|Home | About | Journals | Submit | Contact Us | Français|
The advent of multicellular organisms, some 800 million years ago, necessitated the development of mechanisms for cell-to-cell synchronization and for the spread of signals across increasingly large cell populations [168, 185]. Many structures and mechanisms have evolved to achieve such functions [4, 15]. Among these mechanisms, one which is prominent in both the invertebrate and the vertebrate world, across the entire phylogenetic scale, involves the transmembrane flux of large cytosolic and extracellular molecules [4, 15, 65, 66, 69–71, 121, 128, 129, 147, 154, 163]. These fluxes, in turn, are dependent on the formation of specific channels that in all animal classes are made by tetra-span integral membrane proteins [65, 66, 69–71, 121, 128, 129, 147, 154, 163] (Fig. 1).
Early electrophysiological and electron microscopy studies converged in the realization that gap junctions, the membrane domains that concentrate intramembrane particles at sites of close membrane apposition, were the physical substrate of cell-to-cell communication in both invertebrate  and vertebrate tissues  (Fig. 1). The finding that similar drugs (the long-chain alcohols heptanol and octanol) and conditions (intracellular acidification) inhibited intercellular communication in both invertebrate and vertebrate systems [58, 85] was taken as further support that all gap junctions had a similar structure and function. Still, the different size of intramembrane particles, their different partition into the P- and the E-fracture faces of the cell membrane, and the different width of the gap space delineated by the two interacting membranes, suggested that the proteins making vertebrate gap junctions were different from those making the invertebrate structures . These differences have functional consequences, as most elegantly demonstrated by co-culturing cell lines from different animal species. In these experiments, heterotypic coupling was shown between insect cells, as well as between different types of vertebrate cells, whereas virtually no coupling was observed between cells of phylogenetically distant species .
There is no longer any question that vertebrate gap junction channels are made by various combinations of different connexin proteins [66, 69–71]. The 20–21 isoforms of this family in rodents and man differ in size, but share a similar membrane topography. Thus, all connexins feature four transmembrane domains connected by two extracellular loops, each comprising three highly conserved Cys residues, a cytoplasmic loop, and both N and C termini in the cytosol (Fig. 2). The difference in size of the different connexin isoforms is essentially due to a different length of the cytoplasmic loop and/or the C terminus [39, 64, 65, 69–71, 109]. Current nomenclature (Table 1) designates connexin proteins as nCxZ, where n is the species (e.g., h, m, r for human, mouse, rat, respectively), and Z is the predicted molecular weight, in kDa. The genes encoding connexin proteins are named according to subgroups, in the order of discovery (e.g., Gjb1 encodes Cx32, Gjb2 encodes Cx26, Gja1 encodes Cx43, etc).
Attempts to identify the proteins making invertebrate gap junctions initially revealed the proteins Ogre, Passover, Uncoordinated, and Shaking B in Drosophila and Caenorhabditis (hence the original OPUS acronym to name these proteins) with no primary sequence homology to connexins [128, 129, 195]. Since that time, more than 25 other junctional proteins revealing significant similarities have been identified in C. elegans, and it is clear that many other forms are expressed in other invertebrate species [75, 128, 129, 195] (Table 2). Collectively, these proteins have been named innexins to stress their invertebrate distribution and their role, analogous to that of connexins, in the formation of gap junctions (Fig. 1). Strikingly, innexins also share with connexins a similar structure and membrane topography (Fig. 2), even though the two sets of proteins have no homology in their primary amino acid sequence [75, 128, 129, 195]. Notably, connexins display three conserved Cys within each of the two extracellular loops, whereas innexins only carry two such residues. Thus, it is curious in retrospect that hydra development was reported to be blocked by an antibody prepared against Cx32 , that other antibodies have detected connexin-like proteins in anemone  and other marine invertebrates , and that junctional proteins isolated from hepatopancreas of crayfish and lobster were reported to have a sequence similar to that of rat liver connexins .
Sequencing of mammalian genomes has revealed a third family comprising only three genes that code for proteins with a primary sequence showing about 20% similarity to that of innexins [121, 154, 195]. On this basis, these proteins were thought to represent vertebrate homologs of the innexins, and were termed pannexins (Table 3) to encompass both invertebrate and vertebrate members [121, 154]. Like connexins and innexins, all three pannexins display N- and C-terminal domains within the cytoplasm, large extracellular and cytoplasmic loop domains, and four membrane spanning segments (Fig. 2). Like innexins but in contrast to connexins, pannexins contain two Cys residues in each extracellular loop [121, 154]. Furthermore, and in marked contrast with both innexins and connexins, pannexins display consensus sequences for glycosylation [13, 14, 125] (Fig. 2). The distribution of Pnx1 (Pnx1), the most studied form, is widespread and, in most types of cells and tissues, largely overlaps with that of connexins [17, 8]. While at least some phenotypes resulting from loss of specific connexin species are not compensated by pannexin changes [136, 146], suggesting a different function of the two protein families, in other cases, the effects of Pnx1 transfection mimicked that of Cx43, implicating a comparable/overlapping role of these two proteins .
Expression and deletion studies in a variety of systems have established that connexin hexamers, termed connexons, concentrate at gap junction domains of the cell membrane, where the intercellular space is reduced to a gap 2–3 nm wide. At these sites, the connexons of one cell align with, and strongly bind to the connexons of an adjacent cell, establishing a continuous intercellular hydrophilic pathway (Fig. 1) for the cell-to-cell exchange of multiple types of cytosolic molecules [64, 65, 69–71, 155]. The functional importance of this electrical and metabolic coupling is shown by a variety of striking and tissue-specific phenotypes that can be experimentally induced after overexpression or knock-out of individual connexin isoforms, as well as after the knock-in replacement of one isoform by another [92, 192]. It is further stressed by the identification of a number of diseases that are undoubtedly linked to connexin mutations [43, 57, 90, 94, 127, 143]. A variety of other diseases are thought to be due to altered amounts and/or function of these gap junction proteins [21, 108, 153].
Similarly, innexins oligomerize to form innexons that cluster at gap junctions of invertebrate cells (Fig. 1). Functional expression studies in paired Xenopus oocytes demonstrated that several, even though not all innexins also formed intercellular channels [8, 95, 166], and that at least some innexin mutations give rise to phenotypes expected for lack of gap junction-mediated intercellular communication [10, 25, 27, 37, 179]. In fact, it was on the basis of such dysfunctional phenotypes that the OPUS gene family was first identified.
In contrast, and in spite of an initial report , pannexons appear unable to form sizable amounts of cell-to-cell channels under most conditions [76, 146, 147, 163] (Fig. 1). This lack of formation of functional gap junctions is likely due to the glycosylation of the extracellular loops of pannexins [13, 14, 125], which, as mentioned above, is not observed for either connexins or innexins. Still, forced expression of pannexins in paired Xenopus oocytes increased the conductance of the junctional cell membrane to current carrying ions in a way that can only be accounted for by the formation of pannexin cell-to-cell channels .
The finding of nonjunctional permeability induced by expression of connexins, pannexins, and innexins, together with the biochemical, immunological, and functional studies showing that at least some connexon and innexon types may also inserted in nonjunctional regions of the cell membrane [42, 114, 123] challenged the classical views about junctional proteins, by indicating that, at least under certain conditions, some of these proteins may also form functional channels in domains of the cell membrane that are not involved in cell contact. Most strikingly, these channels were found to allow for the leakage of cytosolic molecules, notably ATP and glutamate, into the extracellular medium and to permit the reverse uptake into cells of large extracellular and membrane-impermeant tracers, notably propidium iodide and Lucifer Yellow, [61, 144, 147, 163, 171]. Given that this dye permeability is a major characteristic of connexin and innexin cell-to-cell channels, and that the bidirectional flux of large molecules across the membrane could be abolished by drugs known to block gap junction channels [73, 91, 98, 103, 134], it was proposed that the nonjunctional channels were half of a gap junction channel, made by a nonpaired connexon or innexon (Fig. 1). The subsequent identification of similar channels in cells (e.g., red blood cells) that do not express connexins or innexins but express pannexins , further indicated that these channels may also be made by unpaired pannexons (Fig. 1). For these reasons, these channels were referred to as hemichannels even though this is an obviously semantically improper term to indicate a fully functional and regulated structure [38, 147, 163]. Still, because the term has now gained usual acceptance, we refer to these structures, whether made by single connexons, innexons, or pannexons, as “hemi-channels”.
There is now undisputed evidence that Cx46 and Cx50 form such “hemi-channels” when expressed in Xenopus oocytes  and probably also in the lens , and it is widely believed that Cx43 can also form such structures in a variety of cell types [65, 99, 103, 144, 148, 163]. Other reports have proposed that this property is shared by several other connexin isoforms, including Cx23, Cx26, Cx30, Cx30.2, Cx 31.9,Cx32, Cx35, Cx36, Cx37, Cx41.8, Cx45, Cx45.6, Cx55.5, and Cx56 [20, 46, 60, 80, 118, 144, 151, 181, 182, 191]. However, for the reasons that shall be discussed below, it has proven difficult to unambiguously demonstrate the presence of functionally open and unpaired connexons made by these proteins [146, 163]. At least a few innexins [8, 27, 179] and one (Pnx1) of the three known pannexin proteins [18, 77, 105, 124, 180] also form functionally open innexons and pannexons, respectively (Fig. 1).
In keeping with their parallel evolution, similar structure, and membrane topography, most “hemi-channels” share a number of similar biophysical properties that are deemed diagnostic of a composition by one of the three aforementioned families of junctional properties. Thus, they generally feature a large unitary conductance (200–500 pS) [38, 65, 147, 163], although the unitary conductance of certain connexin “hemi-channels” is much lower (e.g., those attributed to either Cx31.9 or Cx36 are <20 pS) , a reversal potential near 0 mV when tested with standard internal and external solutions, indicating little selectivity for the major current carrying ions, and a sizable open probability at positive resting potentials (~20% at V≥40 mV) [38, 65, 147, 163]. Most “hemi-channels” show an in–out permeability to ATP and an out–in permeability to propidium iodide, Lucifer Yellow and 6 carboxyfluorescein [36, 65, 147, 163, 167]. Opening of all types of “hemi-channels” is induced by membrane depolarization (V≥40–60 mV). Connexin-based “hemi-channels” have been proposed to be activated by low divalent ion concentrations, which for Cx46 and Cx50 shift activation kinetics to lower voltages . Such shift has not been demonstrated for other putative connexin “hemi-channels.” Cx43 “hemi-channels” have been reported to be opened under ischemic conditions [11, 33, 34, 84, 99, 135, 151], whereas Pnx1 “hemi-channels” have been demonstrated to be activated by a diverse set of experimental conditions, including mechanical stress, e.g., as provided by an osmotic shock [6, 7, 102, 134], strong depolarizations, and activation of purinergic receptors, including P2Y1r, P2Y2r, and P2X7r, by ATP and other agonists [26, 72, 87, 105, 122, 126, 173, 174, 189] (Fig. 3).
Closure of a number of “hemi-channels” can be induced by several drugs, including carbenoxolone, α-glycyrrhetinic acid, flufenamic acid, mefloquine, and alkanols that also turn off most gap junction channels [14, 124, 169, 177]. The effective concentration of these drugs varies depending on the identity of the channel-forming protein, thereby possibly allowing for discrimination of which channel type is involved in ion or molecular flux. For example, Pnx1 channels appear to be 1,000- to 10,000-fold more sensitive to mefloquine than are Cx43 gap junction channels [36, 79] and less sensitive than connexons to uncoupling flufenamates . Peptides corresponding to the extracellular loops of Pnx1 and Cx43 have also been reported to be effective in blocking both current flow and dye uptake through pannexons and connexons [73, 124, 190]. However, one of the peptides reported to most effectively block Cx32 “hemi-channels” corresponded to an intracellular epitope . Furthermore, the specificity of these peptides has been challenged, as a result both of cross-inhibition of Pnx1 and Cx46 “hemi-channels” in Xenopus oocytes, and effective blockade by polyethylene glycols . A more physiological regulation of connexons is seen under conditions of lowered intracellular pH [184, 197] or PKC and MAPK activation [8, 40, 49, 56], which are reported to close most connexin “hemi-channels”, and after exposure to either IL1β or TNFα, which is reported to open them [113, 137, 176]. The last finding is quite surprising, given that IL-1 β treatment has been shown in several studies to cause gap junctions to disappear [22, 44, 83, 111, 145]. Conversely, “hemi-channels” may be opened by FGF-1 
The biophysical properties reported for connexin “hemi-channels” differ from those of the gap junction channels formed by the same connexin. Thus, while the unitary conductance of the main state of connexin “hemi-channels” is about twice that of the same connexons in gap junctions, and the substate conductance seen at high positive potentials is generally more than twice that of the corresponding gap junction channel. Voltage sensitivity is also different in “hemi-channels” and gap junction channels. In invertebrates, gap junctions tend to be sensitive to transmembrane (or “inside–outside”) potential, closing when either or both cells are depolarized [23, 157], whereas in mammals, gap junctions are closed by the potential across the gap junction (nonzero transjunctional voltage), each connexin having a specific sensitivity, as shown by Boltzmann parameters [65, 161]. Moreover, a fraction of the junctional conductance is usually preserved in gap junction channels even at the highest voltages. This minimal conductance ranges from virtually zero to almost unity for different connexins and represents occupancy of a channel substate. The predominant form of voltage dependence seen in pannexons and connexons “hemi-channels” is conspicuously different from the gating seen in either innexin or connexin gap junctions. Thus, “hemi-channels” open in response to membrane depolarization above about +40 mV, a gating which is attributed to the so-called “loop gate” that is exposed to the transmembrane field in unpaired connexons . For Pnx1, current activation increases with voltage without attenuation, whereas for connexons, there is a decrease in current at the highest voltages, due to the gating to a substate caused by the transjunctional voltage sensor. The gating to substate and its absence in a C-terminally GFP-tagged Cx43, has provided the strongest evidence to date that Cx43 connexons can open in mammalian cells, albeit under nonphysiological conditions .
A major difference between pannexons and connexons is that pannexons have been shown to open under physiological conditions, while most connexons have not. Notably, whereas pannexin channels can be opened at normal resting potential and in normal extracellular Ca2+ solutions, by mechanical stretch and by P2 receptor stimulation, connexons have only been demonstrated to open under supra- or pathophysiological conditions (no extracellular Ca2+ or Mg2+, depolarization exceeding +40 mV). Although both pannexons and connexons have been reported to be affected by raising intracellular [Ca2+]i levels, the reality of this regulation is still debated .
Given that innexins, connexins, and pannexins can all form “hemi-channels” which share several common features, including permeability characteristics and sensitivity to blocking drugs, the question arises of which of these protein species actually establishes the functional membrane conductance and permeability. A definitive answer to this question is complicated by the fact that pannexins and connexins have largely overlapping distributions in vertebrates. Also, it is conceivable that, with evolution, different cell types have acquired different types of “hemi-channels” to fulfill a specific role in a specific environment, so that the situation in one cell system might not be applicable to another.
Data from the initial experiments were taken as an indication that “hemi-channels” were made by connexons, mostly because gap junction channels and connexon channels share a conductance for multiple ions, a permeability to large hydrophilic molecules and a sensitivity to the same set of drugs [56, 163, 188]. Furthermore, the conductance of “hemi-channels” was about twice as high as that of gap junction channels made by the same connexin, as would be anticipated if the two connexons joined in series in a gap junction were dissociated from each other. The most compelling evidence for Cx43 “hemi-channels” comes from HeLa cells overexpressing Cx43, in which currents were evoked by membrane depolarization above +20 mV, single channel conductance was about twice that of Cx43 gap junction channels, conductance was minimally affected by extracellular Ca2+ levels, but no channel activity was observed at membrane potentials ≤0 mV . Moreover, the “hemi-channel” formed by Cx43-GFP featured no residual conductance, consistent with properties of gap junction channels formed by this construct, and a N-terminal GFP-Cx43 fusion protein was found not to form “hemi-channels,” again consistent with lack of formation of functional gap junction channels by this construct .
The major concern regarding the assignment of connexons as functional “hemi-channels” is that the lack of opening of connexons at negative resting potentials is not consistent with the dye uptake observed under such conditions in many cell types [38, 147, 163]. This finding, as well as a variety of expression studies, analysis of cells lacking connexins, and experiments conducted under conditions expected to close gap junction channels have clearly documented that proteins other than connexins can form “hemi-channels” in vertebrate cells [38, 104, 146, 163]. For example, in different cell types, the uptake of extracellular and membrane-impermeant tracers occurs under conditions (presence of extracellular Ca2+, limited depolarization) in which open connexon “hemi-channels” cannot be revealed by electrophysiology [32, 38, 146, 163]. These considerations raise the issue of whether most of the literature published on Cx43 “hemi-channels” is actually reporting other Ca2+-dependent phenomena, such as transport, vesicular uptake and release, or other pathways.
The demonstration that pannexins by themselves form “hemi-channels” was provided by expression of Pnx1 in single oocytes, which lead to the appearance of large currents activating at membrane potentials above −20 mV . These currents were both voltage- and time-dependent, being larger and inactivating more rapidly at larger depolarization. The co-expression of Pnx1 and Pnx2 elicited currents that were larger than those induced by Pnx1 alone, whereas Pnx2 by itself did not yield currents. Because activation was delayed in the co-expression experiments, it was hypothesized that pannexins 1 and 2 also form heteromeric nonjunctional channels . By contrast, Pnx3 neither formed channels itself, nor modified those formed by the other pannexins .
It is therefore likely that in a number of cases, the “hemi-channels” attributed to Cx43 were actually due to pannexin activation. One example involves the selection of the J774 macrophage cell line for resistance to ATP-induced cell killing by repeated exposure of the cells to 10 mM ATP. Resistant clones were found to be deficient in Cx43, from which it was inferred that the so-called P2Z receptor or permeabilization pore induced by high concentrations of ATP was the Cx43 “hemi-channel” . This hypothesis was disproved by cloning of the P2X7 receptor , by studies showing that the permeabilization pore was present in macrophages lacking Cx43 , that J774 cells co-express Cx43 and P2X7 receptors , and by experiments in which Cx43 transfection did not induce the permeabilization pore in BHK cells . The demonstration that Pnx1 is found in retinal horizontal cells similarly raises the possibility that the large conductance “hemi-channel” currents recorded from these cells might actually be carried through pannexin channels , rather than through connexin “hemi-channels” [42, 155], since the high conductance of pannexons would be expected to provide substantially more ephaptic current at the restricted horizontal-photoreceptor contact than would retinal connexons.
Two independent groups have recently identified Pnx1 as a part of a complex that provides membrane permeabilization to large molecules, following the activation of the P2X7 receptor by ATP or analogues of the endogenous nucleotide [105, 124, 196]. Thus, initial activation of the P2X7r selective cation channels allows for the permeation of molecules up to 900 Da [50, 116]. Two hypotheses have been proposed to account for this change of the ionotropic receptor. The first involves the dilation of the P2X7 cation channel itself [19, 28, 178], while the second involves the recruitment of a protein, most likely Pnx1, forming a lytic pore [81, 150, 175]. Interestingly, the resulting membrane permeabilization, which allows for a nonselective influx of large molecular weight probes [105, 124], does not correlate with the activation of caspase1, nor with the processing and release of IL-1β , even though blockade of Pnx1 “hemi-channels” with mimetic peptides prevents both the uptake of the tracer YoPro and caspase1 activation . At any rate, the apparent coupling of Pnx1 activation to that of P2Z/P2X7 receptors, raises the issue of whether pannexin channels can be activated by other receptors. Indeed, the metabotropic P2Y1 and P2Y2 receptors can also activate Pnx1 opening . In addition, uptake of the YoPro tracer can be induced by other P2X receptors, including P2X2  as well as P2X4 and P2X5, presumably via a change of the channel pore [Surprenant, personal communication].
A common role of the intercellular channels made by connexins and innexins at gap junctions of vertebrate and invertebrate tissues is to establish direct communication between cells in contact, i.e., to allow for direct and bidirectional exchange of current-carrying ions and other cytosolic, membrane-impermeant molecules [64, 65, 69–71, 109, 158, 192]. Several specific functions have been experimentally demonstrated to be due to these communications, and many more are attributed to gap junctions on the basis of strong, still circumstantial evidence [109, 158]. The in vivo importance of these functions is stressed by the many phenotypes that have been reported after deletion, overexpression, or replacement of specific connexin isoforms in mice [159, 192], as well as by the increasing number of human diseases that have been linked to either connexin mutations or pathogenic single nucleotide polymorphisms [2, 30, 68, 108, 140, 193]. While the role of innexins has been less intensively investigated, the direct intercellular communications mediated by these proteins have been shown to be essential for synapse establishment in the retina and the CNS [158, 159, 192].
More recently, other roles of connexins, that may be independent of the establishment of cell-to-cell channels, have been revealed. Thus, it is now clear that gap junction proteins interact with other membrane and cytosolic proteins and signaling pathways, and may profoundly affect the expression of multiple genes [64, 69–71, 78, 162]. These novel roles call for a reevaluation of the mechanism of some of the previous cell alterations shown to be dependent on gap junction proteins. Whether nonjunctional connexons may also contribute remains to be shown, given that connexin “hemi-channels” can be functionally demonstrated only under supraphysiological or pathological conditions.
There is presently little evidence that pannexins provide direct intercellular communication through gap junctions. Injection of pannexin RNAs indicated that Pnx1, but not Pnx2 and Pnx3, formed gap junctional channels in pairs of Xenopus oocytes, that were not very voltage sensitive [16, 18]. Furthermore, the conductance of these channels was about 15% smaller when Pnx1 was co-expressed with Pnx2, which was interpreted as reflecting the formation of heterotypic Pnx1–Pnx2 channels [16, 18]. However, gap junctional communication has so far been reported only after pannexins had been exogenously expressed , raising the question of whether the endogenous levels of the proteins, if not their structure and notably their glycosylation, prevent the series assembly of 2 pannexons into a full gap junction channel [13, 14, 125]. Future studies should address this issue and test whether pannexins could, for example, contribute to the synchronization of neurons in the inferior olive and hippocampus, where synchronized oscillations are generated. Indeed, while these oscillations were altered after loss of Cx36, the predominant neuronal connexin, the high frequency oscillations of the neuronal networks was surprisingly unchanged [41, 47, 158], suggesting a compensation by another cell-to-cell communication mechanism. Pannexons, however, seem an unlikely candidate because the large single channel conductance would provide very strong coupling even if only a few channels were functional.
In contrast, several functions dependent on indirect cell-to-cell communication might use large “hemi-channels” for the release of ATP , glutamate , and epoxyeicosatrienoic acid (EET) . While the mechanism of this release has been repeatedly attributed to Cx43 “hemi-channels,” the discovery that pannexons are permeable to ATP under physiological conditions  establishes that channels made of pannexins may be involved in such release, that is required for different cell functions.
Two distinct pathways responsible for the intercellular propagation of calcium waves have been identified. One involves the diffusion of calcium-mobilizing second messengers (e.g., Ca2+, IP3, cADP-ribose) from the cytosol of one cell to that of another, through the gap junctional channels made by either connexins or innexins [147, 152]. The other is an extracellular pathway, which involves the diffusion of molecules acting on cell surface membrane receptors [147, 152]. In several cell types, this pathway involves the release of ATP through membrane channels, allowing for the activation of ATP-sensitive purinergic receptors (P2Rs), a family comprised by metabotropic P2Y and ionotropic P2X receptors, on both the very same cell that released ATP (autocrine pathway) and/or the neighboring cells (paracrine pathway) . The latter, extracellular pathway operates in many cell systems [47, 67, 119] and, at least in some of these cases, is dependent on ATP release and on the interaction of the nucleotide with membrane receptors . Many mechanisms could account for the release of cytosolic ATP across the cell membrane, including several types of channels (volume-activated anion channels, VDAC, pore forming P2X7r). Previous studies concluded that Cx43 “hemi-channels” were instrumental to this end between cultured astrocytes exposed to low divalent cation solutions, because drugs blocking gap junctions prevented the intercellular propagation and amplification of the calcium waves [35, 170, 172]. However, further studies showing that amplification of intercellular calcium waves was still present in cultures of astrocytes from Cx43-null mice, and absent in those from P2X7r-null mice , have since demonstrated that the poreforming P2X7 receptor is the most likely candidate to mediate ATP release. More recently, Pnx1 has been shown to be part of the P2X7r complex, providing both a site for ATP release [104, 105], and a mechanism for amplifying the extent to which intercellular calcium waves spread between astrocytes .
It has long been known that erythrocytes release ATP following shear stress and hypoxia [11, 160]. However, it is only recently that the mechanism involved in such release has been shown to depend on Pnx1 “hemi-channels” . Thus, erythrocytes express channels with a unitary conductance (400–500 pS), voltage dependence and mechanosensitivity of “hemi-channels,” and express Pnx1, but no connexins. Pnx1 “hemi-channels” were shown to be permeable to ATP and to large molecular weight dyes, such as carboxyfluorescein. ATP release from erythrocytes was potentiated by high K+ and osmotic shock, and prevented by carbenoxolone at concentrations lower than those usually required to block connexin gap junction channels. The physiological contribution of Pnx1 was proposed to be related to the local control of blood flow, whereby the ATP released from oxygen-deprived or shear-stressed red blood cells would stimulate purinergic receptors on nearby endothelial cells, initiating the intercellular propagation of a calcium wave. In turn, the elevation of intracellular calcium would induce the release of NO onto the vascular smooth muscle, leading to vasodilation and to a consequent increase in perfusion. The ATP released from erythrocytes could also feedback activating the P2X7 receptors of these same cells, that normally mediate the release of EET through Pnx1 and CFTR channels, given that both carbenoxolone and the CFTR inhibitors glibenclamide and niflumic acid prevented basal and ATP-evoked release of EET . Since EET is an important player in the regulation of vasomotion in several vessels , pannexons stand out as crucial channels in this modulation.
Gustatory receptor cells do not display the exocytic machinery involved in neurotransmitter release of most other neuronal cells, and hence cannot release ATP molecules via a vesicular-dependent pathway, yet they are able to convey information about the quality of tastants, via afferent fibers. The taste buds involved in the perception of sweetness and bitterness contains sensory neurons equipped with G-coupled receptors which are involved in the transduction of information, while the buds involved in the perception of saltiness and sourness transduce the information through ion channels . Recent studies have indicated that ATP and serotonin (5-HT) are key mediators in the transduction mechanism of sweet–bitter taste buds, via the ATP release that takes place after tastant presentation. In the extracellular medium, ATP stimulates P2 receptors of the presynaptic taste neurons, leading to the release of 5-HT from these cells. A previous report had implicated Cx43 “hemi-channels” in this mechanism, mostly because conditions known to block gap junction channels, also blocked the outward ATP currents, and in spite of the fact that other conditions acting on gap junctions (e.g., niflumic acid, carbenoxolone, quinine) had no effect on the ATP flux . Blocking conditions included the use of connexin mimetic peptides, which have since been shown to block more efficiently Pnx1 “hemi-channels” than connexin “hemi-channels” . A recent study now indicates that the ATP release takes place through Pnx1 channels of receptor neurons . Furthermore, immunocytochemistry shows that Cx43 is not expressed in receptor cells but in the nearby epithelial cells of taste buds . Thus, it is likely that pannexons are the privileged route for the ATP release which modulates taste sensation.
There is also increasing, still circumstantial evidence for several intracellular functions of pannexons.
Pnx1 was recently proposed to be involved in the immune response, by associating with the P2X7 receptor signaling cascade to form the inflammosome [88, 124]. This complex of cytosolic proteins induces the activation of caspase1, which is necessary for the processing of interleukin IL-1β and IL-18, two important components of the inflammatory response. Activation of P2X7 receptors by ATP is essential for IL-1β release [50, 86, 126], and appears to be dependent on functional Pnx1 “hemi-channels” [105, 124]. Pnx1 is also necessary for the Toll-like receptor-independent formation of inflammosomes comprising cryopirin  which, in turn, is required for the activation of caspase1. It has also been suggested that Pnx1 may trigger the activation of the cryopyrin inflammosome, by mediating the passage of bacterial products from endosomes into the cytosol .
Oxygen and glucose deprivation (OGD) induces neuronal necrosis due to imbalance of intracellular ionic concentrations [63, 99, 101, 199]. Large 500 pS conductance events characteristic of Pnx1 “hemi-channels” but not of the channels that could potentially be made by co-expressed connexins, was activated in freshly isolated mouse hippocampal pyramidal neurons following OGD . The OGD-induced currents were blocked by carbenoxolone and La3+, but not by the rat P2X7 receptor antagonist brilliant blue G. Moreover, after microinjection of calcein green into hipocampal neurons in situ, the efflux of the large fluorescein tracer was only observed following OGD, which was also prevented by carbenoxolone . Although the study does not demonstrate the mechanism whereby OGD induced Pnx1 activity, it was suggested that pannexons contributed to the anoxic depolarization, a frequently observed phenomenon in ischemic conditions, which results in neuronal death . Recently, Pnx1 has also been implicated in the leakage of calcium from the endoplasmic reticulum , which is another event promoting neuronal necrosis following CNS ischemia .
The co-expression of Pnx1 and P2X7 receptors induced zeiosis of Xenopus oocytes following activation of the receptor by extracellular ATP, a phenomenon which was not observed when oocytes were injected either with Pnx1 mRNA alone or with both Pnx1 and the metabotropic purinergic P2Y receptor [105, 106]. These observations suggest that although activation of P2 receptors can induce Pnx1 currents, it is the specific signaling through the P2X7 receptor that leads to cell death.
Recent evidence shows that conditions of oxidative stress, which are expected to favor apoptosis in different cell systems, result in the loss of “hemi-channel” currents, variably attributed to connexons or pannexons [88, 124, 135, 138, 152, 156, 169].
Several studies have provided evidence that the degree of cell coupling as well as the level of connexin expression are reduced in many types of tumor cells and that there is a negative correlation between tumor grade and connexin function and/or expression [89, 97, 131]. Because forced expression of connexins in transformed cells reduces the neoplastic phenotype, it has been proposed that connexins display a tumor suppressor role, by a mechanism likely to be independent of cell coupling and rather related to altered cellular distribution of connexin-binding partners that display transcriptional activity, such as Src, β-catenin and NOV [1, 54, 55, 89, 100 120], and/or to changes in the expression of multiple connexin-dependent genes [29, 78, 97, 115, 162]. Whether and how some of these mechanisms are also dependent on nonjunctional connexons remains to be fully established . Still, clones of glioma C6 cells stably expressing either a myc- or eGFP-tagged Pnx1 had a reduced proliferation and motility compared to cells expressing only eGFP, suggesting a tumor-suppressive role of pannexons . This role was confirmed in vivo, inasmuch as mice injected with Pnx1-expressing C6 cells developed smaller tumors than mice injected with eGFP-expressing C6 cells . Because dye coupling mediated by Cx43 was increased and cell morphology altered in the Pnx1-transfected glioma cells, it was proposed that the tumor suppression action of Pnx1 involved both gap junctional and non gap junctional mediated mechanisms .
Control of the steady-state levels of cytosolic calcium involves a dynamic equilibrium between the active uptake of the cation into the endoplasmic reticulum (ER), which is mediated by the ATP-dependent Ca2+ pumps of the SERCA family, and the passive diffusion of Ca2+ from the ER into the cytosol. Several candidates have been proposed to account for the latter passive diffusion, including the reversal of SERCA, translocon channels, IP3 or ryanodine receptors, Bcl2, Pnx1 and Ca2+ ionophore-like channels [31, 107, 117, 130, 183, 186, 187, 200]. Interestingly, transfection of eGFP-Pnx1 in two cell lines expressing endogenous Pnx1 transcripts lead to the accumulation of the tagged-Pnx1 in the ER, which was associated with a reduced thapsigargin-induced Ca2+ release and with a higher Ca2+ efflux . Conversely, downregulation of endogenous Pnx1 by siRNA strategy, decreased the efflux rate of Ca2+ from the ER . Although these results are consistent with the idea that Pnx1 may provide leak channels and therefore contribute to intracellular calcium homeostasis, several open questions still remain. For instance, it would be expected that Pnx1 be tightly regulated by ER Ca2+ levels, such that Pnx1 channels would open when these levels were high and close when these levels decrease. Such regulation seems unlikely given that the calcium concentration is much higher in the ER (mM range) than in the cytosol (nM range), and that pannexons are opened by micromolar Ca2+ concentrations . At any rate, the Pnx1-mediated leakage of ER Ca2+ could represent a key event in triggering the aforementioned neuronal necrosis following CNS ischemia  and tumor suppression .
Microarray analysis of the transcriptome has revealed that the expression of a wide variety of genes is significantly perturbed after loss of individual connexins [29, 78, 115]. Most of these changes can be to a large degree predicted assuming that in the wild-type tissues, the expression of a connexin gene is functionally linked to that of multiple other genes . Thus, when the coordination profiles of individual genes are compared, the striking likeness of certain gene pairs contrasts with the dissimilarity of others. This approach has revealed that Cx43 and Pnx1 expression is very similarly interlinked to that of all other genes (Fig. 4), over the entire range of coordination values. This striking similarity predicts that an up- or a downregulation of either the Cx43 or the Pnx1 gene would equally affect the cell transcriptome, thus probably resulting in a similar phenotype. The data further suggests that, at the transcriptome level, the overexpression of the Cx43 gene may compensate for the underexpression of the Pnx1 gene, and vice versa. However, this compensation could not be expected to correct those phenotypes resulting from altered direct cell-to-cell communication, given the quite limited efficacy of pannexons to form functional gap junction channels.
The discovery that connexin, pannexin, and innexin proteins may form “hemi-channels” in nonjunctional regions of the cell membrane, in addition to paired innexons and connexons forming the classical cell-to-cell channels at gap junctions, has opened a new field in the area of junctional proteins. First, these findings provide a clear view of some of the mechanisms that mediate paracrine intercellular signaling. Second, they provide innovative insights into the regulation of several other cell functions, for which the occurrence of “hemi-channels” provides a plausible if, in most cases, still putative mechanism. Future work should now expose these novel ideas to direct experimental testing. Even if many of the current speculations we have now about “hemi-channels” would turn out to be erroneous concepts, we are certainly entering a time of exciting, new love romance with junctional proteins [132; Bennett MV, talk at the last International Gap Junction Conference, Helsingør, Denmark, August 2007].
Romance, should not detract from the need of a rigorous and critical assessment of the data, notably with regard to the nature of the proteins making “hemi-channels” in different tissues. As repeatedly stressed in this review, this is still an area of hot debate. Many other questions remain to be addressed. For example, can pannexins form gap junction channels in vivo, e.g., under conditions preventing the glycosylation of their extracellular loops? What is the function, if any, of Pnx2 and Pnx3? Are pannexins also forming functional “hemi-channels” in the membranes of cell organelles? What are their physiological functions in vivo ? Do they play any role in the pathogenic mechanisms that cause human diseases ? If so, can we act on “hemi-channels” to correct these dysfunctions ? A nonambiguous answer to most of these questions implies that we have the experimental tools (drugs, antibodies, peptides, siRNAs, knockout pannexin models, and transgenic animals overexpressing selected pannexins in a cell specific manner) to specifically interfere with the levels and/or the function of selected types of “hemi-channels.” It also requires that cell and animal models lacking or overexpressing “hemi-channels” be available for a direct testing of their function. Normal tissues and cells spontaneously lacking detectable “hemi-channel” function, in spite of expression of Cx36, Pnx1, and the P2X7 receptor, have been recently reported , and a mouse null for Pnx1 is said to be viable, possibly without any obvious phenotype (H. Monyer, R. Bruzzone, personal communication). These models should be helpful in approaching some of the many open questions raised by the physiologically paradoxical plasmalemmal “hemi-channels”.
Work of our teams is supported by grants from the National Institute of Health (HD32573, NS41282, NS52245), the Swiss National Science Foundation (310000–122430), the Juvenile Diabetes Research Foundation (1-2007-158), Novo Nordisk, and the European Union (FP7-222980).