Some species of each of the phospholipid classes analyzed were common to Cx26 and Cx32 hemichannels (Figure 1a, right). Phospholipids specific for Cx26 hemichannels excluded PC (Figure 1a, left) and phospholipids specific for Cx32 hemichannels excluded PS (Figure 1a, middle). Thus, no unique species of PS were associated with Cx32 hemichannels, and no unique species of PC were associated with Cx26 hemichannels.

The proportion of anionic phospholipid species found uniquely associated with Cx26 hemichannels or with Cx32 hemichannels was much greater than that associated with both types of hemichannels, in which approx. 50% of the number of phospholipid species identified were zwitterionic. The proportion of unique ether-linked phospholipid species (those with alkyl ether (e) or vinyl ether/plasmenyl (p) glycerol-alkyl chain linkage; noted in tables and figures as 'e', 'p' or 'e+p') was substantially greater in the Cx32 hemichannel preparations than in the Cx26 hemichannel preparations (Figure 1b).

A small number of zwitterionic phospholipids were common to both Cx26 and Cx32 plaques, consisting of species of PC, PE and PA (Figure 2a, right). Although phospholipids unique to Cx26 and Cx32 plaques included members of all five phospholipid classes, these were in different proportions, with Cx32 plaques favoring anionic lipids (Figure 2a, left and center). Thus, all the identified species of PI and PS in Cx26 or Cx32 plaques were unique to each connexin isoform.

In contrast to the phospholipids common to Cx26 and Cx32 hemichannel preparations (Figure 1a, right), the phospholipids common to Cx26 and Cx32 plaques excluded PI and PS. The proportion of ether-linked phospholipids unique to Cx26 plaques was greater than that unique to Cx32 plaques (Figure 2b). The differences in the phospholipid species found uniquely in Cx26 and Cx32 plaques suggest that the integrity of plaque structures formed by these two connexins have distinct phospholipid requirements.

Graphical representation of the phospholipid species associated with hemichannels, delineated by headgroup, acyl chain length (total number of acyl chain carbons), and number of desaturations in the fatty acid acyl chains, reveals greater detail (Figure 3). A large number of phospholipid species were common to both Cx26 and Cx32 hemichannel preparations (Figure 3a), these having 28 to 38 acyl carbons (with a single exception of 44 carbons; median 36) (Table 6). There was a broad range of saturations, which did not correlate with chain length.

The range of acyl carbons was similar for lipids unique to Cx26 and to Cx32 hemichannels, ranging from 28 to 42 (median 34) for the former, and from 26 to 44 (median 38) for the latter. The phospholipids unique to Cx26 hemichannels tended to be either fully saturated or highly desaturated (Figure 3b). The amount of desaturation in phospholipids unique to Cx32 hemichannels (Figure 3c) was more evenly distributed, and phospholipids with longer chain lengths were more desaturated.

The corresponding graphical representations of the specific phospholipid species associated with plaques are shown in Figure 4. There were many fewer phospholipid species common to Cx26 and Cx32 plaques (Figure 4a) than were in common in the corresponding hemichannel preparations (Figure 3a). The total carbon content of the acyl chains ranged from 30 to 38 (median 36), with the longer chains being more desaturated (Table 7).

A relatively large number of phospholipid species were specific for Cx26 plaques. Their chain lengths were skewed to longer lengths (range 28 to 44; median 40) (Figure 4b), compared with those for the Cx32 plaques (range 22 to 38; median 32) (Figure 4c) or those common to plaques of both connexins. However, phospholipids unique to Cx26 plaques tended to be more desaturated than those in Cx32 plaques. For the lipids unique to Cx32 plaques, the degree of desaturation correlated with chain length.

Tables 6 and 7, respectively, present a summary of the phospholipid differences between Cx26 and Cx32 and their hemichannel versus plaque structural forms. For each connexin isoform, a small number of phospholipid species was found in both hemichannel and plaque preparations (Figure 5; designated by ϕ in Tables 1 to 5). The phospholipids in common in hemichannel preparations and isolated junctional plaques for both connexins were two highly saturated species of relatively short acyl chain length PC, one median acyl chain length polyunsaturated species of PC, and two species of moderately saturated median acyl chain length PE (Figure 5a). Notably, these are all zwitterionic. Unique to Cx26 in both structural forms were two species of polyunsaturated PA, one of short and one of long acyl chain length, and one long acyl chain length polyunsaturated species of PI (Figure 5b). Unique to Cx32 in both structural forms were a species of highly saturated relatively short acyl chain length PA, and a polyunsaturated species of medium acyl chain length PE (Figure 5c). Thus, the anionic phospholipid species common to both the hemichannels and plaques appear to be connexin-specific.

Cx32 hemichannels are associated with unique species of PC, which Cx26 hemichannels are not, and Cx26 hemichannels are associated with unique species of PS, which Cx32 hemichannels are not. The proportion of charged lipid species is much greater among the phospholipids specific for each connexin compared with those lipids in common. The proportion of ether (e+p) glycerol-alkyl chain linkages is greatest for Cx32 hemichannel specific phospholipids. Cx26 phospholipids have a bimodal distribution of desaturations, and for Cx32-specific phospholipids the shorter chain lengths tend to be more saturated.

The data suggest that specific PC species are involved in Cx32-specific structural and/or regulatory interactions but that PCs do not have Cx26-specific interactions. Similarly, the data suggest that specific forms of PS have Cx26-specific structural and/or regulatory interactions, as no PS species have Cx32-specific interactions.

No species of PI or PS are shared between Cx26 and Cx32 plaques. The phospholipids unique to Cx32 plaques tend to have a greater proportion of charged species than those unique to Cx26 plaques. The phospholipids retained in all the plaque preparations have a broad distribution of saturations, but for Cx26 plaques the shorter chains tend to be more saturated. The proportion of Cx26-specific ether-linked phospholipids is higher for the Cx32-specific lipids.

These results suggest that plaques, in general, require phospholipids with a broad distribution of saturations and chain lengths, do not require PS or PI for structural integrity, and may require specific varieties of PC, PE, and PA.

The first compositional analyses of the lipids in gap junctions (using different protocols than this study) were reported for rodent liver [11,74], which contain Cx26 and Cx32. PC was identified as the major phospholipid, with significant PS and lesser amounts of PE and PI. Significant cholesterol was present. Other work reported that cholesterol is specifically enriched in junctional plaques [10]. Cholesterol supplementation of cultured cells was shown to increase gap junction formation [75], which subsequently was attributed to indirect effects [76]. Lens fiber junctions (composed of Cx46 and Cx50), were reported to contain significant sphingomyelin, cholesterol, large amounts of PC and PE and smaller amounts of PS [16]. Our data are consistent with these studies, and provide substantial additional detail about the specific phospholipid species involved. It was recently reported that maturation of gap junctions as they move to the inner cortical regions of the lens is associated with near complete removal of cholesterol from the junctional membranes [77]. The reason for this is unknown, and it may be specific to the lens, and/or to Cx46 and Cx50.

The literature on lipids associated with two-dimensional arrays of proteins in native membranes is largely limited to bacterial and thylakoid membranes. This literature established that the lipids tightly associated with such arrays are structurally and/or functionally important. For example, 'delipidation' of bacteriorhodopsin identified various species of PG, in particular, as being associated with the purple membrane structure [57]. Specific forms of PG are required for two-dimensional crystal formation of the protein [78]. Other work on bacteriorhodopsin, light harvesting complex II, and photosynthetic reaction centers show that PG is present in the three-dimensional crystal structures and is important for function [59,73,79,80]. Bacterial and mitochondrial membranes contain substantial PG, but it is absent from plasma membrane of eukaryotic cells, so one does not expect PG to play an analogous role in gap junction plaques.

As described above, potentially the most stringent criterion for identifying tightly associated phospholipids is their presence in both hemichannel preparations and isolated plaques. In this category, the phospholipids unique to Cx26 are all anionic (PA and PI) and polyunsaturated. In contrast, phospholipids unique to Cx32 in both preparations are both anionic and zwitterionic (PE and PA), the former substantially more saturated and with longer acyl chain length than the latter. It is particularly striking that all of the phospholipids common to both connexin isoforms and structural forms are zwitterionic (PC and PE).

As species of PC and PE are found in common for both connexins, and in both the hemichannel and plaque preparations, one may infer that these phospholipids have strong interactions with both Cx26 and Cx32. As species of PA are uniquely found with each connexin in both hemichannel and plaque preparations, it can be inferred that different PAs interact strongly with Cx26 and Cx32. The differences that distinguish these lipids are that the acyl chains of the PA that interact specifically with Cx26 have fewer carbons and are less desaturated than those that interact specifically with Cx32. The fact that specific PEs are found in common suggests a particular relationship with 'non-bilayer' lipids.

The plasma membrane of eukaryotic cells is asymmetric; the bulk neutral lipid and PC are in the outer leaflet, and anionic phospholipids plus PE are in the inner leaflet (refer to [81,82]). All the phospholipids found uniquely with each connexin isoform in both the hemichannel preparations and plaques (that is, potentially those most tightly and/or stringently associated with connexin) are from the inner leaflet, PA and PI for Cx26, and PA and PE for Cx32. The lipids found in common in both hemichannel preparations and isolated plaques were zwitterionic, PC and PE. The former suggests that the most important phospholipids that are specific for Cx26 or Cx32 are in the inner leaflet, and the latter suggests that there is a shared strong affinity for zwitterionic phospholipids in both leaflets. This kind of specificity of protein-lipid interactions for the phospholipids in the cytoplasmic leaflet is also seen for the KcsA channel, for which the phospholipid interactions in the extracellular leaflet are non-specific and in the cytoplasmic leaflet are specific (for PA and PG) [83].

The data show that Cx32 hemichannels are uniquely associated with substantially more species of ether-linked (e+p) phospholipids than Cx26 hemichannels (Figure 1b), while the reverse is true for plaques (Figure 2b). Ether-linked phospholipids have a variety of biological functions; most is known about the plasmalogens ('p' in the tables), which contain a vinyl ether bond at the sn-1 position [84]. Plasmalogens are membrane-delimited antioxidants and targets for oxidative damage, in addition to being regulators of membrane dynamics and reservoirs for polyunsaturated fatty acids [85,86]. Little is known about the specific membrane biophysics of plasmalogens, with the exception of ethanolamine (PE) plasmalogen, which, in contrast to other plasmalogens, promotes formation of hexagonal lipid phases (and thus, increased curvature stress; see below), to an even greater extent than ester-linked PE [87]. This is likely due to additional attractive forces between the headgroups arising from the decreased hydrophilicity from the lack of a carbonyl oxygen in the sn-1 chain. In this context it is interesting to note that the PE ether-linked lipids in the plaques were unique for Cx26, none being associated with Cx32 plaques (Table 3b). This could indicate that Cx26 plaques, as opposed to Cx32 plaques, have a particular requirement for the influence of non-bilayer lipids.

The activity data revealed a strong dependence on the magnitude of negative charge of the liposome membrane, independent of the specific headgroup contributing the charge. This is an important new finding regarding connexin channel function. It is particularly interesting in light of the finding that the phospholipids uniquely associated with either Cx26 or Cx32 hemichannels are mostly anionic (Table 6, Figures 3b and 3c). In addition, there is a correlation between the greater effect of anionic charge on Cx26 channel activity than Cx32 activity (Figure 6a, bars 3 and 4) and the greater number of anionic phospholipid species identified associated with Cx26 in hemichannels and plaques than with the corresponding Cx32 channels (Table 6 versus 7, Figures 3b and 3c and 4b and 4c).

The density of net positively charged residues in the cytoplasmic domains is greater for Cx26 than for Cx32, both in terms of average charge per length and concentration of positive charge in short segments. The recently crystal structure of Cx26 identifies the two transmembrane helices forming the outer wall of the pore (that is, exposed to membrane lipid) as being TM3 and TM4 [88]. Inspection of the relevant sequences suggests more net positive charges at the cytoplasmic membrane interface than at the extracellular one, but no particular differences between Cx26 and Cx32 in this regard.

Similar to our findings with connexin channels, pure PC membranes do not support significant acetylcholine receptor (AChR) activity, but the presence of anionic phospholipids restores function [89-93]. More recent studies stress the importance of acidic phospholipids as well as cholesterol (see below) in AChR function [94,95].

Surface charge can affect membrane protein function by a variety of mechanisms, in addition to affecting the voltage sensed by the protein. For example, activity of the reconstituted sugar transporter of the human erythrocyte increases with magnitude of surface charge in a manner suggesting it protects the catalytic domain against changes in membrane fluidity [96]. In this case, surface charge modulates the extent to which the state of the bilayer hydrocarbon chains affects transport activity.

The phospholipid compositions tested for their effects on Cx26 and Cx32 hemichannels also vary in their degree of curvature stress. In particular, PE and, to a lesser extent, PA contribute substantial negative curvature stress, while PS tends to oppose it [97]. Curvature stress of a membrane is approximately a linear function of the contribution of each component [98]. However, the distribution of phospholipids in each bilayer leaflet of liposomes is generally presumed to be identical, in which case the lipids themselves would not impose curvature stress. The situation is further complicated by the curvature of the small liposomes involved, and the consequences of that on the distribution of lipids. Thus, plotting the observed TSF activity data as a function of the calculated negative curvature potential of the bulk lipid has little meaning (doing so suggests a weak relationship where less calculated 'curvature' stress correlates with greater activity; not shown); the influence of curvature stress on connexin channel function induced by different phospholipids must be explored by other means.

The data show that connexin channel activity in PC:cholesterol membranes is maximal at approx. 25 mol% cholesterol. Cholesterol interdigitates between bilayer lipids, modifying various physical properties of membranes [99], any or all of which could influence channel activity. Cholesterol may also have direct effects on membrane proteins. The simplest interpretations of the biphasic effect observed with increasing concentration (Figure 7) are either that cholesterol has two effects, one positive and one negative, with increasing concentration, or that a single parameter of its membrane effects has a biphasic modulatory influence.

Prominent among the effects of cholesterol on membrane properties are increased ordering of acyl chains [100], thickening of the bilayer [101] and increased dipole potential [102]. Increased acyl chain order would be expected to have a monotonic inhibitory effect on channel activity. For example, if a conformational change of a protein involved a change in its volume within the bilayer, increased order induced by cholesterol would inhibit that transition [103]. If an effect on acyl chain order contributes to the inhibitory phase of a biphasic effect, the activating phase would need to be caused by another factor, such as direct interaction with cholesterol. This explanation has been proposed for the biphasic effects of cholesterol on other membrane proteins, in which activation is caused by direct interaction with cholesterol and inhibition is caused by increased acyl chain order [104-106]. Alternatively, bilayer thickening could pass through an optimum for channel function as the hydrophobic thickness matches that of the channel protein in one or more of its conformational states. A role for hydrophobic match/mismatch between bilayer and membrane protein in protein function is gaining attention [4,107] (see below).

As the membrane concentration of cholesterol increases, so does the extent of interaction with phospholipid until virtually all the phospholipid capable of forming a condensed complex with cholesterol is interacting directly with it [108-111]. Above this concentration, the effects on membrane thickness level off [101,112,113], and pools of 'free' cholesterol form, that is, unassociated with phospholipid. The concentration at which this occurs varies with phospholipid species present, acyl chain length and degree of saturation, but often occurs at or near the concentration at which the maximal connexin channel activity was observed (approx. 25 mol% in PC). Therefore, in addition to its effects on bilayer structure and dynamics mentioned above, the decrease in channel activity at higher cholesterol concentrations could arise from increased direct protein-lipid interactions, with 'free' cholesterol having an inhibitory effect. The formation of such domains has been suggested to underlie a biphasic effect of cholesterol on membrane dipole potential that peaks approx. 25 mol% in dipalmitoyl-PC [102].

A peak in connexin channel activity at maximal cholesterol-induced membrane thickness could have important biological implications. Membrane thickness steadily increases in the secretory pathway from endoplasmic reticulum to Golgi to plasma membrane, due in part to increasing levels of cholesterol (less than 5 mol% to approx. 30 mol% [114]). Thus, cholesterol-mediated membrane thickening could play a role in connexin channel maturation to a functional state as it reaches the plasma membrane. As previously noted, unapposed connexin channels can exist in plasma membrane microdomains [49-51] that are typically thicker than surrounding membrane because of sphingomyelin and cholesterol enrichment [99]. Several lines of evidence also suggest substantial enrichment in the cholesterol/phospholipid ratio in gap junction plaques [9-16]. In addition, a connexin channel shifting between different microdomains, or between microdomains and plaques, could experience different types of hydrophobic mismatch with its surrounding bilayer, which could affect its function (see below).

Though the weak positive effect of PE on channel activity is consistent with the effects of its small surface charge (Figure 6b), it is possible that PE also affects the channels via its ability to introduce changes in the curvature stress and lateral pressure profile of the bilayer. Forms of PE were found uniquely, and in common, for each connexin and for hemichannel and plaque preparations.

PE is a classic 'non-bilayer' lipid. Its small headgroup relative to the cross section of its acyl chains results in a conical shape that introduces negative curvature strain into membranes, that is, increased lateral pressure in the acyl chain region and decreased lateral pressure between the headgroups. As noted, PA has a similar but weaker effect, PS tends to oppose it [97], and curvature stress of a membrane is approximately a linear function of the contribution of each component [98]. The presence of such forces, primarily due to PE, could affect channel function, depending on the types of intramembrane volume changes that occur during gating [115-119]. Also, there may be a highly local relation between spontaneous curvature of the bilayer and the mean curvature of the lipid/water interface adjacent to the protein, as occurs for rhodopsin [120], a variation of the forces that come into play for 'hydrophobic mismatch', mentioned above and discussed below.

In this context it is interesting to note that the tendency of PE-containing membranes to form hexagonal phases, and the consequent increased membrane curvature stress, is enhanced by virtually all of the lipophiles that inhibit connexin channels (see below). These include certain alkanols [121], highly unsaturated fatty acids [122] (including oleamide, a connexin channel inhibitor), and halothane [123]. These agents may act synergistically with PE and other bilayer factors to affect connexin channel function via effects on the lateral membrane pressure profile. A possible exception is cholesterol, which can have the same effect on PE-containing membranes [124-126], but which enhanced channel activity in PC membranes in our study.

The ability of PE to promote negative curvature correlates with acyl chain length, and to a lesser degree with unsaturation [127]. The median chain lengths of the PE species specific to Cx26 plaque preparations are substantially greater than those in any of the other hemichannel or plaque preparations (approx. 40 versus approx. 36), suggesting a particular requirement for, or sensitivity to, negative curvature stress for Cx26.

In fact, the recent high-resolution crystal structure of a Cx26 channel indicates that each hemichannel is a toroid, approx. 30 Å wider at the cytoplasmic lipid boundary than at the extracellular lipid boundary [88]. This form would impose positive curvature stress to the bilayer, unless PE or other non-bilayer lipids were closely associated with the channel in the cytoplasmic leaflet, or lysolipids in the extracellular leaflet.

The PE-dependent effects on lateral bilayer stress described above are a special case of a more global aspect of membrane-protein energetics. Lateral bilayer stress can be induced and influenced by a variety of factors. Any reagent that partitions into a membrane and does not have a cylindrical shape or does not penetrate completely through the bilayer can affect the lateral stress via effects on intrinsic membrane curvature (the same parameter affected by PE). Of particular interest for connexin channels is the possibility of hydrophobic mismatch between the connexin channel and bilayer thickness, here suggested by the biphasic effect of cholesterol.

Hydrophobic mismatch interacts with the intrinsic membrane curvature to generate an energy that impinges on the channel, the 'bilayer deformation energy'. Consequently, membrane proteins can have optimal activity at specific average acyl chain lengths. Modeling shows that the effects arising from hydrophobic mismatch alone can account for biphasic relations between bilayer thickness and membrane protein activity [107,128].

The precise delimitations of the transmembrane helices of connexins remain uncertain; unlike many membrane proteins, the connexin amino acid sequence does not have a conspicuous clustering of aromatic residues at the boundaries between exposure to lipid headgroups and acyl tail [129,130]. However, the recent crystal structure of Cx26 suggests membrane boundaries of the transmembrane helices and a membrane thickness of approx. 38 Å [88]. The two external helices (TM3 and TM4) form the channel boundary with membrane lipid, and are exposed to lipid for almost all of their lengths that are within the membrane. These helices are tilted relative to each other and to the plane of the membrane. The crystal structure shows intraprotomer contacts between these two helices.

In this situation, change of the hydrophobic thickness of the surrounding bilayer will alter the tilt of the helices relative to each other, and relative to the channel axis. This structural change may be a dynamic and specific mechanism by which changes to the hydrophobic thickness of the surrounding bilayer can be sensed and transduced to modulate channel function. One could imagine, for example, such functional channel 'maturation' arising after increases (by cholesterol, see above, or other means) in hydrophobic thickness, for example, in the secretory pathway from endoplasmic reticulum to Golgi to plasma membrane.

A predominance of the known modulators of gap junction communication are lipophilic and/or show some specificity with regard to membrane lipids [17,18]. Among the lipophiles that inhibit connexin channels are the aliphatic alcohols heptanol and octanol [131,132], volatile anesthetics such as halothane [133], arachidonic acid and other non-saturated fatty acids and their amides, for example, oleamide (cis-9-octadecenamide) [134,135], and the triterpinoid saponins, glycerrhetinic acid and carbenoxolone [136]. These reagents are not selective for connexin channels [17,18], and likely act by partitioning into plasma membrane and affecting membrane physical properties. The present study suggests that the reported differences and apparently contradictory reports of the effects of specific lipophiles on different connexin isoforms [18] may arise, in part, from the distinct sets of lipids that are closely associated with different connexin isoforms.

Aliphatic alcohols produce a rapid and reversible inhibition of gap junctional conductance [131], with potency inversely related to chain length. The effects of heptanol and octanol are specifically correlated with a decrease in the fluidity of the cholesterol-rich domains (that is, plaques) in which junctional channels reside, in spite of an increase in bulk membrane fluidity elsewhere [132]. Cis-unsaturated fatty acids, including oleamide, affect connexin channels in a manner positively correlated with the degree of unsaturation and, like aliphatic alcohols, negatively correlated with chain length [137,138]. The ability of fatty acids to affect the function of membrane proteins is thought to be due to their ability to increase disorder of the lipid acyl chains of the membrane [139].

It is unclear whether cis-unsaturated fatty acids have different effects on cholesterol-rich and cholesterol-poor membranes, as do the aliphatic alcohols. In both cases, the relevant membrane parameter, at least regarding bulk membrane properties, may be modification of the lateral pressure profile rather than acyl chain order/disorder per se. In addition, different fatty acids affect different connexins in distinct ways, perhaps reflecting different efficacies for acyl chain disruption of specific lipid species, consistent with our observed preference of different connexins to associate with different phospholipids.

The inhalation anesthetic halothane rapidly and reversibly inhibits junctional communication [131], but does so with junctional channels formed by heteromeric Cx40/Cx43 hemichannels at significantly lower doses than either of the corresponding homomeric Cx40 or Cx43 hemichannels, demonstrating that halothane sensitivity can be influenced by connexin isoform composition [133]. Halothane binds with high affinity to appropriately sized hydrophobic cavities in proteins to stabilize helix-helix configurations [140]. While this is a possible mechanism for its different effects on these homomeric and heteromeric channels, it could also arise from effects on membrane physical properties, if the lipid micro-environments around homomeric and heteromeric channels are different, as inferred from the present study.