Developing a sense of touch

Like all senses, the sense of touch allows us to gather information about the people and things in the world around us. What sets touch apart is its intimacy, as it requires direct contact with skin, the sensory organ of tactile sensation. Humans constantly rely on tactile feedback for mundane tasks such as fishing keys out of a cluttered pocket as well as essential behaviors such as manipulating food to feed ourselves and our offspring. Touch is also required for normal cognitive development in mammals. Indeed, tactile hyper- or hypo-sensitivity is a hallmark of neurodevelopmental disorders in children, including autism spectrum disorder (see Box 1; Blakemore et al., 2006; Cascio et al., 2008; Tomchek and Dunn, 2007). Despite the profound implications of early tactile input in cognitive function, the mechanisms that mediate mammalian touch receptor development remain mysterious.

The neurons that transform touch stimuli at the skin surface into electrical impulses – the currency of the nervous system – are known as mechanoreceptors. Two main types of mechanoreceptors innervate mammalian skin: low-threshold mechanoreceptors (LTMRs; see Glossary, Box 2) are tuned to respond to gentle touch, whereas nociceptors (see Glossary, Box 2) encode forces in the noxious (harmful) range. LTMRs can be further classified based on functional properties and molecular markers (Table 1; Fig. 1). Aδ-LTMRs, for example, detect hair follicle deflection (Rutlin et al., 2014), whereas some types of Aβ-LTMRs allow for high-acuity sensation (Bai et al., 2015; Li et al., 2011; Wellnitz et al., 2010), and C-LTMRs are proposed to convey information about social touch (Liljencrantz and Olausson, 2014; Liu et al., 2007; Olausson et al., 2010; Vrontou et al., 2013; Wessberg et al., 2003). These classes were originally designated based on classic electrophysiological and histological studies. Aβ- and Aδ-LTMR neurons are distinguished by their medium-to-large diameter, myelinated axons and medium-to-fast conduction velocities (Table 1; Gasser, 1941; Horch et al., 1977). Conversely, C-fibers possess small axonal diameters and exhibit slow conduction velocities. In total, seven subclasses of low-threshold mechanoreceptors, a substantial portion of the 17 subtypes of putative somatosensory neurons (see Glossary, Box 2; Usoskin et al., 2015), have been further distinguished by a combination of genetic, morphometric and physiological approaches. This variety has spurred research into how distinct classes of mechanoreceptors are specified during development.

Although they are diverse, mechanoreceptors and other somatosensory neurons share a basic body plan. The cell bodies of LTMRs are housed with those of other somatosensory neurons in dorsal root ganglia (DRG; see Glossary, Box 2). Each somatosensory neuron bears a pseudounipolar axon, or afferent (see Glossary, Box 2). Peripheral branches innervate skin or other target organs, whereas central branches innervate the spinal cord and/or hindbrain. Thus, afferents carry sensory signals from the periphery to the central nervous system.

At the periphery, distinct complements of LTMRs innervate hair-bearing skin compared with glabrous, or hairless, skin on the palms and soles of feet (Fig. 1; Dell and Munger, 1986; Halata and Munger, 1983; Li et al., 2011; Miller et al., 1958; Munger and Halata, 1983; Munger and Ide, 1988; Vallbo et al., 1999). In primates, glabrous skin contains four principal types of Aβ-LTMRs (Johnson et al., 2000; Miller et al., 1958). Complexes between Merkel cells (see Glossary, Box 1) and Aβ-LTMRs, which are arranged at the base of rete ridges in glabrous skin, detect sustained touch as well as object features such as edges and curvature (Goodwin et al., 1995; Iggo and Muir, 1969; Phillips et al., 1990). Also present are encapsulated endings such as Meissner corpuscles, which respond best to low-frequency vibration (Talbot et al., 1968), Pacinian corpuscles, which encode high-frequency vibrations (Talbot et al., 1968), and Ruffini endings, which detect skin stretch (Johansson and Vallbo, 1979; Johnson et al., 2000). The complement of neuronal endings differs in hairy skin, as it is specialized less for high-acuity sensation and more for sensation of movement along the body surface. In the mouse coat, for example, distinct combinations of C-, Aβ- and Aδ-LTMRs form collars around each type of hair follicle (Fig. 1; Li and Ginty, 2014; Wu et al., 2012). These collars are complex, consisting of both fence-shaped lanceolate endings, and neurons that wrap around these fences, called circumferential endings. Moreover, Merkel cell-neurite complexes form discrete clusters in touch domes (see Glossary, Box 2), which are specialized epidermal structures that surround guard hairs. These structures, which are visible to the naked eye, are raised owing to the presence of a thickened epidermis and columnar-shaped keratinocytes (see Glossary, Box 2) and increased vascularization (Fig. 1; Doucet et al., 2013; Halata et al., 2003; Iggo and Muir, 1969; Moll et al., 1996; Moll et al., 1993; Pinkus, 1902; Reinisch and Tschachler, 2005; Smith, 1977; Woo et al., 2010). The LTMRs that innervate hair follicles and touch domes are exquisitely sensitive to gentle touch stimulation.

Recent studies, many of which have established the hairy skin of the mouse as a tractable model system for studying LTMR afferent targeting, have shown that hair follicles and touch domes are discrete mechanosensory units that are innervated by unique complements of LTMRs. This finding, combined with concurrent advances in the field of skin development, set the stage to reveal how LTMRs establish and maintain appropriate innervation patterns. In this Review, we summarize the current state of knowledge about how skin and sensory neurons develop in unison to enable a sense of touch. We first provide an overview of the cellular and molecular mechanisms that drive specification of sensory neurons. Next, the process of skin development is outlined. We then describe current models for how touch receptors develop, with a particular focus on Merkel cell-neurite complexes and lanceolate endings. Finally, we highlight research in invertebrate model systems that provides clues into the evolutionarily conserved mechanisms of touch receptor development.

All somatosensory neurons, including LTMRs, arise from precursors called neural crest cells (see Glossary, Box 2) at early developmental stages (Weston, 1970). Neural crest cells are transient, multipotent cells that delaminate from the dorsal neural tube and migrate to the periphery in mice between embryonic day (E) 8.5 and E10 (Fig. 2; Serbedzija et al., 1990). A subset of neural crest cells follow a ventromedial path to form clusters on each side of the neural tube (Serbedzija et al., 1990; Thiery et al., 1982). As the embryo continues to grow from E9.5 to E13.5, these cells give rise to somatosensory neurons that coalesce into pairs of DRGs located in the intervertebral foramina (see Glossary, Box 2) at each spinal level (Fig. 2; Frank and Sanes, 1991; Marmigère and Ernfors, 2007).

Somatosensory neuron heterogeneity is first established by temporal waves of cell fate specification (Frank and Sanes, 1991; Ma et al., 1999). All somatosensory neurons require expression of either neurogenin 1 or 2 (Ngn1, 2; also known as Neurog1, 2), which encode basic helix-loop-helix transcription factors (Ma et al., 1999). Neurons specified early (E9.5 to E11.5) express Ngn2. These neurons generally include Aβ and Aδ afferents with large-diameter cell bodies (Fig. 2) and myelinated axons. They are fated to become either LTMRs or proprioceptors (see Glossary, Box 2), which sense joint position and muscle contraction (Banks et al., 2009; Goldstein et al., 1991; Lawson and Biscoe, 1979; Ma et al., 1999; Sherrington, 1906). A second wave of neuronal specification requires expression of Ngn1 and spans from E10.5 to E13.5. Neurons born during this period are unmyelinated C afferents with small-diameter cell bodies. They are fated to become nociceptors, a subset of which respond to thermal or itch stimuli (Box 2; Fig. 2; Bautista et al., 2014; Julius, 2013; Lawson and Biscoe, 1979; Ma et al., 1999; McKemy, 2013; Schepers and Ringkamp, 2010).

Transcription factors that specify mechanoreceptors include MafA (Bourane et al., 2009) and c-Maf, which is now known simply as Maf (Hu et al., 2012; Wende et al., 2012). Maf and MafA are both basic leucine zipper transcription factors that are expressed starting at roughly E11 and actively enhance specification of mechanosensory neurons, presumably by maintaining expression of the neurotrophin receptors Ret and Gfra2 (Bourane et al., 2009; Luo et al., 2009; Wende et al., 2012). Brn3a (Pou4f1) might also play a role in regulating touch receptor fate (Badea et al., 2012; Dykes et al., 2010). Brn3a encodes a POU-domain transcription factor that binds histone H3-acetylated chromatin in the runt-related transcription factor 3 (Runx3) locus and is thought to control Runx3 expression (Dykes et al., 2010). Runx3 promotes proprioceptor fate between E11.5 and E12 (Kramer et al., 2006). At E13.5, 59% of DRG neurons display rapidly inactivating mechanically gated currents, a hallmark of LTMRs (Lechner et al., 2009). A substantial proportion of these neurons are early-born, large-diameter neurons. By 2 weeks of postnatal age, mechanoreceptive neurons are not yet fully mature, but several types of A afferents can be distinguished in ex vivo electrophysiology preparations (Koltzenburg et al., 1997). Thus, mechanoreceptive neurons acquire sensitivity to touch very soon after specification, but require postnatal maturation to acquire their adult physiological properties.

Following their specification, LTMRs and other somatosensory neurons simultaneously innervate the spinal cord and the periphery. During this process, the architecture and molecular environment of skin rapidly changes. The neural tube, which will become the spinal cord, transforms drastically as interneurons are born and specified (Lai et al., 2016). The process by which sensory axons target specific spinal cord subregions has been studied in detail and is the subject of several reviews (Gardiner, 2011; Masuda et al., 2009; O'Malley et al., 2014; Olson et al., 2016). Although the molecular and cellular mechanisms that mediate innervation of peripheral target tissues are less well-understood, recent parallel investigations of skin biology and somatosensory neurons have begun to reveal how skin microdomains and touch receptors develop in unison.

Mammalian skin begins as two layers of cells that transform into a multilayered structure by the time of birth (Fig. 2; Koster and Roop, 2007). The upper compartment consists of layers of keratinocytes that form the epidermis (see Glossary, Box 2), a protective barrier that prevents water from evaporating out of the body and protects internal organs from microbial insults and radiation (Hardman et al., 1998). The deeper layers of skin, by contrast, are composed of loosely packed fibroblasts and a collagen-rich extracellular matrix that together form the dermis (see Glossary, Box 2).

The epidermis derives from the ectoderm (Byrne et al., 1994; Fernandes et al., 2004; Jinno et al., 2010; Liu et al., 2013; Ohtola et al., 2008; Rendl et al., 2005; Rivera-Pérez and Hadjantonakis, 2014; Wong et al., 2006). The initial epidermal layer is the periderm, a protective layer that covers the skin surface and dies by apoptosis as the skin barrier develops in utero (Hardman et al., 1998; M'Boneko and Merker, 1988; Richardson et al., 2014). Between E9.5 and E18.5, the epidermis stratifies into distinct laminae: basal, spinous, granular and cornified (Fig. 2). These laminae are built sequentially, starting from the proliferative basal layer of cells and extending to enucleated keratinocytes in the surface cornified layer, which plays a key role in barrier function (Fig. 2; Byrne et al., 1994).

Beneath the epidermis, the dermis forms from multiple cellular sources, including neural crest and mesodermal cells (Fernandes et al., 2004; Jinno et al., 2010; Ohtola et al., 2008; Wong et al., 2006). This skin compartment also stratifies between E12.5 and E18.5, transforming into an upper papillary dermis and a lower reticular dermis (Driskell et al., 2013; Sorrell and Caplan, 2004). A deep layer called the hypodermis, primarily composed of fat cells, appears at postnatal stages (Driskell et al., 2013).

In summary, skin is highly dynamic when sensory innervation develops. This suggests that specialized mechanisms – perhaps mediated via secreted long-range signals and contact-dependent short-range cues – help to guide axons as they grow toward their cutaneous targets.

Hair follicles, which are the most abundant tactile structures in mammalian skin, begin to develop in embryogenesis. In mice, for example, whisker follicles located on the lateral aspects of the face form at E12 and are innervated soon thereafter (Hasegawa et al., 2007; Munger and Rice, 1986; Schmidt-Ullrich and Paus, 2005). At E13.5, an early hair-follicle marker, keratin 17 (Krt17; also known as K17), dots the epithelium over most of the body's surface (Bianchi et al., 2005). These K17-positive patches form three types of hair follicles in successive waves. The first to develop are guard hairs, also known as tylotrich or primary hairs. These long hairs are sparse, composing roughly 1% of the adult coat hair (Fig. 1; Lechner and Lewin, 2013; Li et al., 2011). At E14.5, guard-hair keratinocyte precursors enlarge and elongate to form hair follicle placodes (see Glossary, Box 2), which are thickened patches of epithelial cells that proliferate and invaginate to form follicles (Fig. 2; Paus et al., 1999). Awl/auchene or secondary hair placodes, which represent roughly 20% of coat hairs (Lechner and Lewin, 2013; Li et al., 2011), arise in a subsequent wave of follicle specification between E15.5 and E16 (Paus et al., 1999). At this time, hairs also emerge on the tail. At later stages of development, ranging from E17 to postnatal ages, zigzag or tertiary hairs tile the body surface (Lechner and Lewin, 2013; Paus et al., 1999). Named after the bent appearance of their shafts, zigzag hairs make up the majority of the downy coat of mice. Although the precise function of each hair type is unknown, electrophysiological surveys of hair follicle LTMRs have shed light on the possible mechanosensory function of each follicle subtype (Adrian and Zotterman, 1926a,b; Bai et al., 2015; Brown and Iggo, 1967; Burgess et al., 1968; Iggo and Muir, 1969; Li and Ginty, 2014; Li et al., 2011; Rutlin et al., 2014; Zotterman, 1939).

Each murine hair type is associated with a different complement of LTMRs (Figs 1 and 2; for a review, see Abraira and Ginty, 2013). Guard hairs, for example, are innervated solely by myelinated Aβ-LTMR endings. The terminals of these neurons include lanceolate endings, which are fence-shaped structures that align with the length of the hair follicle, and circumferential endings that wrap around the lanceolate terminals in thick bundles (Bai et al., 2015; Halata, 1993; Ikeda et al., 2014; Li and Ginty, 2014; Li et al., 2011; Munger and Ide, 1988). In rodents, each guard hair is flanked by a touch dome containing Merkel cells innervated by slowly adapting Aβ-LTMRs. Other hair types also recruit distinct combinations of sensory innervation. Awl/auchene hairs are innervated by Aβ circumferential endings, as well as three molecularly distinct types of lanceolate endings that fall in to Aβ, Aδ and C classes. Finally, zigzag hairs are innervated by Aδ-LTMR lanceolate endings, Aβ-LTMR circumferential endings and C-fiber lanceolate endings. Thus, each hair follicle type is proposed to represent a distinctive mechanosensory unit (Bai et al., 2015; Li and Ginty, 2014; Li et al., 2011; Rutlin et al., 2014).

Two general models can account for how the unique innervation pattern of each hair type is achieved. First, given that hair follicles and DRG neurons are born in waves, the timing of the arrival of each neuronal subtype to the skin could correlate with the timing of hair follicle maturation. For example, zigzag hairs, the last hairs to arise on the skin surface, are innervated by a subset of afferents that excludes Aβ-LTMR lanceolate complexes. This could be due to the fact that zigzag hairs have not been specified at the time when Aβ-LTMR lanceolate endings innervate their targets. An alternative hypothesis is that each type of hair expresses a unique complement of molecular cues and that each LTMR class is programmed to express the cognate receptors. It is likely that a combination of the two strategies is employed to ensure that each hair follicle type receives an appropriate pattern of innervation. As discussed below, the search for cues that guide afferents to their target microdomains during development has begun to reveal how distinct mechanosensory organs form in skin. We first discuss lanceolate endings and Merkel cell-neurite complexes, as recent progress has uncovered key cell types and molecular mechanisms that drive their development. Current concepts about the development of other mechanoreceptors are then summarized.

Aδ-LTMR lanceolate endings, which are also called down-hairs (D-hairs; Brown and Iggo, 1967; Brown et al., 1967; Rutlin et al., 2014), form hemi-collars on the caudal side of hair follicles (Figs 1 and 2; Li et al., 2011; Rutlin et al., 2014; Wu et al., 2012). The polarized organization of Aδ-LTMR endings allows these afferents to preferentially detect rostral hair movements (Rutlin et al., 2014). Aδ-LTMRs express TrkB (Ntrk2), a neurotrophin receptor that binds brain-derived neurotrophic factor (BDNF; Klein et al., 1991), whereas keratinocytes on the caudal side of hair follicles express BDNF and participate in the selective targeting of Aδ-LTMRs to these hair follicles (Rutlin et al., 2014). Indeed, the genetic deletion of keratinocyte-derived BDNF is sufficient to disrupt the organization of lanceolate endings, suggesting that BDNF acts as a target-derived cue that is responsible for promoting the guidance of afferent endings to their proper location (Rutlin et al., 2014). In this case, individual afferents are directionally selective but rostral preference is lost, underscoring the link between appropriate development of touch-receptor morphology and function (Rutlin et al., 2014). This example sets a precedent for specialized skin cells acting to promote accurate neural targeting. Given that the neurotrophin receptor Ret is required for the formation of Aβ lanceolate terminals (Bourane et al., 2009; Luo et al., 2009), it is possible that a subset of cells in guard and awl/auchene follicles express the Ret ligands glial cell line derived neurotrophic factor (GDNF), neurturin, artemin and persephin (Airaksinen and Saarma, 2002). Whether and how specific cell types orchestrate the patterning of circumferential endings and other LTMRs around hair follicles remain open questions.

Whereas hair follicle afferents signal passive touch, Merkel cells and the afferents that innervate them are essential for active, discriminative touch. First described by Friedrich Merkel in 1875 as ‘Tastzellen’ or ‘touch cells’, Merkel cells are epidermal-derived cells that cluster in high-tactile acuity areas of vertebrate skin (Merkel, 1875). Such areas include glabrous finger pads, lips, whisker follicles and touch domes (Halata et al., 2003; Lumpkin et al., 2003; Smith, 1977). The slowly adapting Aβ-LTMR afferents that innervate Merkel cells are specialized to mediate discriminative touch by encoding object features, such as shape and texture, as well as sustained pressure (Iggo and Muir, 1969; Johnson, 2001; Maricich et al., 2012; Wellnitz et al., 2010; Woodbury and Koerber, 2007).

The rarest of the four fundamental cell types that compose the epidermis, Merkel cells have long been shrouded in mystery. Merkel cells are unique because they are the only known neuron-like cells in vertebrate skin. Recent studies have confirmed the long-debated hypothesis that Merkel cells are mechanosensory cells required for tactile sensation (Ikeda et al., 2014; Maksimovic et al., 2014; Wallis et al., 2003; Woo et al., 2014). Like most other epithelial-derived sensory cells, Merkel cells form synapse-like contacts with primary afferents (Iggo and Muir, 1969; Merkel, 1875). Merkel cells also express presynaptic molecules (e.g. piccolo and synaptotagmin I) as well as an assortment of neuronal transcription factors (e.g. Atoh1, Gfi1, Isl1 and Lhx3; Ben-Arie et al., 2000; Haeberle et al., 2004). Intriguingly, Merkel cells share histochemical and morphological features with cells found in a rare and exceedingly aggressive cancer called Merkel cell carcinoma (see Box 3). Because of the unusual properties of Merkel cells, intense efforts have aimed to understand how these cells form and how they become integrated into functional mechanosensory organs.

Merkel cell specification requires Atoh1, a helix-loop-helix transcription factor (Fig. 3). Atoh1 is an ortholog of atonal, a proneural gene required for the development of mechanosensory chordotonal organs in Drosophila (Jarman et al., 1993). Skin-specific ablation of Atoh1 totally abrogates Merkel cell specification (Morrison et al., 2009; Van Keymeulen et al., 2009). Downstream of Atoh1 activation, the transcription factor Sox2 is required to maintain Atoh1 expression (Perdigoto et al., 2014). Accordingly, the ablation of Sox2 results in a decrease in Merkel-cell number (Lesko et al., 2013). Conversely, the Polycomb complex subunits Ezh1 and Ezh2 negatively regulate Merkel cell formation by inhibiting Sox2 (Bardot et al., 2013; Perdigoto et al., 2016). Together, these findings highlight that Atoh1 and Sox2 work in concert to initiate and maintain Merkel cell fate (Fig. 3).

Merkel cells first appear in mice at E15.5 in the dorsal skin of the trunk (Ben-Arie et al., 2000; Morrison et al., 2009; Perdigoto et al., 2014), one day after the initial specification of guard hair follicles. Merkel cell specification depends on ectodysplasin-A (Eda), a tumor necrosis factor family member, and the Wnt and Sonic hedgehog (Shh) signaling pathways (Fig. 3). In Tabby mice, in which Eda signaling is disrupted, guard hair follicles and Merkel cells in touch domes fail to develop (Vielkind and Hardy, 1996; Vielkind et al., 1995). Dermal Wnt signaling activates Eda/Eda-receptor (Edar) signaling in hair follicle placodes, which in turn induces Shh expression in developing guard hair follicles (Xiao et al., 2016). This process is thought to spur expression of Atoh1 and Sox2 in nearby Merkel cell precursors (Xiao et al., 2016). Thus, a complex cascade of focal morphogen signaling presumably leads to the specific expression of Atoh1 in Merkel cell precursors (Fig. 3).

Aβ afferents begin to target Merkel cells during embryonic development (Nurse and Diamond, 1984; Vielkind et al., 1995). In Atoh1 knockout mice lacking Merkel cells, touch domes are innervated by Aβ afferents that are myelinated, have nodes of Ranvier and respond to low-threshold mechanical stimuli; however, these touch-dome afferents exhibit some striking phenotypic differences (Maksimovic et al., 2014; Maricich et al., 2009). For example, in the absence of Merkel cells, touch-dome afferents display a hyperbranched morphology (Maksimovic et al., 2014; Maricich et al., 2009; Morrison et al., 2009). They also show reduced firing rates during dynamic touch and truncation of the slowly adapting firing pattern seen in wild-type mice (Maksimovic et al., 2014). Additionally, the expression of several neurotrophin receptors, including TrkB and TrkC (Ntrk3), is decreased in large-diameter DRG cell bodies (Reed-Geaghan et al., 2016). Together, these studies suggest that Merkel cells express short-range cues that refine touch-dome afferents and influence their neuronal activity and gene expression patterns once they contact touch domes. These findings also suggest that Merkel cells are not the cellular source of long- or intermediate-range target-derived guidance cues for the afferents that contact them; instead, they mediate neuronal maturation and refinement.

Although the precise factors that are responsible for recruiting afferents to Merkel cell-enriched skin areas remain elusive, several lines of evidence point to neurotrophin signaling as playing a key role. Merkel cell afferents express TrkC, the receptor for the neurotrophin NT3 (Ntf3) (Cronk et al., 2002; Niu et al., 2014). Surprisingly, normal numbers of Merkel cell-neurite complexes are present at birth in the touch domes and glabrous skin of Ntf3 homozygous knockout animals (Airaksinen et al., 1996; Szeder et al., 2003); however, Merkel cell numbers decrease dramatically by 2-3 weeks of age (Airaksinen et al., 1996). This postnatal loss of Merkel cells also occurs in the whisker pad and is mediated by increased apoptosis (Halata et al., 2005). Merkel cell loss can be rescued by NT3 overexpression in the epidermis (Krimm et al., 2000, 2004). Thus, these findings suggest that NT3 is not required for target innervation of Merkel cell-enriched skin areas but rather plays an essential role in touch receptor maintenance. Similarly, Merkel cell-derived BDNF is not required for Aβ afferents to target touch domes; instead, Merkel cell-mediated secretion of BDNF at a specific time in embryogenesis (E16.5) influences neuronal firing patterns (Reed-Geaghan et al., 2016). It has also been shown that Merkel-cell afferents are reduced at postnatal day (P) 0 in Ret knockout mice, implicating GDNF family ligands in the embryonic development of these LTMRs (Bourane et al., 2009). These findings underscore the complex influence of neurotrophin signaling on afferent specification and maturation. Together, these studies also highlight that Merkel cells secrete BDNF and possibly other factors that influence neuronal development or activity. Uncovering the roles and downstream effects of specific neurotrophins, morphogens and transcription factors in the development of Merkel cells and LTMRs remain active areas of investigation in the field.

The formation of encapsulated endings such as Meissner corpuscles, Pacinian corpuscles and Ruffini endings has also been studied, although less is known about how these endings arise during development. During Meissner's corpuscle development, mechanoreceptive neurons innervate the glabrous skin dermal papillae, which are zones of the dermis that extend superficially into the epidermis, perpendicular to the skin surface (Cauna, 1956; Miller et al., 1958). At postnatal stages, these neurons become myelinated and their associated terminal Schwann cells transform into the lamellar cells that make up each corpuscle (Renehan and Munger, 1990; Zelená, 1994; Zelená et al., 1990). Although Meissner afferents fall into the early Ret-positive class of LTMRs, Ret is dispensable for their formation (Luo et al., 2009). By contrast, TrkB appears to be required for proper innervation of Meissner's corpuscles, as they are absent in TrkB mutant animals (González-Martínez et al., 2004; Luo et al., 2009; Perez-Pinera et al., 2008).

The mechanisms of development of Pacinian corpuscles and Ruffini endings are less well-understood, in part because these endings are rare in the mouse peripheral nervous system. Although Pacinian corpuscles are not found in mouse skin, Ret signaling is required for development of these corpuscles in mouse interosseous membranes (Luo et al., 2009; Zelená, 1981). Downstream of Ret, neuregulin 1 functions in the innervating axons of Pacinian corpuscles to trigger terminal Schwann cell maturation (Fleming et al., 2016). Ruffini endings are another touch receptor type that has rarely been located in mouse or primate skin (Honma et al., 2010; Paré et al., 2003, 2002; Rasmusson and Turnbull, 1986; Rice and Rasmusson, 2000; Ruffini, 1894). Notably, periodontal Ruffini endings are absent in TrkB mutant mice, highlighting conserved mechanisms of mechanoreceptor development across tissues (Matsuo et al., 2002).

Many developmental mechanisms are modular programs that are used in different contexts throughout evolution (Fig. 4). A prime example is the essential role of Atoh orthologs in cell fate determination in various types of mechanosensory cells, including Merkel cells, Drosophila chordotonal organs and vertebrate inner ear hair cells (see Glossary, Box 2; Ben-Arie et al., 2000; Ben-Arie et al., 1996; Bermingham et al., 1999). As we discuss below, this and other similarities between Merkel cells and other sensory cell types can provide some insight into the development of mechanoreceptors.

Hair cells derive their name from bundles of microvilli-derived stereocilia protruding from their apical surfaces. When these tufts of stereocilia are deflected, hair cells translate sound and vestibular energy into electrical signals that trigger neurotransmitter release onto sensory afferents (Hudspeth and Jacobs, 1979). Hair cells are present in the inner ear and, in aquatic vertebrates, in the lateral line (see Glossary, Box 2). Merkel cells similarly possess apical microvilli (Fig. 4A; Sekerková et al., 2004; Sekerková et al., 2006). Like hair cells, Merkel cells are exquisitely mechanically sensitive (Ikeda et al., 2014; Maksimovic et al., 2014; Woo et al., 2014). Furthermore, both cell types derive from epithelial precursors and require Atoh1 and Sox2 for their specification (Fig. 4B; Bermingham et al., 1999; Dabdoub et al., 2008; Driver et al., 2013; Kiernan et al., 2005; Maricich et al., 2009; Morrison et al., 2009; Neves et al., 2013; Van Keymeulen et al., 2009).

Taste cells (see Glossary, Box 2) also bear similarities with Merkel cells. For example, both cell types arise from stratified epithelia but express keratin 8, a marker of simple epithelia (Fig. 4B). Both also express BDNF, which appears to influence the density and physiological hallmarks of their sensory afferents (Huang et al., 2015; Meng et al., 2015). Finally, like hair cells and Merkel cells, taste cells require Sox2 for proper development (Okubo et al., 2006).

Both hair cells and taste cells develop from placodes (Fig. 4C; Hall et al., 1999; Whitfield, 2015), which are thickened patches of specialized epithelial cells. Placodes also give rise to skin appendages (see Glossary, Box 2), such as hair follicles, sebaceous glands and feathers (Biggs and Mikkola, 2014; Piotrowski and Baker, 2014; Schlosser, 2005; Streit, 2004). The otic placode gives rise to the otic vesicle, which is the source of all inner ear cell types. Similarly, hair cells of zebrafish lateral line neuromasts develop from placodes that migrate along the body axis, depositing precursor cells along the way (Agarwala et al., 2015; Dambly-Chaudière et al., 2003; Metcalfe et al., 1985). Based on the similarities between Merkel cells, taste cells and hair cells, we hypothesize that Merkel cells are also placode-derived cells. The possibility that Merkel cells derive from placodes, in particular hair-follicle placodes, is suggested by several observations. First, Merkel cells initially appear both within and adjacent to the hair peg of guard hair follicles (Fig. 2; Ben-Arie et al., 2000; Morrison et al., 2009; Perdigoto et al., 2014; Vielkind et al., 1995). Second, in mature mice, several hair types contain Merkel cells within their follicles, including whiskers, hairs on the distal limb and a subset of guard hairs (Li et al., 2011; McIlwrath et al., 2007; Rice et al., 1997; Vielkind et al., 1995). Third, polycomb mutations cause Merkel cells to develop within all hair follicle subtypes (Perdigoto et al., 2016).

However, two recent findings throw a wrench in the simple model that Merkel cells derive from hair-follicle placodes. One line of evidence comes from analysis of Shh^Cre;Rosa26^YFP mice, in which all hair follicle-derived cells are marked in a Cre-dependent manner (Levy et al., 2005). This analysis showed that, although Merkel cells arise concurrently with the appearance of developing guard hairs, only 11% of Merkel cells express YFP in Shh^Cre;Rosa26^YFP mice, indicating that they are largely derived from a lineage that is distinct from the hair follicle (Xiao et al., 2016). Second, interfollicular Merkel cells are found in mice harboring an Fgf20 null mutation, which disrupts guard hair development (Xiao et al., 2016). These results indicate that Merkel cells in touch domes are not hair follicle derived. NCAM-expressing cells that localize to hair follicles have been proposed to serve as Merkel cell progenitors (Xiao et al., 2016). Alternatively, an as-yet-identified ‘touch-dome placode’ might give rise to interfollicular Merkel cells during development.

Another open question about Merkel cell progenitors is whether they are responsible for selectively recruiting touch-dome afferents. The lateral line system might provide some clues into this. The sensory neurons that innervate lateral line neuromasts target hair cells during development such that one subset of neurons encodes rostral hair-bundle movement and the other reports caudal hair-bundle movement. Thus, each set of neurons must make specific connections with appropriate hair cells. Live imaging has revealed that newly specified hair cells extend processes down into the deeper layers of skin to form physical contacts with developing neuronal endings (Dow et al., 2015). This result highlights an active role for hair cells themselves during the targeting process. Although Merkel cells are not required to recruit afferents to touch domes (Maricich et al., 2009), Merkel cell precursors or placode cells could actively recruit afferent terminals by similar contact-mediated mechanisms once the afferents have reached the vicinity. The live-imaging techniques and genetic mouse models required to address this question are on the horizon.

Evidence from invertebrate sensory structures also provides hints for how Merkel-cell afferents and other LTMRs might selectively target skin areas during development. With the use of powerful genetic tools, some signaling pathways that mediate somatosensory axon morphology have been dissected in Caenorabditis elegans. For example, it has been shown that menorin (MNR-1) and SAX-7 are required for the development of skin-targeted PVD neurons (Dong et al., 2013; Salzberg et al., 2013). PVD neurons extend processes along the length of the worm body wall that then branch and form a ‘menorah’ pattern (Dong et al., 2013; Salzberg et al., 2013). MNR-1 and SAX-7 influence neuronal morphology by forming a complex with the transmembrane receptor DMA-1. A series of expression analyses and epistasis experiments revealed that MNR-1 and SAX-7 are expressed by the hypodermis and that their combined expression is required for the establishment of the canonical menorah pattern. This evidence supports the hypothesis that there is a ‘combinatorial code’ of ligands and receptors that is required for the development of tactile afferents – a more complex model than a one-to-one pairing of neurotrophins and receptors. These findings also set the stage for investigation of vertebrate homologs of these proteins, which include Fam151 family members (Ruiz-Trillo et al., 2007; Salzberg et al., 2013), in the development of mammalian somatosensory neurons. This is an especially intriguing possibility for the development of morphologically similar lanceolate terminals, which also form fence- or menorah-like projections around hair follicles. Interestingly, lanceolate endings engage in close interactions with terminal Schwann cells, which are required for maintenance and regeneration of their fence-shaped terminals (Li and Ginty, 2014; Woo et al., 2012). The ligands and receptors involved in these interactions remain unknown. It will be interesting to examine whether Schwann cells mediate afferent targeting for Merkel-cell afferents, lanceolate endings or other mammalian touch receptors.

Development of the Drosophila melanogaster somatosensory system has been extensively studied and can also provide clues into vertebrate mechanoreceptor development. Indeed, a number of homologs of key molecular effectors of Drosophila somatosensory development have been implicated in mammals. For example, Frizzled mutations were first described in Drosophila, where they disrupt the planar polarity of sensory bristles and trichones, which are microscopic hair-like structures that decorate the wing (Fig. 5; Gubb and García-Bellido, 1982; Vinson and Adler, 1987). In mice, epidermal but not Merkel cell-derived frizzled 6 (Fzd6) regulates the planar organization of hair follicles and Merkel cells during development (Fig. 5; Chang and Nathans, 2013; Chang et al., 2016; Guo et al., 2004; Hua et al., 2014). In mature mice, hair shafts typically emerge from the skin in the rostral-to-caudal direction along the trunk of a mouse, and Merkel cell clusters form crescent shapes around the caudal mouth of guard-hair follicles. In Fzd6 mutant mice, both patterns are disrupted developmentally, which suggests that the same mechanisms that mediate polarity of hair follicles direct the clustering of Merkel cells on the caudal side of hair follicles (Chang and Nathans, 2013; Chang et al., 2016; Guo et al., 2004; Hua et al., 2014). Interestingly, members of the Wnt signaling family and planar cell polarity pathways (e.g. frizzled, Vangl and Celsr) are also involved in patterning the orientation of stereociliary bundles of inner ear hair cells (Fig. 5; Dabdoub et al., 2003; Montcouquiol et al., 2003; Wang et al., 2005; Wang et al., 2006). Given that Merkel cells and hair cells share many features, it is intriguing to speculate how planar cell polarity genes might play a role in Merkel cell organization and function.

Cutaneous mechanoreceptors are diverse, highly specialized and innervate distinct sensory structures. Thus, targeted innervation of the skin requires precision and accuracy. Although the cellular and molecular mechanisms responsible for recruiting afferents to specific areas of the skin are still largely elusive, recent studies of mouse hairy skin have begun to reveal basic principles of somatosensory development. These studies have highlighted that epithelial cells and placodes, which are crucial in other sensory systems for cell specification and innervation, should be investigated for their role in touch receptor development. This role is particularly interesting in the case of Merkel cell-enriched skin areas, such as touch domes, fingertips, lips and whisker follicles. In addition, given that human and non-human primates primarily use fingertips to actively explore tactile features of objects, future studies directed toward glabrous skin microdomain formation might shed light on the development and maintenance of touch receptors that are especially useful for active touch in humans.

The recent explosion of progress in the field of somatosensation has answered long-standing questions about mechanisms that govern development and function of touch receptors. In particular, this progress has highlighted that mammalian touch-receptor development research will benefit from studies of other model systems. Another avenue for the exploration and generation of new hypotheses will be to interrogate directly the link between defects in tactile sensitivity and abnormal cognitive development. The results of future studies have the power to reveal new methods for enhancing touch, for understanding the links between ASD and tactile processing, and for restoring tactile sensitivity lost after injury or through aging.

We thank Drs David Owens and Yanne Doucet for assistance with conception of this review, members of the Lumpkin laboratory for helpful discussions, and Drs Theanne Griffith, Yalda Moayedi and Cynthia Crawford for providing valuable comments on the manuscript.