Ca²⁺ transients in melanocyte dendrites and dendritic spine-like structures evoked by cell-to-cell signaling

As the pigment-producing cells of the skin, melanocytes have melanosomes, specialized lysosome-related organelles. Melanin pigment is made within the lumen of melanosomes and transferred, via melanocyte dendrites, to neighboring keratinocytes, epithelial cells that comprise the bulk of the epidermis (Marks and Seabra, 2001; Wu and Hammer, 2014). Keratinocytes are known regulators of melanocyte behavior, and much work has been done to understand how keratinocytes influence melanocyte cell proliferation and the production and transfer of pigment throughout the skin (Gordon et al., 1989; Hirobe, 2014). Nonetheless, cell–cell communication between melanocytes and keratinocytes, at the single-cell level, is poorly understood.

Epidermal melanocytes reside within the basal layer of the epidermis in a ratio of ∼1:10 with basal keratinocytes (Adameyko et al., 2009; Erickson et al., 1992; Fitzpatrick and Szabo, 1959). This arrangement has led to the proposal that there are pigmentary units across the epidermis in which one melanocyte, through its dendrites, interacts with ∼30–40 neighboring keratinocytes (Fitzpatrick and Breathnach, 1963; Jimbow et al., 1976). However, it remains unclear how each melanocyte dendrite responds to input from adjacent keratinocytes. Studies of other cell types with branched morphologies have shown that long branched cell extensions are capable of localized cell–cell signal through spatial restriction in second messengers such at Ca²⁺ (Llinás and Hess, 1976; Llinás et al., 1968; Spencer and Kandel, 1961; Yuste and Denk, 1995). Here we explore how melanocytes use their dendrites to interact with and receive information from keratinocytes using a co-culture of melanocytes and keratinocytes and an ultra-structure level interrogation of intact human skin. We show that endothelin and acetylcholine (ACh) secreted by keratinocytes elicit local and compartmentalized Ca²⁺ transients in melanocyte dendrites. In addition, we characterize restricted Ca²⁺ transients in spine-like structures on melanocyte dendrites co-cultured with keratinocytes. The dendritic spines were also seen on melanocytes in intact human skin, where they were surrounded by keratinocytes, which contained vesicles in their cytoplasm adjacent to the melanocyte spines.

In intact human skin, melanocyte dendrites extend throughout the layers of the epidermis (Fig. 1 A and Fig. S1). This allows each melanocyte, at both its cell body and dendrites, to interact with multiple neighboring keratinocytes. We sought to recapitulate this arrangement in an in vitro culture system that would allow us to resolve melanocyte–keratinocyte interactions at the dendritic level.

Building upon previous studies, we created a co-culture of melanocytes and keratinocytes containing both keratin 14 (K14)– and keratin 10 (K10)–positive keratinocytes (De Luca et al., 1988; Lei et al., 2002; Valyi-Nagy et al., 1993). Mature co-cultures were achieved by seeding at a high density (0.5 × 10⁴ melanocytes and 5 × 10⁴ keratinocytes per 78 mm²), to achieve 100% confluence, and increasing the CaCl₂ concentration of the growth media from 0.06 mM CaCl₂ to 1.06 mM CaCl₂ for 2 d. This resulted in cultures that contained a bottom layer of cells comprised of melanocytes and K14-positive keratinocytes (Fig. 1, B and C). K10-positive keratinocytes were present above the bottom layer in one to two additional cell layers, which varied throughout the culture (Fig. 1 C). Mature co-cultures exhibited cell morphologies and cadherin-based cell–cell adhesion (Fig. 1, D and E) consistent with intact skin (Fig. S1, D and E; Tinkle et al., 2004), whereas co-cultures maintained in low (0.06 mM) CaCl₂ growth media did not (Fig. 1 F). E-cadherin (ECAD) and β-catenin colocalized in all layers of the mature co-cultures, while only the bottom layer of the co-cultures expressed high levels of P-cadherin (PCAD; Fig. 1, D and E). Importantly, melanocytes exhibited morphologies similar to those from intact epidermal sheets (Fig. 1, A and B; and Fig. S1 D) and made contact with numerous surrounding keratinocytes within and above the bottom layer of cells (Fig. 1 G).

To visualize melanocytes and keratinocytes in living cultures, the two cell types were cultured separately, and then a subset of each population was transduced with lentivirus encoding a fluorescent protein plasma membrane reporter: EGFPmem (melanocytes) or iRFPmem (keratinocytes). After 24 h, both cell types were combined to initiate the co-culture (Fig. 1 B). Mosaic labeling of each cell type allowed for real-time morphological observation of individual cells in the dense 3D culture. In regions with only a few labeled keratinocytes, we observed processes that extended from the cell surface of the keratinocytes, which contacted and wrapped adjacent melanocyte dendrites (Fig. 2 A). The keratinocyte processes and melanocyte dendrites made a stable interaction, which was maintained over the course of an hour even while each of the processes was moving (Fig. 2, B and C).

In intact human skin, similar interactions were observed from thin transmission EM (TEM) sections. Processes from basal and suprabasal keratinocytes enveloped melanocyte dendrites in multiple layers of overlapping projections (Fig. 2 D), often physically isolating multiple dendrites within the same region (Fig. 2 E). Interestingly, some keratinocytes had pools of small vesicles (44–70 nm diameter) in the cytosol adjacent to the plasma membrane that was juxtaposed to the melanocyte dendrite (Fig. 2 E, right panel). These pooled vesicles were distinct in their aggregated localization from other intracellular vesicles present throughout the cytosol of all cells and suggested that melanocyte dendrites might receive localized signaling from individual keratinocytes. To confirm that the keratinocyte processes observed in situ were indeed wrapping around melanocyte dendrites, focused ion beam scanning EM (FIB-SEM) was used to obtain serial sections of intact neonatal foreskin with a volume of 30 × 27 × 27 µm at 30 nm z resolution and an XY pixel size of 14–18 nm. 3D reconstruction of two keratinocyte processes showed that both processes interacted with and wrap around the melanocyte dendrite (Fig. 2, F–H).

Keratinocytes secrete multiple paracrine factors that influence melanocyte behavior (Gordon et al., 1989; Hirobe, 2014), some of which activate canonical cell–cell signaling cascades that use Ca²⁺ as a second messenger. Thus, we used the genetically encoded Ca²⁺ sensor GCaMP6f (Chen et al., 2013) to monitor cytosolic Ca²⁺ in the melanocytes. To ensure co-cultures were imaged in the absence of exogenous growth factors or serum, cells were washed three times with PBS and then imaged in a modified Dulbecco’s PBS (DPBS) containing 1.06 mM CaCl₂ and no additional growth factors or serum (see Materials and methods).  Unless stated otherwise, all subsequent Ca²⁺ imaging was conducted in the absence of exogenous growth factors or serum. Most melanocytes (64 ± 11%, mean ± SD, n = 22 co-culture cultures), co-cultured with keratinocytes in the absence of exogenous growth factors, had spontaneous Ca²⁺ transients over an observation period of 2.5 min (Fig. 3, A and B; and Video 1). The majority of Ca²⁺ transients were localized fluctuations in Ca²⁺ that occurred in the dendrites (Fig. 3, A–C; and Fig. 4), with few localized fluctuations within the cell body (Fig. 3, C and D) and few global transients throughout the whole cell (Fig. 3, A, C, and E).

To determine if the dendritic transients of Ca²⁺ were intrinsic to melanocytes or were dependent on the presence of keratinocytes, we altered the growth conditions and cell types present in cultures of melanocytes expressing GCaMP6f. Cultures were then imaged in modified DPBS without serum or other exogenous growth factors. The high percentage of melanocytes with Ca²⁺ transients (64 ± 11%) was only observed when melanocytes were co-cultured with keratinocytes. Significantly, fewer melanocytes had Ca²⁺ transients when the keratinocytes were replaced with HEK293T cells (Fig. 3 F, 10 × 293T: 13 ± 6%) or melanocytes were grown in mono-culture (Fig. 3 F, 10 × Mel and Mel alone: 16 ± 2% and 11 ± 4%, respectively), or when melanocytes were separated from keratinocytes by a semiporous membrane (Fig. S2 B, TW 10 × Ker: 9 ± 8% to 20 ± 10%, depending on growth conditions). This was irrespective of growth media formulation (Fig. S2, A and B). In co-cultures, the number of local dendritic transients detected during 2.5 min ranged from 0–70 per melanocyte. Of the melanocytes with local dendritic transients, 91% had 1–9 transients while 9% had >10 transients (Fig. 3 G). In comparison, of the few mono-cultured melanocytes that had dendritic transients, 90% had 1–4 transients and 10% had 5–12 transients (Fig. 3 G, inset). Thus, close proximity to, and possibly direct contact with, keratinocytes was required for the high frequency of Ca²⁺ transients.

Analysis of dendritic Ca²⁺ transients from 53 melanocytes grown with keratinocytes in co-culture media showed a range of detectable Ca²⁺ spread from 8–42 µm with the majority (58%) of transients being 14–22 µm in length (Fig. 3 H). Individual dendrites had transients that either originated from a single point (Fig. 4, A–C; and Video 2) or from multiple distinct locations within a region several microns long of the dendrite (Fig. 4, D and E). In those dendrites that had multiple Ca²⁺ transients over time, repetitive transients initiated from the same region of the dendrite (Fig. 4 F).

To determine the source of Ca²⁺ responsible for the dendritic transients, we quantified the number of Ca²⁺ transients before and after the removal of external CaCl₂ and/or addition of 1 µM thapsigargin, which releases Ca²⁺ from internal stores (Lytton et al., 1991). The percentage of melanocytes with Ca²⁺ transients was reduced by either removal of external CaCl₂ or addition of thapsigargin to the imaging media (0.63 ± 0.04 and 0.30 ± 0.05 fold change from before treatment, respectively) compared with the control (0.88 ± 0.5 fold change from before treatment), with the combination of removal of CaCl₂ and addition of thapsigargin having the greatest effect (0.07 ± 0.4 fold change from before treatment; Fig. 5 A). This was also true for the number of Ca²⁺ transients per cell (Fig. 5, B–E), where removal of external CaCl₂ plus the addition of thapsigarin reduced the number of transients per cell to a level similar to that of mono-cultured melanocytes (Fig. S3 H and Fig. S4 D).

Endothelins, a family of 21 amino acid peptides, play a critical role in melanocyte development, maturation, and homeostasis within the epidermis (Reid et al., 1996). Epidermal keratinocytes produce and secrete endothelin 1 (ET-1; Yohn et al., 1993), which can elicit increases of Ca²⁺ in the cell bodies of melanocytes (Kang et al., 1998). Addition of 10nM ET-1 to co-cultures produced recurring Ca²⁺ transients across the entire melanocyte, including dendrites and the cell body (Fig. 6, A–D; Fig. S3 A; and Video 3). The percentage of melanocytes with a response to ET-1 was dose-dependent (Fig. S3 B), and the number of transients per melanocyte also increased with the addition of 10 nM ET-1 (Fig. 6 B). ET-1–induced increase of Ca²⁺ in melanocytes co-cultured with keratinocytes was blocked by preincubation with a selective inhibitor of endothelin receptor B (ET_B), BQ788, but not an inhibitor of endothelin receptor A (ET_A), BQ123 (Fig. 6 B; Fig. S2, C and D; and Video 4), consistent with previous work that showed many patient-derived human melanocytes do not express ET_A (Eberle et al., 1999). This was also true of melanocytes in mono-culture (Fig. S3, E–H).

The localized Ca²⁺ transients in melanocytes elicited by ET-1 (Fig. 6, C–E; and Video 3) resembled the spontaneous Ca²⁺ transients in co-cultures of melanocytes and keratinocytes without the addition of exogenous ET-1 (Fig. 4 and Video 1). To test if the spontaneous local Ca²⁺ transients in melanocytes was caused by keratinocyte secreted endothelin, we treated co-cultures with endothelin receptor antagonists. Co-cultures treated with the ET_B antagonist BQ788 had a significant reduction in the percentage of melanocytes with Ca²⁺ transients, but no change was observed in Ca²⁺ transients in co-cultures treated with the ET_A antagonist BQ123 (Fig. 7 A). To confirm that the effects were due to inhibition of the ET_B on melanocytes and not the inhibition of ET_B on keratinocytes, we used Dicer-substrate siRNA (DsiRNA) against EDNRA (the gene that encodes ET_A) or EDNRB (the gene that encodes ET_B) to reduce expression of either ET_A or ET_B in melanocytes before co-culture with keratinocytes. Consistent with the endothelin receptor antagonist data, knockdown of ET_B but not ET_A reduced the number of co-cultured melanocytes with Ca²⁺ transients (Fig. 7 B). We next knocked down EDN1, the ET-1 gene, in keratinocytes and found that co-cultures with keratinocytes expressing shRNA against EDN1 had significantly fewer melanocytes with Ca²⁺ transients (Fig. 7, C and D). Together, these data confirmed that keratinocyte-secreted endothelin acts on melanocyte ET_B to produce local dendritic Ca²⁺ transients.

The Ca²⁺ transients were reduced by using antagonists in the bath to inhibit ET_B, shRNA in the melanocytes to reduce expression of the EDNRB, or shRNA in keratinocytes to reduce production of ET-1. However, none of these treatments eliminated the Ca²⁺ transients completely. In addition to production and release of endothelin, keratinocytes also possess the necessary cholinergic machinery to synthesize, release, and degrade the neurotransmitter ACh (Grando et al., 1993). It is also known that melanocytes express muscarinic receptors (mAChR) and respond to the mAChR specific agonist, muscarine (Buchli et al., 2001). Addition of ACh increased the number of Ca²⁺ transients per cell in co-culture (Fig. 6, F and G; Fig. S4 A; and Video 5) and mono-culture (Fig. S4, C and D), with some melanocytes having over 100 transients in 2.5 min, a 10-fold increase compared with control conditions (Fig. 6 G). Interestingly, ACh increased the frequency of Ca²⁺ transients in those cells that already showed activity, but had no effect on the number of melanocytes with Ca²⁺ transients (up to 100 µM; Fig. S4, A and B). This increase in frequency of Ca²⁺ transients, in both co- and mono-cultured melanocytes, was blocked by atropine, a selective muscarinic ACh receptor antagonist (Fig. 6 G and Fig. S4 D). To test if the spontaneous local Ca²⁺ transients in melanocytes resulted from release of ACh by keratinocytes, we treated co-cultures with muscarinic antagonist. Atropine decreased the number of co-cultured melanocytes with transients to 0.54 ± 0.08-fold compared with the results before treatment (Fig.7 E), indicating that endogenous ACh release contributed to melanocyte Ca²⁺ transients. Addition of BQ788 with atropine reduced the number of melanocytes with transients by 0.47 ± 0.06-fold compared with the results before treatment (Fig.7 E), which was similar to the level of reduction observed when BQ788 alone was added. When we used shRNA to reduce expression of choline acetyltransferase (CHAT), the enzyme that synthesizes ACh from acetyl CoA and choline, in keratinocytes (Fig. 7 F), the number of melanocytes with Ca²⁺ transients was significantly less than that of the control. This confirmed that keratinocyte-secreted ACh contributes to local Ca²⁺ transients in melanocytes (Fig. 7 G).

Within co-cultures mosaically labeled with plasma membrane–localized fluorescent protein, we observed discrete spine-like structures protruding from melanocyte dendrites (Fig. 8 A), which was consistent with the small dendritic protrusions we observed on melanocytes injected with Lucifer yellow in epidermal sheets (Fig. S5 A). Spine-like structures present on the dendrites of the co-cultured melanocytes had variable morphologies: mushroom, stubby, thin, and cup-shaped (Fig. 8). These morphologies were distinct and different than the previously described melanocyte filopodia (Scott et al., 2002; Singh et al., 2010), which were also present on melanocyte dendrites (Fig. 8 A). The width of the spine heads ranged from 0.47 to 1.36 µm, and total lengths (from dendrite to tip of spine) were 0.55–2.75 µm (Fig. 8 C). The density of dendritic spine-like structures varied from dendrite to dendrite (1.6 ± 1.3 mean ± SD, spines per 10 µm) with some dendrites having no detectable spine-like structures (Fig. 8, D and E). The dendritic spine-like structures were surrounded by adjacent keratinocyte plasma membrane and contacted keratinocyte processes (Fig. 8, F and G). The morphology of some spines were stable (Fig. 8 G), while others were dynamic, moving 0.33–1 µm over an hour (Fig. 8 H).

In some cases, Ca²⁺ transients originated within and remained confined to resolvable spine-like structures (Fig. 9, A–D, panel 1; and Videos 6 and 7). Additionally, we observed Ca²⁺ transients near the perimeter of a dendrite and contained within a <2-µm region (Fig. 9, A–D, panels 2 and 3). These may be increases in Ca²⁺ within a spine-like structure protruding off the dendrite that could not be distinguishable in XY by fluorescence.

To test for the presence of dendritic spines on melanocyte dendrites in intact human skin at the ultrastructural level, we used TEM. In TEM images, melanocytes and their dendrites were distinguished from other cells by having a lighter, more electron-lucent cytoplasm, the presence of melanosomes within the cell body and dendrites, and a lack of desmosomes and lack of strong keratin filament staining. Melanosomes were also observed in keratinocytes, but keratinocytes were easily distinguished from melanocytes by the other criteria mentioned above. Using manual serial sectioning followed by TEM, we found examples of spine-like structures, which were indistinguishable from those observed in the melanocyte–keratinocyte co-cultures (Fig. 10 A and Fig. S4 A).

Due to the sparse density of melanocyte dendrites compared with the total cell mass of the epidermis, we used FIB-SEM to analyze more dendrites and visualize larger lengths of dendrites. We reconstructed four individual melanocytes among 16 surrounding keratinocytes within a volume of 30 × 27 × 27 µm at 30-nm z resolution and an XY pixel size of 14–18 nm. An example of one such melanocyte is shown in Fig. 2 F. Melanocytes were distinguished from other cell types by their dendritic morphology and lighter, more electron-lucent cytoplasm (Fig. S5 B). Individual melanocytes contacted 7 to 10 keratinocytes within the imaged volume, consistent with previous data (Fitzpatrick and Breathnach, 1963; Jimbow et al., 1976). All four melanocytes had spine-like structures (Table S1) with head widths of 0.14–0.72 µm and total lengths of 0.24–1.53 µm (Fig. 10, D and E). In addition, we found pools of small keratinocyte vesicles surround a stubby spine-like structure (Fig. 10 E and Fig. S5 B).

Through its dendritic arbor, one melanocyte physically interacts with multiple keratinocytes across different layers of the epidermis. How melanocytes conduct and process communication with multiple cells is not understood. Other neural crest–derived cell types with branched cell morphologies are capable of receiving and processing information from neighboring cells at a local level, due, in part, to their cell shape and local biochemical activities including local release of signaling molecules, local receptors, and local elements, such as increases of cytosolic Ca²⁺ that amplify the response (Llinás and Hess, 1976; Llinás et al., 1968; Spencer and Kandel, 1961; Yuste and Denk, 1995). We investigated how melanocyte dendrites interact with neighboring keratinocytes using both an in vitro co-culture system and ultrastructure imaging of intact human skin. Our findings show that melanocytes produce compartmentalized fluctuations of Ca²⁺ in response to local signals from neighboring keratinocytes.

We found that keratinocytes also have small cytoplasmic projections both in situ and in culture, which contacted the melanocytes (Fig. 2). Similar structures were first described in amphibian skin at keratinocyte–keratinocyte junctions (Farquhar and Palade, 1965) and have been referred to as microvilli by other groups when discussing phagocytic events in keratinocytes (Ando et al., 2012). However, not much is known about their function or molecular composition, and it remains to be determined what class of cell extension they belong. The keratinocyte processes we observed wrapped around individual melanocyte dendrites (Fig. 2, A and D) and could be physically separating adjacent dendrites (Fig. 2 E). This local arrangement is similar to that of glial cells of the nervous system (Auld and Robitaille, 2003; Chong et al., 1999). Thus, keratinocyte projections may serve similar functions such as physically supporting the architecture of the melanocyte network across the skin as well as maintaining local signaling niches.

Endothelin and ACh secreted by keratinocytes elicited a localized Ca²⁺ response in melanocyte dendrites. Within the context of a multicellular tissue, the physical envelopment of melanocyte dendrites by the keratinocyte projections could facilitate localized delivery and recycling of secreted factors, similar to what occurs between cells of the nervous system (Auld and Robitaille, 2003; Chung et al., 2015). Further investigation into the molecular components of the keratinocyte projections as well as the stability and time scale of their interactions with melanocyte dendrites would provide further insight into their function.

Dendritic processes in neurons have functional domains and compartments that allow for local response and processing of cell–cell signaling events through compartmentalized changes of cytoplasmic Ca²⁺ in the presynaptic terminal, post-synaptic dendritic spine, and/or dendrite (Katz and Miledi, 1965; Llinás et al., 1968, 1972; Llinás and Hess, 1976; Simon and Llinás, 1985; Spencer and Kandel, 1961; Yuste and Denk, 1995). Cytoplasmic fluctuations of Ca²⁺ within dendrites and dendritic spines regulate a variety of neuronal functions including local synaptic signaling, plasticity, protein translation as well as distal events such as gene transcription (Higley and Sabatini, 2008; Nimchinsky et al., 2002). Astrocytes can also respond to cell–cell signaling transmitters with transient compartmentalized increases of Ca²⁺ within their dendritic processes (Di Castro et al., 2011; Panatier et al., 2011). This type of localized cell–cell communication has been viewed as a hallmark of the nervous system. Dendritic morphologies are not unique to the nervous system, but it is not known if nonneuronal dendrites, such as those on melanocytes, are also capable of compartmentalizing signals received from adjacent cells. Previous work has demonstrated that Ca²⁺ concentration increases within the cell bodies of melanocytes in response to the exogenous addition of keratinocyte-derived paracrine factors (Imokawa et al., 1997; Kang et al., 1998). Here we show that Ca²⁺ transients originated from distinct regions within the melanocyte dendrite when they are co-cultured with keratinocytes. Similar Ca²⁺ transients are seen in the melanocyte dendrites and spines with the addition of ET-1 and ACh (Fig. 7, Fig. S3, and Fig. S4). These data are indicative of functional domains along melanocyte dendrites in which receptors and linked signaling cascades are restricted to, or active at, specific regions of the dendrite. The observed spatial distribution of localized Ca²⁺ transients could be the consequence of, but not limited to, Ca²⁺ buffering proteins, ER Ca²⁺ availability, and spatial distribution of IP3 receptors, as well as store operated calcium and calcium-induced calcium response machinery. Indeed, the spread of the Ca²⁺ response over a 14–22-µm (Fig. 3 H) length across the dendrites resembles the inositol-1,4,5-trisphosphate receptor-mediated dendritic Ca²⁺ transients previously described in pyramidal neurons and Purkinje cells (Finch and Augustine, 1998; Fitzpatrick et al., 2009; Manita and Ross, 2009).

ET-1 modulates, in a dose-dependent manner, melanogenesis, mRNA transcription, and proliferation of melanocytes in mono-culture (Imokawa et al., 1997; Imokawa et al., 1995; Tada et al., 1998). While transcription of mRNA and cell proliferation require transduction of signals into the nucleus, melanogenesis can be regulated locally (within dendrites) or globally (throughout the entire cell). Local regulation is achieved through alterations in the activity of melanin-synthesizing enzymes on melanosomes, whereas global regulation occurs through changes in the amount of melanosome-specific proteins. ET-1 has been shown to both up-regulate activity of tyrosinase, a key enzyme in melanin synthesis, and increase tyrosinase mRNA transcription in melanocytes (Imokawa et al., 1997). The ability to threshold a response to stimulus, such as local versus global Ca²⁺ increases, offers a possible mechanism by which melanocytes can regulate such outcomes. Previous work (Higley and Sabatini, 2008; Nimchinsky et al., 2002) has shown that compartmentalization of second messengers allows for spatial control of biochemical response to external stimuli that provide the cell a mechanism for local regulation of a variety of processes including protein translation, enzymatic activation, and recruitment of membrane receptors. We show that 10 nM ET-1 elicits a strong global change in cytosolic Ca²⁺ within seconds after application (Fig. 6 A, Fig. S3, and Video 3). However, in co-cultures without exogenous addition of ET-1, melanocytes have discrete dendritic transients in response to physiological levels of ET-1 secretion from neighboring keratinocytes (Fig. 3 B, Fig. 4, and Video 1) that are rapidly blocked by antagonists, suggesting an active signaling rather than a slow modulatory effect on transcription. These effects are at a much faster time scale than the previously described work showing global changes in melanocytes, such as transcriptional change, when melanocytes were incubated with 10 nM ET-1 for one or more days (Imokawa et al., 1997).

It should be noted that inhibition of the receptors for endothelin and ACh did not completely eliminate the number of melanocytes with dendritic Ca²⁺ transients in co-culture with keratinocytes (Fig. 7 E). The remaining Ca²⁺ transients in the melanocytes could be the result of other keratinocyte-derived factors or melanocyte autocrine factors or intrinsic signaling. While melanocytes in mono-culture did have a some dendritic Ca²⁺ transients, they were significantly fewer than those in co-culture (Fig. 3 G) and were often below the level seen in the co-cultures in the presence of antagonists of the ET-1 (BQ788) and ACh (atropine) (Fig. 7 E). This suggests there may be an additional secreted factor contributing to the dendritic Ca²⁺ transients in the melanocytes.

Last, we found dendritic spine-like structures on melanocytes in co-culture and in intact human skin. The morphology of the spine-like structures was strikingly similar to what has been described for neuronal dendritic spines (Nimchinsky et al., 2002; Stuart et al., 2007; Papa et al., 1995). While some spines were stable, others underwent morphological changes over time, similar to dendritic spines in neurons of live mice (Berning et al., 2012). We also observed that they were capable of compartmentalizing Ca²⁺, another feature of neuronal dendritic spines. Neuronal dendritic spines are dynamic structures involved in short- and long-term plasticity at synapses. They can provide spatial restriction of cell–cell signaling that is disseminated throughout the cell upon receiving input that surpasses a threshold limit. It is possible that the dendritic spine-like structures on melanocyte dendrites function to integrate signals that determine the states of melanin production while providing a way to adapt to long-term responses to UV radiation. While further work is needed to determine if melanocyte dendritic spine-like structures are involved in cell–cell signaling, the presence of these physical structures on melanocytes provides an opportunity to understand the function and signaling mechanisms in such structures outside of the nervous system.

Melanocytes and keratinocytes were isolated as previously described (Picot, 2005) from neonatal foreskins obtained after routine circumcision, provided by Mount Sinai School of Medicine, New York, NY, or Weill Cornell Medical College, New York, NY, in accordance with the Rockefeller University Institutional Review Board protocol RBE-0721. Tissue was kept in CO₂ Independent Media (Gibco–Thermo Fisher Scientific) with 1× Antibiototic-Antimycotic (Gibco) at 4°C until processed for cell culture. In short, the epidermis was removed from the dermis via 10 mg/ml dispase II, neutral protease, grade II (Roche–Sigma-Aldrich), digestion for 14–17 h at 4°C. Single-cell suspension was achieved by mincing the epidermis and digesting with 0.5% trypsin (Gibco) for 5 min at 37°C. Trypsin was deactivated with soybean trypsin inhibitor (Gibco), and the cells were washed with Hanks’ balanced salt solution, no Mg²⁺, no Ca²⁺ (Gibco), before plating. Cells were grown at 37°C, in 5% CO₂ with 1× Antibioltic-Antimycotic. After the first media replacement, all antibiotics and antifungals were omitted from the growth media.

Melanocytes and keratinocytes were grown in melanocyte growth media (Medium 254 with Human Melanocyte Grow Serum-2; Gibco) and keratinocyte growth media (Epilife with Human Keratinocyte Growth Serum; Gibco), respectively. Only cells with five passages or fewer were used for experiments. 293T-HEK cells were grown in DMEM, high glucose, and pyruvate media (Gibco) supplemented with nonessential amino acids (Gibco) and 10% FBS at 37°C, 5% CO₂.

For melanocyte–keratinocyte co-cultures, melanocytes and keratinocytes were seeded at a 1:10 ratio (0.5 × 10⁴:5 × 10⁴ cells) onto the 10-mm glass window of a 35-mm MatTek dish (P35G-1.5-10-C, MatTek) in keratinocyte growth media (EpiLife media with Human Keratinocyte Growth Serum) and incubated 12–24 h at 37°C, 5% CO₂. Once a confluent monolayer was achieved, the media were changed to keratinocyte growth media containing 1.06 mM CaCl₂ final concentration. Subsequent experiments were performed 48–72 h after switching to 1.06 mM CaCl₂ media.

Within 12 h after harvest, neonatal foreskin (kept at 4°C in DMEM, high glucose, no pyruvate [Gibco] and Antibiotic-Antimytotic or CO₂ Independent media [Gibco] and Antibiotic-Antimytotic) was fixed overnight at 4°C with 4% paraformaldehyde (Electron Microscopy Sciences). After washing with PBS, the fixed tissue was incubated with 30% sucrose in DPBS overnight at 4°C. The sucrose was exchanged with Optimal Cutting Temperature OCT compound (VWR) and frozen on dry ice. In Fig. S1 B, a cryostat section of ∼5–10 µm was used. After immunofluorescence, the section was mounted in ProLong Gold (Molecular Probes) and sealed.

For Fig. S1, the epidermis was removed from the dermis by overnight incubation with dispase II. Freshly isolated epidermal sheets were fixed with 4% paraformaldehyde overnight at 4°C and washed before immunofluorescence staining. Epidermal sheets were imaged in PBS rather than mounting media to prevent compaction of tissue and loss of z resolution.

Cell cultures were washed 1× with PBS, fixed in 4% paraformaldehyde for 10–30 min at room temperature, and then washed before immunofluorescence staining.

After fixation, tissue (either fixed frozen sections or epidermal sheets) or cells were incubated in blocking buffer: 2.5% donkey serum, 2.5% goat serum (Jackson ImmunoResearch Laboratories), 1% bovine serum albumin (Sigma-Aldrich), and 0.1% Triton X-100 (Sigma-Aldrich) for 1–2 h at room temperature. The following primary antibodies were used at the indicated concentration in blocking buffer overnight at 4°C: mouse monoclonal anti-TRP1 1:200 (TA99, ab3312, Abcam), mouse monoclonal anti-K14 1:300 (LL002, ab7800), mouse monoclonal anti-E cadherin 1:100 (ab1416, Abcam), rabbit monoclonal anti-β catenin 1:100 (8480T, Cell Signaling Technology), mouse monoclonal anti-PCAD 1:100 (MAB861, R&D Systems), mouse monoclonal anti-cKit 1:100 (CD11705, Invitrogen–Thermo Fisher Scientific), rabbit polyclonal anti-K10 1:500 (Poly19054, BioLegendA), and rat monocolonal anti-α6 integrin (eBioGoH3 [GoH3], Invitrogen–Thermo Fisher Scientific). Secondary antibodies against mouse IgG, rat IgG, or rabbit IgG conjugated to Alexa Fluor 488, 594, or 680 (Molecular Probes–Thermo Fisher Scientific) were used at a 1:1,000 dilution for 1–2 h at room temperature. Hoechst (Molecular Probes) was used at a 1:1,000 dilution in PBS. Secondary antibodies against mouse IgG2a and mouse IgG1 were used for dual staining of PCAD and TRP1, ECAD and TRP1, and the two-color double anti-TRP1 and anti-cKit immunofluorescence (Fig. S1 A). For all other TRP1/cKit staining, a pan anti-mouse IgG was used to detect both primary antibodies.

Freshly isolated neonatal foreskin epidermis was obtained by an overnight dispase digestion at 4°C and removal from the dermis while on ice. Individual melanocytes were microinjected using a manual microinjection (model MM-88-0, Narashige International) system on an Upright BX61WI microscope with manual syringe pressure application. 2% (wt/vol) Lucifer yellow CH, Lithium Salt (Molecular Probes) was directly injected into the basal surface of the melanocyte at room temperature.

Neonatal foreskin was immersion fixed in 4% paraformaldehyde, 2% glutaraldehyde, 2 mM CaCl₂, and 0.1 M sodium cacodylate at 4°C. Tissue was cut into 1-mm × 500-µm pieces for processing. Tissue was then post-fixed with 2% osmium tetroxide and 1.5% potassium ferrocyanide on ice for 1 h followed by a thiocarbohydrazide amplification step for 20 min and then a subsequent 2% osmium tetroxide incubation for 30 min at room temperature. Tissue was then stained with 1% uranyl acetate overnight at 4°C. Lead aspartate staining was performed for 30 min at 60°C the next day followed by tissue dehydration using sequential ethanol dehydration followed by propylene oxide and embedded in Durcapan resin (Sigma-Aldrich). The sample block was trimmed then mounted on a SEM sample holder using double-sided carbon tape (EMS). Colloidal silver paint (EMS) was used to electrically ground the exposed edges of the tissue block. The entire surface of the specimen was then sputter coated with a thin layer of gold/palladium. The tissue was imaged using back scattered electron mode in a FEI Helios Nanolab 650. Images were recorded after each round of ion beam milling using the SEM beam at 2 keV and 100 pA with a working distance of 2.8 mm. Data acquisition occurred in an automated way using the Auto Slice and View G3 software. The raw images were 2,048 × 1,768 pixels, with 30-nm slices viewed at a −38° cross-sectional angle. Each raw image had a horizontal field width of 30 µm with an XY pixel size of 14–18 nm and 30-nm Z slices. These images were then aligned using the image processing programs in IMOD (Kremer et al., 1996). The aligned images were then use to generate a 3D model in IMOD’s 3dmod image display and modeling program.

Neonatal foreskin was immersion-fixed in 4% paraformaldehyde, 2% glutaraldehyde, 2 mM CaCl₂, and 0.1 M sodium cacodylate at 4°C. Tissue was cut into 1-mm × 500-µm pieces for processing. Tissue was post-fixed with 1% osmium tetroxide on ice for 3 h followed by uranyl acetate staining (1% in 0.05 M maleate buffer, pH 5.2) overnight at 4°C. The next day, tissue was dehydrated using sequential ethanol dehydration followed by propylene oxide and embedded in epon resin (Electron Microscopy Sciences). “Gray sections” for TEM (∼50–100-nm sections) were used. Sections were imaged on a JEM 100CX Transmission Electron Microscope (JEOL USA) with an AMT 2K × 2K digital camera at the Rockefeller University Electron Microscopy Resource Center.

To assess the percent of melanocytes with transients in co-culture and mono-culture conditions, all cultures were seeded onto the 10-mm glass window of a collagen I–coated MatTek dish. The following seeding ratios and growth media were used: “10× ker” is 0.5 × 10⁴ melanocytes expressing GCaMP6f (mel6f) and 5 × 10⁴ keratinocytes, grown in keratinocyte growth media; “Mel alone” is 0.5 × 10⁴ mel6f grown in melanocyte media; “10× Mel” is 0.5 × 10⁴ mel6f and 5 × 10⁴ nontransduced melanocytes in melanocyte media; “10× 293T” is 0.5 × 10⁴ mel6f and 5 × 10⁴ HEK293T cells; and “TW” is 0.5 × 10⁴ mel6f seeded on the glass bottom of a collagen I–coated MatTek dish with 5 × 10⁴ keratinocytes seeded into a collagen I–coated Transwell culture insert, 3 µm pore (CORNING), above the glass window, grown in keratinocyte media. For all culture conditions, cultures were incubated overnight at 37°C, 5% CO₂, and media were replaced with fresh media containing 1.06 mM CaCl₂. Cultures were imaged 48 h after media change. To compare the effect of different growth conditions for co-cultures and mono-cultures, 10× Ker, Mel alone, 10× Mel, 10× 293T, and TW 10× Ker were grown in triplicate with one dish for each media condition: keratinocyte media, melanocyte media, and 50/50 media (1:1 melanocyte: keratinocyte media). Keratinocyte-conditioned media were obtained by harvesting the supernatant from a skin donor–matched keratinocyte culture at each step of the culturing process mentioned above.

Images were acquired using cellSens software (Olympus) on an IX83 microscope (Olympus) with a UPLSAPO 60× 1.3 NA silicone oil objective (Olympus) and an Orca-ER CCD digital camera (Hamamatsu Photonics). Deconvolution was done with Autoquant (Media Cybernetics) standard settings for each filter set using a blind adaptive point-spread function. Image analysis was done in Fiji (http://fiji.sc/).

Cultures in MatTek dishes were imaged (fluorescence and infrared–differential interference contrast) on an Upright BX61WI microscope (Olympus) with a UMPlan FL 60× 1.0 NA water dipping objective (Olympus), an Orca Flash 4.0 digital CMOS camera (Hamamatsu) using MetaMorph image acquisition software (Molecular Devices), or a DeltaVision system (GE Healthcare) at the Rockefeller University Bio-Imaging Resource Center with a 60× NA 1.52 oil objective and SoftWoRx software (DeltaVision). Deconvolution was performed using the adaptive point spread setting in Autoquant (Media Cybernetics) or, for images acquired on the DeltaVision, deconvolution was performed using the measured point-spread function in the SoftWoRx software. All imaging was conducted at room temperature. Image analysis as described below was done in Fiji.

Cultures were imaged in a modified DPBS (10 mM Hepes, 10 mM D-glucose, 0.1 µM glycine, 1.06 mM CaCl₂, and 0.5 mM MgCl₂, pH 7.1–7.3) without exogenous growth factors. Exceptions were Ca²⁺ source, agonist, and antagonist assays, which are described below. Prior to imaging, melanocyte–keratinocyte co-cultures were washed 3× with modified DPBS. Cultures were imaged in streaming acquisition mode in Metamorph Software (Molecular Devices) with a 250-ms exposure time on the Upright BX61WI microscope at room temperature using UMPlan FL 10×, 0.3 NA water dipping objective, or UMPlan FL 60×, 1.0 NA water dipping objective. GCaMP6f was used to measure relative cytosolic Ca²⁺ levels for all Ca²⁺ imaging experiments except for co-cultures, where keratinocytes were infected with EDN1 or CHAT shRNA. Darmacon GIPZ shRNA lentivectors contain turboGFP, and therefore we could not use the green calcium reporter GCaMP6f in melanocytes. Instead, a red calcium reporter, RCaMP1h, was used.

Number of cells with transients was quantified by manually counting the cells with at least one transient Ca²⁺ signal during the 2.5-min imaging period. For each positive cell, presence of local or global Ca²⁺ transients was manually determined.

Number of transients per cell was quantified by manually counting individual peaks in fluorescence by eye using line scans along the dendrite, or a region of interest (ROI) was drawn at initiation sight and a plot of the fluorescence intensity over time was used to determine the number of peaks and thus the number of transients. It should be noted that the use of a 0.3-NA, 10× objective limited the analysis to the brightest fluorescence signals, and therefore our final count is likely an underestimate of the frequency.

Spatiotemporal assessment of transients was done on images acquired in streaming mode with 50 ms or 250 ms exposure using a UMPlan FL 60×, 1.0 NA, water dipping objective. The distribution of the spread in Ca²⁺ signal was determined using data acquired in streaming mode with a 250-ms exposure time using a UMPlan FL 10×, 0.3 NA water dipping objective.

For calculating ΔF/F0, F0 was determined by averaging 10 consecutive fluorescence intensities, within the same ROI, at the lowest fluorescence intensity point during the imaging sequence. ΔF was determining by: ΔF = Fn – F0, where Fn is the fluorescence intensity at that given time point. For ΔF/F0 images, 10 consecutive frames were averaged, and the resulting image was used as F0. The ΔF image was obtained by subtracting the F0 image from the frame corresponding to the peak Ca²⁺ signal in the corresponding ΔF/F0 plot. The final ΔF/F0 image was obtained by dividing the ΔF image by the F0 image.

Co-cultures in modified DPBS, 1.06 mM CaCl₂, were imaged for 2.5 min in streaming mode, 250 ms exposure, to determine the baseline percentage of cells with transients and the number of transients per cell. After baseline imaging, media were replaced with modified DPBS, containing 0 mM CaCl₂ and/or 1 µM thapsigargin, 1 µM BQ788, 1 µM BQ123 (Tocris–BioTechne), 100 µM atropine sulfate (Sigma-Aldrich), or 100 µM atropine sulfate with 1 µM BQ788. To ensure full replacement of the imaging media with the treatment solution, 10 times the volume of the imaging chamber was perfused through the imaging chamber of a MatTek dish using manual syringe-driven delivery through a modified open perfusion insert (model RC-37F; Warner Instruments). The same field of view imaged for the baseline analysis was imaged after a 20-min incubation with the perfused solution. The fold change in percent melanocytes with one or more Ca²⁺ transients was determined by dividing the number of cells with transients during the treatment imaging period by the number of cells with transients during the baseline imaging period. For agonist stimulation, ACh chloride (Tocris) or endothelin-1 acetate salt (Bachem) was perfused into the chamber at 10× excess volume, to ensure complete replacement of imaging media, at the indicated time during the streaming acquisition.

EGFPmem was made by replacing ECFP with EGFP in the Clontech vector pECFPmem (Clontech–Takara Bio USA). EGFPmem was inserted into the HIV-1–based lentivirus vector pLVX Phi3 (Takacs et al., 2017a) using restriction enzymes BamHI and XbaI and Quick Ligation Kit (New England BioLabs).

Annealed sense and antisense DNA oligos containing the sequence for the first 20 amino acids of neuromodulin (mem) flanked by 5′ XhoI and 3′ NotI restriction enzyme sites were inserted into pLVX Phi3 by restriction digest and subsequent ligation using the Quick Ligation Kit (New England BioLabs). Near-infrared fluorescent protein (iRFP; Filonov et al., 2011) was PCR-amplified from the plasmid piRFP (a gift from V. Verkhusha, Addgene plasmid 31857; http://n2t.net/addgene:31857; RRID:Addgene_31857) using forward primer: 5′-CAA​AAG​ATC​GCG​GCC​GAA​GGA​TCT​GTC​GCC​AGG-3′, reverse primer: 5′-GCG​TCC​GGA​GCG​GCC​TTA​CTC​TTC​CAT​CAC​GCC​G-3′, and then inserted into the NotI-digested Phi3-mem plasmid after the neuromodulin plasma membrane binding domain using the In-Fusion HD Cloning Kit (Clonetech).

mCherry (from pmCherry-N1; Clonetech) was inserted in PLVX Zhi3 (Takacs et al., 2017b) using restriction enzymes MluI and BamHI and a Quick Ligation Kit (New England Biolabs).

GCaMP6f from pGP-CMV-GCaMP6f (a gift from D. Kim) was excised from Addgene plasmid 40753 using restriction enzymes BglII and XbaI and inserted into BamHI- and XbaI-digested PLVX Phi3 (Takacs et al., 2017a) using the Quick Ligation Kit.

RCaMPh1 was amplified from pRSET-RCaMP1h, a gift from L. Looger (Addgene plasmid 42874; http://n2t.net/addgene:42874; RRID:Addgene_42874), using forward primer: 5′-TAG​ATC​TCG​AGA​ATT​GCC​ACC​ATG​GGT​TCT​CAT​CAT​CAT​CA-3′, reverse primer: 5′-GAG​CGG​CCG​CGA​ATT​TTA​CTT​CGC​TGT​CAT​CAT​TTG-3′ and inserted into ECoRI-digested PLVX Phi3 using Cold Fusion ligation independent cloning (Systems Bioscience).

Low-passage HEK293T cells (60–80% confluent) were cotransfected using polyethyleneimine (Sigma-Aldrich) at a 5:5:1 ratio of a lentiviral plasmid, a HIV-1 GagPol plasmid, and a VSVg plasmid, respectively. To obtain viral stocks, 88 µg of total DNA was used to transfect a 150-mm dish of 70% confluent cells. Virus-producing HEK293Ts were maintained in DMEM, high glucose, pyruvate media (Gibco) supplemented with nonessential amino acids (Gibco) and 3% FBS (Sigma-Aldrich). Virus-containing media were collected at 24 h, 48 h, and 72 h after transfection, filtered through a 0.45-µm filter (EMD Millipore), and concentrated with Lenti-X Concentrator (Clontech). Concentrated virus was resuspended in Hanks’ balanced salt solution or phosphate buffered saline and stored in aliquots at 20× final concentration. High titer virus stocks were stored at −86°C for future use.

DsiRNAs were generated from DsiRNA sequences predesigned by IDT (Integrated DNA Technologies). Duplexed sequences were as follows: EDNRA 13.1 antisense: 5′-GGA​CAA​GAA​CCG​AUG​UGA​AUU​ACT​T-3′, sense: 5′-AAG​UAA​UUC​ACA​UCG​GUU​CUU​GUC​CAU-3′, EDNRA 13.2 antisense: 5′-GCA​ACC​UUC​UGC​AUU​CAU​AAA​UCT​T-3′, sense: 5′-AAG​AUU​UAU​GAA​UGC​AGA​AGG​UUG​CUA-3′, EDNRB 13.1 antisense: 5′-CAU​GUC​AGU​AUC​AUG​UUC​UCU​AAT​T-3′, sense: 5′-AAU​UAG​AGA​ACA​UGA​UAC​UGA​CAU​GGA-3′, EDNRB 13.2 antisense: 5′-AGU​AUU​GAC​AGA​UAU​CGA​GCU​GUT​G-3′, sense: 5′-CAA​CAG​CUC​GAU​AUC​UGU​CAA​UAC​UCA-3′.

On day 0, DsiRNA were introduced to melanocytes using Lipfectamine RNAiMAX transfection reagent (Invitrogen–Thermo Fisher Scientific), following the recommended transfection procedure for 35-mm dishes. On day 1, melanocytes were infected with Phi3-GCaMP6f and incubated overnight. On day 2, melanocytes were seeded with keratinocytes to generate co-cultures as described above.

Knockdown of CHAT and EDN1 was achieved through transduction of keratinocytes with the GIPZ shRNAs EDN1: V3LHS_151308, V3LHS_370736, V3LHS_370739 CHAT: V3LHS_393516, V3LHS_393518. Keratinocytes were grown for 48 h after transduction and then underwent bulk selection using 2 μg/ml puromycin for 48 h. Cultures were maintained in 1 μg/ml puromycin until seeded for co-cultures.

Keratinocytes were harvested in 1 ml TRIzol Reagent (Invitrogen–Thermo Fisher Scientific), and RNA was purified following the manufacturer’s protocol. cDNA synthesis was done using the SensiFAST cDNA Synthesis kit (Bioline) according to the manufacturer’s protocol. Subsequent qRT-PCR was performed using SensiFAST SYBER NO-ROX Kit, with the manufacturer’s suggested two-step protocol using the follow primers for EDN1: forward 5′-CAG​AAA​CAG​CAG​TCT​TAG​GCG-3′, reverse 5′-GTC​CAG​GTG​GCA​GAA​GTA​GAC-3′; CHAT: forward 5′-CGG​TCC​TCG​TGA​AAG​ACT​CC-3′, reverse 5′-GGA​CTT​GTC​GTA​CCA​GCG​AT-3′; housekeeping gene GPI for internal normalization: forward 5′-CCA​ACA​AGG​ACC​GCT​TCA​AC-3′, reverse 5′-CAT​CAC​GTC​CTC​CGT​CAC​C-3′.

Two-sample t test, one-way ANOVA with Tukey means comparison, and Kruskal–Wallis ANOVA statistical analyses were performed using standard settings in OriginPro software (OriginLab) as indicated in the figure legends and main text. We used the customary threshold of P < 0.05 to declare statistical significance. Sample size and statistical details can be found in the figures, legends, and main text.

Fig. S1 shows melanocyte morphology in intact neonatal foreskin. Fig. S2 shows that growth media do not influence the percent of melanocytes with dendritic Ca²⁺ transients. Fig. S3 shows melanocyte response to exogenous ET-1 in co-cultures and mono-cultures. Fig. S4 shows melanocyte response to exogenous ACh in co-cultures and mono-cultures. Fig. S5 shows the ultrastructure of melanocyte dendritic spine-like structures and pooled vesicles in adjacent keratinocyte. Table S1 shows quantification of melanocyte dendrites and spines from FIB-SEM reconstruction. Video 1 shows Ca²⁺ transients in co-cultured melanocytes. Video 2 shows dendritic Ca²⁺ transients. Video 3 shows the co-cultured melanocyte response to exogenous 10 nM ET-1. Video 4 shows abrogation of Ca²⁺ transients in co-culture melanocytes in the presence of BQ788 but not BQ123. Video 5 shows co-cultured melanocyte response to exogenous ACh. Video 6 shows Ca²⁺ transients in a co-cultured melanocyte dendrite and dendritic spines. Video 7 shows Ca²⁺ transient in co-cultured melanocyte dendritic spines.

We thank the Departments of Obstetrics and Gynecology at Mt. Sinai School of Medicine and Cornell-Weill for their assistance in the collection of the neonatal foreskins. We thank the Rockefeller Electron Microscopy Resource Center, the Rockefeller University Bio-Imaging Resource Center, and the Simons Electron Microscopy Center for assistance. We thank Pascal Maguin for assistance with cloning and Marina Bleck, Constantin Takacs, Michelle Itano, Sohail Tavazoie, and Elaine Fuchs for helpful discussions.

The FIB-SEM data were acquired at the Simons Electron Microscopy Center and National Resource for Automated Molecular Microscopy located at the New York Structural Biology Center, supported by grants from the Simons Foundation (349247), NYSTAR, and the National Institutes of Health, National Institute of General Medical Sciences (GM103310), with additional support from National Institutes of Health grant S10 RR029300-01. This work was supported in part by a National Center for Advancing Translational Sciences grant (UL1 TR001866; National Institutes of Health Clinical and Translational Science Award program), National Institute of General Medical Sciences of the National Institutes of Health under award T32GM066699, and a Rockefeller University Women in Science Fellowship to R.L. Belote.

The authors declare no competing financial interests.

Author contributions: R.L. Belote performed the experiments, and R.L. Belote and S.M. Simon designed the experiments and wrote the manuscript.