Intercellular calcium signaling in a gap junction-coupled cell network establishes asymmetric neuronal fates in C. elegans

Many cell types communicate through gap junctions during embryonic development to regulate cell-fate determination, proliferation and morphogenesis (Levin, 2007). Gap junctions form intercellular channels that assemble from head-to-head docking of hemichannels, one provided by each of two contacting cells (Bennett and Zukin, 2004). Each hemichannel consists of subunits called connexins and pannexins in vertebrates, and innexins in invertebrates. Gap junction-mediated intercellular communication allows rapid and direct transfer of small molecules and ions across a field of cells without the need for specialized ligands and receptors. Gap junction-mediated calcium waves divide the mammalian neocortex into distinct neuronal domains (Yuste et al., 1995; Bennett and Zukin, 2004). Spontaneous gap junction-dependent calcium waves are also observed in the developing retina (Kandler and Katz, 1998). In addition, gap junction communication through movement of serotonin contributes to orienting the left-right body axis in both frog and avian models (Fukumoto et al., 2005a; Fukumoto et al., 2005b; Levin et al., 2006). Importantly, most studies that defined roles for gap junctions in development relied on broad loss-of-function or inhibition of individual subunits, or global inhibition of molecules that are predicted to pass through gap junctions. Therefore, the roles that individual cells within gap junction networks may play in vivo to drive cell-fate determination or other patterning events are almost completely unknown.

We showed previously that the innexin nsy-5 (inx-19 – WormBase) is required for asymmetric differentiation of the C. elegans AWC olfactory neuron pair (Chuang et al., 2007). The left and right AWC neurons are morphologically symmetric (White et al., 1986), but express different genes to convey the ability to sense different chemicals (Pierce-Shimomura et al., 2001; Wes and Bargmann, 2001). The two AWC neurons communicate through Notch-independent signaling during late embryogenesis to generate asymmetric cell fates: induced AWC^ON, which expresses the reporter gene str-2p::GFP, and default AWC^OFF, which does not (Troemel et al., 1999; Chuang and Bargmann, 2005). Each wild-type animal generates one AWC^ON and one AWC^OFF; 50% of the animals induce AWC^ON on the left whereas the other 50% induce AWC^ON on the right (Troemel et al., 1999). The default AWC^OFF fate is executed by the cell-autonomous activity of a calcium/calmodulin-dependent protein kinase II (CaMKII) and mitogen-activated protein kinase (MAPK) cascade (Sagasti et al., 2001; Chuang and Bargmann, 2005). NSY-5 gap junctions and NSY-4 claudins function in parallel to induce AWC^ON by inhibiting the CaMKII cascade (VanHoven et al., 2006; Chuang et al., 2007).

nsy-5 is expressed during embryogenesis in both AWC neurons and in at least 17 pairs of other neurons (Chuang et al., 2007) (supplementary material Fig. S1). nsy-5-dependent gap junctions transiently connect AWCs and adjacent neurons in embryos. nsy-5 functions primarily autonomously in AWCs to promote AWC^ON fate. However, nsy-5 activity in ASH, AWB and AFD neurons promotes or inhibits AWC^ON (Chuang et al., 2007). These results suggest that individual neighboring neurons in the NSY-5 network, which is defined by the 18 pairs of nsy-5-expressing neurons that are likely to be linked by gap junctions (Taylor et al., 2010), communicate with each other and with AWCs to fine-tune the signaling that establishes precise AWC asymmetry. However, the nature of intercellular signaling across the NSY-5 network and how individual non-AWCs in the network influence AWC asymmetry is not known.

Components of the voltage-gated calcium channel (the pore-forming α1 subunit UNC-2 and the regulatory α2δ subunit UNC-36) are also required upstream of CaMKII for communication between AWCs to ensure a precise 1AWC^ON/1AWC^OFF decision (Bauer Huang et al., 2007). This leads to the hypothesis that intercellular calcium signaling, through NSY-5, mediates communication between AWCs and non-AWCs, and that changes in calcium levels in neurons that are coupled with AWCs through NSY-5 influence AWC fate choice. Here, we develop a unique system to address the roles of specific cells within the NSY-5 network in AWC asymmetry by using cell type-specific expression of calcium-buffering proteins. We show that calcium in non-AWCs of the NSY-5 network is required for the induction of 1AWC^ON/1AWC^OFF asymmetry. In addition, calcium in AWB and ASH neurons confers different side biases of AWC^ON induction. Our results support a model in which calcium within the NSY-5 network has dual roles in establishing AWC asymmetry: autonomous promotion of AWC^OFF and non-autonomous promotion of AWC^ON.

Wild-type strains were C. elegans variety Bristol, strain N2. Strains were generated and maintained using standard methods (Brenner, 1974). Integrated transgenic lines and mutations used in these experiments included: kyIs140 [str-2p::GFP; lin-15(+)] I (Troemel et al., 1999), nsy-5(ky634) I (Chuang et al., 2007), lfe-2(sy326) I, kyIs323 [str-2p::GFP; ofm-1p::GFP] II, plc-3(tm1340) II, tph-1(mg280) II, unc-36(e251) III, itr-1(sy331) IV, itr-1(sy290) IV, itr-1(sy291) IV, itr-1(sy327) IV, itr-1(sy328) IV, nsy-4(ky627) IV (VanHoven et al., 2006), unc-43(n498gf) IV, eri-1(mg366) IV, unc-76(e911) V, unc-2(e55) X, unc-2(zf35gf) X, lin-15b(n744) X, ipp-5(sy605) X, kyIs136 [str-2p::GFP; lin-15(+)] X, vyIs18 [nsy-5p::mCherry-nsy-5 3′].

Transgenes maintained as extrachromosomal arrays included: vyEx498, 499 [odr-3p::calbindin D28K; odr-1p::DsRed; ofm-1p::DsRed], vyEx699, 700 [odr-3p::parvalbumin; odr-1p::DsRed; ofm-1p::DsRed], vyEx1066, 1068, 1069 [odr-3p::calbindin D9K; odr-1p::DsRed; ofm-1p::DsRed], vyEx567, 571 [nsy-5p::calbindin D28K; odr-1p::DsRed; ofm-1p::DsRed], vyEx724, 725, 726 [nsy-5p::parvalbumin; odr-1p::DsRed; ofm-1p::DsRed], vyEx1063 [odr-3p::cmd-1; odr-1p::DsRed; ofm-1p::DsRed], vyEx484 [nsy-5p::calbindin D28K::mCherry; ofm-1p::DsRed], vyEx992, 993, 994, 997 [odr-3p::rParv; odr-1p::DsRed; ofm-1p::DsRed], vyEx990, 991, 996 [odr-3p::rParvCDEF/AV; odr-1p::DsRed; ofm-1p::DsRed], vyEx521, 522 [str-1p::calbindin D28K; odr-1p::DsRed; ofm-1p::DsRed], vyEx562, 564 [sra-6p::calbindin D28K::SL2::mCherry; ofm-1p::DsRed], vyEx568, 569, 570 [str-1p::nsy-5IR::SL2::mCherry; ofm-1p::DsRed], vyEx1320, 1321 [str-1p::mCherry; ofm-1p::DsRed], vyEx1322, 1323 [sra-6p::SL2::mCherry; ofm-1p::DsRed].

M. musculus calbindin D28K cDNA from pCMV6-calbindin-28K (Origene) was subcloned to make nsy-5p::calbindin D28K, odr-3p::calbindin D28K, nsy-5p:calbindin D28K::mCherry, str-1p::calbindin D28K and sra-6p::calbindin D28K::SL2::mCherry. M. musculus parvalbumin cDNA from pCMV6-parvalbumin (Origene) was subcloned to make odr-3p::parvalbumin and nsy-5p::parvalbumin. Calbindin D9K cDNA was synthesized from mixed embryonic day (E) 14.5 and postnatal day (P) 0 mouse brain total RNA and subcloned to make odr-3p::calbindin D9K. cmd-1 cDNA was obtained by RT-PCR from L1 stage worms and subcloned to make odr-3p::cmd-1. odr-3p::rParvCDEF/AV was constructed using the Quikchange II XL Kit (Stratagene). The original construct, rat ParvCDEF, contains four substitutions that replace the first (D51A, D90A) and last (E62V, E101V) charged amino acids of the two EF-hand domains responsible for calcium binding (Pauls et al., 1994; John et al., 2001). We mutated the remaining charged amino acids in the second EF hand domain (K92V, D93A, D95A, K97V and E100V). To make nsy-5IR, antisense and sense transcripts of nsy-5 were inserted upstream and downstream, respectively, of an 888-bp linker sequence as previously described (Chuang and Meyerowitz, 2000). This construct was subcloned to make str-1p::nsy-5IR::SL2::mCherry.

odr-3p::calbindin D28K (100 ng/μl), odr-3p::parvalbumin (100 ng/μl), odr-3p::calbindin D9K (100 ng/μl), nsy-5p::calbindin D28K (100 ng/μl), nsy-5p::parvalbumin (50 ng/μl), odr-3p::cmd-1 (50 ng/μl), nsy-5p::calbindin D28K::mCherry (100 ng/μl), odr-3p::rParv (40 ng/μl), odr-3p::rParvCDEF/AV (40 ng/μl), str-1p::calbindin D28K (100 ng/μl), sra-6p::calbindin D28K::SL2::mCherry (100 ng/μl), str-1p::nsy-5IR::SL2::mCherry (100 ng/μl), str-1p::mCherry (100 ng/μl), sra-6p::SL2::mCherry (100 ng/μl), odr-1p::DsRed (15 ng/μl) and ofm-1p::DsRed (30 ng/μl) were injected into animals as previously described (Mello and Fire, 1995).

Chemotaxis assays were performed as previously described (Bargmann et al., 1993). Odors were diluted in ethanol and tested at standard concentrations (1:1000 for butanone and 1:10,000 for 2,3-pentanedione). About 50-200 animals were assayed for each strain and each odor in each individual assay. All assays were performed four independent times. Relative chemotaxis index was calculated by dividing each absolute chemotaxis index by the highest wild-type chemotaxis index observed for each odor for each independent trial.

Cells were isolated from C. elegans embryos as previously described (Christensen et al., 2002) and resuspended to a final concentration of 3-7 × 10⁶ cells/ml. For dye transfer assays, cells were cultured in rotating eppendorf tubes for 24-48 hours at 22°C prior to assays. For calcium imaging, cells were seeded on 25-mm round cover glasses coated with 0.5 mg/ml peanut lectin (Sigma) and cultures were maintained for 21-72 hours at 22°C in a humidified chamber prior to imaging.

C. elegans embryonic cells were washed twice with HBS (Gibco) containing 10 mM HEPES (pH 7.35) and 5.5 mM glucose, then loaded with 3 μM NPE-HCCC2/AM (plus 0.015% puronic acid in 0.015% DMSO) for 45 minutes in the dark at 22°C. Loaded cells were washed with HBS and seeded on 15-mm slides that had been precoated with peanut lectin (Sigma), in 35-mm dishes (MatTek). Seeded cells were incubated in HBS for 15 minutes to allow complete hydrolysis of acetoxymethyl (AM) esters. Localized ultraviolet (UV) uncaging of NPE-HCCC2 and fluorescence imaging were performed on an inverted Zeiss Axiovert 200 with a Hamamatsu ORCA-ER cooled CCD camera and 100× oil immersion objective (NA 1.4) as previously described (Dakin et al., 2005). Images were acquired and analyzed with the Openlab integrated imaging software (Improvision). All of the five wild-type cell pairs that were analyzed showed dye transfer, whereas none of the five nsy-5(lf) cell pairs assayed had dye transfer.

C. elegans embryonic cells were loaded with Fluo-4 calcium-indicator dye as previously described (Marsh, 1995). Briefly, cells were washed three times with physiological saline solution (PSS; 145 mM NaCl, 5 mM KCl, 1 mM MgSO₄, 10 mM HEPES, pH 7.4, 10 mM glucose, plus 2 mM CaCl₂), then loaded with 1.5 μM Fluo-4, AM (plus 0.015% puronic acid in 0.015% DMSO) for 45 minutes in the dark at 22°C. Cells were washed three times with PSS (2 mM CaCl₂), then recovered in the final wash for 30-60 minutes in the dark. Ionomycin (in PSS plus 10 mM CaCl₂) was added on top of cells to a final concentration of 10 μM. EGTA in PSS was added on top of cells to a final concentration of 6 mM. Optical recording was performed using an Olympus IXS1 inverted microscope equipped with a Hamamatsu ORCA-AG camera. Frames were acquired at 2-second intervals for a total of 4 minutes. Images were processed using ImageJ software. At each time point, Fluo-4 fluorescence was normalized to mCherry fluorescence.

Mosaic analysis and statistical analysis were performed as previously described (Sagasti et al., 2001; VanHoven et al., 2006; Chuang et al., 2007). For experiments using the str-1 and sra-6 promoters, bicistronic constructs were generated using the SL2 trans-splicing signal to drive both the gene of interest and mCherry, and mCherry expression determined retention of the transgenes. Previous studies showed that the nsy-5 promoter, odr-3 promoter, str-1 promoter, sra-6 promoter, odr-1p::DsRed and various co-injection markers had no effect on str-2::GFP expression (Sagasti et al., 2001; Chuang and Bargmann, 2005; VanHoven et al., 2006; Bauer Huang et al., 2007; Chuang et al., 2007). At least two independent lines of the same transgene were analyzed. Expected numbers of mosaic animals were predicted from non-mosaic animals of the same transgenic line.

Consistent with its cell-intrinsic and network functions in AWC asymmetry, NSY-5 can form both functional hemichannels and intercellular gap junction channels in Xenopus oocytes (Chuang et al., 2007). However, the permeability properties of NSY-5 gap junctions in C. elegans have not been examined. To determine the permeability of NSY-5 gap junctions to small molecules in C. elegans embryonic cells, we used a photoactivatable (caged) fluorescent dye and the uncaging/imaging technique (Dakin et al., 2005; Dakin and Li, 2006). Embryonic neurons were isolated from nsy-5p::mCherry transgenic animals, in which neurons of the NSY-5 network were labeled with mCherry (Shaner et al., 2004; Shaner et al., 2005). Cultured neurons were loaded with the cell-permeable caged fluorophore NPE-HCCC2/AM (Zhao et al., 2004; Dakin et al., 2005). After entering the cell, the acetoxymethyl (AM) esters are hydrolyzed and the caged dye (NPE-HCCC2) becomes trapped inside the cell. When uncaged by UV light, the parent fluorophore HCCC2 is released. As the molecular weight of HCCC2 is 450 Da, well below the ~1.2 kDa molecular passage limit of typical gap junction channels (Kumar and Gilula, 1996), HCCC2 diffuses to its neighboring cells through gap junctions (Fig. 1A). We focused on pairs of nsy-5p::mCherry-expressing cells that made direct contact (Fig. 1B,C,I,J), and uncaged the fluorophore in one cell. In wild-type cell pairs, HCCC2 transferred rapidly from the uncaged cell (donor) to its neighboring cell (recipient) (Fig. 1E-H). By contrast, in cell pairs isolated from nsy-5 loss-of-function (lf) embryos, in which nsy-5-dependent gap junctions were absent (Chuang et al., 2007), fluorescent HCCC2 was only detected in the uncaged cell (cell 1) and no fluorescence was observed in its neighboring cell (cell 2) (Fig. 1L-O). These results indicate that NSY-5 gap junctions can mediate small molecule transfer between coupled C. elegans neurons.

To determine the requirements for intracellular free calcium in specific cells within the NSY-5 network, we established transgenic lines expressing the vertebrate calcium-buffer proteins calbindin D28K, parvalbumin or calbindin D9K in various subsets of neurons. Calcium-buffer proteins contain EF-hand domains (four functional domains in calbindin D28K and two each in parvalbumin and calbindin D9K) responsible for binding to and sequestering intracellular free calcium, preventing calcium from activating downstream targets such as CaMKII (Ikura, 1996). There are no homologs of these calcium buffer genes in the C. elegans genome (C. elegans Sequencing Consortium, 1998).

Expression of each buffer under the control of the odr-3 promoter (Roayaie et al., 1998), which is strongly expressed in AWC and weakly in AWB neurons, led to str-2::GFP expression in both AWC neurons (2AWC^ON phenotype) (Fig. 2D,E; supplementary material Table S1), in contrast to the 1AWC^OFF/1AWC^ON phenotype observed in the wild-type population (Fig. 2A,E). The 2AWC^ON phenotype is identical to the loss-of-function mutant phenotype of the calcium channel α2δ subunit gene unc-36 (Troemel et al., 1999) (Fig. 2C,E; supplementary material Table S1), consistent with each buffer strongly inhibiting calcium signaling. Mutations of conserved charged residues at the beginning and end of the two EF-hand calcium-binding domains of parvalbumin reduce calcium binding in vitro (Pauls et al., 1994). To determine whether parvalbumin generates a 2AWC^ON phenotype owing to its ability to bind calcium, we mutated all charged resides in the second EF-hand domain. This resulted in a significant reduction of 2AWC^ON phenotype (P=0.006, n=245), suggesting that the 2AWC^ON phenotypes are specifically due to calcium-binding activity.

nsy-5 mutants have two AWC^OFF cells (2AWC^OFF phenotype) (Chuang et al., 2007) (Fig. 2B,E), consistent with the requirement of gap junction-mediated signaling for AWC^ON induction. Calbindin D28K or parvalbumin expression in the entire NSY-5 network resulted in 2AWC^ON, but also significantly increased the percentage of 2AWC^OFF compared with expression primarily in AWCs (Fig. 2E; supplementary material Table S1), suggesting that intracellular calcium signals in non-AWCs of the NSY-5 network promotes the wild-type 1AWC^ON/1AWC^OFF phenotype.

odr-3p::calbindin D28K animals chemotaxed normally to butanone, which is sensed by the AWC^ON cell, but did not chemotax to 2,3-pentanedione, which is sensed by the AWC^OFF cell (Fig. 2F). These results confirm that the AWC^OFF cell fate is lost in odr-3::calbindin D28K animals, and that the AWC^ON cells are capable of performing an appropriate physiological function. Because calbindin D28K has the highest number (four) of potential calcium-binding sites among the calcium-buffer proteins, we focused primarily on developing this buffer as a tool to examine the role of calcium in specific cells in the NSY-5 network.

Calcium-buffer proteins share the EF-hand calcium-binding domain with calcium sensors, which bind to calcium and promote signaling by activating downstream kinase cascades. Although it is well established that calbindin D28K and parvalbumin function as calcium buffers in vertebrate systems and transgenic fly neurons (Chard et al., 1993; Grabarek, 2006; Harrisingh et al., 2007), their calcium-buffering ability has not been tested in any type of C. elegans cells. unc-43/CaMKII gain-of-function (gf) mutants have a 2AWC^OFF phenotype (Troemel et al., 1999) (Fig. 3A; supplementary material Table S1). To confirm that calbindin D28K does not promote calcium signaling, we compared the abilities of calbindin D28K and C. elegans CMD-1/calmodulin, a calcium sensor that activates CaMKII, to induce a 2AWC^OFF phenotype. Neither protein when expressed alone in AWCs under the odr-3 promoter strongly induced a 2AWC^OFF phenotype (Fig. 3A; supplementary material Table S1). Calcium concentration could be one of the limiting factors in allowing cmd-1 to activate downstream targets. unc-2(zf35) is a gain-of-function allele of the neuronal P/Q type calcium channel (J.K.P. and M.J.A., unpublished), which should lead to a rise in intracellular calcium. Although unc-2(zf35gf) mutants had wild-type AWC asymmetry (Fig. 3A; supplementary material Table S1), they may provide a sensitized background for a 2AWC^OFF phenotype. When the same odr-3p::calbindin D28K or odr-3p::cmd-1 transgenes were expressed in unc-2(zf35gf) homozygous mutants, odr-3p::cmd-1 generated a strong 2AWC^OFF phenotype (Fig. 3A; supplementary material Table S1), indicative of promoting calcium signaling. Importantly, odr-3p::calbindin D28K did not promote 2AWC^OFF in the unc-2(zf35gf) background (Fig. 3A), indicating that calbindin D28K does not activate downstream targets and supporting the role of calbindin D28K as a calcium buffer rather than a calcium sensor.

To confirm that calbindin D28K reduces the amount of intracellular free calcium, we performed calcium-imaging experiments in C. elegans primary cell culture. We targeted neurons in the NSY-5 network by expressing mCherry or calbindin D28K::mCherry translational fusion protein under the control of the nsy-5 promoter. The nsy-5p::calbindin D28K::mCherry transgene allowed us to determine relative levels of calbindin D28K expression in each cell, which we broadly categorized into ‘dim’ cells, which had lower levels of calbindin D28K, and ‘bright’ cells, which had higher levels of calbindin D28K. Wild-type mCherry-expressing neurons displayed a typical increase in intracellular calcium measured by the calcium indicator Fluo-4 following ionomycin (a calcium ionophore) application (Fig. 3B,C). ‘Dim’ calbindin D28K::mCherry neurons had a similar increase to wild-type neurons (Fig. 3B,C). However, the increase in intracellular calcium was significantly reduced in ‘bright’ neurons, which expressed high levels of calbindin D28K::mCherry, indicating that calbindin D28K is able to bind and sequester intracellular calcium. Taken together, these results are consistent with calbindin D28K acting as a calcium buffer in the NSY-5 network to disrupt AWC asymmetry.

The 2AWC^ON phenotype caused by calcium buffer expression in AWCs resembles that seen in loss-of-function mutations of unc-2 and unc-36 calcium channels and of the downstream calcium-activated unc-43 (CaMKII)/tir-1 (Sarm1 adaptor)/nsy-1 (MAPKKK) kinase pathway (Troemel et al., 1999; Sagasti et al., 2001; Chuang and Bargmann, 2005; Chang et al., 2011) (Fig. 4A,B). This similarity suggests that the calcium buffer-induced 2AWC^ON phenotype could result from an inhibition of the unc-2/unc-36 calcium signaling pathway.

To determine the relationship between the calcium-buffer target and other genes affecting AWC asymmetry, we crossed odr-3p::calbindin D28K or odr-3p::parvalbumin transgenic animals (2AWC^ON) with 2AWC^OFF mutants. Loss-of-function mutations in the unc-76 axon outgrowth gene, the nsy-5 innexin gene and the nsy-4 claudin-like gene, or gain-of-function mutations in the unc-43 CaMKII gene cause a 2AWC^OFF phenotype (Troemel et al., 1999; VanHoven et al., 2006; Chuang et al., 2007) (Fig. 4A,B). Calbindin D28K or parvalbumin expression strongly suppressed the 2AWC^OFF phenotype of unc-76(lf), nsy-5(lf) or nsy-4(lf) mutants (Fig. 4A), indicating that calcium buffers function downstream of axon outgrowth, nsy-5 and nsy-4 (Fig. 4B). By contrast, the 2AWC^ON phenotype resulting from calbindin D28K or parvalbumin expression was greatly suppressed by unc-43 gain-of-function mutants (Fig. 4A), suggesting that calcium buffers act on a target predominantly upstream of unc-43 (Fig. 4B).

These results suggest that calcium buffers affect the AWC signaling step downstream of nsy-5 and nsy-4, and upstream of unc-43. Previous genetic studies suggest that unc-2 and unc-36 calcium channels also act downstream of nsy-5 and nsy-4, and upstream of unc-43 (Troemel et al., 1999; VanHoven et al., 2006; Chuang et al., 2007). Thus, we examined the genetic interaction between the calcium-buffer target and unc-2. The unc-2(e55) mutant had a mixed 2AWC^ON and 2AWC^OFF phenotype due to functional redundancy of unc-2 with the additional α1 subunit egl-19 and inhibition of egl-19 activity by unc-2, respectively (Troemel et al., 1999; Bauer Huang et al., 2007) (Fig. 4A). Calbindin D28K expression significantly suppressed the 2AWC^OFF phenotype of unc-2(e55) mutants (Fig. 4A). Together, these results are consistent with the antagonistic effect of calcium buffers on the unc-2/unc-36 calcium signaling pathway (Fig. 4B).

Based on previous results that calcium-signaling components (unc-43, tir-1, nsy-1) function cell-autonomously to execute the AWC^OFF fate (Sagasti et al., 2001; Chuang and Bargmann, 2005), we hypothesized that the calcium buffers would cell-autonomously inhibit the calcium-signaling pathway to promote AWC^ON. To test this hypothesis, we performed genetic mosaic analysis using calbindin D28K transgenes expressed primarily in AWCs and AWBs using the odr-3 promoter. Transgenes are inherited as extrachromosomal arrays that are unstable and subject to spontaneous mitotic loss in random cell lineages. Mosaic animals were identified by loss of the co-injection marker odr-1p::DsRed, which is expressed in AWC left (AWCL), AWC right (AWCR), AWB left (AWBL) and AWB right (AWBR). These four cells represent four lineage groups within the NSY-5 network (Chuang et al., 2007) (supplementary material Fig. S1), and thus can serve as indicators of which cells within the network retain or lose calbindin D28K transgenes. Seventy percent of non-mosaic calbindin D28K animals had 2AWC^ON (Fig. 5B; Table 1, row d). Non-mosaic 1AWC^ON animals showed no side bias of AWC^ON induction. In >90% of mosaic animals that retained the calcium buffer only in one AWC, the calcium-buffered AWC became AWC^ON and the wild-type AWC became AWC^OFF, independent of expression of the buffer in AWB cells (Fig. 5C; Table 1, rows e,f). This suggests that intracellular calcium is required cell-autonomously to execute AWC^OFF cell-fate specification (Fig. 6A).

Based on our observation that expressing the buffers in the entire NSY-5 network led to a moderate 2AWC^OFF phenotype (Fig. 2E), we hypothesized that calcium within non-AWC cells is required to promote 1AWC^ON/1AWC^OFF. To test this hypothesis, we performed mosaic analysis with the nsy-5p::calbindin D28K; odr-1p::DsRed transgenic extrachromosomal arrays to determine whether buffering calcium only in non-AWC neurons within the NSY-5 network was able to disrupt asymmetry. As 15 out of the 34 non-AWC cells expressing nsy-5 are closely related to either AWB by lineage (Chuang et al., 2007) (supplementary material Fig. S1), their genotypes can be inferred by following the odr-1p::DsRed expression in AWB. If calcium signaling in non-AWC cells is required for AWC^ON induction, buffering of calcium signals in the AWB lineage would disrupt intercellular calcium signaling within the NSY-5 network and lead to a defect in the induction of AWC^ON fate. Thus, we examined mosaic animals that had odr-1p::DsRed expression in both AWB cells but not in either AWC, which would have buffered calcium levels in ~50% of non-AWCs in the NSY-5 network. Interestingly, 79% of this type of mosaic animals had 2AWC^OFF (Fig. 5D; Table 1, row c). In addition, we examined odr-3p::calbindin D28K mosaic animals that retained the array only in both AWC cells, and found that only 38% of animals had 2AWC^ON (Table 1, row g), compared with 70% when calcium was buffered in both AWCs and AWBs (Table 1, row d). These results support the hypothesis that intracellular calcium in non-AWC cells of the NSY-5 network is required to promote 1AWC^ON/1AWC^OFF (Fig. 6B; supplementary material Fig. S2A).

To refine the requirements of calcium in the AWB lineage further, we examined the role of calcium in AWBs. We examined mosaic animals that lost the odr-3p::calbindin D28K array only in AWBL or AWBR, and determined whether AWCL or AWCR becomes AWC^ON. Of the 1AWC^ON mosaic animals that lost the odr-3p::calbindin D28K array in AWBR, 74-77% had a significant bias towards the opposite side AWCL becoming AWC^ON, regardless of the presence of the transgene in both AWCs (Table 1, rows h,j). Because AWB inhibits AWC^ON fate via NSY-5 (Chuang et al., 2007), these results are consistent with the hypothesis that normal calcium levels in AWBR inhibits AWCR from becoming AWC^ON. In the nsy-5(lf) mutant, mosaic animals that lost the array in AWBR no longer had a bias towards AWCL^ON (Table 1, rows j,m), suggesting that calcium in AWBR requires gap-junction communication within the NSY-5 network to inhibit AWCR^ON. In addition, this result supports the notion that calcium travels through NSY-5 gap junctions within the network.

To probe the role for AWB further, we used the str-1 promoter (Troemel et al., 1997) to drive expression of calbindin D28K only in AWB. Expression of str-1p::calbindin D28K in both AWBL and AWBR did not disrupt AWC asymmetry (Table 1, row n; supplementary material Table S1), whereas the similar class of animals containing the nsy-5p::calbindin D28K array did disrupt AWC asymmetry (Table 1, row c). These results suggest that other cells in the AWB lineage of the NSY-5 network also contribute to promoting the wild-type 1AWC^ON/1AWC^OFF fate. Mosaic animals that lost str-1p::calbindin D28K in AWBR had a significant bias towards AWCL^ON (Table 1, row o), confirming that the negative influence on AWCR^ON via calcium is partly, if not completely, derived from AWBR (supplementary material Fig. S2B). Control mosaic experiments showed that str-1p::mCherry did not influence AWC^ON side bias (supplementary material Table S2). To test whether nsy-5, like calcium, in AWBR generates AWCL^ON side bias, we made a tissue-specific inverted repeat (IR) RNAi construct, str-1p::nsy-5IR. Knocking down NSY-5 in both AWBL and AWBR did not disrupt AWC asymmetry or generate an AWC^ON side bias (Table 1, row q). Consistent with our str-1p::calbindin D28K results (Table 1, row o), 62% of 1AWC^ON animals that lost str-1p::nsy-5IR in AWBR had a significant bias towards AWCL^ON (Table 1, row r). These results suggest that the negative influence of AWBR on AWCR^ON is dependent on both calcium and nsy-5. In addition, 70% of 1AWC^ON animals that lost str-1p::nsy-5IR in AWBL had a significant bias towards AWCR^ON (Table 1, row s). As the same class of mosaic animals containing the str-1p::calbindin D28K did not show a bias towards AWCR^ON (Table 1, row p), these results suggest that other signals from AWBL might go through NSY-5 gap junctions to negatively influence AWCL^ON.

ASH neurons are derived from the same lineage as AWCs. To examine whether calcium in ASH generates AWC^ON side bias, we used the sra-6 promoter (Troemel et al., 1995) to drive calbindin D28K expression primarily in ASH, in addition to ASI neurons. Expression of sra-6p::calbindin D28K in both ASH and ASI did not disrupt AWC asymmetry nor did it generate AWC^ON side bias (Table 1, row t; supplementary material Table S1). Mosaic animals that lost the array in ASHL had a significant bias towards AWCL^ON (Table 1, row v). Control mosaic experiments showed that sra-6p::SL2::mCherry did not influence AWC^ON side bias (supplementary material Table S2). ASH was previously shown to promote AWC^ON fate via NSY-5 (Chuang et al., 2007), suggesting that calcium in ASHL promotes AWCL^ON fate (supplementary material Fig. S2C). In addition, mosaic animals that lost the sra-6p::calbindin array in ASIR had a significant bias towards AWCR^ON (Table 1, row w). It has not been determined whether ASI has a positive or negative influence on AWC^ON fate. ASIR is derived from the same lineage as AWBR and AFDR (supplementary material Fig. S1), which both inhibit AWC^ON (Chuang et al., 2007), so it is possible that calcium in ASIR inhibits AWCL^ON fate.

Intercellular calcium waves can be propagated by movement of inositol trisphosphate (IP3) through gap junctions (Sáez et al., 1989). In addition, serotonin may move through gap junctions to coordinate the left-right body axis and modulate intercellular calcium signaling, and ADF neurons of the NSY-5 network synthesize serotonin (Blomstrand et al., 1999; Duerr et al., 1999; Zimmermann and Walz, 1999; Sze et al., 2000; Fukumoto et al., 2005a; Fukumoto et al., 2005b; Peters et al., 2005; Levin et al., 2006). To address potential roles of IP3 and serotonin in mediating AWC asymmetry, we examined AWC^ON induction in mutants that disrupt components of each signaling pathway. We analyzed mutants of the following genes required for IP3 signaling: itr-1 (a single IP3 receptor), lfe-2 (IP3 kinase), ipp-5 (5-phosphatase) and plc-3 (phospholipase C) (Clandinin et al., 1998; Dal Santo et al., 1999; Sze et al., 2000; Bui and Sternberg, 2002). IP3 signaling mutants did not affect AWC asymmetry (supplementary material Table S3). C. elegans has a single tryptophan hydroxylase (the key enzyme for serotonin biosynthesis) gene, tph-1, and tph-1(mg280lf) mutants do not synthesize serotonin (Sze et al., 2000). tph-1(mg280lf) mutants had wild-type AWC asymmetry (supplementary material Table S3). Together, these results suggest that IP3 and serotonin are not involved in nsy-5-mediated intercellular signaling for AWC asymmetry.

Discovering the nature of intercellular signaling is an important step towards understanding how groups of cells coupled by gap junctions coordinately communicate to diversify cell fates. We have taken advantage of well-characterized vertebrate calcium-buffer proteins and powerful C. elegans genetic tools to manipulate intracellular calcium levels selectively in individual cells or small groups of cells within the NSY-5 gap junction network to determine the non-cell-autonomous role of calcium in AWC fate diversification across the left-right axis. The genetic mosaic results revealed that calcium in non-AWC neurons of the NSY-5 network promotes wild-type 1AWC^ON/1AWC^OFF cell fate decision. In addition, calcium in AWB, ASH and ASI cells influences AWC^ON side biases. To our knowledge, this is the first in vivo study that directly reveals novel roles for intercellular calcium signaling within a gap junction network and dissects roles for specific cells in the network in stochastic cell-fate choice. In addition, the calcium-buffer transgenic lines represent a unique system in which to rigorously test the actual role of intracellular calcium in various neurons in directing cell-fate choice, rather than simply correlating calcium dynamics with cell-fate outcomes.

Based on the mosaic results, we propose a model in which calcium regulates AWC asymmetry by providing both cell-autonomous inputs from AWCs and non-cell-autonomous inputs from other neurons in the NSY-5 network (Fig. 6). Prior to their interaction, the two AWC neurons have spontaneous activity in order to maintain voltage-gated calcium channel-mediated signaling and the default AWC^OFF fate (Troemel et al., 1999; Sagasti et al., 2001; Tanaka-Hino et al., 2002; Chuang and Bargmann, 2005; VanHoven et al., 2006; Bauer Huang et al., 2007; Chuang et al., 2007) (Fig. 6A). Once the AWC cells sense each other's presence through intercellular calcium signaling between AWCs and other neurons within the NSY-5 network, calcium channels in the two AWC cells may be differentially regulated by an initial stochastic signaling event that breaks AWC symmetry. Voltage-gated calcium channels generate both calcium and voltage signals, and are subject to both calcium- and voltage-dependent activation and inactivation (Catterall, 2000). As calcium-activated signaling pathways are very sensitive to the temporal pattern of calcium signals (Thomas et al., 1996; West et al., 2001), transient differences in calcium influx between two cells can potentially generate sustained differences in calcium-regulated signaling outputs through positive- and negative-feedback mechanisms. Our results support the model that NSY-5 gap junctions transmit or modify calcium signals across the neural network, leading to differential regulation of calcium channel activity in the two AWC cells and the subsequent asymmetric differentiation of AWC cells (Fig. 6B). This extends the previous model of NSY-5 function by identifying calcium as a major mediator of intercellular communication in the NSY-5 network.

Gap junction-dependent intercellular calcium signaling has been observed in a number of different cell types and organ systems, including the neocortex, retina, airway epithelia, hepatocytes and osteoblastic cells (Sáez et al., 1989; Boitano et al., 1992; Yuste et al., 1992; Lee et al., 1994; Yuste et al., 1995; Kandler and Katz, 1998; Singer et al., 2001). One mechanism for transduction of these calcium waves is the spread of the second messenger IP3 through gap junctions, which then triggers the release of intracellular calcium stores in neighboring cells. Our results suggest that AWC asymmetry does not rely on IP3 signaling, implying an alternative mechanism for propagation of calcium signaling in the NSY-5 network. One possibility is that calcium itself travels through gap junctions to mediate intercellular communication. This type of communication is found in coupled hepatocytes, in which calcium diffusion through gap junctions may help to equalize hormonal responses across a field of cells (Sáez et al., 1989). In addition, membrane potential is a possible signal that mediates NSY-5 gap junction intercellular communication. In support of this notion, SLO-1 and EGL-2, which represent two different classes of voltage-regulated potassium channels, also affect AWC asymmetry (Troemel et al., 1999; Sagasti et al., 2001; Davies et al., 2003).

It is clear from these results that AWC neurons integrate signals from a number of different cells both to induce AWC^ON and to influence AWC^ON side bias. In particular, the mosaic analysis in nsy-5(lf) mutants revealed that the calcium input from AWBR requires the presence of NSY-5 gap junctions to inhibit AWCR^ON, suggesting that there is a flow of information from AWBR to AWCR through gap junctions. AWBs do not form direct gap junctions with AWCs (Chuang et al., 2007), thus the influence from AWBs on AWCs might go through other cells in the NSY-5 network. However, it is possible that AWCR or AWCL receive inputs from additional neurons that were not included in this analysis. AWCR can respond to nsy-5 directly but AWCL requires nsy-5 function in multiple cells of the network (Chuang et al., 2007), suggesting that AWCL requires more inputs from neighboring cells to become AWC^ON. The result that AWCL^ON is promoted by inputs from AWBR and ASHL, whereas AWCR^ON has only the cell-autonomous input, is consistent with this.

Side-biased calcium-wave dynamics have been implicated in establishing the left-right axis in vertebrate systems. Intracellular calcium elevation on the left side of the mouse embryonic node, Hensen's node in chick, or zebrafish Kupffer's vesicle is the earliest asymmetric molecular event that is functionally linked to organ laterality (McGrath et al., 2003; Sarmah et al., 2005; Garic-Stankovic et al., 2008). Calcium may move through gap junctions to reach left side target cells, leading to activation of downstream transcription (Levin and Mercola, 1999; Hatler et al., 2009). In zebrafish, left-sided elevation of intracellular calcium transiently activates CaMKII phosphorylation, which is required for proper left-right organ placement (Francescatto et al., 2010). A similar situation might occur in AWC asymmetry. It is possible that side biases of calcium dynamics in non-AWC cells of the NSY-5 network cause elevation of calcium in one AWC and reduction of calcium in another AWC, leading to increased and decreased CaMKII activity in the future AWC^OFF cell and the future AWC^ON, respectively.

Stochastic cell-fate acquisition in the nervous system is a conserved but only partly understood phenomenon in all species (Johnston and Desplan, 2008; Losick and Desplan, 2008; Johnston and Desplan, 2010). This study suggests a novel mechanism by which stochastic neuronal diversification is established. We propose that stochastic AWC asymmetry is coordinated through competition of different AWC^ON side biases generated by intracellular calcium signals in non-AWCs of the NSY-5 network, in addition to previously identified intrinsic opposite biases of nsy-4 and nsy-5 activity in AWCs (Chuang et al., 2007).

We thank Cori Bargmann for valuable discussions and comments on the manuscript. We also thank David Richards for advice on Fluo-4 calcium imaging; Yan Zou for worm cDNA; Yutaka Yoshida for mouse cDNA; Andy Fire for C. elegans vectors; James Lechleiter (UT Health Science Center, San Antonio, TX) for Parv and ParvCDEF plasmids; the C. elegans Genetic Center for strains; and WormBase.