Gap junction proteins are not interchangeable in development of neural function in the Drosophila visual system

Gap junctions (GJs) allow cytoplasm-to-cytoplasm communication via ions or other small molecules (Kumar and Gilula,1996; Nicholson and Bruzzone,1997). In vertebrates, GJs are composed of multimers of connexins(Swenson et al., 1989), and the connexin family has at least 25 different members(Simon and Goodenough, 1998). In invertebrates, innexins, which have no sequence homology to connexins, are the structural molecules for GJs (Krishnan et al., 1993; Starich et al.,1996; Phelan et al.,1998). The innexins are a large family with approximately 25 genes in Caenorhabditis elegans and eight genes in Drosophila. Mammalian genes with a very low homology to innexins have been reported, which suggests that this may be a universal family(Panchin et al., 2000).

The reason for the large number of connexin genes seems to be related to their variable permeability and electrophysiological properties, as well as their ability to regulate cellular coupling through selective connexin mating(White and Bruzzone, 1996)(also see Discussion). Connexins are sometimes interchangeable, and sometimes not, as shown in mice when one connexin gene is replaced with another via genetic knock-in (Plum et al.,2000). Specifically, there is tissue-specific variability in the ability of Cx43 or Cx32 to functionally replace Cx40. Differences in intercellular communication requirements are suggested to be the main reason for failure of one connexin to substitute for another, although the exact causes of such failure are generally unknown(Plum et al., 2000). A large number of studies on connexin chimeras have implicated discreet portions of connexin proteins as responsible for the observed functional variability, from mating compatibility to ion selectivity (see Discussion).

Systematic studies examining innexin interchangeability have not been done as innexins are a relatively new GJ protein family. However, different innexins behave differently in experimental situations, suggesting that they have distinct properties. First, some innexins form GJs in a paired Xenopus oocyte assay, whereas others do not(Phelan et al., 1998;Landesman et al., 1999;Stebbings et al., 2000)(K.D.C., T. White, R.J.W. and D. Paul, unpublished). Those innexins that fail to form GJs in this assay, nevertheless apparently code for GJs as their mutation leads to a loss of GJs between specific cells in animals(Krishnan et al., 1993;Starich et al., 1996) (Y.-A. Sun and R.J.W., unpublished). Second, pan-neural expression of GJ transgenes causes a variety of phenotypes. Such expression of two different, but highly homologous, splice variant products of shakingB (shakB), shakB(L) and shakB(N)(Krishnan et al., 1995;Zhang et al., 1999) leads,respectively, to lethality or to no obvious phenotype(Stebbings et al., 2000;Curtin et al., 2002). Pan-neural expression of D-inx-2 is lethal, whereas expression of D-inx-7 has no obvious phenotype (K.D.C. and R.J.W., unpublished). Pan-neural expression of ogre (optic ganglia reduced) causes cell loss in the eye, whereas expression of D-inx-2 in the eye along with ogre reduces ogre-induced cell loss (K.D.C. and R.J.W., unpublished).

These differences in innexin function could be due to specificity in innexin-innexin interactions. Some innexins form heteromeric junctions: the hemi-junction is composed of more than one innexin(Stebbings et al., 2000). Innexins may also form heterotypic junctions: hemi-junctions on apposing cells are made from different innexins. The observed overlapping innexin expression patterns in flies would allow such mixing and matching(Curtin et al., 1999;Stebbings et al., 2000;Zhang et al., 1999). Innexins could also differ in crucial permeability or biophysical properties. By whichever mechanism, innexin specificity must be mediated by protein sequence differences.

We examined innexins interchangeablity in the Drosophila visual system. We have recently shown that shakB and ogre are required for the development of normal neural information flow between retinal photoreceptors and their lamina synaptic partners in the adult(Curtin et al., 2002). Mutations in either gene eliminate neural transmission from the retina to the lamina (Homyk et al., 1980;Lipshitz et al., 1985; Curtin et al.,2002). Ogre is required in retinal neurons, whereas shakB(N) is required, at a minimum, in lamina neurons(Curtin et al., 2002). Expression of both proteins is required during development(Curtin et al., 2002).

During development, temporary GJs have frequently been observed between cells that will later form chemical synapses with each other(Fischbach, 1972;Lopresti et al., 1974;Allen and Warner, 1991). However, tools have not been available to determine whether this relationship is causal or coincidental. Our work suggests that developmental GJs, possibly between pre- and post-synaptic cells, are needed for normally functioning connections to form in the Drosophila visual system(Curtin et al., 2002). Specificity in innexin function could determine which cells form GJs and regulate what signals are exchanged. In connectivity/synaptogenesis, such molecular specificity could influence which cells interact to form functional chemical synapses, thus contributing, directly or indirectly, to synaptic specificity.

Here we show that innexins, though highly homologous(Curtin et al., 1999), are not interchangeable in their role to promote normal neural development. In addition, we show that we can map protein regions required for such innexin specificity and we discuss what specific residues may mediate the observed specificity.

A 4.5 kB XhoI-HindIII fragment from upstream of the ogre message was cloned into the Gal4 vector pGatB(Brand and Perrimon, 1993). The XhoI site at the 5′ end was the same site as that in the 5′ end of the clone used by Watanabe and Kantel for ogre rescue(Watanabe and Kankel, 1990). The HindIII site is within the 3′ end just inside the ogre message start site. ShakB^N1 and ogre cDNAs were amplified by PCR and cloned into pUAST(Brand and Perrimon, 1993). Constructs were introduced into flies by P-element-mediated transformation(Spradling et al., 1982). The elav-Gal4 enhancer trap line, C155, was obtained from Corey Goodman. ShakB² was recombined with C155, and this recombinant was used in rescue experiments by crossing to UAS-geneX lines. Gene X refers to any one of several innexins,connexin43 or chimeric genes used for rescue.

Ogre^cb8/Y; ogre-promoter Gal4; UAS-geneX males were tested for rescue.

We made chimeric genes by making two half-genes by PCR and ligating them together. Since we wanted to make a breakpoint in the coding part of each gene, we used pfu polymerase, which gives bluntended products (i.e. it does not tack on extra nucleotides). pfu also has 10×higher fidelity than taq. To make shakB(N)^1-97ogre^108-end we used a 5′ primer that was homologous to the 5′ untranslated portion of shakB(N). This primer included sequences for an EcoRI restriction site at the very 5′ end. The other shakB(N)-specific primer was in a reverse direction and complementary to the codons for amino acids 97, 96 etc. (21 bp long). This primer contained a 5′ phosphate. These two primers were used to amplify shakB(N) sequences from the beginning of the gene through to amino acid 97 using a cloned shakB(N) gene as the template. Similarly, ogre was amplified from amino acid 108 to the end. The forward primer was homologous to ogre codons for amino acids 108, 109 etc. (20 bp long). This primer contained a 5′phosphate. The reverse primer was homologous to sequences in ogre just 3′ to the stop codon. This primer contained at it's 5′ end an EcoRI restriction site. These two PCR products were gel purified and cut with EcoRI. They were then ligated into bluescript cut with EcoRI in a three-way ligation, including a blunt-end ligation at the`joint' between shakB(N) and ogre. IPTG/X-gal selection allowed us to detect clones that contained inserts. Three inserts are possible, and we used restriction analysis to choose clones with the correct insert. These were sequenced to insure that they were correct. ShakB(N)^1-172ogre^183-end was made by introducing an EcoRI site at the junction between shakB(N) and ogre. Though neither gene has an EcoRI site at this position normally, ogre's sequence is one base different from an EcoRI site. Introducing an EcoRI site involved changing a base in the wobble position leading to no change in amino acid 185, which remains as F. The ogre primer at this spot was complementary to amino acids 184, 185 etc. (21 bp long), with the one base change at the wobble position for 185(F) to create the EcoRI site. The primer to the 3′ non-coding sequences of ogre was designed with a BglII site at its end. These two primers were used to amplify ogre from the codon from amino acid 183 to the end. To amplify shakB(N) from the beginning to the codon for amino acid 172, we designed a reverse primer that contained an EcoRI site with homology to codon 172, 171 etc. (21 bps long). The forward primer was homologous to the 5′ non-coding end of shakB(N), with a BglII site at the 5′ end. These two products were ligated via a three-way ligation into bluescript cut with BglII. Again, we used X-gal/IPTG screening to select for clones with inserts,followed by restriction analysis to find clones with the correct insert. These were subsequently sequenced. The inverse product ogre¹⁸²shakB(N)¹⁷³ was made with the same basic strategy of introducing an EcoRI site into both genes at this position. In the forward primer for shakB(N), the EcoRI site replaces codons for amino acids 174 and 175, causing a sequence change from GAG CTG to GAA TTC. This changes residues E L in shakB(N) to the ogre sequence E F. The breakpoint for shakB(N)²⁴¹ogre²⁵² was created by using an existing PstI site in ogre at the codon for amino acid 249/250. This restriction site was introduced into the corresponding place in shakB(N) by including PstI sequences in the primer used for PCR. A three-way ligation was done and clones selected as previously described. To make clones with more than one breakpoint, the same general strategy was used, only this time chimeric genes containing one breakpoint were used as templates for the second PCR reaction. All chimeric genes were subcloned into pUAST and introduced into flies by P-element-mediated transformation.

Flies were immobilized in a plastic pipet tip with the head protruding from the tapered end. Flies were oriented with one eye facing a fiber optic lamp positioned behind a shutter. Glass capillaries (1 mM diameter, with filament,Precision Instruments M1B100F-4) were pulled and filled with 130 mM NaCl, 50 mM KCl. A recording electrode was inserted just inside the cornea. A ground electrode was placed into the back of the head. A Grass stimulator was used to trigger an oscilloscope, and after a delay, to open a shutter to expose flies to the fiber optic light for 4 seconds. The response was recorded on an oscilloscope, the screen photographed on polaroid film and the pictures scanned into a computer.

Electroretinograms (ERGs) record the presynaptic receptor potential and the postsynaptic response of the lamina. Fig. 1A shows the ERG from a wild-type (Canton-S) strain of Drosophila. ERG measurements were taken in a dimly lit room. Immobolized animals were subjetced to a bright light pulse from a fiber optic lamp for four seconds. As can be seen inFig. 1, wild-type ERGs have at least three distinct phases. The cornea-negative (downward) response that lasts for the duration of the light pulse (the receptor potential) records the depolarization of the photoreceptor cells. In addition, there is a cornea-positive (upward) on-transient (arrow,Fig. 1A). A second downward transient at lights-on is usually masked by the receptor potential and is thus not further considered. When the lights go off there is a short cornea-negative (downward) transient (arrow,Fig. 1A). Subsequently, the receptor potential recovers to its pre-stimulus baseline.

The on- and off-transients are responses of the second order lamina cells when they are synaptically driven by the photoreceptors(Alawi and Pak, 1971;Heisenberg, 1971;Burg et al., 1993). In addition, mutants that disrupt neurotransmitter synthesis in the retina or eliminate synaptic vesicle release abolish transients but do not affect the receptor potential (Hotta and Benzer,1969; Pak et al.,1969; Burg et al.,1993; Stowers and Schwarz,1999). These synapses are purely chemical - there are no gap junctions between pre- and postsynaptic cells in the adult.

The ERG phenotypes of both ogre^cb8 and shakB² mutants indicate a failure of transmission between the retinal photoreceptor neurons and their post-synaptic partners, the lamina monopolar neurons (below) (Homyk et al.,1980; Lipshitz and Kankel,1985; Curtin et al.,2002). For both mutants, the receptor potential was normal(Fig. 1B,C), indicating normal photoreceptor cell transduction and depolarization. However, both mutants show an absence or a reduction in transients, indicating that the lamina is not receiving its normal input from the retina. For ogre^cb8animals, both transients were missing entirely(Fig. 1C,Table 1; see alsoCurtin et al., 2002). The phenotype for shakB² was less severe(Fig. 1B;Table 1) (see alsoHomyk et al., 1980;Curtin et al., 2002). ShakB² flies (Fig. 1B; Table 1 showed on-transients that were approximately 47% of wild-type, whereas the off-transients were approximately 25% of wild-type(Table 1).

Though ogre^cb8 has optic ganglia that are reduced in size (Lipshitz and Kankel,1985), the failure of transmission in this mutant is independent of the reduced ganglia. The ERG transients are absent even in mosaic animals in which just the retina is mutant for ogre and the optic ganglia is genotypically and morphologically normal(Curtin et al., 2002).

The ERG readings shown for ogre^cb8(Fig. 1C) also show delayed recovery of the receptor potential. Our ogre^cb8 flies also carry mutations that reduce eye pigmentation. Flies with reduced screening pigment are more sensitive to blue light, which causes an after-depolarization that leads to a delayed recovery of the photoreceptor potential(Cosens and Briscoe, 1972). The delay in recovery we see can be eliminated by screening out blue light with an orange filter, which shows that this is due to the lack of screening pigment and not to the ogre mutation (data not shown).

To rescue both shakB² and ogre^cb8,we made use of the versatility of the Gal4/UAS expression system developed previously (Brand and Perrimon,1993). Briefly, GAL4 is a yeast transcription factor protein that binds to a sequence known as the upstream activating sequence (UAS). Binding turns on expression of any gene that is placed immediately downstream of the UAS. Gal4 is cloned behind a Drosophila promoter that will turn on in the cells of interest, and the gene that we want to express in those cells is cloned behind UAS. These constructs are introduced separately into flies by P-element-mediated transformation. Both constructs are combined in the same fly by mating. Gal4 is expressed in the cells of interest, it binds to UAS and the target gene is turned on.

To rescue shakB², an elav-Gal4 enhancer trap,which expresses in all neurons from neural birth through adulthood, was used to drive expression of wild-type shakB(N). The on-transient was rescued to 95%of wild-type levels; the off-transient was rescued to 76% of wild-type levels(compare Fig. 2D,Table 2 with wild-type animals in Fig. 1A,Table 1 (see alsoCurtin et al., 2002). As a negative control we used a parent with the driver but without the UAS-shakB(N), that is, elav-Gal4, shakB² males(Fig. 2C); these exhibited no rescue and were identical to shakB²(Fig. 1B).

For rescue of the ogre^cb8 ERG phenotype, we used an ogre-promoter-Gal4 driver (Curtin et al., 2002), which should contain all the upstream elements needed for ogre rescue, to drive expression of wild-type ogre. This driver expresses in a subset of retinal neurons beginning in the early pupal stage (Curtin et al., 2002). We found that double copies of both the ogre-promoter-Gal4 driver and UAS-ogre (or UAS-geneX) were needed to effect adequate rescue (see Materials and Methods). Animals that contained two copies of both the UAS-ogre and the ogre-promoter-Gal4 showed rescue of on-transients that are approximately 80% of wild-type and off-transients that are approximately 70% of wild-type levels(Fig. 2B,Table 2). Control ogre^cb8 mutants containing the driver alone show no rescue(Fig. 2A).

Once we demonstrated rescue of the ogre and shakB² mutants with their respective wild-type genes, we went on to test the ability of other innexin genes, as well as connexin43, to rescue ogre^cb8 and shakB² using the same drivers.

The innexin genes we tested are shown in the pileup inFig. 3. These proteins are all highly related (Curtin et al.,1999). They also show partially overlapping expression patterns. For example, D-inx-2, D-inx-3 and ogre all have largely overlapping expression patterns in the embryo (Curtin et al.,1999; Stebbings et al.,2000). ShakB(N) and D-inx-2 are both expressed in adult lamina neurons (Curtin et al., 1999;Zhang et al., 1999). ShakB(N)also has partially overlapping expression patterns with other innexins in the embryo (Zhang et al.,1999).

ShakB(N) and shakB(L) are made from alternately spliced mRNAs and are identical except for an approximately 20 amino acid difference in the first approximately 105 amino acids (Krishnan et al., 1995) and a slightly longer N-terminus in shakB(L) (see Discussion). Fig. 4 compares this divergent portion of shakB(N) and shakB(L). Mutations in shakB(N) lead to a loss of GJs in the Giant Fiber system(Phelan et al., 1996;Sun and Wyman, 1996), whereas mutations that eliminate shakB(L) lead to lethality(Krishnan et al., 1995). ShakB(L) also forms homotypic GJs in the paired Xenopus oocyte assay,whereas shakB(N) does not (Phelan et al.,1998).

Ogre^cb8 mutants were rescued by wild-type ogre driven by the ogre-promoter-Gal4, as described above. However, there was no rescue when this construct was used to drive expression of shakB(N), shakB(L),D-inx-7 [prp7 (Curtin et al.,1999)] or D-inx-2 [prp33(Curtin et al., 1999)] in an ogre^cb8 background[Table 2; for shakB(L) data,see Curtin et al., 2002(Curtin et al., 2002)]. Thus wild-type ogre protein rescues ogre mutants, but the four other innexins tested do not.

ShakB² was rescued when shakB(N) was driven in all neurons via elav-Gal4, as described above. ShakB²was not rescued at all by ogre using this same driver(Table 2). The failure of ogre to rescue is not due to an inadequate amount of ogre protein made via the elav-Gal4 driver as we know we can rescue ogre^cb8with this same driver (Curtin et al.,2002). Interestingly shakB² was partially rescued by D-inx-7 (prp7), which rescues the on- and off-transients to levels that are approximately 55% and 74%, respectively, of the rescue seen with shakB(N) (Table 2, seeFig. 5 legend for calculation scheme). We were unable to test whether D-inx-2 (prp33) or shakB(L) can rescue shakB² because expression of either of these homotypic gap junction proteins in all neurons via the elav-Gal4 driver is lethal(Stebbings et al., 2000)(K.D.C. and R.J.W., unpublished). We have been unable to find a driver that expresses in the developing visual system that does not lead to lethality when driving either of these two homotypic GJ proteins.

Cx43 does not rescue either mutant. However, immunohistochemistry shows that Cx43 is not stable, or not expressed well, in the visual system when expressed in all neurons (Z.Z. and R.J.W., unpublished).

The failure of each of the above tested innexins to replace shakB(N) or ogre in rescue experiments is not due to a failure to express a stable,functional protein. We have direct functional evidence that each protein is made in the visual system. ShakB(N) and D-inx-7 both rescue shakB² and ogre rescues ogre, showing that these three proteins are all functionally expressed in this assay. We have also demonstrated, by immunohistochemistry, expression of shakB(N) in the nervous system of animals expressing shakB(N) via elav-Gal4(Curtin et al., 2002). In addition, ogre overexpression in the eye with several different drivers causes a rough eye phenotype, and co-expression of D-inx-2 reduces this roughness,indicating that D-inx2 is also functional when expressed in the eye (K.D.C. and R.J.W., unpublished). Lastly, we know that expression of both shakB(L) and D-inx-2 via elav-Gal4 is lethal(Stebbings et al., 2000)(K.D.C. and R.J.W., unpublished). We have no way to directly compare the amount of each protein made in each rescue paradigm since most of the proteins do not include an antibody tag. Thus we cannot formally rule out the possibility that inadequate amounts of protein are made for rescue with some Gal4-UAS combinations.

One might imagine that the absence of ERG transients in the above rescue experiments could be due to the exogenous proteins polymerizing with and thus disrupting the function of endogenous innexin proteins. However, when we express any of these proteins in non-mutant animals using either elav-Gal4 or ogre-promoter-Gal4 the ERGs are normal (data not shown). This shows that the failure of one innexin to rescue another is not a false negative caused by interference.

Clearly, innexins are usually not interchangeable despite fairly high homology. For example, shakB(N) and ogre, which cannot substitute for each other, are 64% similar and 54% identical, with the homology being fairly evenly distributed across the protein. Next, we determined whether identifiable regions of the protein were crucial for functional specificity by making chimeras between shakB(N) and ogre and testing them for their ability to rescue ERGs for each mutant. This kind of information may be a first step in understanding what kinds of molecular functions are important for innexin specificity, that is, extracellular interactions, properties of the transmembrane pore, protein polymerization etc. No portion of innexin proteins have been found to be homologous to other protein families, thus innexins do not contain known structural motifs other than the hydrophobic transmembrane domains.

Barnes and Hekimi describe a putative tertiary structure for innexin proteins, predicting four transmembrane domains with the N and C termini both inside the cell (Barnes and Hekimi,1997). Such folding would lead to two extracellular and one intracellular loop as well as a short N-terminal tail and a long C-terminal tail. Putative transmembrane domains are marked in the pileup inFig. 3. We have some experimental evidence to suggest that the C-terminus is inside the cell as the model predicts. Specifically, when shakB(L) protein, with a C-terminal myc tag attached, is expressed in Drosophila salivary gland cells, the myc tag cannot be detected by antibody unless a detergent is added to permealize the cell membrane and thus allow access of antibody to the cell cytoplasm(data not shown).

We used the Barnes and Hekimi model as a guide to design chimeras between shakB(N) and ogre in which putative physical regions of these two proteins,such as extracellular loops etc, are exchanged. We used this model because it was based on the analysis of several family members from both Drosophila and C. elegans. Chimeras were tested to see if they could rescue ogre^cb8 or shakB²using the drivers and protocols already described. The three main `break points' for our chimeras are marked on the pile-up inFig. 3. These break points represent places where the coding sequence from one protein has been exchanged with a coding sequence from the other. There are gaps in the numbering only because shakB(N) is slightly shorter at the N-terminus than ogre so that, for example, amino acid 97 of shakB(N) is in a homologous position to residue 107 of ogre. The break points are as follows: (1) amino acid 107 of ogre (97 of shakB(N)), marking the end of the first putative extracellular loop; (2) amino acid 182 of ogre (172 of shakB(N)), located just after the presumed intracellular loop; and (3) amino acid 251 of ogre (241 of shakB(N)), which is about three quarters of the way through the presumptive second extracellular loop. Chimeras are named along the following lines: ogre^1-107shakB(N)^98-172 ogre^183-end encodes ogre from the N-terminus through amino acids 107, shakB(N) from amino acids 98-172 and ogre from amino acid 183 to the end. The chimeras are shown schematically inFig. 5 where it is easy to see what portions of the proteins have been swapped.

First, it is important to note that all the chimeras were able to partially rescue either ogre^cb8 or shakB², so all must be stable and make a functional protein. In addition, none of these proteins had any negative interfering effects; expressing each chimeric transgene in wild-type animals had no effect on the ERGs. Strikingly, although each chimera rescued either shakB² or ogre^cb8, no chimera rescued both. This result suggests that there are distinct regions of each innexin protein that have diverged to provide specific functions. Such protein regions may mediate defined molecular functions, such as innexin-innexin interaction or pore permeability.

A chimera whose first half is ogre and second half is shakB(N),ogre^1-182shakB(N)^173-end rescued ogre^cb8, but did not rescue shakB²(Fig. 2E,G,Table 2, and schematically inFig. 5F). The reverse chimera whose first half is shakB(N) and second half is ogre,shakB(N)^1-172ogre^183-end did not rescue ogre^cb8 but did rescue shakB²(Fig. 2F,H, alsoTable 2,Fig. 5B). This chimera rescued shakB² effectively, rescuing the on-transient 89% as effectively as shakB(N) protein and rescuing the off-transient 92% as effectively as shakB(N) (Fig. 5B). These results show that residues in the first half of shakB(N) are essential for shakB(N)-specific function.

We constructed and tested other chimeras to narrow down the portion of shakB(N) required for specific function. A chimera that contains only the second transmembrane domain, the intracellular loop, and the first half of the third transmembrane domain of shakB(N)(ogre^1-107shakB(N)^98-172ogre^183-end) rescued shakB² to almost 100% of the level of full-length shakB(N)protein (Table 2,Fig. 5C). This would suggest that all the sequences that determine specificity of shakB(N) activity (the`protein identity region') are in this intracellular loop or in the second transmembrane domain. Though a shakB(N)-specific first extracellular loop was not needed for specificity of development of retina-lamina transmission, it may be needed for specificity in other tissues where shakB(N) may make other protein-protein contacts (see Discussion). In addition, we saw some rescue(30-38%) with shakB(N)1-97ogre^108-end(Table 2,Fig. 5D). This construct rescued approximately one third of animals to a nearly complete level and failed to rescue approximately two thirds of animals at all. This suggests that shakB(N) sequences from the N-terminus through the presumptive first extracellular loop are capable of some degree of function. It also suggests that the requirement for sequences between 98-172 for shakB(N) specificity is not absolute.

Sequences between residues 98 and 172 contribute to shakB(N) specificity. In addition, D-inx-7 rescues shakB² but ogre does not. Therefore, to identify possible key residues important for shakB(N) identity,we looked for residues between 98-172 where D-inx-7 and shakB(N) are chemically similar, but ogre is chemically different(Fig. 3, asterisks). There are six such residues. Four of these residues are found clustered just after the second transmembrane domain. In addition, five of the six represent a change in charge. These residues may turn out to be important for shakB(N)specificity.

An ogre-shakB(N) chimeric protein that contains only the first half of ogre, ogre^1-182shakB(N)^173-end rescued ogre^cb8 ERG phenotype to an appreciable degree, retaining 50% to 66% of the function of the full-length ogre protein(Table 2,Fig. 2E,Fig. 5F). It did not rescue shakB² (Fig. 2G, Table 2). This construct contains ogre sequences from the N-terminus through to the first few residues of the third transmembrane domain and thus includes the first extracellular loop, the first two transmembrane domains and the intracellular loop (Fig. 5F). Conversely, the chimera in which the first half of the protein encodes shakB(N),shakB(N)^1-172ogre^183-end, does not rescue ogre^cb8 at all (Table 2, Fig. 2F,Fig. 5B). These results suggest that some residues in the first half of ogre are essential for the specificity of ogre function.

We made chimeras that subdivide the first half of ogre in an attempt to narrow down the sequences needed for ogre rescue. A chimera that contains shakB(N) sequences from the beginning of the molecule through the presumptive first extracellular loop,shakB(N)^1-97ogre^108-end failed to rescue ogre^cb8 (Table 2, Fig. 5D). Likewise, the chimera ogre^1-107shakB(N)^98-172ogre^183-end, which contains shakB(N) residues from the second transmembrane domain through to the first few residues of the third transmembrane domain (this includes the intracellular loop), failed to rescue(Table 2,Fig. 5C). This suggests that some residues on both sides of the second transmembrane domain breakpoint are essential for ogre-specific function.

Though some residues in the first half of ogre are essential for ogre rescue, chimeras that contained amino acids 1-182 rescued a third less than full-length ogre does, suggesting that some sequences in the last half of ogre may contribute to, but are not essential for, ogre specificity. In addition, ogre^1-182shakB(N)^173-end(Fig. 5F) and ogre^1-182shakB(N)^173-241ogre^252-end(Fig. 5G) rescued to the same level (Table 2), suggesting that amino acids 252-362 are not contributing to specificity at all. This means that the difference in activity between these chimeras and full-length ogre is due to replacement of ogre residues within amino acids 183-251 with shakB(N) residues. These sequences are in the putative second extracellular loop.

Lastly, since we know that the internal tail is not conferring unique identity on shakB(N) or ogre, we wondered if it was required at all. We cloned a truncated form of ogre that stops just after the last transmembrane domain. This protein was not able to rescue ogre^cb8 ERGs(Table 2,Fig. 5H). Thus the internal tail is essential for innexin function.

Although they show high sequence similarity, shakB(N) and ogre, as well as other innexins, are generally not functionally interchangeable. We have shown that mutations in ogre are rescued by ogre but not by shakB(N),shakB(L), D-inx-2 or D-inx-7. Mutations eliminating shakB(N) are rescued by shakB(N) and D-inx-7, but not by ogre. Hence, there must be some functional specialization of different innexins. In addition, we have mapped regions of innexins that are required for shakB(N) and ogre specificity.

Specificity in innexin function could be the result of differing biophysical properties. Evidence from Xenopus oocyte experiments shows that innexins differ in voltage gating and conductance(Phelan et al., 1998;Landesman et al., 1999;Stebbings et al., 2000). Junctions composed of different innexins may also allow passage of different sets of second messengers. Studies on connexins show that they show distinct preferences for ions of different charge (Beblo and Veesntra, 1997;Wang and Veenstra, 1997) as well as different permeabilities for non-charged molecules and second messengers (Bevans et al.,1998; Cao et al.,1998). Connexins also show notable differences in biophysical properties (White et al.,1994; Verselis et al.,1994).

Innexins may also engage in unique intermolecular interactions, with other innexins or with other classes of proteins. Two innexins form heteromeric GJs(Stebbings et al., 2000). In addition, although some form homotypic GJs, others do not(Phelan et al., 1998;Landesman et al., 1999;Stebbings et al., 2000)(K.D.C., T. White, R.J.W. and D. Paul, unpublished). The Drosophilainnexins that form homotypic junctions in Xenopus occytes, shakB(L)and D-inx-2, are lethal when expressed pan-neurally in Drosophila(Stebbings et al., 2000)(K.D.C. and R.J.W., unpublished), although those that do not form homotypic junctions do not cause lethality (D-inx-7, shakB(N) and ogre) (K.D.C. and R.J.W., unpublished). Overexpression of ogre in the eye, but not other innexins, causes a rough-eye phenotype, indicating cell loss or death (K.D.C. and R.J.W., unpublished). One likely explanation for these differences is that each innexin can polymerize or mate only with particular other innexins to form GJs. Most connexins are capable of making homotypic junctions in the Xenopus oocyte assay (White and Bruzzone, 1996), whereas data from both the Xenopusoocyte assay and from the ectopic expression studies in the animal just described suggest that more than half of innexins do not form homotypic junctions.

Since neither ogre nor shakB(N) seems to make homotypic GJs, it is possible that they form GJs by mating with other innexins. Both show highly overlapping expression patterns with other innexins in animals(Curtin et al., 1999;Zhang et al., 1999), making their involvement in heteromeric or heterotypic GJs a reasonable possibility.

The molecular functions that distinguish different innexins may be mediated by distinct protein regions. We tested this by making chimeras between shakB(N) and ogre and testing these for rescue of the shakB² or ogre^cb8 ERG phenotypes.

In order to function in this assay, both ogre and shakB² need some gene-specific sequences from the second transmembrane domain or the putative intracellular loop. These sequences could play a role in polymerizing innexins to form a hemichannel. Both may normally form homomultimers so that replacing these sequences with residues from the other protein could prevent self-association. Alternatively, the sequences could specify polymerizing interactions with other innexins in the cell membrane to make a heteromeric channel. As a possible example of a heteromeric channel, we know that both shakB(N) and D-inx-2 are expressed in the lamina(Curtin et al., 1999;Zhang et al., 1999). We also know that shakB(N) expression in the lamina is required for normal retina-lamina synapses to form (Curtin et al., 2002;), and this may also be true for D-inx-2 (K.D.C. and R.J.W., unpublished). Thus shakB(N) could form heteromeric GJs with D-inx-2 in lamina neurons. A third possibility is that the sequences specify interactions with non-innexin proteins in the cell.

Intracellular or transmembrane sequences could also affect biophysical properties or channel permeability to second messengers. For example, chimeric studies suggest that the first transmembrane domain of Cx46 confers its unique channel kinetics and single channel conductance properties(Hu and Dahl, 1999). Intracellular sequences of Cx26 and Cx30 also influence biophysical properties(Manthey et al., 2001).

Some of the crucial residues within shakB(N) that are likely to be required for shakB(N) identity are marked with asterisks inFig. 3. Five of the six represent a change in charge. Four of these residues are found clustered just after the second transmembrane domain, with three of the four having a difference in charge. These differences in charge close to the transmembrane domain may be more consistent with a role for these sequences in biophysical or permeability differences between the proteins.

Interestingly, the total sequence difference between shakB(N) and ogre in the first 97 amino acids is much larger, even though these sequences are not crucial for shakB(N) `identity'. Therefore overall homology is not a good indicator of what portions of these proteins are functionally important for specificity. As a further example of this, the innexin proteins, as a family,differ most in the C-terminal tail, which plays no role in specificity in this assay.

For both shakB (N) and ogre, sequences in the first half of the protein,but not the second half, are essential for rescue. Possibly crucial residue differences between the two proteins in the intracellular loop have been identified (see above). In addition, some sequences N-terminal to the second transmembrane domain are also essential for ogre identity. We proposed that ogre-containing junctions in the retina mate with junctions in the lamina(Curtin et al., 2002). In that case, one might expect extracellular sequences in ogre to be crucial for specificity in mating with apposing hemi-junctions.

Other evidence suggests that mating specificity or innexin multimerization may be mediated by the first extracellular loop. ShakB(L) forms GJs when expressed alone in the paired Xenopus oocyte assay and is lethal when expressed panneurally (Stebbings et al.,2000; Phelan et al.,1998) (K.D.C. and R.J.W., unpublished). ShakB(N) does not form junctions when expressed alone in Xenopus oocytes and is not lethal when expressed pan-neurally (Phelan et al., 1998). Thus it seems that shakB(L) can form homomeric/homotypic GJs whereas shakB(N) cannot. Because shakB(N) and shakB(L)only differ from the beginning of the protein through the first putative extracellular loop (see Fig. 4)(Krishnan et al., 1995), some residues within this portion of protein probably mediate specificity of mating or polymerization.

Although shakB(N) and shakB(L) differ in the length of the N-terminal intracellular tail (see Fig. 4), this difference in length is not likely to contribute significantly to shakB(N) function. ShakB(N+16) is 16 amino acids longer at the very N-terminus than shakB(N), but otherwise identical and it is interchangeable with shakB(N) in the ERG rescue assay (data not shown). However, specific residues within the N-terminal tail could play a role in homo-multimerization of shakB(L) compared with shakB(N).

In the first extracellular loop, there are 20 residues that are dissimilar between shakB(N) and shakB(L). Eight of these are clustered together immediately prior to the second transmembrane domain and four of these eight lead to a change in charge (Fig. 4). A comparison of shakB(N) and ogre in this same stretch shows that 12 out of 15 residues preceding the transmembrane domain are different in the two proteins, and three of these represent a difference in charge(Fig. 3). It should be noted that shakB(N) and D-inx 7 also show divergence in this short stretch.

Neither ogre or shakB(N) proteins make homotypic GJs in the Xenopus oocyte assay (Phelan et al., 1998) (K.D.C., T. White, R.J.W. and D. Paul, unpublished). They also do not lead to lethality when expressed pan-neurally as do the known homotypic innexins, shakB(L) and D-inx-2(Stebbings et al., 2000;Curtin and Wyman, 2002). Lastly, they do not make heterotypic or heteromeric GJs with each other in oocytes (K.D.C., T. White, R.J.W. and D. Paul,unpublished). Hence ogre and shakB(N) probably polymerize with other innexins to form gap junctions. Since they are not interchangeable, the required gene-specific sequences may be involved in heteromeric or heterotypic interactions. The first extracellular loop may be crucial for homo/heteropolymer formation within the cell membrane, or it may help determine which innexins recognize and mate with each other on apposing cells. In chimera studies on connexins, it is the second extracellular loop that has been suggested to play an important role in mating specificity to regulate what connexins can form heterotypic GJs(White et al., 1994).

Although the C-terminal tail of the innexin proteins is not required for functional specificity in this assay, it is required for function. An ogre truncation lacking this tail fails to rescue mutants. When we overexpress shakB(N) in the nervous system with the elav-Gal4 driver it localizes to the lamina neuropil roughly where synapses form(Curtin et al., 2002). Synaptic localization often requires an interaction of the C-terminal tail of a channel protein with an anchoring protein, such as a PDZ protein or other scaffold protein (e.g. Kim et al.,1995). Thus truncated ogre may not be correctly localized. Since we lack good antibodies we were unable to address this hypothesis directly. The C-terminal tail could also affect channel biophysics. Truncations of the cytoplasmic tail of Cx43 alter junctional conductance(Fishman et al., 1991), and the C-terminal tail of the shaker channel is important for voltage-gated inactivation (Miller,1991).

The lack of interchangeability of these proteins is an important finding. Developmental GJs, possibly occurring between pre-and post-synaptic cells, are required for development of normal neural transmission in the Drosophila visual system (Curtin et al., 2002). Specifically, both shakB(N) and ogre are required during development, with ogre required presynaptically and shakB(N) required,at a minimum, postsynaptically (Curtin et al., 2002). Both mutants show a functional failure of retina to lamina neural transmission as determined by the ERG as well as some retinal axon projection errors (Curtin et al.,2002). One hypothesis to explain this is that developmental GJs between pre- and postsynaptic cells may be necessary for final target selection and/or the development of functional chemical synapses(Curtin et al., 2002). During development, while searching for their final synaptic targets, retinal cell axons contact several neural types in the optic lamina. They interact with each other, with their synaptic partners and with other neurons and glia in the optic lamina. The selective formation of GJs between neurons via selective interaction or activity of innexin proteins could thus have important effects on target selection and synaptic specificity.

We are greatly indebted to Yi-An Sun for technical expertise and wise advice. We also thank Haig Keshishian for assistance and support. This work was funded by NIH grant # GM57889 and by the Charles E. Culpeper Foundation.