, primitive streak formation and ectodermal patterning have all been modelled in two-dimensional (2D) settings (Britton et al., 2019; Martyn et al., 2018; Warmflash et al., 2014), whereas the blastula and gastrula as a whole have been modelled in three-dimensional (3D) structures named blastoids and gastruloids, respectively (Kagawa et al., 2021; Moris et al., 2020). These 3D cultures model their respective developmental stages both morphologically and in terms of cell composition, making them amenable to the study of human-specific developmental processes at these very early stages. A great value of PSC-based models is the close connection they show to species-specific developmental processes, a characteristic that has been employed, for example, to study the differences in developmental timing and tempo between species (Ebisuya and Briscoe, 2018; Rayon et al., 2020), and to dissect molecular and cellular mechanisms underlying the unique scale of expansion and folding of the human cerebral cortex (Benito-Kwiecinski et al., 2021; Kanton et al., 2019; Pollen et al., 2019).

Although the field of human central nervous system (CNS) development has already been enriched by the employment of hPSC-based models (Chiaradia and Lancaster, 2020; Kelley and Paşca, 2022; Sidhaye and Knoblich, 2021; Velasco et al., 2020), hPSC-based models of peripheral nervous system (PNS) development are at a less mature stage. Particularly relevant in PNS development are the cranial placodes, ectodermal thickenings arising soon after neural tube closure in the anterior portion of the vertebrate ectoderm, which generate greatly diverse derivatives, including portions of the PNS. Such derivatives are components of paired sense organs (i.e. the olfactory epithelium, the inner ear and the eye lens), cranial ganglia (i.e. the trigeminal, geniculate, petrosal and nodose ganglia) and the anterior portion of the pituitary gland, also named the adenohypophysis (Fig. 1A) (Schlosser, 2006). Thus far, placodal development has been studied almost exclusively in animal models, providing a detailed description of the gene regulatory networks and signalling pathways instructing the development of such diverse tissues (Brugmann et al., 2004; Kwon et al., 2010; Litsiou et al., 2005; Schlosser and Ahrens, 2004). Only recently has knowledge regarding placode development in animal models been successfully applied to differentiate hPSCs into placodal progenitors and placode-derived tissues (Chen et al., 2012; Leung et al., 2013; Ozone et al., 2016; Tchieu et al., 2017; Zimmer et al., 2018), offering an exciting opportunity to translate the findings obtained in vivo to a human setting, and to study placodes and their development in our species. Additionally, the dysregulation of placodal derivatives is at the basis of several debilitating human pathologies, including trigeminal neuralgia, migraine, hypopituitarism and hearing loss (Maarbjerg et al., 2017; Messlinger and Russo, 2019; Qiu and Qiu, 2019; Yeliosof and Gangat, 2019). Thus, modelling placode derivatives in vitro from hPSCs has high translational potential, paving the way for the identification of human-specific pathogenic mechanisms, for the discovery of new drugs and therapeutic targets, and for personalised cell replacement therapies (CRTs).

In this Review, we highlight the close connection that exists between placode development in vivo and placode differentiation in vitro, provide insight into how in vitro models may differ from in vivo development and discuss how human stem cell models can facilitate the study of human placode development and disease in the future.

During the early stages of vertebrate development, the ectoderm subdivides into four domains (Fig. 1B), which can be broadly divided into (1) the neural plate, which is located medially and is the precursor of the CNS; (2) the non-neural ectoderm (NNE), which is located laterally and gives rise to the skin, as well as further dividing into two other domains at later stages of development: (3) the neural crest, which is located posteriorly on both sides of the neural plate and gives rise to tissues including bones of the head, parts of the enteric nervous system, dorsal root ganglia and a fraction of the cranial sensory ganglia; and (4) the pre-placodal ectoderm (PPE), which surrounds the neural plate anteriorly and is the origin of all cranial placodes across species (Bailey et al., 2006; Bhattacharyya et al., 2004; Jacobson, 1963; Kozlowski et al., 1997; Pieper et al., 2011; Schlosser and Ahrens, 2004; Streit, 2002). These domains are not delineated by distinct borders, as progenitor cells expressing markers of multiple fates exist over time, suggesting that the distinction between them is less sharp than initially thought (Roellig et al., 2017; Streit, 2002).

Ectodermal patterning is initially dependent on bone morphogenetic protein 4 (BMP4) and its antagonists, whereby high BMP4 signalling specifies the NNE laterally, and medial inhibition of BMP4 signalling triggers neural plate specification (Fig. 1B) (Wilson and Hemmati-Brivanlou, 1995). Three different models have been proposed for the subsequent specification of PPE and neural crest, each providing a different perspective regarding whether these ectodermal domains have a common origin: the ‘binary competence’ model, the ‘neural plate border’ model and the ‘gradient border’ model (Box 2) (Pieper et al., 2012; Roellig et al., 2017; Schlosser, 2006; Streit, 2002; Thiery et al., 2022 preprint). Despite the debate on the precise origin of the PPE and its developmental relationship to the neural crest, the transcriptional programmes driving PPE specification and the signalling pathways instructing them are, overall, well defined (reviewed by Grocott et al., 2012; Moody and LaMantia, 2015; Saint-Jeannet and Moody, 2014). A comprehensive description of early placodal development in a species-specific manner lies beyond the scope of this Review, and as such, unless otherwise stated, we use as reference model the anamniote Xenopus laevis (reviewed by Baker and Bronner-Fraser, 2001; Grocott et al., 2012; Moody and LaMantia, 2015; Saint-Jeannet and Moody, 2014; Schlosser, 2010).

As previously mentioned, all placodes arise from a common domain, the PPE, which surrounds the anterior neural plate and forms a posterior ‘boundary’ with the neural crest (Fig. 1B) (Pieper et al., 2011; Schlosser and Ahrens, 2004). The acquisition of a PPE fate can occur only in ectoderm that has previously gained pre-placodal competence, which, in Xenopus laevis, is marked by the expression of transcription factors (TFs), such as Tfap2a, Gata2 and Dlx3 (Maharana and Schlosser, 2018; Pieper et al., 2012; Woda et al., 2003). These TFs are expressed in a vast domain of the NNE and, as such, their initial expression has been proposed to be dependent on BMP4 signalling, from which these TFs are directly or indirectly activated (Feledy et al., 1999; Friedle and Knöchel, 2002; Luo et al., 2001). Ectoderm that expresses such competence factors, can later be specified into PPE, which itself is marked by the expression of the TF Six1 (and other Six gene family members) and Six1 co-factors Eya1/2 and Dach1 (Brugmann et al., 2004; David et al., 2001; Schlosser and Ahrens, 2004). The PPE is specified only in the region surrounding the anterior neural plate, where, following exposure to BMP4 signalling granting pre-placodal competence, inhibition of BMP4 and activation of fibroblast growth factor (FGF) allows expression of PPE genes, such as Six1 (Ahrens and Schlosser, 2005; Pieper et al., 2012). Additionally, the mutually exclusive antero-posterior patterning of PPE and neural crest is instructed by the high concentration of Wnt posteriorly and high concentration of Wnt antagonists anteriorly, as Wnt signalling strongly inhibits PPE onset (Brugmann et al., 2004; Groves and LaBonne, 2014; Pieper et al., 2012).

Because the staggeringly diverse placodal derivatives all originate from the SIX1⁺ pre-placodal ectoderm region, when obtaining these tissues in vitro the first crucial step is to pattern hPSCs into SIX1⁺ placodal progenitors, which have been successfully differentiated with varying degrees of purity (Dincer et al., 2013; Leung et al., 2013; Tchieu et al., 2017). Despite considerable variability between differentiation paradigms, published strategies to derive SIX1⁺ cells rely on the modulation of two key signalling pathways. First, transforming growth factor β (TGFβ) inhibition prevents mesoderm and endoderm specification and skews the differentiation towards ectodermal lineages (Smith et al., 2008). Second, BMP4 signalling needs to be activated to obtain PPE-like cells, as seen during development. Whether BMP4 is provided exogenously (Tchieu et al., 2017), or endogenously expressed by the cells in protocols employing non-chemically defined medium (Dincer et al., 2013; Leung et al., 2013), lack of activation of this signalling pathway abolishes the differentiation towards SIX1⁺ placodal progenitors. This highlights a conserved role of BMP4 in the acquisition of PPE identity in human development. Moreover, mirroring in vivo development (Ahrens and Schlosser, 2005), FGF stimulation at the time of BMP4 discontinuation leads to a near fivefold increase in the percentage of hPSC-derived SIX1⁺ cells (Tchieu et al., 2017), also pointing to a conserved role of FGF signalling in human placode development. However, interesting differences exist between how FGF and BMP4 signalling are required for placode differentiation in vitro compared with in vivo. Firstly, whereas in Xenopus and chick FGF8 can drive PPE specification (Ahrens and Schlosser, 2005; Litsiou et al., 2005), for hPSC differentiation it is FGF2 that leads to an increase in SIX1⁺ cells, and FGF8 has little to no effect on placodal progenitor differentiation (Tchieu et al., 2017). Moreover, in vivo, FGF activation is accompanied by BMP4 inhibition during PPE-fate acquisition, but exogenous inhibition of BMP4 signalling does not seem to be necessary for placode progenitor differentiation from hPSCs (Leung et al., 2013). The precise timing at which such inhibition might lead to increased placodal differentiation is, however, difficult to untangle and the complete abrogation of BMP4 endogenous signalling might be challenging to obtain in vitro. Such differences may derive either from human-specific developmental mechanisms or from artefacts of the in vitro system and could be addressed by the direct comparison of PSCs from different species and their differentiation towards SIX1⁺ cells.

Despite these differences in signalling requirements for placode specification, the sequential activation of transcriptional programmes during SIX1⁺ cell differentiation in vitro seems to recapitulate in vivo development well. Indeed, exposure of hPSCs to BMP4 first triggers the expression of TFs that grant pre-placodal competence in different species (Bhat et al., 2013; Kwon et al., 2010; Pieper et al., 2012), including TFAP2A/C, DLX3/5, FOXI1 and GATA3. Only after the onset of their expression, and almost exclusively in cells where such pre-placodal competence has been acquired, SIX1 is expressed in differentiating cells, marking their acquisition of a PPE-like fate (Dincer et al., 2013; Ealy et al., 2016; Leung et al., 2013; Tchieu et al., 2017). Interestingly, knocking out TFAP2A in hPSCs strongly reduces their ability to acquire a placodal fate, although it does not completely abolish it (Tchieu et al., 2017). The inability of TFAP2A knockout to completely prevent differentiation of SIX1⁺ cells might be due to the activity of other competence factors partially compensating for its absence, as observed in vivo in zebrafish, where the absence of a single competence factor is not sufficient to abolish PPE formation (Bhat et al., 2013; Kwon et al., 2010). All this argues in favour of conserved mechanisms for placodal competence across species and shows that sequential activation of transcriptional programmes during placode development can be replicated in similar ways in vitro in hPSC-based models to recapitulate similar tissue types found in vivo. Moreover, knockout of TFAP2A is a good example of how genetic perturbations in hPSCs can be used to investigate the role of genes in human development and translate findings obtained in animal models to a human setting and vice versa.

The great scalability of hPSCs makes them an ideal platform for performing high-throughput genetic screens and high-content chemical screens for factors able to enhance differentiation towards a given cell type. The application of chemical screens to optimise placode progenitor differentiation has resulted in the identification of the metalloprotease inhibitor phenanthroline as an enhancer of such differentiation efficiency (Tchieu et al., 2017). Small-molecule screens also have the potential to identify in an unbiased manner signalling pathways involved in the development of specific tissues in a human context [for a non-placode related example, see Majd et al., 2022 preprint, where the authors identify platelet-derived growth factor (PDGF) receptor inhibition as an enhancer of nitrinergic neuron differentiation].

Overall, the development of SIX1⁺ placode progenitors can be modelled in vitro starting from hPSCs in a way that mirrors the early steps of placodal development observed in animal models, serving as a starting point for the subsequent differentiation of placode derivatives and as a platform to identify human-specific mechanisms underlying placodal development.

At the time of PPE specification, the entire region has equal potential to develop into all placodes. Soon thereafter, distinct regions of the PPE acquire more restricted placogenic potential and are ultimately fated towards differentiation into a single placode (Jacobson, 1963; Pieper et al., 2011). Initially, the PPE is patterned into an anterior and a posterior area (Fig. 1B), marked by the expression of Otx2 and Gbx2, respectively (Steventon et al., 2012). After this binary subdivision, different portions of the PPE start to express specific Pax genes, in an anterior to posterior position-dependent manner, according to the so-called ‘Pax code’ (McCauley and Bronner-Fraser, 2002), so that three distinct regions with different competence to generate placodal derivatives arise (Fig. 1B). In particular, Pax6 is expressed in the anterior region of the PPE, Pax3 in the intermediate region, and Pax2 and Pax8 in the posterior region (Schlosser and Ahrens, 2004).

The Pax6-expressing anterior portion of the PPE generates lens, anterior pituitary and olfactory placodes (Pieper et al., 2011; Schlosser and Ahrens, 2004). In chick, the lens is the default fate of all placodal progenitors, which is fulfilled only in cells that are not subjected to FGF (Bailey et al., 2006). By contrast, studies in chick and zebrafish show that FGF8 alone skews anterior placodal progenitors towards olfactory placode fate, whereas FGF in combination with sonic hedgehog (SHH) skews them towards anterior pituitary placode fate (Bailey et al., 2006; Herzog et al., 2003). Both the default lens state and the dependence of anterior pituitary development on FGF and SHH are well recapitulated in hPSC-based models, as described below (Ozone et al., 2016; Tchieu et al., 2017; Zimmer et al., 2016).

Posterior to this Pax6-expressing region, the intermediate placodal area is characterised by the expression of Pax3, which, in chick, is dependent on Wnt signalling (Lassiter et al., 2007) and generates the trigeminal placode. Similarly, PAX3 expression is induced by Wnt signalling in hPSCs differentiating towards trigeminal neurons before the acquisition of the post-mitotic neuronal fate (Zimmer et al., 2018).

In the posterior-most portion of the PPE, FGF signalling specifies a region termed the ‘otic-epibranchial progenitor domain’ (OEPD), which expresses both Pax2 and Pax8, from which the otic placode (precursor of the inner ear) and the epibranchial placodes (precursors of the geniculate, petrosal and nodose cranial ganglia) originate (Schlosser and Ahrens, 2004). Consequently, zebrafish and chick studies have shown that continuous FGF signalling specifies the epibranchial placodes, whereas FGF inhibition makes part of the OEPD responsive to subsequent Wnt stimulation, determining otic placode differentiation (Freter et al., 2008; Nechiporuk et al., 2007). Although epibranchial neurons have not yet been generated in vitro from hPSCs, the otic placode and its development have been modelled both in 2D and 3D hPSC-derived cultures (Chen et al., 2012; Ealy et al., 2016; Koehler et al., 2017). Again, this process recapitulates in vivo development, as Wnt stimulation induces PAX8⁺ otic placode-like cells in otic organoids (Koehler et al., 2017). Interestingly, the timing of otic placode differentiation in the organoids follows its in vivo developmental pace in a species-specific manner. In vivo, the human cranial placodes arise between day 18 and 24 post-conception. From hPSCs, which date approximately to 12 days post-conception, the otic placode originates between day 6 and 12 of in vitro differentiation, showing that developmental timing is closely mimicked. The differentiation tempo of the inner ear organoid cells is significantly faster in organoids derived from murine PSCs, mirroring the species-specific temporal differences seen during in vivo development (Koehler et al., 2013, 2017).

After neural tube closure and Pax gene expression, placodal progenitors are localised in different portions of the embryo's head and, from this point on, the development of each individual placode is independent from the others (Pieper et al., 2011; Schlosser and Ahrens, 2004). The progenitors of each placode are initially located on the ectodermal surface, and only later migrate, delaminate or invaginate to occupy the final position of the differentiated tissue (Breau and Schneider-Maunoury, 2014; Schlosser and Ahrens, 2004). We now focus on the placodal derivatives that have been modelled using hPSCs (Fig. 2), describing similarities and differences between in vivo development and hPSC differentiation.

Derived from the anterior Pax6⁺ placodal ectoderm, the eye lens is a transparent structure allowing the passage and focusing of light on the retina. In vivo, the lens placode invaginates to form the lens vesicle, in which posteriorly located cells elongate, lose their nuclei and cytoplasmic organelles and are filled with crystallin proteins (CRYAA and CRYAB), thus becoming transparent fibre cells. In contrast, anteriorly located cells acquire an epithelial nature (reviewed by Bassnett and Šikić, 2017).

In vitro, SIX1⁺ placode cells obtained through stimulation of BMP4 and FGF2, when terminally differentiated, express CRYAA and CRYAB at day 30, consistent with lens being the placode default state (Tchieu et al., 2017). Other studies have shown that, even upon continuous stimulation of FGF2 and BMP4/7 signalling between day 6 and 15 (Fu et al., 2017), crystalline proteins start to be expressed. Consistent with the role of these signalling pathways in promoting lens maturation at later developmental stages in mice, after the lens placode has separated from the PPE (Stump et al., 2003; Zhao et al., 2008), activating FGF2 and Wnt together at later differentiation stages enhances maturation and differentiation efficiency towards lens fate. In these conditions, lens cells form so-called ‘lentoid bodies’ – aggregates that recapitulate lens features from a morphological, cell composition and functional perspective. Indeed, an inner cell population arises from the lens progenitors within the lentoid bodies, composed of fibre-like cells, whereas the outer population is composed of epithelial cells. Fibre cells progressively lose their nuclei through a process of fragmentation, which is similar to that witnessed in vivo, and autophagic activity leads to organelle loss in the cytoplasm, so that the cells are filled with crystallins (Fu et al., 2017; Yang et al., 2010b). Strikingly, not only are lentoid bodies similar to the lens regarding cell composition and morphology, but they also have a magnification ability that is comparable to that of a lens isolated from adult rat (Fu et al., 2017).

The anterior pituitary gland plays a major role in the vertebrate endocrine system, with the production of hormones relevant for growth, sexual maturation, temperature control and metabolism. This gland forms from the anterior pituitary placode, characterised in mice by early markers, such as PITX1, and later by LHX3 (Sheng et al., 1996; Suh et al., 2002). When in contact with the developing hypothalamic neuroectoderm, the anterior pituitary placode invaginates to form a structure named Rathke's pouch, which gives rise to six types of hormone-releasing cells, including adrenocorticotropic hormone (ACTH)- and thyroid-stimulating hormone (TSH)-releasing cells (Schlosser, 2006). Anterior pituitary hormone secretion is controlled by hormones released by the hypothalamus, beneath which the anterior pituitary gland is ultimately located. Pituitary hormones, in turn, trigger the secretion of downstream hormones from target organs. Defects in hormonal secretion by the anterior pituitary gland can result in a severe condition named hypopituitarism (Yeliosof and Gangat, 2019), which could potentially be treated by CRTs based on the differentiation of hPSCs towards hormone-releasing pituitary cells.

Anterior pituitary development has been modelled in vitro in both 2D (Fig. 2) and 3D cultures, first from murine PSCs (Suga et al., 2011), and subsequently from hPSCs (Ozone et al., 2016; Zimmer et al., 2016). In particular, anterior pituitary cells have been obtained in 3D cultures by juxtaposing hPSC-derived hypothalamic cells, which derive from the neural plate in vivo, and hPSC-derived non-neural ectodermal cells. In these conditions, SHH activation leads to the differentiation of cells expressing the early anterior pituitary markers PITX1 and LHX3 at the interface between the two tissues: upon stimulation of FGF signalling, these cells invaginate into the hypothalamic tissue to form a Rathke's pouch-like structure during in vitro differentiation (Ozone et al., 2016), although this invagination does not occur in all organoids. Therefore, as during in vivo development (Zhu et al., 2007), FGF and SHH signalling pathways play a pivotal role in the differentiation of PITX1⁺/LHX3⁺ cells and in their successful invagination in a 3D culture system. In all hPSC-based organoid models, most hormone-releasing cells are ACTH⁺ and show physiological functional activity. Indeed, not only are they able to release ACTH, but this release can be physiologically fine-tuned according to the same feedback and feedforward mechanisms that exist in vivo (Dallman et al., 1987; Ozone et al., 2016). Moreover, transplantation of these cells in an anterior pituitary gland-resected mouse results in their functional integration into the endocrine circuitry (Ozone et al., 2016). Despite attempts at enhancing development of distinct hormone-releasing cells, this has only resulted in a modest increase in the in vitro differentiation of other pituitary lineages, with yields of around 8% of PITX1⁺ cells for growth hormone-expressing cells, and only 1-2% for prolactin- and TSH-expressing cells. Although early anterior pituitary development and Rathke's pouch invagination can be modelled using these hPSC-based models, they fail to recapitulate the full diversity of hormone-releasing cells at a similar level as seen in vivo. The low efficiency of pituitary cells within the organoids, and also their limited diversity, currently limits their applicability for CRTs.

An alternative approach for the differentiation of pituitary cell populations in 2D is arguably more suitable for this purpose because the employment of chemically defined culturing conditions makes 2D culture amenable for the manufacturing practices required for therapeutic employment (Zimmer et al., 2016). Again, SHH and FGF8/10 stimulation after placodal progenitor specification leads to the differentiation of LHX3⁺/PITX1⁺ pituitary-like cells at day 30, yet without the need for hypothalamic conditioning. Already at this stage, 50% of the cells express at least one hormone at the transcriptional level, a percentage that increases to 80% at differentiation day 60. In vivo, high dorsal FGF8 concentrations and high ventral BMP2 concentrations create two distinct dorsoventral gradients in the anterior pituitary gland, the integration of which results in the specification of the six hormone-releasing populations in stereotypical positions in the gland (Treier et al., 1998). Accordingly, fine-tuning the concentrations of FGF8 and BMP2 in vitro, based on their developmental role in vivo, leads to the differentiation of the expected hormone-releasing populations (Zimmer et al., 2016). Similar to the 3D differentiation paradigm, differentiation efficiency obtained for different hormone-releasing cells is highly variable, with ACTH-releasing cells being the ones obtained most consistently and with highest efficiency. After xenotransplantation of cells obtained in these conditions in hypophysectomised rats, the graft secretes pituitary hormones and shows signs of maturation in the weeks following transplantation. This 2D differentiation paradigm shows that hPSC-derived anterior pituitary cells can be obtained independently of the hypothalamic lineage. However, a direct comparison with 3D differentiation models would provide insight into how the hypothalamic anlage is important for complex structure formation (e.g. Rathke's pouch invagination) and maturation of the cells obtained through these differentiation strategies. Future studies that further optimise 2D and 3D differentiation strategies will provide the opportunity to study human pituitary cell development in vitro and potentially generate diverse hormone-producing cells on a large scale, opening new roads for CRTs.

The trigeminal ganglion hosts the somas of trigeminal somatosensory neurons, which transmit sensory information of pain, temperature and touch from the face to the CNS: these neurons are organised into the ophthalmic, maxillar and mandibular branches. From these facial areas, trigeminal neurons transmit information to the trigeminal sensory nuclei of the brainstem, from which sensory information is relayed to the thalamus, and consequently to the somatosensory cortex (Erzurumlu et al., 2010). Facial somatosensory information is particularly relevant for vertebrate physiology, also considering that, in the somatosensory cortex, the number of neurons receiving facial somatosensory information is disproportionately high compared with the actual surface of the face itself; moreover, facial somatosensory circuitry is highly species specific, considering that different facial sensory organs are present in different mammals (e.g. mouse whiskers) (Erzurumlu et al., 2010). The facial trigeminal circuitry originates from the intermediate placodal ectoderm, which gives rise to the trigeminal placode. In chick, trigeminal placode development is dependent on Wnt and PDGF, whereas subsequent neuronal differentiation and delamination is dependent on FGF (Lassiter et al., 2007, 2009; McCabe and Bronner-Fraser, 2008). Trigeminal neurons have been successfully differentiated in vitro from hPSCs, by inducing NNE differentiation through activation of BMP4 signalling, and then skewing it towards an intermediate placodal fate through subsequent activation of Wnt signalling (Fig. 2). This approach leads to the differentiation of SIX1⁺/PAX3⁺ cells within 10 days, and of neurons positive for the sensory neuron marker ISL1 after 20 days (Sun et al., 2008; Zimmer et al., 2018). This differentiation setting gives rise to all three sensory neural populations found in the trigeminal ganglion: nociceptive, mechanoreceptive and proprioceptive neurons. Furthermore, three distinct nociceptive populations arise and become progressively distinct as the neurons mature: one is responsive to heat, one to cold and one to inflammatory pain (Zimmer et al., 2018). The in vitro differentiation of trigeminal neurons captures the cellular diversity and physiological function of the ganglion, as is apparent from the responsiveness of the neurons to diverse stimuli.

A human trigeminal-specific characteristic that these neurons recapitulate well is the susceptibility to herpes-simplex virus-1 (HSV-1) infection; indeed, after having infected the labial epithelium, HSV-1 is retrogradely transported to the nucleus of trigeminal neurons, where it remains latent indefinitely, and sporadically gets reactivated in a subset of patients (Wilson and Mohr, 2012). Cortical neurons, however, are normally not susceptible to HSV-1 infection (Lafaille et al., 2012; Lafaille et al., 2019). This difference in cell-intrinsic immunity is, at least partially, due to the presence of a constitutive toll-like receptor 3 (TLR3)-dependent antiviral immunity in cortical neurons, which is absent in trigeminal neurons. This cell type-specific viral susceptibility is well recapitulated in cortical and trigeminal neurons obtained from hPSCs (Zimmer et al., 2018), adding to the high degree of similarity that the hPSC-derived neurons have to their in vivo counterpart, which is not limited to their physiological development, but includes neuronal subtype-specific responses to pathogenic insults.

After leaving the PPE, the otic placode invaginates to form the otic vesicle, giving rise to the inner ear and sensory neurons of the vestibulocochlear ganglion, which transmits auditory information to the CNS (Bruska et al., 2009). The inner ear comprises the sensory epithelium, composed of hair cells with apical protrusions that sense auditory and vestibular stimuli and transmit them to bipolar sensory neurons of the vestibulocochlear ganglion (Lim and Brichta, 2016). Loss of either hair cells or sensory neurons can lead to auditory impairment and deafness.

Auditory sensory neurons and hair cells have been differentiated in vitro first from murine PSCs (Koehler et al., 2013), and later from hPSCs, in 2D (Fig. 2) and 3D models (Chen et al., 2012; Ealy et al., 2016; Koehler et al., 2017; Lahlou et al., 2018; Matsuoka et al., 2017; Ohnishi et al., 2015). FGF3 and FGF10 are key cues driving hPSCs toward a PAX2⁺/PAX8⁺ otic placode fate (Chen et al., 2012; Lahlou et al., 2018), leading to the activation of transcriptional programmes related to inner ear development after 12 days of FGF treatment (Chen et al., 2012), consistent with the role of this signalling molecule in ear development in vivo (Wright and Mansour, 2003). In addition, ‘posteriorising’ signals, such as Wnt and retinoic acid, have a role in otic placode induction in vivo (Freter et al., 2008; Hans and Westerfield, 2007). As such, their stimulation increases purity and Pax code specificity in cells differentiating towards otic placode in vitro (Ealy et al., 2016). Interestingly, the transcriptional programmes sequentially activated during differentiation of hPSCs to otic placode cells are very much consistent with those observed in vivo during development, with cells at 12 days of differentiation showing the highest degree of transcriptomic similarity with the murine otocyst at embryonic day (E) 10.5 (Ealy et al., 2016), supporting the notion that the transcriptional changes occurring during hPSC differentiation toward a given cell fate in vitro closely recapitulate transcriptional changes seen during in vivo development (Camp et al., 2015; Sridhar et al., 2020). In 3D cultures, the formation of otic placode-like cells is followed by their spontaneous invagination into pits first and vesicles later, in a manner reminiscent of otic vesicle formation in vivo (Koehler et al., 2017). Such morphological and mechanical recapitulation makes otic organoids amenable to modelling of the inner ear as a whole, a complex tissue in which function is based on the interplay between distinct, yet developmentally related cell types.

Both in 2D and in 3D, otic placode cells have been differentiated in two distinct cell populations equally essential for the transmission of auditory and vestibular stimuli: hair cells and vestibulocochlear neurons, which transmit sensory information from hair cells to the CNS. Hair cells differentiated in vitro from otic placode-like cells express typical markers found in vivo in human and mouse, such as ATOH1, BRN3C (POU4F3) and MYO7A (Locher et al., 2013; Xiang et al., 1997; Yang et al., 2010a), present apical bundles, and show electrophysiological behaviour comparable to native hair cells (Chen et al., 2012). Functional hair cells also arise in otic organoids, albeit in only 20% of them, indicating that we are yet to fully understand the processes underlying their development. Although hair cells are electrophysiologically functional, they remain immature after 60 days of differentiation (Koehler et al., 2017), highlighting one of the biggest challenges in the human stem cell field: that exceedingly long-term differentiation and maturation is needed for hPSC-derived cells to reach an adult stage (Gordon et al., 2021).

In a similar way, hPSC-derived vestibulocochlear sensory neurons (Chen et al., 2012; Matsuoka et al., 2017) show electrophysiological activity consistent with that witnessed in vivo in immature neurons and, when co-cultured with rat brainstem explants containing the cochlear nucleus, they exhibit migratory potential towards the nucleus and form synapses with brainstem neurons (Matsuoka et al., 2017). Preliminary studies attempting xenotransplantation of either in vitro-generated human PPE-like cells into murine hair cell-depleted cochlea (Takeda et al., 2021) or hPSC-derived vestibulocochlear sensory neurons into a neuropathic deafness gerbil model (Chen et al., 2012) have shown signs of graft survival and functional integration. However, these initial studies show considerable variability in their differentiation efficiency and require further validation before such methods can be considered viable for translation to CRTs in a clinical context.

In summary, these examples show that the developmental steps leading to the formation of placodal derivatives are, to some degree, recapitulated from both a transcriptional and morphological perspective during human stem cell differentiation in vitro, opening new roads for the study of human placode development. Such recapitulation of development results in the generation of terminally differentiated cells that resemble the native tissue from a functional point of view, which therefore can be employed to model such tissues in physiological and pathological conditions, and ultimately be used for CRTs.

Placodes give rise to highly diverse tissues, including both nonneural and neural derivatives, guided by patterning events during embryonic development. The steps of placodal development and regionalisation can, in part, be recapitulated in vitro starting from hPSCs, leading to the differentiation of placodal derivatives in both 2D and 3D. This allows for the study of intrinsic molecular mechanisms and extrinsic factors that influence cell fate specification, providing an invaluable platform for the developmental biology field to study human placodal development in vitro. Understanding the immense cellular heterogeneity of placode-derived tissues, and advancements in generating these terminally differentiated cells in vitro, is essential to both disease modelling and regenerative medicine. The possibility of modelling placodal development starting from hPSCs comes from the resemblance that cells differentiating to placodal fates in vitro bear towards placodes developing in vivo. First, the transcriptional programmes sequentially activated by differentiating cells are both temporally and qualitatively similar to those described in vivo during placodal development (Ealy et al., 2016). Second, the morphology of placodal derivatives is recapitulated both at the level of individual cells in 2D cultures (e.g. otic neurons are bipolar and hair cells show apical bundles) (Chen et al., 2012), and, to some degree, at the level of multicellular tissue organisation in 3D cultures (e.g. inner ear organoids form otic vesicles and anterior pituitary organoids form Rathke's pouches) (Koehler et al., 2017; Ozone et al., 2016). Lastly, the first proof-of-principle studies have provided evidence that hPSC-derived cells are functional to the point that their transplantation in vivo can partially compensate for the loss of endogenous cells by integrating into the native circuitry (Fig. 3A) (Chen et al., 2012; Zimmer et al., 2016).

Although considerable progress has been made in using hPSC technology to model human development, and the first attempts to study placodal development have been promising, technical challenges remain. First, and perhaps most importantly, the great diversity of cell types that arise from placode progenitors in vivo are not yet fully recapitulated in the currently established differentiation paradigms in vitro. This emphasises that we are yet to understand all the factors that guide human placode development, which is both a challenge and an opportunity for the stem cell and developmental biology field. Another key limitation is the variability that is present in directed differentiation methods between cell lines, protocols, and even between independent differentiation experiments using the same cell line and protocol. It will be crucial to continue to optimise differentiation paradigms and characterise differentiated cells at a single-cell resolution to ensure protocols become even more robust. In vitro hPSC-based models are, by definition, reductionist model systems. This allows for specific perturbations and spatiotemporal analyses at high resolution. However, placodal development in vivo does not occur in isolation and yields a high diversity of tissues. The development of 3D organoid models has allowed for more complexity and yielded unique opportunities to study developmental processes at both a cellular and tissue level, but it will be important to optimise these 3D systems by increasing reproducibility using specialised bioreactors (Qian et al., 2016), by integrating signalling centres to ensure faithful spatial patterning (Cederquist et al., 2019) and by allowing for significant scaling of organoid differentiations by employing biocompatible polymers (Narazaki et al., 2022 preprint).

The possibility of performing high-throughput analyses at a single-cell level has greatly increased the resolution at which tissue complexity and function can be dissected, making it possible to identify rare cell populations and infer their physiological role. This has had a significant impact on both developmental and stem cell biology, with the possibility to characterise developing tissues at different stages thoroughly (Briggs et al., 2018; Cao et al., 2019; Farrell et al., 2018; Pijuan-Sala et al., 2020; Wagner et al., 2018), as well as differentiating hPSC-derived cultures (Camp et al., 2015; He et al., 2022). The combination of high-throughput, single-cell techniques with in vitro hPSC-derived 3D models has been successfully employed in the context of CNS development, where the complex molecular and cellular changes occurring at the single-cell level during cortical development are well recapitulated by cortical organoids (Camp et al., 2015; Coquand et al., 2022 preprint; Pollen et al., 2019). The combination of single-cell and spatial transcriptomics with CRISPR/Cas9 scarring (i.e. the random insertion and deletions occurring in the genome upon Cas9 DNA cutting, which can be used as individual cell barcodes) has provided further insights into the lineage relationships within cortical organoids (He et al., 2022).

In the field of placode development, the potential of high-throughput, single-cell analyses has not been fully exploited yet, particularly in combination with hPSC-based models. In this context, lineage tracing of hPSCs developing towards placodes could help shed light on placodal origin within the ectoderm, possibly resolving the debate on which model best describes early placodal and neural crest development in humans (see Box 2). Additionally, obtaining 3D cultures of different placode-derived tissues and analysing them using single-cell omics techniques would increase our understanding of the complexity of these tissues in a human setting, providing insight into their precise cellular composition, and possibly benchmarking them with single-cell data from human primary tissue (Yang et al., 2022). When performed in parallel with PSC cultures derived from different species, such high-throughput analyses could be a powerful tool to identify species-specific developmental and functional processes. Finally, from a translational perspective, with the possibility to perform induced PSC-based CRTs and in vitro disease modelling, single-cell omics methods will be key in optimising differentiation strategies to obtain highly pure placode derivatives that better resemble their in vivo counterparts. The unique scalability of hPSC models makes them ideally suited for high-content chemical screens and whole-genome CRISPR screens, allowing for the unbiased identification of molecules, signalling pathways and gene regulatory networks relevant to human placode development and disease (Fig. 3B).

In summary, the recapitulation of development witnessed during the differentiation of hPSCs towards placodes and their derivatives makes them an exciting new tool to translate findings obtained in animal models to the human condition, and to identify species-specific differences in placode biology. Their great scalability and the ease with which they can be genetically perturbed confer to hPSC-based models the potential to significantly increase our knowledge of human placode development and disease, shedding light on the processes guiding the development of these diverse and fascinating tissues.

We thank Sumru Bayin, Nereo Kalebic, Polina Oberst and all members of the Harschnitz group for their feedback on the manuscript. E.C. is a PhD student within the European School of Molecular Medicine (SEMM).