Research ArticleCell Biology

Chemical synapses without synaptic vesicles: Purinergic neurotransmission through a CALHM1 channel-mitochondrial signaling complex

See allHide authors and affiliations

Sci. Signal.  08 May 2018:
Vol. 11, Issue 529, eaao1815
DOI: 10.1126/scisignal.aao1815

Tasting with ATP

The savory taste of umami, the sweetness of sugar, and the bitterness of quinine are transduced by type II taste cells. Unlike most other receptor cells, type II taste cells release their neurotransmitter, ATP, through voltage-gated CALHM1 channels instead of neurotransmitter-containing vesicles. Romanov et al. found that the source of ATP was unusual, large mitochondria, closely opposed to clusters of CALHM1 channels within the plasma membrane of type II taste cells. This arrangement enables an alternate method of chemical neurotransmission that does not rely on vesicles.

Abstract

Conventional chemical synapses in the nervous system involve a presynaptic accumulation of neurotransmitter-containing vesicles, which fuse with the plasma membrane to release neurotransmitters that activate postsynaptic receptors. In taste buds, type II receptor cells do not have conventional synaptic features but nonetheless show regulated release of their afferent neurotransmitter, ATP, through a large-pore, voltage-gated channel, CALHM1. Immunohistochemistry revealed that CALHM1 was localized to points of contact between the receptor cells and sensory nerve fibers. Ultrastructural and super-resolution light microscopy showed that the CALHM1 channels were consistently associated with distinctive, large (1- to 2-μm) mitochondria spaced 20 to 40 nm from the presynaptic membrane. Pharmacological disruption of the mitochondrial respiratory chain limited the ability of taste cells to release ATP, suggesting that the immediate source of released ATP was the mitochondrion rather than a cytoplasmic pool of ATP. These large mitochondria may serve as both a reservoir of releasable ATP and the site of synthesis. The juxtaposition of the large mitochondria to areas of membrane displaying CALHM1 also defines a restricted compartment that limits the influx of Ca2+ upon opening of the nonselective CALHM1 channels. These findings reveal a distinctive organelle signature and functional organization for regulated, focal release of purinergic signals in the absence of synaptic vesicles.

INTRODUCTION

A synapse, defined originally in 1897 for the nervous system by Foster and Sherrington (1), is a point of cell-to-cell contact specialized for rapid signaling between cells. This term has also been applied to the signaling complex formed at the point of contact between T cells and antigen-presenting cells of the immune system (2). In the nervous system, synapses may be either electrical or chemical in nature. An electrical synapse involves physical contact between cell membranes enabling direct transmission of electrical signals between cells. Chemical synapses entail release of neurotransmitter by the presynaptic, signaling cell into a gap between the cells followed by activation of specific receptors on the postsynaptic cell to evoke a cellular response. In a conventional chemical synapse, the neurotransmitter molecules lie within synaptic vesicles, which fuse to the adjacent presynaptic membrane after Ca2+ influx. Here, we describe a different type of chemical synapse by which taste receptor cells signal to the sensory nerve fibers.

Taste buds, the sensory end organs for gustation, comprise 50 to 80 specialized epithelial cells residing in distinctive papillae of the tongue and elsewhere in the oropharynx. When a taste substance stimulates the apices of the taste receptor cells, the cells release neurotransmitter onto the sensory nerve fibers innervating the taste bud. The key neurotransmitter released in this system is adenosine 5′-triphosphate (ATP) (3, 4) because either pharmacological blockade or genetic deletion of the neural ATP receptors eliminates nearly all responses in the taste nerves (35). Consistent with the necessity for purinergic transmission in this system, all gustatory nerve fibers have P2X-type purinergic receptors (68).

The functional contacts between taste cells and nerve fibers differ according to the type of taste cell involved. Type I taste cells are glia-like and do not display specialized points of contact with nerve fibers (9), whereas type III cells, which transduce sour (acid) and perhaps other ionic qualities (911), form conventional chemical synapses complete with voltage-gated Ca2+ channels (12, 13), pre- and postsynaptic membrane thickening, and synaptic vesicles with their associated SNAP receptor (SNARE) complex proteins (14). In these features, type III cells are similar to axonless receptor cells in other sensory systems, such as hair cells and photoreceptors. In contrast, type II taste cells, which transduce sweet, umami, or bitter tastes, lack neuronal SNARE proteins and synaptic vesicles (12, 15) but nonetheless release ATP as a neurotransmitter in a regulated fashion (1618). This output from type II cells is unconventional because it does not involve the Ca2+-dependent exocytosis of vesicles but relies on ATP release through voltage-gated ATP-permeable channels (16, 17). The transduction cascade in these taste cells begins with G protein–coupled receptors whose activation evokes release of Ca2+ from intracellular stores; the increases in Ca2+ trigger the transduction channel TrpM5 (transient receptor potential cation channel subfamily M member 5) to initiate an action potential in the taste cells (9, 19). Taruno et al. (20) demonstrated that these action potentials trigger the opening of the transmembrane protein calcium homeostasis modulator 1 (CALHM1), which forms voltage-gated, ATP-permeable channels responsible for neurotransmission from type II cells to the taste nerve. The channel-dependent ATP release exhibits a steep dependence on membrane voltage with a threshold of about −10 mV (16, 21, 22). Accordingly, type II taste cells, which generate action potentials above the ATP release threshold (20, 22), entrain afferent output reliably as a function of the magnitude of stimulation [(18); see also (22) for discussion]. Thus, similar to quantal release in classical synapses, neurotransmission in type II taste cells is pulsatile because ATP release is driven by discrete action potentials (16, 18, 20). Although ATP secretion is a common feature of other tissues using purinergic autocrine or paracrine regulation, ATP secretion in those systems is long-lasting and continuous unlike in taste buds (23).

The CALHM1 channels underlying the ATP release from taste cells are large-pore (≈14 Å) (19, 20), slow-acting channels that accommodate the outrush of ATP molecules and influx of numerous ionic species. However, the distribution of these channels across the taste cell membrane is unclear. If widely distributed, opening of such large-pore channels would permit wholesale influx of Ca2+ into the cells, possibly triggering adverse events such as apoptosis. The ongoing functionality of the receptor cells, even after maximal activation, suggests that Ca2+ influx is somehow limited. We report that, unlike other channels implicated in taste transduction (24, 25), CALHM1 was tightly restricted to points of contact with afferent nerve fibers. These points of contact between type II taste cells and nerve fibers often exhibit large, so-called “atypical” mitochondria with enlarged tubular cristae (7, 26). Note that we retain the term “atypical” as defined previously for this system (26), even while recognizing that mitochondria assume various morphologies depending on their location and utilization. Mitochondria with tubular cristae also occur in diverse locations including photoreceptors and steroid-producing cells (27). We investigated the possibility that these atypical mitochondria were closely associated with the CALHM1 channels, thereby not only offering a source of ATP for release but also forming a restricted compartment for controlling Ca2+ influx.

RESULTS

Ca2+ flux and CALHM1 channels

The main candidate for the release of ATP from taste cells is the nonselective, large-pore ion channel CALHM1, with a pore size in excess of about 14 Å (20). Being nonselective, the CALHM1 channel is permeable not only to ATP but also to various ions, including Ca2+. It therefore might be expected that appropriate depolarization of type II cells should trigger local Ca2+ influx with concommitant rise in intracellular Ca2+, as is the case after activation of type III taste cells, which express voltage-gated Ca2+ channels but not CALHM1 (28, 29). In our experiments, type II taste cells, which were identified physiologically as those showing large, nonselective outward currents and ATP release upon depolarization (16, 21, 28), were assayed simultaneously with patch-clamp and Ca2+ imaging techniques (Fig. 1A). Depolarization-elicited ATP release was not always accompanied by marked increases in intracellular Ca2+. Rather, of 40 type II cells examined, 23 exhibited no evident change in cytoplasmic Ca2+ upon depolarization to 0 mV and higher (Fig. 1, B and C). In the remaining 17 cells, depolarization elicited both robust ATP release (Fig. 1D, red trace) and detectable Ca2+ transients, which, however, were small and returned to baseline within few seconds (Fig. 1D, black trace), suggesting that these transients were local.

Fig. 1 Depolarization-evoked Ca2+ signals in type II taste cells.

(A) Preparation for simultaneous monitoring of intracellular Ca2+ and ATP release. (B) Calcium responses (black dots; means ± SD; n = 7 cells; see table S1 for details) and voltage dependence of ATP release [red triangles; data from Romanov et al. (22)]. AU, arbitrary units. (C and D) Representative of recordings showing the effect of depolarization on (C) intracellular Ca2+ (n = 23 cells) and (D) voltage-related Ca2+ signals (n = 17 cells). Black arrows indicate Ca2+ responses, whereas the red line shows ATP sensor responses. (E) Effect of altering extracellular calcium on Ca2+ responses. Data are representative of five cells for each alteration. (F) Effect of decreasing extracellular pH on currents and Ca2+responses. Data are representative of three cells. (G) Effect of applying extracellular 100 μM Gd3+ (gray bar) on current and Ca2+ responses (n = 4 cells). P < 0.001, two-way ANOVA. Bottom trace shows control preparations. (H) Effect of applying negative pressure through the patch pipette on calcium events. Data are representative of three of five cells.

Several findings point to CALHM1 as being responsible for the small Ca2+ transients observed in these depolarized type II cells. First, in those cells showing a Ca2+ signal, the magnitude of the Ca2+ influx was proportional to the amount of ATP released (Fig. 1B and table S1), suggesting that these processes are regulated by a common mechanism. Second, these Ca2+ signals varied with bath Ca2+, in that they were abolished at zero Ca2+ and markedly increased at 10 mM Ca2+ (Fig. 1E). This finding demonstrated that Ca2+ influx rather than release from intracellular stores was the basis for the voltage-gated intracellular Ca2+ signals. In addition, CALHM1 is structurally related to the connexin and innexin family of channels, all of which exhibit decreased conductance with low pH (3032). We found that lowering extracellular pH from 7.2 to 5.8 decreased Ca2+ influx by about 50% (Fig. 1F). Finally, extracellular Gd3+ (100 μM), which inhibits various channels including CALHM1 (33), strongly decreased Ca2+ influx and CALHM-mediated voltage-gated outward currents in type II taste cells (Fig. 1G). Together with previous findings, these results implicate CALHM1 as the primary conduit for Ca2+ influx into type II taste cells along with its previously demonstrated role in ATP efflux from these cells (19).

Why depolarization capable of driving a large ion current through CALHM channels resulted in a negligible Ca2+ signal in most type II cells remained enigmatic. We hypothesized that Ca2+ influx might be limited by compartmentalization of CALHM channels into a signaling complex with an atypical mitochondrion, which could provide both space exclusion and mitochondrial Ca2+ uptake, thereby preventing the transformation of CALHM-mediated Ca2+ influx into a large and global Ca2+ transient. If this compartmentalization was the case, then mechanical disturbance of type II cells might disrupt the restricted compartment, thereby enhancing measureable Ca2+ responses on depolarization. Following this idea, we applied relatively strong negative pressure (40–60 cm H2O) to five type II cells, which increased Ca2+ responses in three of them (Fig. 1H). These findings are consistent with the hypothesized special compartmentalization of CALHM channels and atypical mitochondria playing a role in restriction of Ca2+ influx.

Localization of CALHM1 channels

To assess the relationship between the CALHM1 channels, mitochondria, and purinoceptive nerve fibers, we performed high-resolution immunohistochemistry of taste buds. Immunohistochemistry for CALHM1 using an antibody mapping to residues 318 to 328 on the intracellular domain of human CALHM1 (fig. S1A) revealed small (1 to 2 μm), elongate patches of reactivity associated with the plasma membrane of type II taste cells identified by expression of green fluorescent protein (GFP) driven by the promoter of the taste transduction channel TrpM5 (Fig. 2, A to E). Such patches are not seen in tissues prepared from CALHM1 knockout animals (fig. S1B). Furthermore, triple-label immunohistochemistry showed a close relationship between these CALHM1 patches and zones of contact between the type II cells and nerve fibers expressing the P2X2 purinergic receptors at the light microscopic level (Fig. 2, A to F). Such restriction of channels to a focal basolateral compartment has not been noted for other taste transduction components of type II taste cells including TrpM5 (24) and various Na channels (25).

Fig. 2 CALHM1 lies at points of contact between taste cells and nerve fibers.

(A to E) Immunostaining for CALHM1 (red), nerve fibers (P2X2; green), and type II taste cells (TrpM5-driven GFP; blue). Scale bars, 10 μm (A) and 5 μm [(B) to (E)]. Data are representative of multiple taste buds from each lingual papilla in more than five animals. Zeiss Airyscan images are shown in (D) and (E) (fig. S2); others are conventional laser scanning confocal microscope. Data are representative of four Airyscan images. (F) A line plot of a point of contact shown in (E). See also fig. S2. CALHM1 staining, red curve; type II cell cytoplasm, blue curve; nerve membranes, bimodal green line.

To assess whether CALHM1 channels were localized appropriately to serve as the release channels associated with atypical mitochondria, we used immunohistochemistry to compare the distribution of CALHM1 channels and mitochondria, which were identified with cytochrome C (CytC) immunostaining. The CALHM1 clusters localized to sites displaying patches of CytC activity appropriate in size to be atypical mitochondria (Fig. 3, A to C). Together, these results suggest a close association between the CALHM1 channel and atypical mitochondria and between the CALHM1 channel and purinoceptive nerve fibers. Ultrastructural localization of the CALHM1 channels showed specific localization of the channel to the membrane subjacent to the atypical mitochondria (Fig. 3, D and E) and always adjacent to a nerve fiber profile. The dark immunoreaction product (Fig. 3E, arrows) lies at the interface between the taste cell membrane and the atypical mitochondrion, consistent with the intracellular location of the CALHM1 epitope recognized by the antibody. No aggregates of CALHM1 label occurred at other locations in the taste cells.

Fig. 3 CALHM1 lies adjacent to large mitochondria.

(A to C) Colocalization of CALHM1 (red) and CytC (green) marking mitochondria. Type II taste cells (TrpM5-driven GFP) are rendered in blue. Scale bars, 2 μm [(A) and (B)] and 10 μm (C) with insets at 2× digital magnification and Unsharp Mask of 7.5 pixels at 85%. See fig. S1 for validation of the CALHM1 antibody. Data are representative of multiple taste buds from each lingual papilla in more than five animals. (D and E) Electron micrographs of a taste cell–neurite contact immunoreacted for CALHM1 with a peroxidase-based detection system. (E) An enlargement of the boxed area in (D). Scale bars, 500 nm (D) and 200 nm (E). The nerve fiber process is shaded yellow in (D) to facilitate visualization of the different structures in this unstained section. Data are representative of three taste cells in the CV papilla of one animal.

Relationship of atypical mitochondria to CALHM1 and nerve fibers

We investigated the possibility that the CALHM1 channels were specifically associated with previously described atypical mitochondria and that both localized to points of contact between type II cells and nerve fibers. To date, it has been unclear whether the atypical mitochondria occur only in type II cells, and then only at taste-cell neurite contacts, or are distributed more widely. In many cells, mitochondria are polymorphic with different-appearing mitochondria situated at specific locations in a cell. For example, in photoreceptors, densely clustered mitochondria partition the cell into separate domains: cell body and outer segment (34). In that context, the mitochondria are crucial in maintaining distinct Ca2+ compartments in the two domains.

To test whether the atypical mitochondria in taste cells are specifically associated with neuronal contacts, we used ultra-resolution confocal microscopy (Fig. 2, D and E, and fig. S2) on specimens stained for CALHM1 and P2X2 or P2X3 receptors. In addition, we used serial blockface scanning electron microscopy (sbfSEM) to reconstruct the relationships between taste cells and nerve fibers in four taste buds from the circumvallate (CV) papillae of three mice. CALHM1 immunoreactivity was localized to small spots along the surface of type II taste cells identified by GFP immunofluorescence (Fig. 2, A to E). As demonstrated by the line plot through CALHM1 puncta (Fig. 2F and fig. S2), a P2X-immunoreactive nerve fiber was in contact with the taste cell for each of these CALHM1 spots. In our sbfSEM samples, of the more than 125 atypical mitochondria identified, ranging in size from 0.5 to 3 μm in length, all except 1 lay at the point of contact between a type II taste cell and an afferent nerve fiber (Fig. 4, A to D, and movies S1 and S2). Individual taste cells often had multiple atypical mitochondria at the contact points with a single adjacent nerve fiber (Fig. 4, A, B, and D, and movies S1 and S2). Moreover, conventional mitochondria did not lie closely apposed to the membrane of type II taste cells at points of contact with nerve fibers, although we did identify three mitochondria with features of both atypical and conventional organization of cristae.

Fig. 4 Reconstructions of typical mitochondria from sbfSEM data.

(A) Type II taste cell (green), a nerve fiber (yellow), and a type I cell (gray) showing the relationship of atypical mitochondria (red) and sensory nerve fibers. Data are representative of 29 type II taste cells in three CV taste buds from two mice. See movie S1. (B) Reconstruction of a type II taste cell (pale green; nucleus, dark green) showing typical (blue) and atypical (red) mitochondria in relation to the sensory nerve fiber (yellow). Scale bar, 10 μm. Data are representative of three reconstructed cells. See movie S2. (C) Enlarged bracketed region from (B), showing separation of atypical (red) from typical (blue) mitochondria. Scale bar, 1 μm. See movie S3. (D) Single image from an sbfSEM data set showing atypical mitochondria (red) and the afferent nerve terminal (yellow). Boxed area enlarged at right showing the tubular cristae within the atypical mitochondria. Scale bars, 1 μm. Data are representative of greater than 100 identified atypical mitochondria. (E and F) 3D reconstructions of the pair of typical (E) and atypical (F) mitochondria in (D). See movie S4.

Reconstructions of all mitochondria in two of these cells showed that the atypical mitochondria were distinct organelles, essentially disconnected from the network of conventional mitochondria within the same cell (Fig. 4, C to F, and movie S3). The relationship between the atypical mitochondrion, the taste cell plasma membrane, and neurite membrane showed a rigid dimensionality (Fig. 5A) with the estimated distance between mitochondria and taste cell plasma membrane being ~20 to 30 nm (Fig. 5, A and B) and a corresponding intercellular distance between nerve and taste cell (synaptic cleft) of ~10 to 15 nm. The uniform spacing between the mitochondrial outer membrane and the plasma membrane strongly suggests the presence of a scaffolding to maintain this relationship. On the basis of three-dimensional (3D) reconstructions (Fig. 5B) and analysis of mitochondrial sizes (Table 1), we estimated the volumes of these spaces for a medium-sized atypical mitochondrion (1.4 μm across and 2 μm long) to be about 0.04 to 0.08 fl.

Fig. 5 Cell membrane specialization at the CALHM1-mitochondrial complex.

(A) Conventional transmission electron micrograph illustrating the relationship between the membrane of the atypical mitochondrion, the plasma membrane of the taste cells, and the nerve ending. Inset on the lower right is an enlargement of the boxed area in the main image. The inset has been enlarged (3×), and the edges has been enhanced by application of an Unsharp Mask filter (60% at five pixels). The arrow points to the synaptic cleft between the taste cell and nerve fiber. Scale bars, 100 nm for the main image and 33.3 nm for the inset. Data are representative of a minimum of five taste buds in CV papillae of more than five mice. (B) 3D reconstruction of a contact between type II taste cell and the nerve ending (yellow), including visualizations of conventional mitochondria in the nerve ending (magenta), and an atypical mitochondrion (red) and conventional mitochondria (cyan) inside the taste cell (green). Putative submitochondrial active zone is highlighted in blue. Detail to the right shows the method for approximating the volume of the different synaptic compartments.

Table 1 Measurements (in pixels) of mitochondrial features.

Typical and atypical mitochondria were compared in three different general size classes: small, medium, and large.

View this table:

Ultrastructure of atypical mitochondria

The atypical mitochondria at taste cell–nerve fiber contacts had unusually large tubular cristae in contrast to the thin parallel cristae of conventional mitochondria (Figs. 4, E and F, and 5A, and movie S4). The structure of cristae within mitochondria differs according to many factors including bioenergetic state, ATP synthase dimerization, lipid composition, and fission-fusion events (35). In mitochondria, ATP synthase on the crista membranes generates ATP that is rapidly moved by ATP translocase from the mitochondrial matrix into the luminal space that is continuous with the intermembrane space between the outer mitochondrial membrane and the walls of the cristae. Once in the intermembrane space, ATP leaves the mitochondrion through VDAC (also known as porin) channels, which are abundant in the outer membrane. Our measurements of conventional mitochondria (Table 1) within the type II taste cells showed that the intermembrane space represented about 10 to 12% of the total volume of the mitochondrion, as it does in many other mitochondrial systems (9.8 to 20%) in the nervous system (27, 33). In contrast, the intermembrane volume of the atypical mitochondria (Table 1) accounted for about 38% of the total mitochondrial volume. Although ATP in the intermembrane space should readily diffuse through VDAC channels on the external bounding membrane of the mitochondrion, the large size of the atypical mitochondrion coupled with its larger intermembrane space may offer a relatively large reservoir of newly synthesized ATP available to buffer local release. For a mid-sized atypical mitochondrion, we estimated that the volume of the intermembrane space would be about 0.35 fl, considerably larger than the 0.04- to 0.08-fl volume of the restricted space between the atypical mitochondrion and the plasma membrane of the cell. Together with our data on the consistent spacing of the mitochondria and the plasma membrane of the taste cell, these findings suggest the possibility of a unique presynaptic compartment that ensures robust and specific nonvesicular ATP storage and release onto nerve fibers through gated release channels. In addition, the restricted space between the mitochondrial outer membrane and the plasma membrane of the taste cell coupled with the ability of mitochondria to take up free Ca2+ may account for the limited diffusion of Ca2+ into the taste cell upon opening of the CALHM1 channel (36).

Role of mitochondria in ATP release

Because type II taste cells release ATP in response to depolarization, but lack synaptic vesicles, a likely source of ATP is the atypical mitochondria associated with the taste cell–nerve fiber contact. To test the necessity for active production of ATP in the release process, we used oligomycin, an inhibitor of ATP synthase, to block production of ATP by oxidative phosphorylation in mitochondria, allowing for the buildup of adenosine 5′-diphosphate (ADP) within the matrix and intermembrane spaces but leaving intact the ability of the cell to generate ATP by direct glycolysis. Using ATP biosensors to monitor efflux of ATP from oligomycin-treated cells, we found that ATP efflux lasted 3 to 5 min but declined rapidly thereafter (Fig. 6, A and B, and fig. S3A), although the taste cells themselves and the biosensor cell remained physiologically functional in terms of generating Ca2+ responses to exogenous activators even in the presence of oligomycin. Furthermore, after depletion of the mitochondrial ATP store by oligomycin, the taste cells then release ADP (Fig. 6, C and D, and fig. S3, B and C), which likely originates from the accumulation of ADP in the intermembrane space of the mitochondria in the absence of ATP synthase activity and the resulting cessation of translocase activity. These results indicate that taste cells release ATP produced in mitochondria and suggest the existence of a pool of releasable ATP that is poorly exchangeable with the pool of cytosolic ATP. The ultrastructure of the atypical mitochondria and junctional relationships between the taste cell and the nerve fiber are consistent with this hypothesis. Assuming that the pool of ATP in the cytoplasm in the 0.04- to 0.08-fl space between the mitochondrion and the plasma membrane represents the readily releasable pool, then the 0.35-fl volume within the mitochondrial intermembrane space would represent the reservoir by which the releasable pool is refilled during and after channel opening.

Fig. 6 Blocking ATP production in mitochondria blocks release of ATP.

(A) ATP biosensor cells were CHO cells transfected with mRNAs encoding P2X2 and P2X3 ionotropic purinergic receptors, which are responsive to ATP but not to ADP. (B) Depolarization-evoked ATP release from taste cells in the absence (blue trace) or presence (red trace) of oligomycin. The bar graph shows a quantitative comparison of the responses (5 min with oligomycin; n = 4 experiments with individual taste cells; means ± SEM; *P < 0.05). (C) ATP/ADP biosensor cells were COS-1 cells, which endogenously express P2Y receptors and are responsive to both ATP and ADP. (D) Depolarization-evoked release of ATP and ADP from taste cells in the absence (blue trace) or presence (red trace) of oligomycin. The bar graph shows a quantitative comparison of the responses (5 min with oligomycin; means ± SEM; n = 6 experiments with individual taste cells; *P < 0.05, t test). Scale bars, 10 μm [(A) and (C)]. Validation of biosensor sensitivity and functionality in the presence of oligomycin given in fig. S3.

Carbenoxolone (CBX) inhibits this release, which was previously attributed to a block of pannexin (Pnx1) channels, then believed to be the route of ATP release (1618, 31). If CALHM1, not Pnx1, is the crucial ATP release channel, then why should CBX interfere with ATP release? One possibility is that CBX can collapse the mitochondrial potential (37, 38), which would halt production of ATP, thereby reducing ATP release. To test this notion, we loaded the mitochondrial potential dye (Mito-ID) into taste cells in preparations of nondissociated taste buds (fig. S4A). When Mito-ID was applied simultaneously with 10 μM CBX (which is sufficient to inhibit ATP secretion in taste cells) (17, 18), intramitochondrial orange fluorescence exhibited a bell-shaped distribution as a function of incubation time (fig. S4B), suggesting that CBX treatment substantially reduced mitochondrial potential and consequently reduced ATP synthesis. Thus, the reduction in ATP release upon application of CBX may not be a reliable measure of channel activity but may be a measure of mitochondrial efficiency. Accordingly, our hypothesis of a mitochondrial origin of ATP and efflux through CALHM1 channels resolves the apparent inconsistency between the effects of CBX and the pharmacological profile of CALHM1 [see (39) for discussion].

DISCUSSION

Here, we provided evidence for the existence of a synaptic signaling complex in type II taste receptor cells formed by clustered ATP-permeable CALHM1 channels and atypical mitochondria apposed to points of contact with purinoceptive nerve endings. This ATP release machinery is fast enough and capable of releasing detectable levels of ATP, the afferent taste neurotransmitter, into the intercellular space, even in response to short-term taste stimulation accompanied by generation of several action potentials (18). Here, we provide experimental support for the idea that ATP-permeable CALHM1 channels are clustered and that the corresponding submembrane compartment is physically separated from the bulk of the cytoplasm by atypical mitochondria (Fig. 7). The unique organization of this compartment protects the cell from metabolite leakage through the large-pore channel and allows for specific and precise nonvesicular, regulated neurotransmission.

Fig. 7 Schematic diagram of a mitochondrial/CALHM1 synapse between a type II taste cell and an afferent nerve terminal.

The CALHM1 channels lie within the plasma membrane of the taste cell (light blue) between the atypical mitochondrion (pink) and the nerve terminal (green. (A) In the resting state, CALHM1 channels (blue) are closed and ATP levels in the presynaptic compartment will be close to levels in the intermembrane space of the mitochondrion due to diffusion of ATP (red circles) through the porin channels of the outer mitochondrial membrane. (B) Taste evoked action potentials in the taste cell open the CALHM1 channels, allowing ATP to flow into the synaptic cleft between the taste cell and the afferent nerve terminal. (C) The ATP in the intercellular space binds to and gates open the P2X receptors (green bars) in the neural membrane, permitting influx of sodium ions (Na+), which depolarizes the afferent terminal. (D) As the taste cell repolarizes, CALHM1 channels close and extracellular levels of ATP return to prestimulus levels.

CALHM1 is a nonselective channel permeable to many ionic species, with relative permeabilities following the sequence Ca:Na:K:Cl ~ 11:1:1.2:0.6 (40). On the basis of this permeability sequence, we estimate, using the Goldman-Hodgkin-Katz current equations (41), a fractional current carried by Ca2+ ions to be nearly 30% at −50 mV but less than 1% at 50 mV. Such CALHM1 or CALHM1-like currents have been recorded in taste buds from all three main taste fields in the tongue: CV, foliate, and fungiform papilla (16, 20, 4244). CALHM1 channels are largely responsible for large (~1 nA) inward tail currents in type II cells (43), allowing one to estimate CALHM1-dependent conductance to be nearly 20 nS at 50 mV. With this steady-state conductance and for characteristic times of CALHM1 activation and deactivation at 50 mV of about 20 and 10 ms, respectively (22), action potentials should stimulate loading of about 106 Ca2+ ions. This would result in a rise in Ca2+ concentration by nearly 10 mM in the 0.08-fl volume of the submitochondrial compartment. That we observe only negligible or small Ca2+ signals generated due to activity of CALHM1 channels in type II cells (Fig. 1D) argues for the existence of a mechanism that provides limitation of influx and/or effective clearance of Ca2+. Because TrpM5 in type II taste cells is gated by intracellular Ca2+, wholesale increases in intracellular Ca2+ would likely open these channels, thereby further depolarizing the taste cells. That such positive feedback does not occur suggests that Ca2+ influx and diffusion must be limited. In conventional synaptic terminals (45), Ca2+ is cleared by both mitochondrial uptake and plasma membrane Ca2+-ATPase (PMCA). The large reservoir of ATP in the atypical mitochondrion of taste cells should offer a ready supply of ATP required to drive PMCA-mediated removal of Ca2+ as well as provide a local source of ATP for release.

The intimate relationship between the atypical mitochondrion and the CALHM1 channels offers an unconventional means for quantal-like release of ATP in the absence of synaptic vesicles. Each action potential in a type II taste cell stimulates the release of a near-constant amount of neurotransmitter (16, 18), leading to a scaled system translating the magnitude of the taste cell response into a proportionate neural response. Because of the large intramitochondrial reservoir of ATP adjacent to the release channels, this process is likely independent of the short-term overall cellular metabolic status, including the activities of other energy-consuming intracellular processes. At the immunological synapse, mitochondria are also necessary for the production and release of ATP crucial for regulation of T cell function (46).

Other axonless sensory cells, including rod and cone photoreceptors, hair cells, and even type III taste cells, use classical chemical synapses to release neurotransmitters. In addition, photoreceptors also exhibit nonvesicular release of neurotransmitter through reversal of neurotransmitter transporters (47, 48). However, release through these transporters is graded and not quantal, and occurs diffusely across widespread areas of cell membrane, rather than at a specialized synaptic focus. It is currently unclear why type II cells adopt a specialized channel-based release system rather than the typical SNARE-based vesicular exocytotic mechanism used by other receptor cells. CALHM1-mediated release of ATP and its association with atypical mitochondria that we found in taste buds may be more common in other neuronal systems but not fully recognized. CALHM1 is expressed in hippocampal neurons, and a polymorphism in this channel influences the onset of Alzheimer’s disease and affects Ca2+ homeostasis (49). In neurons, CALHM1 plays an important role in Ca2+ homeostasis and flux through the membrane, especially in response to low extracellular Ca2+ (50). Whether CALHM1 in central neurons also forms clusters associated with specialized mitochondria to regulate release of ATP requires further investigation.

The situation of the atypical mitochondria only at points of contact with nerve fibers suggests the presence of signaling between the afferent fiber and the taste cell similar to the signal between antigen-presenting cells and T cells (2). In that immunological system, a key signal is intercellular adhesion molecule 1 along with other factors. Upon recognition of the signal, the cytoskeleton and mitochondria of the T cell reorganize to form the specialized immunological synapse. In the taste system, it is likely that a cell surface molecule on the nerve fiber, such as the neural cell adhesion molecule, may trigger the clustering of the CALHM1 channel and reorganization of the local mitochondria. The exact nature of the intercellular signals in the taste system and the subsequent cascade of events leading to the development of the mitochondrial CALHM1 synapse remain to be elucidated.

MATERIALS AND METHODS

Animals

Mice were used in all experiments involving live animals. All experimental protocols were in accordance with local regulatory bodies and the European Communities Council Directive (86/609/EEC) and approved by the regional ethical committee (Stockholms Norra Djurförsöksetiska Nämnd; N512/12) and by the Animal Care Committee of the Institute of Cell Biophysics (Pushchino) or by the local Animal Care and Use Committees at The Feinstein Institute for Medical Research, Northwell Health, or at the University of Colorado School of Medicine. Particular effort was directed to minimize the number of animals and their suffering during the experiments.

Tissue preparation, immunohistochemistry, and imaging

Wild-type and TrpM5-GFP [Tg(Trpm5-EGFP)tm1Sdmk and RRID:MGI:3615507] transgenic mice (to reveal type II taste cells) were perfused with a fixative composed of 4% paraformaldehyde and 0.05% glutaraldehyde in 0.1 M phosphate buffer (PB) (pH 7.4) that was preceded by a short rinse with physiological saline [anesthesia, 5% isoflurane or Fatal-Plus Solution (pentobarbital); Vortech Pharmaceuticals]. After postfixation in the same fixative (3 hours to overnight) and cryoprotection in 10 to 20% sucrose for 12 to 48 hours, tongues were cryosectioned at 5 to 20 μm thickness and dried onto positively charged glass slides.

In some specimens, antigen retrieval was carried out using either Dako solution (Dako S1699) or 10 mM sodium citrate (pH 9 for 10 min at 85°C) before exposure to antisera to enhance antigen accessibility. Other sections were exposed to antibodies without antigen retrieval. Sections were then exposed for 16 to 72 hours at 4°C to select combinations of primary antibodies diluted in PB or phosphate-buffered saline to which 0.1 to 2% normal donkey serum and 0.3% Triton X-100 had been added. After extensive rinsing in PB, fluorescent secondary antibodies were applied for 2 hours at 22° to 24°C: donkey anti-rabbit and anti-mouse [carbocyanine (Cy)2, 3 or 5-tagged at 1:200; Jackson ImmunoResearch Laboratories], donkey anti-mouse IgG2a A555 (Molecular Probes), and goat anti-rabbit A649 or donkey anti-chicken A488 (Jackson ImmunoResearch Laboratories) at 1:400 dilution each. After three additional washes, sections were placed on coverslips with glycerol-based solution or with Fluoromount-G (Southern Biotechnology Associates). Primary antibodies used were as follows: mouse monoclonal antibody 32C2 directed against the C-terminal domain of CALHM1 at 1:50 dilution (51), rabbit polyclonal antibody against CytC (sc-7159; AB_2090474; Santa Cruz Biotechnology), rabbit anti-P2X3 at 1:1000 dilution (APR-016; AB_2341047; Alomone Labs), and chicken anti-GFP at 1:1000 dilution (Aves Labs; AB_10000240). All staining reported here was absent when the primary antibody was omitted.

For ultrastructural immunohistochemistry, tissues were fixed as above and then sectioned on a vibratome at 80 μm and collected in PB. Sections containing taste buds were incubated with 1% NaBH4 for 10 min, rinsed, and then transferred to 10 mM sodium citrate (pH 6) and 0.05% Tween 20 for 10 min at 85°C. Afterward, sections were placed for 15 min into an avidin-biotin blocking system involving exposure of the tissue first to avidin D followed by incubation in unlabeled biotin (Vector SP-2001) containing 2% normal donkey serum plus AB media and then exposed to unlabeled goat anti-mouse immunoglobulin G (IgG) Fab (1:25; cat. no. 115-007-003; Jackson ImmunoResearch Laboratories) for 1 hour to block nonspecific binding of secondary antisera to endogenous IgG. The sections were then incubated with primary antibody monoclonal CALHM1 (1:20) for four nights. After washing, the sections were incubated overnight to biotinylated rat anti-mouse IgG2a (1:1000; cat. no. 04-6240; Life Technologies) and then incubated with ABC (Vector Laboratories) for 2 hours. The sections were treated for 10 min in 0.05 M tris buffer (pH 7.3) containing 0.05% diaminobenzidine HCl (DAB). The label was visualized by floating the sections for 2 to 4 min in the fresh DAB mixture with hydrogen peroxide (0.002%).

Images were acquired either on an Olympus Fluoview Confocal microscope or on Zeiss LSM 700, LSM 710, or LSM 780 confocal laser-scanning microscopes with maximal signal separation or spectral scanning. Some images of Fig. 2 were processed with Zeiss Airyscan. Composite figures were assembled in CorelDraw X5 or Photoshop CC 2014 (Adobe).

32C2 antibody specificity assessment

Peptide array for epitope mapping was used to determine the exact epitope sequence of the mouse IgG2a monoclonal antibody (32C2) against CALHM1. Ten-residue-long peptides, with an offset of three residues covering the entire cytosolic C-terminal end of human CALHM1, were dotted on nitrocellulose membrane. The membrane was then processed for Western blot using 32C2, as previously described (51). The identified epitope peptide was used for antibody blocking. HT-22 cells were transiently transfected with empty vector or human CALHM1. Protein extracts were then processed by SDS–polyacrylamide gel electrophoresis and transferred on nitrocellulose membranes. 32C2, preincubated 1 hour at room temperature with the indicated peptides (0.2 μg/ml) in 5% milk Tween-TBS, was then used for Western blots.

sbfSEM and transmission EM

Methods for 3D EM imaging and volumetric analysis are slightly modified from those described previously (52, 53). Mice were anesthetized with Fatal-Plus Solution and perfused with 0.1% NaNO2, 0.9% NaCl, and 200 U of sodium heparin in 100 ml of 0.1 M PB (pH 7.3) at 35°C followed by 2.5% glutaraldehyde and 2% formaldehyde with 2 mM CaCl2 in 0.025 M cacodylate buffer (pH 7.3) at 35°C for 10 min. Tissues were removed and placed in the same fixative for 2 to 3 hours on ice and then cut into 200-μm-thick vibratome sections.

For conventional transmission EM, some thick sections prepared as for sbfSEM were rinsed in buffer and then stained with 2% osmium tetrooxide in 0.05 M sodium cacodylate buffer for 30 min. After rinsing, the sections were placed overnight in 1% uranyl acetate in double-distilled H2O and then stained en bloc in Walton’s lead at 60°C for 40 min before embedment in Luft’s Epon. Thin sections (90 to 120 nm) were cut with a diamond knife on a Reichert Ultracut E ultramicrotome, examined with a FEI Tecnai G2 Biotwin Transmission Electron Microscope, and photographed with a Gatan Ultrascan 1000 digital camera.

Vibratome sections (200 μm thickness) for sbfSEM were rinsed with 0.025 M cacodylate buffer (pH 7.3) containing 2 mM CaCl2 and then incubated for 1 hour at 0°C in a solution containing 3% K4[Fe(CN)6] in 0.025 M cacodylate buffer (pH 7.3) with 2 mM CaCl2 combined with an equal volume of 4% aqueous OsO4. After the first heavy metal incubation, the sections were washed five times with H2O at room temperature for 3 min each and then placed in 1% thiocarbodydrazide solution for 20 min at room temperature. After washing, the sections were placed in 2% OsO4 for 30 min at room temperature. After this second exposure to osmium, the tissues were washed five times with H2O at room temperature for 3 min each, placed in 1% UO2(CH3COO)2·2H2O, and left in a refrigerator overnight. The next day, en bloc Walton’s lead aspartate staining was performed for 30 min at 60°C in 0.066 g of Pb(NO3)2 in 10 ml of aspartic acid stock and pH-adjusted to 5.5 with 1 N KOH. Sections were dehydrated five times for 3 min each using an increasing series of ice-cold alcohol solutions before transfer into propylene oxide and final embedment in Lufts Epon 3:7 at 60°C overnight.

The tissues were then trimmed and mounted on an aluminum pin, coated with colloidal silver paste around the block edges, and then examined in a Zeiss Sigma VP system equipped with a Gatan 3View in-chamber ultramicrotome stage with low-kilovolt backscattered electron detectors optimized for 3View systems. Samples were routinely imaged at 2.25 kV, at 5 to 10 nm per pixel resolution (30-μm aperture, high-current mode, and high vacuum), with field sizes between 80 and 250 μm in x,y, and approximately 500 slices with 80 nm thickness were generated. The resulting image stacks were aligned and montaged in Photoshop (Adobe Systems) and ImageJ. Segmentation and reconstruction were accomplished using Reconstruct software.

Measurement of mitochondrial intermembrane space

Atypical and typical mitochondria were selected from virtual sections derived from sbfSEM image stacks as viewed in Reconstruct. These mitochondrial types were distinguished on the basis of several morphological features including overall size of the mitochondrion and exaggerated tubular shape of the cristae. Images were imported into Amira 5.6.0 (FEI Company) for volumetric analysis. A total of 20 atypical mitochondria and 18 typical mitochondria from a total of three different taste buds were analyzed. Typical mitochondria were selected to match approximately in terms of overall profile area of the atypical mitochondria measured. Either three or four profiles were analyzed per mitochondrion according to the total number of profiles through the particular mitochondrion. Using the “MaterialStatistics” measurement function, we measured the total area of each mitochondrial profile as well as the area of the intermembrane space including the area within the tubular cristae. We then calculated the ratio of intermembrane space to total mitochondrial area for each mitochondrial profile. Within each class of mitochondrion, typical and atypical, no significant differences existed between samples or for different mitochondrial sizes. Accordingly, data from the different sizes of mitochondria were pooled for a statistical comparison using an unpaired t test on GraphPad (www.graphpad.com/quickcalcs).

Taste cell isolation

For isolation of taste buds or individual cells, 8- to 10-week old C57Bl6 mice were euthanized with CO2 followed by cervical dislocation before tongues were removed. Taste cells were isolated from mouse (NMRI, 6 to 8 weeks old) CV papilla. Tongues were injected between the epithelial and muscle layers with collagenase B (0.7 mg/ml), dispase II (1 mg/ml), elastase (0.2 mg/ml) (all from Roche Diagnostics), and trypsin inhibitor (0.5 mg/ml; Sigma-Aldrich) dissolved in a solution [140 mM NaCl, 20 mM KCl, 0.3 mM MgCl2, 0.3 mM CaCl2, and 10 mM Hepes-NaOH (pH 7.4)]. The tongue was incubated in an oxygenated Ca-free solution [120 mM NaCl, 20 mM KCl, 1 mM MgCl2, 0.5 mM EGTA, 0.5 mM EDTA, and 10 Hepes-NaOH (pH 7.4)] for 20 to 30 min. The epithelium was then peeled off from the underlying muscle, pinned serosal side up in a dish covered with Sylgard resin, and incubated in the Ca-free solution for 10 to 30 min. The isolated epithelium was kept at room temperature in a solution [130 mM NaCl, 10 mM NaHCO3, 5 mM KCl, 1 mM MgCl2, 1 mM CaCl2, 10 Hepes-NaOH (pH 7.4), 5 mM glucose, and 2 mM Na pyruvate]. Taste cells were removed from the CV papilla by gentle suction with a fire-polished pipette with an opening of 70 to 90 μm and then expelled into an electrophysiological chamber.

ATP and ATP/ADP biosensors and calcium imaging

COS-1 cells (which endogenously express P2Y receptors that mobilize Ca2+) and Chinese hamster ovary (CHO) cells were transfected with P2X2 and P2X3 expression constructs and used as cellular sensors for monitoring ambient nanomolar ATP and ADP concentrations. The bath solution for cellular physiology experiments contained 140 mM NaCl, 2.5 mM KCl, 1 mM MgSO4, 1.3 mM CaCl2, 1.2 mM NaH2PO4, 10 mM glucose, 5 mM pyruvate, and 10 mM Hepes-NaOH (pH 7.4). For calcium imaging, ATP-sensitive cells were preloaded with 4 mM Fluo-4AM + Pluronic (1.5 mg/ml) (both from Molecular Probes) for 30 min at 23° to 25°C. Cell fluorescence was excited with a computer-controlled light-emitting diode (Luxion) at 480 nm and recorded at 535 nm. Sequential fluorescence images were acquired every 0.5 to 2 s using a fluorescent Axioscope-2 microscope, an EMCCD Andor iXON camera (Andor Technology), and Workbench 6.0 software (INDEC Biosystems). Cells were stimulated by bath application of compounds. All chemicals were from Sigma-Aldrich. Experiments were carried out at 23° to 25°C.

Electrophysiology and calcium imaging

Taste cells were assayed with the patch-clamp technique using the perforated patch (with 400 mg/l of amphotericin B in the recording pipette) or whole-cell configuration. Ion currents were recorded, filtered, and analyzed using an Axopatch 200B amplifier, a DigiData1322 interface, and the pClamp8 software (Axon Instruments). Intracellular solution contained 100 mM CsCl, 40 mM KCl, 1 mM MgATP, 1 mM EGTA, and 10 mM Hepes-NaOH (pH 7.4). The bath solution contained 140 mM NaCl, 2.5 mM KCl, 1 mM MgSO4, 1.3 mM CaCl2, 1.2 mM NaH2PO4, 10 mM glucose, 5 mM pyruvate, and 10 mM Hepes-NaOH (pH 7.4).

For calcium imaging, cells were loaded with 4 μM Fluo-4AM or FURA-2AM + Pluronic (1.5 mg/ml) (both from Molecular Probes) for 30 min at 23° to 25°C. For Fluo-4–loaded cells, fluorescence was excited with a computer-controlled light-emitting diode (Luxion) at 480 nm and recorded at 535 nm. Sequential fluorescence images were acquired every 0.5 to 2 s using a fluorescent Axioscope-2 microscope, an EMCCD Andor iXON camera (Andor Technology), and Workbench 6.0 software (INDEC Biosystems). To measure Fura-2 signals, recordings were done using a VisiChrome monochromator and VisiView software (Visitron Systems) on an AxioExaminer.D1 microscope (Zeiss) equipped with a CoolSnap HQ2 camera (Photometrics). Cells were stimulated by bath application of compounds. All chemicals were from Sigma-Aldrich. Experiments were carried out at 23° to 25°C.

Statistical analysis

Physiological data were analyzed using SigmaPlot (Systat Software Inc.). Data were expressed as means ± SEM. For all quantitative data, P < 0.05 was considered statistically significant and was calculated by Student’s t test (paired or unpaired as appropriate), two-way analysis of variance (ANOVA), or Mann-Whitney rank sum test.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/11/529/eaao1815/DC1

Fig. S1. Validation of the CALHM1 antibody.

Fig. S2. Localization of CALHM1 immunoreactivity in taste buds.

Fig. S3. Calcium responses in sensor cells.

Fig. S4. Effect of CBX on Mito-ID fluorescence in living taste cells.

Table S1. Amplitudes of calcium responses in type II taste cells from Fig. 3B.

Movie S1. 3D visualization of the relationship between a type II taste cell and the innervating nerve fiber shown in Fig. 1A.

Movie S2. 3D visualization of mitochondria and taste cell shown in Fig. 1B.

Movie S3. Relationship of atypical and typical mitochondria.

Movie S4. Tubular cristae in atypical mitochondria.

REFERENCES AND NOTES

Acknowledgments: We thank N. Shultz and M. Li for assistance with immunohistochemistry, R. Russell and R. K. Johnson for segmentations and reconstructions from sbfSEM data, and S. Fore (Carl Zeiss Microscopy, LLC) for assistance in acquiring the Airyscan images. We also appreciate the following for review of this manuscript at various stages in its preparation and finalization: F. A. Chaudhry (University of Oslo, Oslo, Norway), K. Beam and S. C. Kinnamon (University of Colorado School of Medicine), and E. Liman (University of Southern California). Funding: S.S.K. was supported by the Russian Academy of Sciences (Program #7) and the Russian Foundation for Basic Research (grant no. 13-04-40082). I.A. was supported by the Swedish Research Council, the Bertil Hallsten Research Foundation, the Jeassons Foundation, the Strategic Regeneration Foundation, and the Knut and Alice Wallenberg Foundation (CLICK). T.H. was funded by the Swedish Research Council, the Swedish Brain Foundation, the Novo Nordisk Foundation, the Petrus and Augusta Hedlunds Foundation, a European Research Council (ERC) Advanced Grant (“Secret-Cells,” ERC-AdG-2015-695136), and the European Commission Integrated Project “PAINCAGE.” R.A.R. was a European Molecular Biology Organization (EMBO) long-term research fellow (ALTF 596-2014) co-funded by the European Commission FP7 (Marie Curie Actions, EMBOCOFUND2012, GA-2012-600394). R.A.R. was also supported by the Ministry of Education and Science of the Russian Federation (agreement no. 14.575.21.0074 with Immanuel Kant Baltic Federal University; project leader, V. Kasymov). This study was supported by grants from the National Institute for Deafness and Communicative Disorders of the NIH (USA) (to T.E.F., R.S.L., B.H., R.Y., G.J.K., and J.C.K.), grants 1R21DC013186 and R01DC014728 (to T.E.F.), and grant P30DC004657 to D. Restrepo (University of Colorado School of Medicine). P.M. was supported by a grant from the National Institute on Aging of the NIH (R01AG042508). Author contributions: S.S.K., T.E.F., and R.A.R. designed the study. R.A.R., T.E.F., and R.Y. performed immunostaining. R.A.R. performed calcium imaging and electrophysiological experiments. R.A.R., O.A.R., and M.F.B. performed experiments with cellular ATP and ATP/ADP sensors. H.Z. and P.M. developed and characterized the antibody against CALHM1. V.V.R., T.E.F., B.H., R.S.L., R.Y., L.E.S., G.J.K., A.L., and J.C.K. performed EM experiments and made 3D reconstructions. G.D.C. and O.A.R. performed experiments with effects of CBX. R.A.R., T.E.F., I.A., T.H., P.M., and S.S.K. analyzed data and contributed to writing the paper. All authors commented on the manuscript and approved its submission. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All the data required to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. The Calmh1 knockout mice require a material transfer agreement from P. Marambaud (The Feinstein Institute for Medical Research).
View Abstract

Navigate This Article