Derivation and analysis of human somatic sensory neuron subtypes facilitated through fluorescent hPSC reporters

Summary

Peripheral sensory neuropathy (PSN) is associated with several devastating neurological conditions, yet effective strategies to prevent or alleviate the consequences of PSN are nearly non-existent. A major challenge in the development of better therapeutic interventions is the lack of appropriate human model systems. Human induced pluripotent stem cell (hiPSC)-derived somatosensory neurons present a promising strategy to overcome this issue but remain of limited translational utility, in part due to the low efficiency and lack of sensory subtype selectivity of the existing sensory neuron derivation protocols. To improve upon iPSC-based somatosensory disease models, we here describe the generation and validation of a genetic toolset to fluorescently label all or distinct (nociceptor, low threshold mechanoreceptor, and proprioceptor) somatosensory subtypes. These new resources will be transformative for hPSC-based approaches in PSN disease modeling—a critical step toward translating new findings into clinically relevant therapeutic strategies.

Highlights

• •An AVIL:tdT reporter permits targeted analyses of hPSC-derived sensory neurons
• •Reporters for NTRK1, MAFA, and RUNX3 facilitate hPSC derivation of sensory subtypes
• •Protocols for the generation of human nociceptors and mechanoreceptors
• •Use of machine learning in determining sensory neuron differentiation efficiency

Peripheral sensory neuropathy (PSN) has devastating consequences for quality of life, yet effective strategies to prevent or alleviate this neurological condition are non-existent. Here, de Nooij and colleagues report the generation of hPSC reporter lines that facilitate the derivation and analysis of human nociceptor, mechanoreceptor, or proprioceptor sensory subtypes. These resources are transformative for hPSC-based approaches in PSN disease modeling.

Introduction

Peripheral sensory neuropathy (PSN) is a neurological condition caused by heritable or idiopathic metabolic disease, exposure to toxic substances (including chemotherapeutic agents), or physical nerve damage (Gomatos and Rehman 2022). PSN patients suffer from chronic pain, paresthesia, or impaired touch sensation or sensory-motor control—conditions that can severely impact quality of life and often result in disability. Strategies to prevent or alleviate the consequences of PSN are essentially non-existent. In part, this is due to the heterogeneous nature of the trigeminal and dorsal root ganglion neurons that provide somatosensory feedback (Usoskin et al., 2015; Nguyen et al., 2017; 2021; Bhuiyan et al., 2024). In addition, in both preclinical mouse models and post-mortem human dorsal root ganglia (DRG) tissue, individual sensory subtypes are often too scarce for rigorous analyses. Together, the neuronal heterogeneity and the limited experimental material make it difficult to examine the mechanistic basis of disease or to systematically evaluate potential interventions. A second major issue is the difficulty in translating animal studies into effective therapeutics for human subjects. This problem has been attributed to the differences in gene expression profiles between mouse and human somatosensory subtypes (Han et al., 2015; Nguyen et al., 2021). These observations have emphasized the need for better, and in particular, human-based pre-clinical model systems for sensory neuropathies.

Despite isolated successes, the translational utility of in vitro-derived somatosensory neurons has remained limited. The reasons for this are severalfold. First, existing strategies to derive or transdifferentiate sensory neurons from induced pluripotent stem cells (iPSCs) or fibroblasts, respectively, often do not yield enough mature sensory neurons to permit the screening of new therapeutic compounds at an industrial level (Chambers et al., 2012; Blanchard et al., 2015). Second, although neuropathies often show pathological consequences in select sensory subtypes, most sensory differentiation protocols result in a mixture of sensory neurons (Chambers et al., 2012; Blanchard et al., 2015; Saito-Diaz et al., 2021; Deng et al., 2023). While this perhaps better represents the normal physiological condition, it complicates the phenotypic analysis of specific sensory subtypes. A lack of validated immunological reagents that reliably distinguishes between human sensory neuron subtypes further compounds this problem. Some of these difficulties have in part been addressed by newer differentiation strategies that rely on the transient expression of transcription factors in fibroblasts or iPSCs, often in combination with the addition of specific cytokines, to promote a generic or subtype selective sensory neuron identity (Tsunemoto et al., 2018; Nickolls et al., 2020; Dionisi et al., 2020; Schrenk-Siemens et al., 2022; Hulme et al., 2024). While these induced sensory neurons offer some advantages, they present a limitation for sensory developmental disorders, including neurocristopathies, given that they bypass early developmental stages. Another drawback of such models is the need for doxycycline to induce the transcription factors. Given that doxycycline can adversely affect cell proliferation and mitochondrial function, the use of this agent may inadvertently modify disease phenotypes (Luger et al., 2018; De Boeck and Verfaillie 2021).

To address the limitations in existing iPSC-based somatosensory disease models, and to facilitate the optimization of sensory neuron differentiation protocols that best resemble their nascent developmental trajectory, we here describe the generation and validation of a genetic reporter toolset for human embryonic stem cell (hESC)/iPSC-derived sensory neurons. These reporters include an AVIL:tdTomato (AVIL:tdT) reporter, designed to mark all sensory neurons, and NTRK1:tdT, MAFA:tdT, and RUNX3:tdT reporters, to label nociceptive/thermoceptive, low-threshold mechanoreceptive, and proprioceptive lineages, respectively. NTRK1 (encoding the TRKA protein), MAFA, and RUNX3 represent the signature molecular markers of these three sensory lineages and are expressed in a non-overlapping pattern in DRG. In contrast, the AVIL gene, encoding ADVILLIN, is expressed in all DRG neurons and thus serves as a pan sensory marker. The reporter lines can be used in a variety of applications, ranging from optimization of differentiation protocols and visualization of the neurons in multiplex organoids (innervated skin or bone), to isolating the specific sensory subsets using fluorescence-activated cell sorting (FACS) for detailed molecular or biochemical analyses. Thus, our collection of hPSC sensory reporter lines offers a versatile new toolkit to optimize sensory neuron differentiation strategies and to permit systematic phenotypic investigations of select sensory neuron subtypes under pathological conditions.

Results

Generation of a generic sensory neuron AVIL:tdTomato hESC-reporter line

Somatic sensory neurons that reside in spinal DRG all derive from the neural crest, a transient progenitor population that delaminates from the dorsalmost aspect of the neural tube (Simões-Costa and Bronner, 2015). The proximity of the neural tube and DRG means that the neurons in these two tissues are exposed to many of the same external growth factors. This similarity is reflected in in vitro protocols that aim to generate dorsal spinal or sensory neurons and that rely on many of the same signaling molecules (Chambers et al., 2012; Saito-Diaz et al., 2021; Duval et al., 2019; Gupta et al., 2022). In addition, early sensory progenitors express several transcription factors also found in developing spinal neuron subsets (e.g., ISLET1, POU4F11, NGN1, and NGN2) (Ma et al., 1999; Dykes et al., 2011; Andersen et al., 2023). To help distinguish between spinal and sensory lineages during the in vitro differentiation of somatic sensory neurons, we developed an hESC fluorescent reporter line based on the expression of the AVIL gene, which encodes ADVILLIN, an actin regulatory protein of the Gelsolin/Villin family (Marks et al., 1998). Advillin is expressed in most if not all rodent and human DRG neurons but not in spinal cord (Figures S1A–S1C) (Hasegawa et al., 2007; Zurborg et al., 2011; Andersen et al., 2023).

To manipulate the AVIL locus in RUES2 hESCs, we used a CRISPR/Cas9-mediated gene editing approach to insert a Cre recombinase transgene (Knott and Doudna 2018; Garcia-Diaz et al., 2020). To ensure high levels of reporter expression, we coupled AVIL:Cre expression to Cre-dependent GAGGS-promoter-driven reporter expression (Brault et al., 2007). In addition, we selected tdT as our reporter to enable immediate visualization of expression using most standard fluorescent microscopes (i.e., without a need for immunological signal amplification). The AVIL:Cre driver was generated by targeting the Cre coding sequence (cloned in frame with a P2A internal ribosomal entry site sequence) to exon 6 of the AVIL gene, just before the stop codon (Figures S1D and S1E) (see methods for details). One hESC AVIL:Cre clone was subsequently used to target a CAGGS:lxp-STOP:lxp-tdTomato reporter into the PPP1R12C locus (also known as the AAVS1 safe harbor locus) again using a CRISPR/Cas9 strategy as described previously (DeKelver et al., 2010; Garcia Diaz et al., 2020).

Validation of the AVIL:tdT reporter

To validate the AVIL:Cre;AAVS1:tdTomato reporter (hereafter AVIL:tdT), we differentiated RUES2 hESCs harboring this reporter into sensory neurons. To do so, we used a differentiation strategy based on previously described protocols but with a few modifications (Chambers et al., 2012; Maury et al., 2015). In this protocol, hESCs are exposed to various signaling molecules (including the SMAD inhibitors LDN and SB and modulators of the Wnt and Notch signaling pathways) to emulate the in vivo differentiation trajectory of native sensory neurons (Figure 1A; Tables S1 and S2) (Simões-Costa and Bronner, 2015; Hari et al., 2002; Lee et al., 2004). We noted that for RUES2 cells, the addition of BMP4 at 4 days of in vitro culture (DIV4) was essential to induce the expression of PAX3 (a marker of dorsal neuroepithelium) and neural crest identity (Figures S2A–S2G) (Cimadamore et al., 2011; Duval et al., 2019). Under the influence of the various signaling factors, the RUES2 embryoid bodies (EBs) went through several transcriptional stages, marked by the expression of SOX2 (indicative of a general neural identity), PAX3 (indicating a dorsal neural fate), and AP2α and SOX10 (markers for a neural crest cell identity) (Figures 1A, 1B, and S2D) (Cimadamore et al., 2011; Simões-Costa and Bronner, 2015). During these early stages, we also detected expression of the transcription factors NEUROGENIN (NGN) 1 and NGN2, known to be expressed in somatosensory progenitors, as well as of HOXC5 and HOXC8, known to be associated with brachial spinal segmental levels (Figure S2H) (Ma et al., 1999; Miller and Dasen, 2024). The latter observation suggests that our EBs exhibit a spinal neural crest identity, instead of a placodal/cranial crest identity.

Figure 1.

Protocol to derive human sensory neurons from embryonic and induced pluripotent stem cells

(A) Schematic describing the main growth factors/cytokines and their time points of application to differentiate human ESC toward a somatic sensory neuron fate. Transcription factors (TFs) that mark specific stages of the differentiation process are indicated at the time point at which they are generally first detected. TF expression extends beyond this stage consistent with the variable differentiation time scales for individual progenitors within the EBs).

(B) Expression of the neural crest markers AP2α and SOX10 and the neural progenitor marker SOX2 in DIV14 EBs. Image on the right is a magnification of the boxed area in the image on the left.

(C) Percentage of AP2α⁺, SOX10⁺, and AP2α⁺SOX10⁺ cells from the total number of cells in DIV11 and DIV14 EBs. Percentages are estimates based on a machine learning approach (see supplemental methods for details). Data derived from 7 (DIV11) or 11 (DIV14) independent differentiations with at least 2 representative EB images used per data point (differentiation).

(D) Expression of BRN3A and ISLET, and the general neuronal Nissl stain (using NeuroTrace [NT]) in DIV14 EBs. Image on the right is a magnification of the boxed area in the image on the left. (The ISLET antibody used in these experiments recognizes both ISLET 1 and 2.)

(E) Percentage of BRN3A⁺, ISLET⁺, and BRN3A⁺,ISLET⁺ neurons from the total number of cells in DIV11 and DIV14 EBs. Percentages are estimates based on a machine learning approach (see supplemental methods for details). Data derived from 8 (DIV11) or 13 (DIV14) independent differentiations with at least 2 representative EB images used per data point.

(F) Expression of ISLET and TUJ1 at DIV17.

(G) Expression of BRN3A, ISLET, and PERIPHERIN at DIV17.

(H) Percentage of BRN3A⁺, ISLET⁺, and BRN3A⁺, ISLET⁺ post-mitotic sensory neurons in DIV17 cultures. Data obtained from 5 independent differentiations; total numbers of neurons counted/experiment were 166, 203, 556, 268, and 278.

Boxplots in figures show the median, 25th, and 75th percentile and whiskers extend to the most extreme data point less than 1.5 times the interquartile range. Significance was accepted for p < 0.05 and denoted as ^∗p < 0.05 or ^∗∗p < 0.01. Scale bars: 50 μm in (1B and 1D left), 10 μm in (1B and 1D right), and 20 μm in (1F and 1G). In (C, E, and H) different colored data points represent small modifications to the general protocol (see Table S9 for details).

By DIV11–14, EBs start to express the transcription factors POU4F1 (encoding BRN3A) and ISLET1/2 (Figure 1D). Unless specified otherwise, the ISLET antibody used in our experiments recognized both ISLET1 and ISLET2. Co-expression of BRN3A and ISLET is known to mark a post-mitotic sensory identity (Dykes et al., 2011). To estimate the relative differentiation efficiencies, we developed a FIJI-based machine learning (ML) approach (see supplemental methods for details) to quantify the numbers of SOX10-, AP2α-, BRN3A-, and ISLET-expressing cells within EB tissue sections. Based on these ML analyses, we estimate that at DIV11, 15.7% (±3.0% SEM) of EB cells express AP2α, 26.9% (±5.7%) express SOX10, and 8.8% (±2.2%) co-express both (Figure 1C). Co-expression of BRN3A and ISLET1 is still low at that stage: 3.9% (±1.7%) of cells express BRN3A, 15.2% (±6.6%) express ISLET, and just 1.1% (±0.5%) co-express BRN3A and ISLET (Figure 1E). However, by DIV14, EBs on average have 39.1% (±5.3%) BRN3A⁺ cells, 45.9% (±5.2%) ISLET⁺ cells, and 24.1% (±3.4%) of cells that co-express BRN3A and ISLET (Figures 1D and 1E). At DIV14, we also still detect considerable numbers of cells that co-express AP2α and SOX10 (29.8% ± 2.7%), while we never observed co-expression between AP2α and ISLET (Figure 1C, and data not shown). This indicates that the differentiation from neural progenitor to neural crest and post-mitotic neurons progresses dynamically over several days and that the cumulative percentage of neural crest cells and sensory neurons will be higher than our estimates based on the DIV11 and DIV14 snapshots. We observed similar levels of AP2α/SOX10 and BRN3A/ISLET expression in four other stem cell lines (hESC or iPSC) that we differentiated using similar protocols (Figures S3A and S3B).

Upon dissociation of the EBs and replating the differentiated cells at DIV14, we found that most cells assume a neuronal morphology and extend neurites shortly after plating. In addition, by DIV17, the majority of BRN3A⁺ and ISLET⁺ neurons co-express class III β-TUBULIN (TUJ1), a neuronal-specific tubulin, and PERIPHERIN, an intermediate filament protein predominantly expressed in peripheral neurons (Figures 1F and 1G). At DIV17, the percentage (±SEM) of TUJ1⁺ or PERIPHERIN⁺ neurons positive for BRN3A⁺ neurons is 68.3% ± 4.3%, the percentage of ISLET⁺ neurons is 68.0% ± 8.3%, and the percentage of BRN3A⁺ISLET⁺ neurons is 56.6% ± 5.6% (Figure 1H). Bulk RNA sequencing of these neurons at DIV60 confirms that their molecular identity most resembles that of developing SIX1⁺ human sensory neurons rather than SOX10⁺ sensory progenitors (Figure S3C). Based on these observations, we conclude that our protocol differentiates hESCs into BRN3A⁺ISLET⁺ sensory neurons with relative high efficiency.

Employing this sensory neuron differentiation strategy, we next validated our newly generated AVIL:tdT hESC reporter line. Consistent with the expression of ADVILLIN in differentiated mouse and human neurons (Figures S1A and S1B), we found that starting from DIV20 to DIV23, AVIL:tdT hESC-derived sensory neurons express tdT with the number of tdT⁺ neurons increasing with extended culturing (Figures 2A–2C). At early stages, very few neurons show tdT expression (0.6% at DIV18). By DIV30, 24.5% (±2.7%) of neurons express tdT, and by DIV60, this increases to 48.4% (±5.6%) (Figures 2B and 2C). We also explored if small modifications to our protocol (e.g., the concentration of BMP4 added, the addition of retinoic acid [RA], and the timing and concentration of growth factors, such as neurotrophin 3 [NT3], brain-derived neurotrophic factor [BDNF], glial-derived neurotrophic factor [GDNF], or nerve growth factor [NGF]) would result in higher yields of AVIL:tdT⁺ neurons, but none of these alterations consistently did so (Figure 2C). Using our standard protocol, we find that nearly 100% of tdT⁺ neurons co-express both BRN3A and ISLET, confirming their somatosensory neuron identity (Figures 2D and 2E). Interestingly, although all tdT⁺ neurons co-express BRN3A and ISLET, not all BRN3A⁺ISLET⁺ express tdT; at DIV30, we observed tdT in 57.7% (±5.9%) of BRN3A⁺ISLET⁺ neurons (Figures 2D and 2F).

Figure 2.

Validation of the AVIL:tdT hESC reporter line

(A) Genetic strategy to mark sensory neurons with a tdT fluorescent protein.

(B) Expression of tdT in (i) DIV23 and (ii) DIV30 sensory neurons. Boxed area in (ii) is enlarged in the rightmost panels to visualize tdT⁺ neurites in addition to sensory neuron cell bodies.

(C) Percentage of tdT⁺AVIL:tdT hESC neurons at DIV30 and DIV60. DIV30 includes data from DIV28–32 and DIV60 includes data from DIV58–60. Data obtained from 19 (DIV30) or 5 (DIV60) independent experiments using two independent gene-edited clones (C8 and C10). Total number of neurons counted per experiment ranges between 48 and 940 (average/experiment for DIV30 is 362.7 ± 51.7 neurons and average for DIV60 is 196.6 ± 35.5 neurons).

(D) Expression of tdT, BRN3A, and ISLET in AVIL:tdT⁺ hESC-derived sensory neurons at DIV30. Box indicates enlarged area in the rightmost panel. Arrows indicate BRN3A⁺, ISLET⁺ neurons that lack tdT expression.

(E and F) Percentages of tdT⁺ neurons that co-express BRN3A and ISLET (E) and of BRN3A⁺ISLET⁺ neurons that co-express tdT (F) at DIV30.

(G) Representative whole-cell current clamp recordings showing a subthreshold depolarization evoked by a 2-ms current injection overlaid with an action potential evoked by a higher-amplitude current. Current injections at DIV30 elicit action potentials in a majority of neurons.

(H) Overlay of membrane potential traces from the same cell following a 1-s hyperpolarizing or depolarizing current injection. A train of action potentials is evoked by the depolarizing current step. A “sag” in membrane potential is clearly visible when the cell is hyperpolarized.

(I–K) Electrophysiological characteristics of hESC-derived sensory neurons treated with 10 or 40 ng/mL BMP4 at DIV4-7. Individual neurons are represented as circles. Neurons derived from EBs treated with 40 ng/mL BMP4 at DIV4–7 are further segregated based on status of tdT expression: tdT^off (open red circles) and tdT⁺ (solid red circles). Solid black circles (RA) represent neurons derived from a differentiation protocol using RA.

Boxplots in figures show the median, 25th, and 75th percentile and whiskers extend to the most extreme data point less than 1.5 times the interquartile range. Significance was accepted for p < 0.05 and denoted as ^∗p < 0.05 or ^∗∗p < 0.01. Scale bars: 200 μm in (Bi), 100 μm in (Bii), and 20 μm in (D). In (C, E, and I–K), different colored data points represent small modifications to the general protocol (see Table S9 for details).

To determine if the absence of tdT in BRN3A⁺ISLET⁺ neurons is a consequence of an immature sensory neuron identity, we explored the physiological properties of tdT⁺ and tdT^off neurons. To do so, we performed whole-cell patch clamp recordings on DIV30 sensory neurons and measured various physiological features that develop with increasing neuronal maturation, including membrane resting potential, membrane resistance, and repetitive firing properties (Figures 2G–2K and S4A–S4E). We observed an average membrane resting potential of −52.4 (±0.8) mV and found that the vast majority (55/58) of neurons produced an action potential (AP) upon current injection (Figures 2G and 2I). In addition, most (>75%) of the responding neurons showed repetitive firing with an average of 17.5 (±1.7) spikes during a 1-s current step at an amplitude that elicited maximum firing (Figures 2H and 2J). Yet, while we noted a significant difference in repetitive firing and AP duration between neurons that were exposed to higher levels of BMP4 (40 vs. 10 ng/mL at DIV4–7), we were unable to detect a significant difference in firing rate, membrane resting potential, membrane resistance, or the presence of an H-current when specifically comparing tdT⁺ and tdT^off neurons (all exposed to 40 ng/mL BMP4) (Figures 2I, 2J, S4A, and S4C). However, DIV30 tdT⁺ neurons showed a significantly shorter AP duration (p < 0.05, Mann-Whitney rank-sum test) (Figure 2K). Thus, the presence of tdT^off, BRN3A⁺ISL⁺ neurons possibly relates to the maturation state of their membrane or intrinsic firing properties (Lechner and Lewin 2009). We also observed an increase in repetitive firing and shortened AP duration in neurons that were exposed to RA (100 nM at DIV2–4) during the differentiation process (Figures 2J and 2K). While we have not yet correlated this with increased AVIL:tdT expression, this could indicate that RA may accelerate the neuronal maturation process.

Interestingly, when performing these recordings, 76% (40/62) of neurons exhibited a small shoulder on the downstroke of the AP (a transient decrease in the rate of repolarization; Figures S4B and S4C), a feature more often associated with small-diameter nociceptive or temperature-sensitive neurons (Traub and Mendell 1988; Lechner and Lewin 2009). Considering that the differentiation protocol we used resembles the protocol by Chambers et al., which yields mostly nociceptor-like neurons (Chambers et al., 2012), we also tested responses to typical nociceptive stimuli, including ATP, capsaicin, and menthol. However, while we observed robust responses to potassium chloride and α,β-methylene-ATP, a P2X3 receptor agonist, neither AVIL:tdT⁺ nor AVIL:tdT^off neurons displayed any discernable neural excitation in response to capsaicin or menthol (Figures S4D and S3E, and data not shown). Consistent with these observations, bulk transcriptomic analyses of these hESC-derived neurons show poor expression of the relevant receptors (TRPV1 and TRPM8, respectively) (Figure S3C). Nevertheless, these studies demonstrate the utility of the AVIL:tdT reporter in optimizing the generation, as well as the phenotypic and electrophysiological analysis of in vitro hESC-derived somatic sensory neurons.

Generation and validation of a NTRK1:tdT reporter line

Small-diameter nociceptive, thermoceptive, and pruritogenic sensory neurons together constitute the largest DRG sensory neuron subset and are associated with acute pain, itch, and many chronic pain disorders (Woolf and Ma 2007). The present need for non-opioid analgesics has exacerbated a push for human model systems to explore and evaluate such new pain medications. To help facilitate the development of better iPSC-based human models, we also generated a tdT reporter marking small-diameter nociceptive/temperature-sensitive sensory neurons.

In both mice and humans, developing nociceptive neurons initially express the tropomyosin receptor kinase A (TrkA) protein, which is the receptor for NGF and is encoded by the NTRK1 gene (Figure 3A) (Chen et al., 2006; Lu et al., 2024). This suggested that the NTRK1 locus could offer an opportunity for a genetic reporter insertion in hESCs to mark developing nociceptors. In mice, TrkA expression is ultimately extinguished from the non-peptidergic nociceptive population and only maintained in peptidergic nociceptive neurons (Chen et al., 2006). Therefore, to best reflect NTRK1 expression dynamics, we coupled tdT expression directly to the NTRK1 transcript (using a P2A internal ribosomal entry sequence) (Figure 3B). Similar to AVIL:tdT, NTRK1:tdT hESCs were generated using CRISPR/Cas9 gene editing, and NTRK1:tdT clones were identified by genotyping PCR (Figure 3B). Considering that TrkA signaling is essential for nociceptor/thermoceptor survival, we inserted the tdT just before the translational stop codon such that the TRKA coding sequence remained nearly intact (Figure 3B).

Figure 3.

Generation and validation of a NTRK1:tdT hESC reporter line

(A) Expression of TrkA, cMaf, and PV in P0 mouse DRG. Box in the left panel shows area enlarged in panels on the right.

(B) Strategy to generate the NTRK1:tdT reporter using CRISPR/Cas9 gene editing. Genomic location, guide RNA sequence (blue font), and PAM site (underlined bold font) to insert a P2A-tdT reporter into the 3′ UTR of the NTRK1 locus are shown. STOP codon is indicated in red font. Genotyping primers and results are indicated. For details, see methods and Tables S3–S5.

(C) NTRK1:tdT⁺ sensory neurons in DIV21 and DIV28 cultures.

(D) Percentage of tdT⁺NTRK1:tdT hESC neurons at DIV20 and DIV30 across various differentiation protocols. DIV20 includes data from DIV17–22 and DIV30 includes data from DIV28–30. Data obtained using three different NTRK1:tdT hESC clones (F5, G2, and H7) from 13 (DIV20) or 8 (DIV30) independent experiments, with the total number of neurons counted per experiment ranging between 78 and 2,035 (average/experiment for DIV20 is 696.8 ± 155.7 neurons and average for DIV30 is 483.1 ± 156.7 neurons). Note that at DIV30, the two red solid data points were from clones G2 and H7, while all other data points were from clone F5.

(E) NTRK1:tdT⁺ neurons co-express ISLET and BRN3A.

(F) Percentage of ISLET⁺DAPI⁺ neurons in DIV30 NTRK1:tdT⁺ sensory neurons. DIV30 includes data from DIV30 and 31. Data obtained using one NTRK1:tdT hESC clone (F5) from 3 independent experiments, and total number of neurons counted per experiment ranges between 176 and 472 neurons.

(G) Percentage of ISLET⁺ neurons that co-express BRN3A or tdT in DIV30 NTRK1:tdT hESC-derived sensory neurons. DIV30 includes data from DIV30 and 31. Data obtained using one NTRK1:tdT hESC clone (F5) from three (BRN3A) or two (tdT) independent experiments, and total number of neurons counted per experiment ranges between 97 and 209 (BRN3A) or 88 and 209 (tdT).

Boxplots in figures show the median, 25th, and 75th percentile and whiskers extend to the most extreme data point less than 1.5 times the interquartile range. Significance was accepted for p < 0.05 and denoted as ^∗p < 0.05 or ^∗∗p < 0.01. Scale bars: 20 μm in (A and E), 50 μm in (C, left), and 100 μm in (C, right). In (D, F, and G), different colored data points represent small modifications to the general protocol (see Table S9 for details).

Upon differentiation of NTRK1:tdT hESCs into sensory neurons (using the same general protocol as described for AVIL:tdT neurons) (Figure 1A), we detected tdT fluorescence as early as DIV17, and the average percentage of tdT⁺ neurons was 35.4% ± 5.8% by DIV20 (Figures 3C and 3D). For one NTRK1:tdT hESC line, the percentage of tdT⁺ neurons subsequently declined to just 2.6% ± 1.1% of neurons by DIV30, but in other lines, the percentage remained the same or increased (Figures 3C and 3D). Loss of TRKA expression has similarly been observed in other sensory neuron differentiation protocols (Chambers et al., 2012; Deng et al., 2023) and possibly reflects a switch to a non-peptidergic nociceptive/thermoceptive phenotype (Chen et al., 2006). Although this remains to be tested, transcriptome analyses of hESCs differentiated following a similar “nociceptor” protocol supports this idea: the expression levels of RUNX1 and several pruritogenic molecular markers are relatively elevated compared to NTRK1 levels (Figure S3D). Alternatively, given that some of our lines did maintain TRKA expression at later stages, it is also possible that intrinsic differences between hESC lines may influence the ability of some hESC clones to continue to develop into mature TRKA⁺ neurons. Expression of tdT was always restricted to ISL1⁺ or POU4F1⁺ISLET⁺ neurons (Figures 3E and 3F). The number of NTRK1:tdT⁺ neurons increased when detecting tdT using immunological staining (Figure 3G), suggesting that at later stages, many neurons may still express tdT but at lower levels. Consistently, many neurons from a NTRK1:tdT hESC line that showed little to no native tdT fluorescence at this stage, were tdT⁺ after immunolabeling. Together, these data indicate that the NTRK1:tdT reporter reliably marks the presence of TRKA⁺ somatosensory neurons. Considering the relative inefficiency of current nociceptor differentiation strategies with respect to their mature physiological properties, more work is required to optimize these protocols. The NTRK1:tdT reporter we describe here should help to facilitate this process.

Generation and validation of mechanoreceptor and proprioceptor reporter lines

Efforts to develop or test new therapeutics that control or reduce neuropathic pain vastly outnumber studies of neuropathies that also, or primarily, involve large-caliber sensory neurons. Many of these disorders, such as hereditary sensory and autonomic neuropathy type III, multiple sclerosis, diabetic neuropathy, chemotherapy-induced peripheral neuropathy, or Friedreich ataxia, have similarly devastating consequences for quality of life. An added complication in modeling these large-fiber neuropathies is the scarcity of these sensory subtypes in DRG. On average, just 5%–8% of MafA⁺ (“mechanoreceptor”) or Parvalbumin (PV)⁺ (“proprioceptor”) neurons are found in DRG (Figure 3A) (Bourane et al., 2009; Lecoin et al., 2010; Wu et al., 2019; Nguyen et al., 2021). Therefore, protocols to derive mechanoreceptor or proprioceptor neuronal subtypes for disease modeling are as much in need as nociceptor differentiation protocols. To facilitate the development of hESC/iPSC differentiation protocols that bias sensory neurons toward a mechanoreceptor or proprioceptor phenotype, we also developed MAFA:tdT and RUNX3:tdT reporter lines, respectively.

MafA, a basic leucine-zipper transcription factor of the AP1 family, marks multiple classes of low-threshold non-nociceptive mechanoreceptive sensory neurons in mouse and human, including Meissner afferents, Merkel cell afferents, and longitudinal lanceolate endings (Bourane et al., 2009; Lecoin et al., 2010; Wende et al., 2012). MafA is not expressed in sympathetic neurons but labels a small subset of spinal interneurons. Runx3, in turn, is a Runt-domain transcription factor, which, within the nervous system, is nearly exclusively expressed in proprioceptive muscle afferents that supply muscle spindles and Golgi tendon organs (Wu et al., 2019). Runx3 expression persists in adult proprioceptive neurons and is not expressed in spinal or sympathetic neurons. Based on these prior observations, we hypothesized that MAFA:tdT and RUNX3:tdT reporters should similarly delineate a majority of human low-threshold mechanoreceptors and proprioceptors, respectively.

Generation of MAFA:tdT and RUNX3:tdT reporters followed the same general strategy as described for NTRK1:tdT, by coupling tdT expression to the endogenous MAFA or RUNX3 transcript using a P2A IRES sequence and with minimal disruption of the MAFA- or RUNX3-coding sequences (Figures S5A and S5B). To validate the MAFA:tdT and RUNX3:tdT reporter lines, we explored several adaptations to our “nociceptor” protocol to bias neurons toward a mechanoreceptive/proprioceptive fate. Previous studies had noted that reduced WNT signaling coupled to early DAPT exposure increased expression of NTRK2, a marker typically associated with low-threshold mechanoreceptors (Chambers et al., 2012). This led us to explore these modifications in our own protocol. When implementing these changes, we obtained similar efficiencies with respect to POU4F1⁺ISLET⁺ sensory neurons in DIV11 EBs when compared to our previously described protocol (Figures 4A, 4B, and S5C–S5F). In addition, we noted elevated expression of several molecular markers typically associated with low-threshold mechanoreceptive or proprioceptive neurons (e.g., PIEZO2, RET, and SHOX2) in DIV60 neuronal cultures (Figure S3E). When we next tested our MAFA:tdT and RUNX3:tdT reporters using this new differentiation protocol, we detected tdT-expressing neurons for both reporter lines, albeit with vastly different efficiencies (Figures 4C, 4D, 4F, and 4G). MAFA:tdT⁺ neurons were readily observed by DIV18 and continued to increase in number to DIV30 (Figures 4C, 4D, and S5H). While we did not have a working MAFA antibody, all MAFA:tdT⁺ neurons co-expressed the close family member cMAF. cMAF is also expressed in mechanoreceptive sensory neurons, therefore indirectly validating the low-threshold mechanoreceptive identity of the MAFA:tdT⁺ neurons (Figures 4E and 4I) (Wende et al., 2012). Consistent with this, nearly all MAFA:tdT⁺ neurons co-expressed TRKB (Figures 4I and 4J).

Figure 4.

Generation and validation of hESC reporter lines for low-threshold mechanoreceptors and proprioceptors

(A) Schematic of differentiation protocol to bias neurons toward a mechanoreceptor and/or proprioceptor sensory identity. See methods for details.

(B) Expression of BRN3A and ISLET in DIV11 EBs derived from MAFA:tdT (mechanoreceptor) and RUNX3:tdT (proprioceptor) hESCs.

(C) Expression of tdT in DIV18 and DIV30 mechanoreceptor sensory neurons derived from MAFA:tdT hESCs.

(D) Percentage of tdT⁺ neurons in DIV20, DIV30, and DIV≥40 neural cultures derived from MAFA:tdT hESCs directed toward a mechanoreceptor identity. DIV20 includes data from DIV18–22, DIV30 includes data from DIV28–30, and DIV≥40 includes data from DIV37–50. Data obtained using two different MAFA:tdT hESC clones (F10 and G10) from 4 (DIV20), 4 (DIV30), or 5 (DIV≥40) independent experiments. Total number of neurons counted per experiment was between 42 and 360 (average/experiment for DIV20 is 131.5 ± 3.5 neurons, average for DIV30 is 52.5 ± 10.2 neurons, and average for DIV≥40 is 161.8 ± 50.4 neurons). Different colored data points represent small modifications to the general protocol (see Table S9 for details).

(E) MafA:tdT⁺ neurons co-express ISLET and cMAF at DIV30.

(F) Expression of tdT in DIV21 and DIV39 proprioceptor sensory neurons derived from RUNX3:tdT hESCs.

(G) Percentage of tdT⁺ neurons of the total number of neurons in DIV20, DIV30, and DIV≥40 neurons derived from Runx3:tdT hESCs directed toward a proprioceptor identity. DIV20 includes data from DIV17-22, DIV30 includes data from 28–31, and DIV≥40 includes data from DIV37-45. Data obtained using three different RUNX3:tdT hESC clones (A2, C10, and H6) from 11 (DIV20), 13 (DIV30), or 5 (DIV≥40) independent experiments. Total number of neurons counted per experiment was between 21 and 1,009 (average/experiment for DIV20 is 257.5 ± 107.9 neurons, average for DIV30 is 270.0 ± 82.4 neurons, and average for DIV≥40 is 345.8 ± 70.6 neurons). Different colored data points represent small modifications to the general protocol (see Table S9 for details).

(H) RUNX3:tdT⁺ neurons co-express ISLET and RUNX3 at DIV30.

(I) Percentages of tdT⁺ISLET⁺ (left graph) and MAFA⁺tdT⁺ and TrkB⁺tdT⁺ (right graph) neurons at DIV22. Data obtained using one MAFA:tdT clone (F10) from two independent experiments. Total number of neurons counted per experiment ranged between 90 and 271 (ISLET) and 67 and 213 (tdT). Colored data points represent differentiations as in (D); see Table S9.

(J) MAFA:tdT⁺ mechanoreceptive sensory neurons co-express ISLET and TRKB.

(K) Percentages of tdT⁺ISLET⁺ (left graph) and RUNX3⁺tdT⁺ (right graph) neurons at DIV30. DIV30 includes data from DIV29–30. Data obtained using one RUNX3:tdT hESC clone (C10) from 4 (tdT) or 3 (RX3) independent experiments. Total number of neurons counted per experiment ranged between 48 and 329 (average/experiment for tdT is 183 ± 61.3 neurons and average for RX3 is 228.0 ± 58.9 neurons). Colored data points represent differentiations as in (G; see Table S9).

Boxplots in figures show the median, 25th, and 75th percentile and whiskers extend to the most extreme data point less than 1.5 times the interquartile range. Significance was accepted for p < 0.05 and denoted as ^∗p < 0.05 or ^∗∗p < 0.01. Scale bars: 20 μm in (B, E, H, and J) and 100 μm (C and F).

Although expression of tdT is sparse in the RUNX3:tdT reporter line, all tdT⁺ neurons co-expressed RUNX3 (Figures 4F–4H and S5G). This suggests that the reporter works as designed but that we do not yet have the correct protocol for the generation of this sensory subtype. This is supported by our preliminary experiments in which we cultured our EBs in the presence of muscle-conditioned media (Figure S5G). Under these conditions RUNX3:tdT⁺ cells were strongly enriched in small patches suggesting that a local signaling source (perhaps generated in response to the muscle-conditioned media) was providing an element that we are currently missing in our protocol. Together these data confirm that our MAFA:tdT and RUNX3:tdT reporters recapitulate the endogenous MAFA and RUNX3 expression patterns. Future work is needed to increase the yields of these in vitro-generated neuronal subsets, in particular for RUNX3 proprioceptors, and to compare their molecular and physiological properties with those of their nascent in vivo counterparts. The MAFA:tdT and RUNX3:tdT reporters described here should help advance these studies. Last, we also tested our ability to use our sensory neuron hESC reporters to purify the generated neurons for downstream analyses using FACS. We find that dissociated MAFA:tdT⁺ neurons, as well as NTRK1:tdT⁺ neurons, can be isolated and replated for axonal regeneration studies (Figures S5I–S5K), thus further expanding the utility of our sensory neuron genetic reporters.

Discussion

Somatic sensory neurons comprise many subtypes that are broadly grouped into high-threshold nociceptive-, temperature- or itch-sensitive neurons, and low-threshold touch mechanoreceptors or proprioceptors and are each marked by select molecular determinants that underlie their distinct functions. Despite the urgent need for better preclinical sensory neuropathy models, robust protocols that reliably generate large quantities of the individual human sensory neuron subtypes remain lacking. We, here, describe the generation and validation of four genetic tools designed to help offset these inherent difficulties in the in vitro derivation of somatic sensory neurons from human ESCs and iPSCs. These resources should facilitate the optimization of current sensory neuron differentiation protocols and will aid in the phenotypic analyses of the individual subtypes in sensory neuropathic disease models.

Optimizing sensory neuron derivation protocols for disease modeling

Protocols for the derivation of somatic sensory neurons have been described previously (Chambers et al., 2012; Maury et al., 2015; Blanchard et al., 2015; Deng et al., 2023; Hulme et al., 2024). Generally, they are based on chemical strategies that involve the successive addition of small molecules or growth factors to mimic the in vivo environment of developing sensory neurons, sometimes in combination with the directed expression of sensory neuron transcription factors. However, many of these protocols yield mixed populations of sensory neurons, rendering individual subtypes difficult to isolate for analysis. These limitations present a challenge in disease modeling, given that sensory neuropathies often affect specific sensory neuron subtypes. The sensory neuron reporter lines that we describe here can mitigate these issues at multiple levels. First, hPSC sensory neuron differentiation protocols can be optimized in real-time based on the direct readout of the tdT fluorescent reporter. Small molecules, neurotrophic factors, and extracellular matrix substrates can all be applied in multiple combinations and their effectiveness assessed through the number of tdT⁺ neurons obtained. In similar manner, myriad other signaling factors can be tested to optimize differentiation protocols for specific classes of TRKA⁺ somatosensory neuron subtypes that project to bone, gut, muscle, or visceral organs instead of skin. Likewise, low-threshold mechanoreceptors and proprioceptors may be further directed into distinct subtypes with repeated testing of candidate signaling molecules.

Another common issue with many hPSC differentiation protocols is stem cell line-specific features. Small intrinsic differences in the properties of ESCs or patient-derived iPSCs (e.g., endogenous BMP4 or Wnt expression levels) may diminish the effectiveness of a given differentiation protocol. As such, genuine improvement in sensory neuron protocol development means a robust protocol that works with at least 50% efficiency for all iPSC/ESC lines. The tdT reporters we describe can be generated for any other ESC/iPSC line through a nucleofection with the donor, guide RNA (gRNA), and Cas9 expression plasmid (all available upon request). Such an approach permits assessing differentiation efficiencies across multiple ESC/iPSC lines. Once a robust protocol is achieved, the reporter itself may no longer be needed, except for specific applications. Strategies that promote a more “natural” differentiation of sensory neuron subtypes would also be desirable over the use of methods that rely on the doxycycline-mediated inducible expression of transcription factors, given that patient-derived iPSCs in particular could be negatively influenced by the exposure to doxycycline.

A third advantage of the hESC/iPSC reporter lines we developed is that even if a differentiation protocol initially yields a limited number of sensory neurons of the desired subtype, one can isolate these for analysis using FACS or can focus histological (or other) analyses on the tdT⁺ cell bodies/neurites in the mixed culture. Indeed, the tdT reporters used in these studies are sufficiently bright to permit the use of automated plate-reader-based analyses in axonal regeneration studies, toxicity analyses, or neuronal activity measurements using calcium imaging. A related application of the sensory neuron tdT reporter lines is their use in multiplex 3D co-cultures such as innervated skin or bone. Expression of a tdT reporter enables the direct visualization of sensory axons in such systems.

Limitations of the individual sensory neuron tdT reporter lines

The reporters we describe here do not mark a single sensory subtype but rather represent broader populations of presumptive TRKA⁺ nociceptors, thermoreceptors, or puriceptors; MAFA⁺ Pacinian-, Meissner-, or Lanceolate-ending mechanoreceptors; or RUNX3⁺ group Ia, Ib, or II proprioceptors or Merkel cell afferents (de Nooij et al., 2013). In addition, TRKA is also expressed in sympathetic neurons, and as such, this reporter will require additional validation to ascertain that the tdT⁺ neurons constitute somatic DRG sensory neurons. Thus, these reporters cannot by themselves be used to definitively mark a select sensory subtype but require secondary validation through alternative strategies. Another potential issue with the reporters is that the small genomic alterations we introduce within the loci of genes that are crucial in the generation of their sensory subclass identity may interfere with the normal differentiation of the neurons. While we cannot yet fully exclude this possibility, the notion that we detect cMAF and RUNX3 protein in MAFA:tdT⁺ and RUNX3:tdT⁺ neurons, respectively, indicates that insertion of the tdT reporter does not interfere with their expression. A validated anti-TRKA antibody will be required to similarly evaluate this for the NTRK1:tdT⁺ neurons.

The AVIL:Cre;CAGGlxp-stop-lxp:tdT (AVIL:tdT) hESC line we generated appears most useful for mature neurons given its gradual increase in expression with prolonged culture. Its Cre/loxP-dependent tdT reporter offers both advantages and disadvantages. For instance, the CAGGS promoter drives very high levels of tdT expression, further facilitating the visualization of neuronal cell bodies and axons. Another advantage is the versatility of the Cre/loxP system given that AVIL:Cre can easily be coupled to other reporters such as a GCaMP calcium indicator or any specific experimental modulator (e.g., small interfering RNA) to test new hypotheses. A potential disadvantage of the AVIL:tdT line is that the high levels of tdT accumulation could influence neuronal health, an important consideration when used in disease modeling. In addition, the nature of this Cre/loxP system is such that it permanently labels a neuron even if ADVILLIN expression were to be transient and not maintained.

Applications for NTRK1, MAFA, and RUNX3 reporters beyond somatic sensory neurons

Considering that NTRK1, MAFA, and RUNX3 serve important roles in other developmental processes and tissues, the reporters we developed can also be utilized for the in vitro modeling of a range of other cell types. As described above, sympathetic neurons similarly rely on NGF for their survival and development and express NTRK1 (Smeyne et al., 1994). Likewise, MAFA represents a critical molecular marker for pancreatic beta cells, in which it is required to drive insulin expression (Aramata et al., 2007). Last, while RUNX3 is largely selective for proprioceptive sensory neurons, outside of the nervous system, its expression is required in CD4 and CD8 T cell lineage selection, is found in hair cell follicles, and has a role in spermatogenesis (Egawa et al., 2007; Raveh et al., 2005; Rahmawati et al., 2023). Clearly, the current set of sensory neuron fluorescent reporter lines will not just be relevant for the somatosensory field but also may serve as a valuable resource for in vitro hPSC differentiation and/or disease modeling of various other biological tissues.

Methods

hESC/iPSC culture and maintenance

hESC and iPSC lines used for this study include RUES2 (NIHhESC-09-0013; Rockefeller University), H9 (NIHhESC-09-0022; Harvard University), FA0000011 (Patel et al., 2020), FRDA4676, and FRDA68 (both corrected FRDA lines from the Friedreich’s Ataxia Cell Line Repository; UT Southwestern) (Li et al., 2016; Misiorek et al., 2020). Experiments conducted with these ESC and iPSC lines were approved by Columbia University’s Human Embryo and Embryonic Stem Cell Research Committee and the Institutional Review Board from Columbia’s Human Research Protection Office (protocol #AAAT9363). hESCs/iPSCs were cultured on mouse embryonic fibroblasts or feeder-free on plates coated with Matrigel or Cultrex (see Table S1 for reagent details). Cells on feeders were fed daily with hESC media (DMEM-F12, 20% KnockOut Serum Replacement, 1X non-essential amino acids [NEAA], 1X penicillin/streptomycin (P/S), 1X glutamine, 0.1 mM 2-mercaptoethanol, and 4 ng/mL fibroblast growth factor [FGF] 2). Feeder-free cultures were fed with mTeSR Plus or mTeSR1, with media changes every other day, and supplemented with additional FGF2 (5 ng/mL) StemBeads for over weekend (o/w) culture. When using mTeSR Plus, o/w cultures were provided with 2× the normal feeding volume. Cells were passaged using 0.5 mM EDTA and mechanical trituration. Rock inhibitor ([RI] 10 μM), 1X CloneR2, or 4 μL/mL CEPT (Chroman 1, Emricasan, Polyamines, Trans-ISRIB) (Chen et al., 2021) was added after passaging the cells to increase survival. All hPSCs were tested routinely for Mycoplasma using the e-Myco plus Mycoplasma detection Kit (Intron #25234). Cells were kept in a normoxic incubator, at 37°C, 5% CO₂.

hESC/iPSC differentiation

hESCs/iPSCs were differentiated into sensory neurons based on a protocol adapted from Chambers et al. (2012); Maury et al. (2015). hESCs/iPSCs used in differentiations were expanded to 70%–80% confluency. Cells were dislodged using 0.5 mM EDTA and mechanically triturated in 1 mL of N2B27 media (50% advanced DMEM/F12, 50% Neurobasal, 1X P/S, 1X glutamine, 0.1 mM 2-mercaptoethanol, 1X B27 minus Vitamin A, 1X N2-Supplement-B) (see Table S1 for reagent details). Cells were diluted in DIV0 medium (N2B27 supplemented with 10 μM ascorbic acid (AA), 20 μM SB431542, 0.1 μM LDN193189, 3 μM CHIR 99021, 10 ng/mL FGF2, and 10 μM RI), and plated in ultra-low attachment plates (Corning #3471) at a concentration of 1 × 10⁵ cells/mL (for hESCs) or 2 × 10⁵ cells/mL (for iPSCs) in either 6-well (2 mL/well) or 12-well (0.5 mL/well) plates to promote the formation of EBs. Except when noted otherwise, 100 nM RA was added on DIV2 and 10–40 ng/mL BMP4 was added on DIV4. Depending on the target sensory neuron subtype, 3 μM CHIR, 10 μM DAPT, 50 ng/mL NGF, 20 ng/mL BDNF, 10 ng/mL GDNF, or 25 ng/mL NT3 were added to the media on DIV2, DIV4, DIV7, DIV9, or DIV11 (as detailed in Table S2). On DIV 11 (mechanoreceptor and proprioceptor protocols) or DIV14 (nociceptor protocol), EBs were dissociated into single-cell suspension and plated as 2D cultures. To dissociate, EBs were collected in 15-mL conical tubes, washed once in PBS, and incubated in 1 mL 0.05% trypsin supplemented with 25 μg/mL DNase I (Roche #11-284-932-001) for 15 min at 37°C. EBs were mechanically dissociated by gentle trituration after which the digestion reaction was stopped with 2 volumes of heat-inactivated fetal bovine serum (HI-FBS) and 2 volumes of complete trituration wash media (PBS, 2.5% HI-FBS, 0.5 mM EDTA, 0.1% BSA, 25 mM glucose, 2 mM MgCl₂, 1X B27 minus Vitamin A, 1X N2 Supplement B). The cell suspension was spun and resuspended in NDM sensory neuron media (Neurobasal, 1X P/S, 1X L-glutamine, 1X NEAA and N2 Supplement B, 1X B27 minus Vitamin A, 0.1 mM 2-mercaptoethanol, 10 μM AA, and 10 ng/mL insulin growth factor-1). Cells were seeded on poly-ornithine ([PO] 1:5 diluted in H₂O to 20 μg/mL) and Laminin (1 mg/mL) or Cultrex-coated plates at 50,000 cells/cm². Medium was supplemented with the relevant neurotrophic factors (see Tables S1 and S2 for details). On the day of plating, 25 μM L-glutamic acid, 1 μM uridine, and 1 μM fluorodeoxyuridine were also added to the culture media. NDM sensory neuron media was changed every 2–3 days. When using Laminin as culture substrate, NDM media was supplemented weekly with 1 μg/mL Laminin.

Cloning of gRNA and donor plasmids and generation of hESC/iPSC reporter lines

gRNAs were designed using the DeskGen online algorithm (https://www.deskgen.com). Sequences of gRNAs used in this study are listed in Table S3. Sense and antisense sequences were annealed and cloned into the U6 gRNA expression vector (Addgene #41824; Mali et al., 2013) using either Gibson Assembly Master Mix (NEB #E2611) or InFusion HD In-Fusion HD Cloning Plus (Takara #638920). The AAVS1 gRNA plasmid was obtained from Addgene (#41818; Mali et al., 2013). AVIL, NTRK1, MAFA, and RUNX3 gRNAs were tested using the T7 endonuclease-I method as described previously (Garcia Diaz et al., 2020). Cas9 plasmids used in experiments were pCAGGS-Cas9-mCherry (Jacko et al., 2018) and pCAGGS-Cas9-mScarlet (Jacko, Wichterle, and Zhang, unpublished). The AVIL, NTRK1, MAFA, and RUNX3 donor plasmids were generated by amplifying the relevant areas from RUES2 genomic DNA (see supplemental methods for details and Table S4 for primer sequences). Genomic regions were combined with P2A-iCre or P2A-tdT reporters using various molecular cloning techniques. The AAVS1:loxSTOPlox:tdTomato donor plasmid was described previously (Mali et al., 2013; Garcia Diaz et al., 2020). hESC reporter lines were generated as described previously (Garcia Diaz et al., 2020) and as detailed in the supplemental methods. Crispr off-target analyses were performed by cloning the chromosomal locations with the highest probability for gRNA mistargeting as defined through the Synthego online guide validation tool (http://www.synthego.com) (Table S6). Identified cut-sites and surrounding areas (∼300-bp regions) were cloned by PCR amplification from two individual clones per reporter line (for primer sequences see Table S6). Cloned off-target areas were sequenced (Plasmidsaurus.com). None of the cloned regions showed deletions and/or rearrangements.

Immunohistological analyses

EB staining

On day of collection, EBs were washed once in PBS and fixed in 4% paraformaldehyde for 15 min on ice. Following fixation, EBs were washed in PBS, equilibrated in 30% sucrose for at least 2–3 h, and mounted in Tissue Tek OCT (FisherScientific #4585) in Peel-A-Way molds (Polysciences #18985-1). EBs were sectioned at 30 μm on a cryostat (Leica CM1850). Sections were stained in primary antibodies (see Table S7 for details) in PBS, 1% BSA, and 0.1% Triton X-100, overnight at 4°C. Following incubation in primary antibodies, sections were washed in PBS (2 × 6 min at RT) and incubated in Cy3-, fluorescein isothiocyanate-, or Cy5-conjugated secondary antibodies (all Jackson ImmunoResearch Labs) for 2 h at RT. Sections were washed in PBS (2 × 6 min at RT) and cover-slipped with VECTASHIELD (Vectorlab #h-1000-10) or Fluoromount (SouthernBiotech #0100-010.

Staining of 2D neuronal cultures

Dissociated EB progenitor neurons were plated on coverslips (ThemoScientific #3323) coated with PO/Laminin, Matrigel, or Cultrex. On day of collection, cells on coverslips were gently washed once in PBS and fixed in 4% paraformaldehyde (in PBS) for 10 min at 4°C. Neurons on coverslips were stained with primary and secondary antibodies as described above. Images were acquired on LSM510 Meta, LSM700, or LSM900 (all Zeiss) confocal microscopes.

Calcium imaging and electrophysiology

Calcium imaging was carried out using Fluo-4 calcium-sensitive dye and epifluorescence microscopy as previously described (Bosco et al., 2024). Briefly, neurons were bulk-loaded with Fluo-4 a.m. and imaged at 2 Hz. Images were acquired from the same cell field during two consecutive 60-s recording periods, separated by a 2-min interval. In the first period, a 2-s application of either capsaicin (1 μM; Tocris) or L-menthol (100 μM; Sigma-Aldrich) was delivered. In the second period, cells were exposed to 2-s application of 30 mM KCl. Fluorescence intensity changes were measured from manually drawn regions of interest encompassing neurons identified by their large round cell bodies or by tdTomato expression. ΔF/F was calculated by dividing the average pixel intensity of a region of interest at each time point by that of a baseline period prior to addition of each drug. AP and passive membrane properties were assessed using conventional whole-cell current clamp technique as detailed in the supplemental methods. In brief, DIV11 or DIV14 neurons were plated on 15-mm-diameter coverslips at a density of 50,000 cells per well in a 24-well culture plate and maintained until ∼DIV28–32 prior to recording. Membrane potential recordings were performed using a Multiclamp 700B amplifier and a Digidata 1550 digital-to-analog converter. Signals were recorded at a 10-kHz sample rate using pClamp 10 software (all molecular devices). APs (single or trains) were evoked using depolarizing current steps and defined as a transient depolarization of the membrane that had a minimum rise rate >10 mV/ms and reached a peak amplitude >0 mV. Recordings were all performed at RT.

FACS of dissociated sensory neurons

Cell dissociation and FACS of neuronal cultures was essentially as described in Wu et al., 2019 (see supplemental methods for details). Neurons isolated by FACS were directly deposited in Cultrex-coated 96-well culture plates. Cultures were thrice-weekly fed using standard NDM media supplemented with the appropriate growth factors. Sorted neurons were maintained for 7–14 days post-sort, after which they were fixed and processed for immunological analysis as described above.

Semi-quantitative RT-PCR

RNA was isolated using the Qiagen RNeasy Miniprep Kit (Qiagen #74104). First-strand cDNA was generated using the ThermoScientific RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific #K1691). cDNA was diluted in water and used at 2 ng/reaction in PCR (see Table S8 for primer sequences). Reactions were allowed to proceed for 35 cycles and run on a 2% agarose gel before imaging.

Transcriptome analysis

RNA was isolated from DIV60 neurons from one well of a 6-well plate/sample using Trizol (Invitrogen #15596026) and chloroform extraction. Glycogen (20 μg; Sigma #10901393001) was added to the aqueous phase during isopropanol precipitation and 75% ethanol wash. Isolated RNA was treated with DNAse (ThermoFisher Scientific #AM2222), followed by phenol chloroform (Acros Organics #327111000) extraction and ethanol precipitation. RNA was resuspended in ultrapure water, and RNA concentration and integrity was assessed by Bioanalyzer (samples included had RIN numbers between 7.9 and 9.9). The Clontech Ultra Low v4 kit was used for cDNA synthesis, and libraries were prepared using NexteraXT. Data were processed as described in supplemental methods.

Neuronal counts and statistical analysis

Counts of neuronal nuclei of confocal images of EB tissue sections (30 μm) were performed using an ImageJ-based ML approach as detailed in the supplemental methods. Unless specified otherwise, data were acquired from at least two independent gene-edited clones. Averages were derived from a minimum of three biological replicates, each with 3–8 technical replicates. Graphs were generated using R. Average counts/values and SEM were calculated using SigmaPlot. Whenever used, boxplots in figures show the median, 25th, and 75th percentile and whiskers extend to the most extreme data point less than 1.5 times the interquartile range. Statistical analysis on counted neuronal populations (Student’s t test or Mann-Whitney U test) or to compare gene expression levels (expressed in transcripts per million) between neurons derived using nociceptor, mechanoreceptor, or proprioceptor differentiation protocols (one-way ANOVA) was performed using SigmaPlot. Significance was accepted for p < 0.05 and denoted as ^∗p < 0.05 or ^∗∗ p < 0.01.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Joriene de Nooij (sd382@cumc.columbia.edu).

Materials availability

Key plasmids generated in these studies are available through Addgene (Plasmid ID 246652–246659). The hESC reporter lines generated in this study are available upon a completed materials transfer agreement with Columbia University. Request for materials should be sent to Joriene C. de Nooij (sd382@cumc.columbia.edu).

Data and code availability

Generated bulk RNA-seq datasets from hESC derived sensory neuron subsets have been deposited at NCBI’s GEO: GSE310583.

Acknowledgments

We thank Carmen Birchmeier (Max Delbruck Center, Berlin) for the cMaf antibody, Mike Kissner and the CSCI flow core for help with FACS, Vilas Menon for advising on transcriptome analyses, Hynek Wichterle for advise during the initial project stages, Marek Napierala (UT Southwestern), and the Friedreich’s Ataxia Cell Line Repository for the corrected FRDA4676 and FRDA68 lines. We also thank Luke Hammond for discussions on the EB neuronal counting methodology, and Grace Shin and Callie Barber (both The Ohio State University) for critical reading of the manuscript. These studies were supported by the Thompson Family Foundation Initiative on CIPN, the Friedreich's Ataxia Research Alliance (CU17-1234), and the CDMRP/DOD (HT94252310092) (all J.C.d.N.). This study was also funded in part through the NIH/NCI Cancer Center Support Grant P30CA013696 and used the Genomics and High Throughput Screening Shared Resource.

Author contributions

E.M.-G., K.M.O., S.C., and B.T.A. performed sensory neuron differentiations, RNA isolations, and/or data analyses, E.M.-G., K.M.O., and A.G.-D. performed gene-editing of hPSCs, D.W. performed electrophysiological experiments, J.A.H. analyzed bulk RNA sequencing data, B.C. advised on the project, and J.C.d.N. conceived of the study, performed experiments, and analyzed data. J.C.d.N. wrote the manuscript with input from all other authors.

Declaration of interests

The authors declare no competing interests.

Published: January 2, 2026