In Vitro Sensitive Skin Models: Review of the Standard Methods and Introduction to a New Disruptive Technology

: The skin is a protective organ, able to decode a wide range of tactile, thermal, or noxious stimuli. Some of the sensors belonging to the transient receptor potential (TRP) family, for example, TRPV1, can elicit capsaicin-induced heat pain or histamine-induced itching sensations. The sensory nerve ﬁbers, whose soma is located in the trigeminal or the dorsal root ganglia, are able to carry signals from the skin’s sensory receptors toward the brain via the spinal cord. In some cases, in response to environmental factors, nerve endings might be hyper activated, leading to a sensitive skin syndrome (SSS). SSS affects about 50% of the population and is correlated with small-ﬁber neuropathies resulting in neuropathic pain. Thus, for cosmetical and pharmaceutical industries developing SSS treatments, the selection of relevant and predictive in vitro models is essential. In this article, we reviewed the different in vitro models developed for the assessment of skin and neuron interactions. In a second part, we presented the advantages of microﬂuidic devices and organ-on-chip models, with a focus on the ﬁrst model we developed in this context.


Introduction
Skin is a protective organ that decodes a wide range of tactile, thermal, and noxious stimuli. The perception of sensory stimuli involves various sensory receptors, such as mechanoreceptors (mechanical stimuli), thermoreceptors (warm or cold temperature), nociceptors (noxious stimuli) [1], and pruriceptors (itching sensations) [2]. All these receptors contain the axonal endings of sensory neurons, which can be divided into three types depending on their diameter and speed of impulse. The highest velocity is found in Aβ fibers, which are the biggest and are surrounded by several myelin sheaths. These nerve fibers are connected to hair follicles and mechanoreceptors (Merkel disks, Meissner, Pacinian, and Ruffini corpuscles) [3]. The small fibers are either surrounded by a thin myelin sheath that is lost in the dermis (Aδ fibers) or unmyelinated (C fibers), leading to slower nerve conduction, especially in C-fibers [1]. Their free nerve endings, located in the dermis (Aδ fibers) or the epidermis (C fibers), also largely innervate hair follicles [4]. They are both involved in temperature, pain [1], and itching sensations [5,6]. Some C-fibers also detect gentle mechanical stimuli, especially in skin on the face [7]. Both C and Aδ free nerve endings can detect a large panel of physical and chemical stimuli [1], being either specifically activated by mechanical, thermal, or chemical stimuli (unimodal fibers) or responding to several stimuli (polymodal fibers) [8]. Their sensitivity depends on the properties of their membrane sensors and associated proteins. The sensors mostly belong to the transient receptor potential (TRP) family and can be studied using various agonists and antagonists. For example, TRPV1 can elicit capsaicin-induced heat pain [9] or a histamine-induced itching sensation [10], while TRPV4 is known as a heat sensor [11]. The menthol-activated receptor TRPM8 is responsible for the detection of innocuous and noxious cooling [8,12], and the wasabi-activated receptor TRPA1 is involved in a wide panel of stimuli, including chemo-nociception [12], thermal sensations [13,14], and itch [14].
All these sensory nerve fibers, for whom the cell body is located in the trigeminal or the dorsal root ganglia (DRG), can carry signals from skin sensory receptors toward the appropriate integration center of the brain via the spinal cord [8,15]. Besides this afferent sensory information transfer, some stimulated nociceptive C-fibers, called peptidergic C-fibers, can produce and release various neuropeptides, such as substance P and calcitonin-gene-related peptide (CGRP) at the skin level [16], thus provoking vasodilation and neurogenic inflammation [12]. To a lesser extent, this is also true for small myelinated Aδ fibers [1].
Sensitive skin syndrome (SSS) has been defined as "the occurrence of unpleasant sensations (stinging, burning, pain, pruritus, and tingling sensations) in response to various stimuli that normally should not provoke such sensations" [17]. A significant correlation between SSS and small-fiber neuropathies, especially those affecting unmyelinated C-fibers, has been established [18]. A decrease in the heat-pain threshold detection is currently considered to be responsible for the hyper-reactivity of nerve endings in response to environmental factors and the induction of neuropathic pain [11]. This detection threshold directly depends on the sensor's presence and properties. TRPV1 was found to be significantly increased in hypersensitive skin after nerve injury and repair [19]. TRPV1 also seems to play a key role in heat hyperalgesia following burn or inflammation [8], while TRPM8 and TRPA1 are involved in injury-evoked cold hypersensitivity [12]. TRPV1, TRPV4, and TRPA1 also contribute to mechanical hyperalgesia [8]. Studying the sensory perception mechanisms of sensory neurons is, therefore, essential for the better diagnosis, prevention, and treatment of sensitive skin. Nevertheless, some studies revealed that keratinocytes also express diverse sensory receptors, such as TRPV1, TRPV3, and TRPV4 [8,11,20]. Mainly expressed in keratinocytes, TRPV3 can stimulate CGRP secretion, which induces local vasodilation that plays a role in body thermoregulation [21]. The selective activation of the keratinocyte-located TRPV1 and TRPV4 sensors is sufficient to induce pain and itch, respectively [11]. It has been shown that the activation of keratinocyte TRPV1 elicits a calcium signal, leading to intraepidermal nerve-ending activation through synaptic-like contacts. Interestingly, the density of these synaptic-like contacts between keratinocytes and sensory neurons was found to be significantly higher in patients with sensitive skin compared to unsensitive controls [22]. Thus, sensitive skin not only depends on the sensory neurons' functionality but also on their physiological interactions with keratinocytes through those synaptic-like contacts.
As sensitive skin affects about 50% of the population [23], it is essential to find efficient active molecules and, thereby, develop relevant in vitro models for the cosmetical and pharmaceutical industries.

Sensory Neuron Culture and Coculture
Skin nerve sensitivity relies primarily on sensory neurons. Thus, many teams have focused on 2D cultures of sensory neurons to better understand their function in correlation with skin sensitivity [24][25][26]. As the compositions and properties of primary neurons are much more similar to those of sensory neurons in tissue than those of immortalized cell lines [26], most in vitro models were developed using DRG sensory neurons. Such models have successfully demonstrated that histamine can induce itching by activating TRPV1 [10], which sheds light on the molecular mechanisms that explain how IL-31 secreted by T helper cells induces itching by activating a small population of TRPV1+/TRPA1+ sensory neurons [27] or how keratinocyte-secreted TSLP elicits itching through TRPA1 in atopic dermatitis [28].

2D model
Animal [29] Tri-compartmented culture with (1) rat spinal cords cells; (2) rat DRG; (3) epidermal cells (keratinocytes, Merkel cells and melanocytes) Hot medium in the epidermal/axonal compartment: Electrophysiological recording in the soma compartment by patch clamp showed spikes registered at 37 and 45 • C but not at 22 • C. [44] Bi-compartmented culture with (1) rat DRG neurons; (2) rat primary epidermal keratinocytes Unlike skin cells, which are easy to obtain from human skin biopsies, human primary neurons are difficult to collect for practical and ethical reasons, and they do not proliferate in vitro [45,46]. For this reason, most of the models were created by using human skin cells in association with rat [32,37,40], pig [33,34,38], or mouse DRG neurons [35,36], which are easier to obtain. Nevertheless, for regulatory reasons, pig and rodent DRG neurons cannot be used for cosmetic testing. Moreover, it has been shown that neurons from rodent DRG differ in their receptor types or expression levels, genes, and signaling pathways compared to human DRG neurons [46], which makes them less relevant than human neurons for use in innervated human skin models.
The disruptive technology that creates induced pluripotent stem cells (iPSCs) [47] has allowed us to further obtain human sensory neurons derived from human iPSCs. Depending on the culture condition, human iPSCs can be differentiated into various neuron-like cells sharing characteristics with sensory neurons such as nociceptors, mechanoreceptors, and proprioceptors [48,49]. These human sensory neurons derived from iPSCs can produce substance P or CGRP under capsaicin activation and have been successfully included in innervated or reinnervated skin models [31,39]. They have been used, for example, to demonstrate the positive effect of an active ingredient on sensitive skin [39].
To better mimic the anatomical distance and respective micro-environment between, on one side, neuronal cell bodies (somas) and, on the other side, axonal endings and skin cells, these cell types should be seeded and cultivated in distinct compartments. These compartments can be connected by microchannels [29,33] to study the behavior of a specific cell type. Thanks to the microchannels, Chateau et al. established a functional synapticlike connection between two compartments containing neurons and skin cells [29]. With this model, they demonstrated the possibility of activating the soma by only applying heat-stimulation in the axon-keratinocyte compartment.
Considering the key role of small sensory fibers and TRPV1, sensitive skin is usually mimicked by activating TRPV1 in an innervated skin model. TRPV1 can be directly activated by temperatures above 43 • C [9], pH below 6 [52], endogenous molecules such as anandamide [53], and exogenous agonists such as capsaicin (the main pungent component in hot chili peppers) or its analog, resiniferatoxin [9,54,55]. Many other molecules, including histamine, indirectly activate TRPV1 [10,56]. As TRPV1 is a non-selective calcium channel, this results in calcium influx, leading to neuron depolarization, which is responsible for both the afferent signal genesis and the local secretion of neuropeptides [57]. To validate innervated skin functionality and/or mimic sensitive skin, several teams have, therefore, evaluated the reactivity of their model to a temperature increase from 22 to 37 or 45 • C [29], capsaicin added in the medium [31,35,39] or topically applied [37,40], and/or histamine added in the medium [39] (Table 1). Active ingredients have been considered to have positive effects when they could reduce the temperature-, histamine-, or capsaicin-induced activation of innervated skin [39].
Based on the review of the existing models (Table 1), it seems that the ideal model to mimic sensitive skin and activate the TRPV1 pathway would be a compartmentalized model in which neurons and skin cells are independently cultured but interconnected by microchannels, with both cell types being derived from humans. In addition, for the drug screening or toxicity assessment, this model should also be standardized to ensure reproducibility and large-scale production.

Introduction to Microfluidic Devices and Organ-on-Chip Models
The organ-on-chip (OoC) model consists of an in vitro microfluidic platform with micrometer-sized chambers that allow the combination of different cell types in a 3D structure to recreate the functionality of an organ close to that of a human. The dynamic culture of cells inside and their interaction allow the characterization of the organ's properties, its interaction with other tissues, and its behavior in certain physiological and pathological conditions. These models aim to replace animal testing and be more predictive, while, with miniaturization, they could reduce the costs of experimental setups by reducing the use of tissue or specific cells, especially non-proliferative cells (e.g., iPSC-derived neurons), medium, growth factors, and drugs. Another advantage of microfluidic devices is the control of the interaction between cells or tissues with the possibility to use microchannels, which enables the interconnection between independent compartments-for example, to isolate the soma of a neuron to its axonal endings [58]. The fluidic isolation created by the microchannels is of great interest to limit diffusion from one compartment to another, which allows the use of different optimal culture media for each cell type, the selective application of stimuli (e.g., actives) in a compartment, or the sampling of the supernatant in a specific compartment. Microfluidic devices usually consist of microchannels to guide neurites' growth and membranes to create cellular barriers or air-liquid interfaces. Many applications and new perspectives arise from microfluidic device technologies, such as functional studies, mechanistic studies, drug screening, toxicology, and pharmacokinetics.
In the past decade, there has been an increasing number of reports published about the development of microfluidic devices for testing various organs, such as the liver, kidney, brain, intestine, and heart [59]. The development of an OoC involving skin is more recent and essentially consists of using microfluidic technology, channels, and porous membranes to reconstruct the different layers of the skin [60,61]. These first models may include endothelial cells to study drug toxicity, inflammation [62], skin permeability, or the immune response to the presence of bacteria on the skin [63]. Few studies have focused on the coculture of skin with other organs, such as the liver [64][65][66][67][68], intestine [66], kidney [66], and hair [69]. The main objectives were to observe cell viability or to study the crosstalk between the skin, liver, or kidneys for drug toxicity.
In recent years, the interaction between cutaneous nerves, the neuroendocrine axis, and the immune system has been established. It has been found that neurocutaneous interactions influence various physiological and pathophysiological functions, including cell growth, immunity, inflammation, pruritus, and wound healing [70,71]. Thus, compartmentalized microfluidic devices represent a relevant model to assess the interaction of these cells' networks; however, to our knowledge, only one article reported the development of a real microfluidic device to assess the interaction between skin and neurons. Tsantoulas et al. cocultured keratinocytes and neurons in a microfluidic model and assessed whether their coculture interacted with nerve activation induced by capsaicin [44]. The results were promising; however, the main limitation was the use of animal cells, which limited the transferability of the results to humans.

Interest and Objectives
Psoriasis, atopic dermatitis, eczema, wound healing, skin irritation, herpes infection, hyperhidrosis, neuropathy, and skin sensitivity syndrome are conditions involving both skin cells, i.e., keratinocytes and neurons. Preclinical research is limited by in vitro models, which are not always relevant either because they often use rodent neurons cells or because the cocultures of skin cells and neurons do not fully recapitulate the anatomical structure, i.e., a neuronal cell body compartmentalized from the skin. The association between compartmentalized microfluidic device technology and human-derived sensory neurons may address these limitations. The interest in such a model is double: 1. to evaluate the impact on neurons of a stimulus applied on the skin, and 2. to evaluate the impact on skin of a stimulus applied on neurons. Stimuli are various and might be physical (UV, mechanical forces, electrical stimulation of neurons), chemical (toxins or drugs), or molecular (capsaicin, oxytocin, substance P, histamine, bradykinin, etc.). In the field of dermatology and cosmetics, the applications are varied and include, for example, drug efficacy screening, pharmacokinetic studies, skin penetration and drug diffusion, drugdrug interaction, and toxicological studies (sensitization, irritation, phototoxicity). This model is even more relevant in cosmetics that avoid animal testing by using human cells in vitro.

Design of an Innervated Skin-on-Chip
OoC models developed by Netri™ are versatile technologies that allow the 3D coculture of different cell types in isolated compartments interconnected by microchannels, which allow the unidirectional or bidirectional growth of neurites. Thus, the chip's architecture is tailored according to the objective of the trial. These devices are made with polydimethylsiloxane (PDMS), a biocompatible silicon material that has several advantages: 1. Its transparency allows the use of standard analytical methods, such as optical, immunefluorescence, or confocal microscopies. 2. It is permeable to gas, and thus, no pump or mechanical stirrer is needed, and the culture is compatible with standard incubators. 3. Its suppleness allows biopsy and standard histology. These associated features contribute to the long-term culture and survival, up to several months.
Thanks to the microfluidic technology and compartmentalization, we developed a microfluidic model including human sensory neurons and keratinocytes (Figure 1). The device consisted of two isolated channels linked by microchannels in which only axons can pass through over dendrites. Such a configuration allows the creation of a selective barrier while also connecting the skin compartment with keratinocytes and the spinal cord compartment with the neuronal cell body. Human-iPSC-derived sensory neurons were seeded in one channel, and human primary keratinocytes were seeded in the other. Cell seeding density and timing between the two cell types were optimized to ensure proper cell maturation and proliferation. Thanks to compartmentalization and fluidic isolation, cells were cultivated with their respective culture medium, allowing optimal growth. This model can be used to mimic sensitive skin, for example, by adding capsaicin to axonal endings and assessing its effect on either the skin compartment or the neuronal cell body.

From Standard to Innovative Specific Readouts
The versatility of the OoC developed by Netri™ allows the use of a large variety of readouts. Cells and tissues might be visualized by optical, immunofluorescence, or confocal microscopy, and their viability is assessed using standard methods (MTT, live/dead assay, etc.). The real-time follow up of cell migration is also possible. Moreover, collecting the cells or the supernatant allows the dosage of molecules, mediators, or biomarkers to be determined by standard methods such as ELISA, PCR, etc. PDMS suppleness allows slicing and standard histology on every compartment, either skin cell or neuron. In addition to these standard methods, neural activity might also be specifically evaluated. Calcium imaging is a common method and enables the visualization of the activity of hundreds of individual neurons simultaneously using fluorescent activity sensors. Changes in fluorescence indicate fluctuations in intracellular calcium, which is an indirect indicator of neural activity. This technique may induce long-term cytotoxicity and disruption of cell behavior. In addition, the field of view of the sample is limited. On the other hand, a micro-electrode array (MEA) has more advantages because, being not cytotoxic, it allows a repeated and continuous electrophysiological recording of functional activity. The MEA is an array of microscopic electrodes distributed over a small surface area that provides a macroscopic view of neural networks by recording the electrical activity of brain slices, dissociated tissues, organoids, spheroids, or the 2D neuronal cell culture. An MEA records

From Standard to Innovative Specific Readouts
The versatility of the OoC developed by Netri™ allows the use of a large variety of readouts. Cells and tissues might be visualized by optical, immunofluorescence, or confocal microscopy, and their viability is assessed using standard methods (MTT, live/dead assay, etc.). The real-time follow up of cell migration is also possible. Moreover, collecting the cells or the supernatant allows the dosage of molecules, mediators, or biomarkers to be determined by standard methods such as ELISA, PCR, etc. PDMS suppleness allows slicing and standard histology on every compartment, either skin cell or neuron. In addition to these standard methods, neural activity might also be specifically evaluated. Calcium imaging is a common method and enables the visualization of the activity of hundreds of individual neurons simultaneously using fluorescent activity sensors. Changes in fluorescence indicate fluctuations in intracellular calcium, which is an indirect indicator of neural activity. This technique may induce long-term cytotoxicity and disruption of cell behavior. In addition, the field of view of the sample is limited. On the other hand, a micro-electrode array (MEA) has more advantages because, being not cytotoxic, it allows a repeated and continuous electrophysiological recording of functional activity. The MEA is an array of microscopic electrodes distributed over a small surface area that provides a macroscopic view of neural networks by recording the electrical activity of brain slices, dissociated tissues, organoids, spheroids, or the 2D neuronal cell culture. An MEA records the field potential electrical activity from the extracellular space of a population of neurons, whereas patch-clamp electrophysiology records the action potential of electrical activity from the intracellular space of a single neuron. MEA technology associated with compartmentalized microfluidic devices allows the monitoring of the electrical shift of neuron activity induced by skin cells.

Future Perspectives
Current limitations in traditional in vitro skin models are the correlation between skin cell culture or reconstructed skin and the clinical conditions. Our future development involves introducing an air-liquid interface to recreate conditions close to in vivo to enable the differentiation of the stratum corneum and the topical application of products or microbiota on the skin. To study specific pathological conditions, such as psoriasis or eczema, a patient's skin biopsy might be directly inserted into the chip and colonized by neurites of human sensory neurons, as Lebonvallet et al. have previously demonstrated [40].
Intense noxious stimuli and tissue inflammation produce a pain hypersensitivity resulting from peripheral and central sensitization. This hypersensitivity of the receptor might be explained by the fact that the activation threshold of TRPV1 might be lowered by a pro-inflammatory environment [72]. Thus, to better mimic sensitive skin and TRPV1 hyperexcitability, pro-inflammatory cytokines might be added into the OoC.
As recently popularized, the skin, gut, and microbiota are closely linked, and their interactions directly influence both normal physiology and disease processes, such as atopic dermatitis [28,33,34] or psoriasis. Therefore, a multi-organ-on-chip combining these entities seems relevant to better understand their interaction and develop new treatments.
To go one step further in the predictability of these models, the future could involve a personalized and innervated skin-on-chip. From the subject's skin biopsy, it would be possible to derive neurons that can be associated together with the subject's skin cells into a unique and personalized chip. This personalization is especially relevant in rare diseases or in fragile populations such as the pediatric population, pregnant women, the elderly, or immunocompromised patients.

Conclusions
The prevalence of sensitive skin varies according to gender, ethnicity, and environmental factors (e.g., climate, exposures to products and chemicals, and cultural influences), but it is generally agreed that this condition affects a substantial portion of the population. Recently, the importance of skin and nerve interaction was highlighted by the 2021 Nobel Prize in physiology. Indeed, David Julius and Ardem Patapoutian shared the Nobel Prize for the identification of receptors that enable cells to detect temperature and touch. Dr. Julius worked on TRPV1 and capsaicin, the latter of which is currently considered to be the reference marker of sensitive skin.
Several in vitro models have been developed to assess the interaction between the skin and neurons, essentially to mimic sensitive skin. Most of them consisted of the coculture of both cell types in the same system. However, to better reproduce the physiological distance between neuron cell bodies (soma) and skin cells, these cell types should be cultivated in distinct compartments. Thus, compartmentalized microfluidic devices such as the OoC represent a relevant alternative.
In the next few years, it can be envisioned that animal studies might be at least in part replaced by OoC technology, leading to more reliable results close to clinical studies. Progress in iPSC manipulation during these last years is a key element that may contribute to the development of all-human in vitro models. In dermatology and especially in cosmetology, the combination of compartmentalized microfluidic technology and iPSC-derived human sensory neurons to create a standardized all-human in vitro model of innervated skin makes sense. In addition, nociception, itch, or inflammation might be recreated and controlled in these models by adding specific modulators such as capsaicin, menthol, cytokines, and drugs. Furthermore, besides the standard readouts, microfluidic devices can be coupled to MEA technology to allow the recording of the electrical signals' of the soma after axonal molecular stimulation.
Innervated skin-on-a-chip represents several advantages at the different stages of drug or cosmetic development. The model is expected to reduce R&D costs by avoiding animal costs through reducing the sample size and duration of the trial. All in all, these predictive preclinical models aim to better discriminate new drug candidates and, thus, to reduce the risk of clinical study failures (phases I and II).