Synaptic circuits in the retina transform visual input gathered by photoreceptors into messages that retinal ganglion cells (RGCs) send to the brain. Processes of retinal interneurons (amacrine and bipolar cells) form synapses on dendrites of RGCs in the inner plexiform layer (IPL). The IPL is divided into at least 10 parallel sublaminae; subsets of interneurons and RGCs arborize and form synapses in just one or a few of them1,2,3. These lamina-specific circuits determine the visual features to which RGC subtypes respond3,4,5. Here we show that four closely related immunoglobulin superfamily (IgSF) adhesion molecules—Dscam (Down’s syndrome cell adhesion molecule), DscamL (refs 6–9), Sidekick-1 and Sidekick-2 (ref. 10)—are expressed in chick by non-overlapping subsets of interneurons and RGCs that form synapses in distinct IPL sublaminae. Moreover, each protein is concentrated within the appropriate sublaminae and each mediates homophilic adhesion. Loss- and gain-of-function studies in vivo indicate that these IgSF members participate in determining the IPL sublaminae in which synaptic partners arborize and connect. Thus, vertebrate Dscams, like Drosophila Dscams11,12,13,14,15,16,17,18,19, play roles in neural connectivity. Together, our results on Dscams and Sidekicks suggest the existence of an IgSF code for laminar specificity in retina and, by implication, in other parts of the central nervous system.
We previously showed that Sidekick-1 and Sidekick-2, large transmembrane adhesion molecules of the IgSF, are expressed by complementary subsets of amacrine, bipolar and ganglion cells in chick retina10. Double-label in situ hybridization revealed that only 10–15% of RGCs were positive for each Sidekick (Supplementary Table 1). We speculated that Sidekick-negative retinal neurons might express genes encoding related molecules. Homology searches of genomic databases revealed that their closest relatives are the Dscams6,7,8,9, which, like Sidekicks, contain multiple immunoglobulin and fibronectin type III domains, plus a carboxy terminus predicted to bind PDZ domains (Fig. 1a and Supplementary Fig. 1). Moreover, vertebrate Dscams are related to Drosophila Dscams, which have been implicated in neural specificity11,12,13,14,15,16,17,18,19, although they lack the complex pattern of alternative splicing found in Drosophila Dscam-1. We therefore cloned chick Dscam and DscamL, the orthologues of mammalian Dscam and Dscam-like-1, respectively, and assessed their expression in retina. Dscam and DscamL, like Sidekicks, were expressed by non-overlapping subsets of retinal neurons between embryonic day (E) 12 and hatching (E21) (Fig. 1b and Supplementary Fig. 2). None of the Dscam-positive neurons expressed either Sidekick (Fig. 1c and Supplementary Fig. 3). Thus, Dscams and Sidekicks mark four distinct sets of presynaptic (amacrine and bipolar) and postsynaptic cells (RGCs) that arborize in the IPL.
To localize Dscam and Sidekick proteins, we generated antibodies specific for each (Supplementary Fig. 4) and used them to immunostain retinal sections. We follow the convention1,2,3 of dividing the IPL into 5 parallel slabs of equal width, S1–S5; each, however, may contain 2 or more sublaminae defined physiologically4,5. Each IgSF protein was concentrated in a distinct set of IPL sublaminae: Dscam in S5, DscamL in S1, S2 and S4, Sidekick-1 predominantly in the inner half of S4 with low expression in the inner half of S2, and Sidekick-2 predominantly in the inner half of S2 with low expression in the inner half of S4 (Fig. 1d). Confocal imaging showed that Dscams were frequently associated with vesicle-rich synaptic sites in the IPL (not shown), as shown previously for Sidekicks10. In situ hybridization of retinae in which individual cells had been filled with GFP (green fluorescent protein) by transposon-mediated gene transfer (see below) showed that neurites of cells expressing each Dscam or Sidekick gene arborize in IPL sublaminae containing the corresponding Dscam or Sidekick protein (Supplementary Fig. 5, ref. 10 and data not shown; summarized in Fig. 1e). Thus, Dscams and Sidekicks mark four distinct synaptic pathways in the retina.
Dscams, like Sidekicks, might not only mark lamina-specified neuronal subsets but also play roles in establishing or maintaining the sublaminar specificity of their synapses. We predicted that if this were the case, (1) each Dscam and Sidekick would mediate homophilic but not heterophilic adhesion; (2) depletion of Dscam or Sidekick would selectively perturb arbors of neurons in particular IPL sublaminae; and (3) ectopic expression of a Dscam or Sidekick would reroute arbors of the expressing cell to normally Dscam- or Sidekick-positive IPL sublaminae. We tested these ideas in turn.
To assess adhesion, we transfected heterologous cells with complementary DNAs encoding Dscams or Sidekicks, then allowed them to aggregate. Cells expressing any one of these four IgSF members formed aggregates but untransfected cells did not (Fig. 2a and data not shown). Moreover, Dscam and Sidekick proteins became concentrated at sites of contact between transfected cells (Fig. 2b and data not shown). Thus, as shown previously7,8,10, Dscams and Sidekicks mediate homophilic adhesion. To assess heterophilic adhesion, we labelled transfected populations separately with red and green fluorescent dyes. When both populations expressed the same adhesion molecule, red and green cells co-aggregated to form mixed clusters, but when the two populations expressed different adhesion molecules, mixed aggregates were rare (Fig. 2a, c). Thus Dscams and Sidekicks do not detectably interact with each other under the conditions of our assay.
To decrease Dscam and Sidekick levels in retinal neurons, we identified interfering RNAs20,21 that targeted Dscam, Sidekick-1 and Sidekick-2 effectively (Supplementary Fig. 6) and incorporated them into retroviral vectors22 that also expressed GFP (Fig. 3a and b). Infection of retina with these vectors selectively decreased expression of the corresponding protein (Supplementary Fig. 7a). To assess effects of depleting Dscam or Sidekicks, we needed independent markers of the neurons that normally express them, so we could identify appropriate cells following depletion. We tested previously described markers of RGC subsets10,23,24, and found that most Dscam-, Sidekick-1- and Sidekick-2-positive RGCs also expressed R-cadherin, cadherin-7 and calbindin, respectively (Supplementary Fig. 8 and data not shown). Thus, we could use these markers to assess lamina-specific connectivity in the absence of Dscam or Sidekicks.
Depletion of Dscam disrupted the laminar patterning of R-cadherin-positive processes in S5. Likewise, depletion of Sidekick-1 and Sidekick-2 disrupted laminar patterning of cadherin-7- and calbindin-positive processes, respectively, in S4. In all three cases, processes appeared defasciculated and extended beyond the boundaries of the sublamina in which they were normally confined (Fig. 3c). Several observations indicated that these effects were specific (Fig. 3c, Supplementary Fig. 7 and data not shown). First, no disruptions were seen in uninfected areas of the same retinae. Second, depletion of each IgSF member affected only the RGC subset marked by the coexpressed gene. Third, we saw no displacement of processes formed by choline acetyltransferase- and substance P-positive cells, which do not express Dscam or either Sidekick, although a small number of choline acetyltransferase-positive somata were sometimes displaced into the IPL following Dscam depletion (Supplementary Fig. 9). Fourth, R-cadherin and cadherin-7 were present not only in Dscam- and Sidekick-1-positive processes in S4 and S5, but also in Dscam- and Sidekick-1-negative processes in S2; only the former were affected by the interfering RNAs (see Supplementary Fig. 7 for examples and quantification). Finally, similar results were obtained with each of two interfering RNAs that depleted Dscam (Supplementary Fig. 6). Thus, Dscam, Sidekick-1 and Sidekick-2 are necessary for lamina-specific arborization of neurites in the IPL (Fig. 3d).
Next, we asked whether expression of Dscams or Sidekicks was sufficient to specify the laminar position of neurites in the IPL. Retroviral vectors such as those used to deliver interfering RNAs could not be used for this purpose because the size of the required cassette (8–9 kb) exceeds the packaging capacity of available vectors22. Electroporation of plasmids10 was also problematic because expression levels declined greatly during the 2-week interval between gene transfer (on E2) and scoring (on E17). We therefore used an alternative method of gene transfer utilizing PiggyBac transposon and transposase25,26. We adapted this insect-derived system to chicks, generated transposons encoding Dscams or Sidekicks plus GFP (Fig. 4a), introduced them into retina, and scored the laminar distribution of GFP-positive processes emanating from groups of GFP-expressing cells. When only GFP was introduced, processes were evenly distributed among IPL sublaminae, but introduction of a Dscam or Sidekick diverted processes to the sublamina in which the corresponding endogenous adhesion molecule is normally concentrated (Fig. 4b–e and data not shown). Cells ectopically expressing Dscam, for example, predominantly arborized in S5 (Fig. 4e). At least some of these processes had been destined for Dscam-negative laminae, as evidenced by their expression of choline acetyltransferase, which is normally confined to S2 and S4 (Fig. 4d). Likewise, cells that ectopically expressed DscamL avoided S3, which is normally DscamL-negative, and cells that ectopically expressed Sidekick-1 or Sidekick-2 sent processes predominantly to S2 and S4 (Fig. 4e, ref. 10 and data not shown). In all cases, both RGCs and interneurons were affected. Thus Dscams and Sidekicks are sufficient to strongly bias the laminar choices of retinal neurons (Fig. 4f).
These results suggest that Dscams and Sidekicks act as attractive or adhesive cues in the IPL, whereas Drosophila Dscams act as repulsive factors14,17,18,19. To explore this difference, we expressed Dscams or Sidekicks, along with GFP, in cultured neural cells. The transfected cells extended neurites that contacted each other, often accumulated synaptic vesicles at sites of contact, and sometimes fasciculated (Fig. 4g and h and Supplementary Fig. 10). The difference between the behaviour of vertebrate and Drosophila Dscams is likely to reflect context-dependence of signalling; many family of guidance molecules, including ephrins and netrins, act as attractants in some cells and repellents in others27,28.
In summary, we have shown that Dscams and Sidekicks mark and help to specify four parallel pathways through the inner retina. They mediate homophilic adhesion and are both necessary and sufficient to ensure sublamina-specific arborization in the IPL. We do not yet know the steps at which they act; they might regulate arborization of pre- and postsynaptic processes, target recognition, or stabilization of synaptic contacts. Although positioning of some amacrine cells is perturbed in Dscam mutant mice29, we do not believe that the defects we have seen are indirect consequences of such perturbations, in that RNA-interference-mediated loss of Dscam in chick does not lead to significant alterations in the distribution of amacrine cells that normally express this gene (data not shown). Moreover, positioning of amacrines is normal in early postnatal mouse Dscam mutants, at the time that synapses are forming in the IPL29. Although Dscam may play slightly different roles in the two species, we favour the idea that it has distinct effects on the two processes.
Dscams and Sidekicks mark only ∼60% of RGCs and a similar fraction of retinal interneurons (Supplementary Fig. 3 and Supplementary Table 1). Other adhesion molecules may exist that mark Dscam- and Sidekick-negative retinal neurons, subdivide groups of cells expressing each of these four IgSF members, and specify the tectal sublaminae in which axons of defined RGC subsets24 terminate. The retinotectal system may be useful for testing the idea that a complex network of recognition molecules mediates lamina-specific synaptic connectivity. Importantly, laminar specificity is a common feature of many circuits in the vertebrate brain and spinal cord30. As Dscams and Sidekicks are expressed by neuronal subsets in the brain (refs 6 and 9, and M.Y., unpublished results), similar networks may be used beyond the retina.
Monoclonal antibodies to chick Dscam and DscamL were generated using bacterially expressed recombinant proteins as immunogens. Full length cDNAs encoding Dscams and Sidekicks were cloned as described in Methods. For RNA interference, the DNA sequences specified in Methods were inserted along with GFP into the retroviral vector RCAS-BP (B). PiggyBac plasmids25 were obtained from M. J. Fraser (University of Notre Dame) and modified as described in Methods.
Gene transfer in ovo
RCAS plasmids (for loss-of-function assays) or a 10:1:1 mixture of transposon, transposase and dsRed plasmids (for gain-of-function assays) were injected into the optic vesicles of stage 9–11 chick embryos. Electroporation was with three square pulses of 7 V, 25 ms. Embryos were fixed and sectioned at E17.
For adhesion assays, HEK293 cells (ATCC) were transfected with plasmids encoding Sidekick- or Dscam-GFP fusion proteins (EGFP-N1, Clontech). Stably expressing cells were selected, labelled with Cell tracker dyes (Invitrogen), and rotated at room temperature for 1 h. For assays of neurite outgrowth, human neuroblastoma cells (SH-SY 5Y; ATCC) were plated on glass cover slips, transfected with Sidekick- or Dscam-IRES-GFP plasmid, then fixed and immunostained 3 days later.
Homologies were calculated using EMBOSS ‘water’ pairwise alignment algorithms (http://www.ebi.ac.uk/emboss/align/index.html). Phylogenetic relationships of protein sequences were obtained from the Treefam database (http://www.treefam.org/). Multiple sequence alignment was performed with the ClustalW program (http://www.ebi.ac.uk/clustalw/index.html). In all cases, alignments and calculations of similarity were done on the entire protein sequence.
Mouse Dscam-like-1. A fragment of mouse Dscam-like-1 cDNA was amplified from the mKIAA1132 plasmid (Kazusa DNA Research Institute) using the following primers: GGCCGCGGCCGCCCGCACGGCCGGCCACAGGACCACC and CGCGGTCGACAGGTCCCAGTGTGGAGCCCTTCTCC. This fragment was cloned into the NotI/SalI sites of pCMVscript (Stratagene). To complete its open reading frame, a BglII fragment of this plasmid was replaced by a cDNA amplified from mouse brain using the following primers: GGCCAGATCTCCGCACGGCCGGCCACAGGACCACC and CGCGAGATCTGTTGCCGTTGTGCTTGAGGGAGAC. The complete open reading frame from this construct was transferred to piggyBac transposon vector pXLBacII-RfA along with an IRES2-EGFP fragment (Clontech). To construct a Dscam-1-GFP fusion protein, the sequence corresponding to extracellular plus transmembrane domains was amplified using primers CGCGCTCGAGCCGCACGGCCGGCCACAGGACCACC and GGCCGAATTCGCATTTCTGCCAAACTCTTTGCATC. The product was inserted between the EcoRI and XhoI sites of the EGFP-N1 vector.
Mouse Dscam. An AccI-NotI fragment from IMAGE clone 6410759 (ATCC) was cloned into the SalI/NotI sites of EGFP-N1. The AccI-AccI fragment was then replaced with cDNA amplified from mouse brain using primers GGCCGTCGACGTGGCTCGCTCGCTGGCTCGCTGGCT and CGACTCCAGCCGAGTTGTTGGCAGT. Transfer to the piggyBac transposon and construction of a GFP fusion construct were as above. The Dscam fragment for the fusion protein was generated using primers CGCGCTCGAGTGGCTCGCTCGCTGGCTCGCTGGC and GGCCGGGCCCGCATGAGCATTTCAGCTAAACTCTT.
Mouse Sidekick-1. A Sidekick fragment was amplified from an IMAGE clone and inserted into pCMVscript. The sequence GGCCGCGGCATGGCCCGCGCCCGGCCCTCGGT GGCGGGCGGCGGAGTCGCGGCGCCCCCTGAGC GAGCGGGGCCCGGGC was then inserted 5′ of this fragment to complete the open reading frame. The sidekick-1-GFP fusion construct was made using primers CGCGCTCGAGAATTAACCCTCACTAAAGGGAAC and GGCCAAGCTTATCCAGGGTCACTGTCTCCTCCAT.
Mouse Sidekick-2. The NdeI-NotI fragment obtained from mKIAA1514 plasmid was cloned into the XhoI and NotI sites of EGFP-N1, and the XhoI-NdeI fragment was replaced by a cDNA amplified with primers GGCCCTCGAGTTATAAAGCAACTCGCAAAGAGG and CTGTTGTAGGCAGCCACCTCGATC. Transfer to the piggyBac transposon and construction of a GFP fusion construct were as above. The Sidekick-2-GFP fusion construct was made using primers CGCGCTCGAGCACTTTAGAGGACCCCTCGGTCA and GGCCGAATTCGACTGCTTTCATCCAAGCTTAGCAT.
Chicken Dscam. A fragment was amplified as above using primers ATCTTCAGCCTCCACCGTCTTTGT and TGCTTCTTGCTCGATCTGTGTCCT, and cloned into pGEM-Teasy vector. For bacterially expressed fusion proteins, extracellular and cytoplasmic domains were amplified using GCGAATTCGTGTCCTGGCTCCCTCCCCT plus CGCTCGAGAGATGTTCGCTGAGCAGTTG, and GGGCGAATTCACCGCATCCCCTCCCAAACCTCA plus GGCCCTCGAGGTAGGGGTTGTTTCCTTTCAAGTG. For ectopic expression, a cDNA amplified from E17 chicken retina total RNA using GCGCGGTCCGGAGATCTACACGCCAGTCTATATTTTGTCAATGCA and CGCGCTCGAGGCCATGCTCTTTTCGTTATACCAG was cloned into BspEI/XhoI site of APtag5 or FLAG-tagged APtag5 (GenHunter).
Chicken DscamL. For in situ hybridization, a 2 kb fragment was amplified from E17 chick retina using primers GAAACATGCGAAAGGCAGTGGGAA and GCTCGGAAGAGCTGATGAAGACAT. The fragment was cloned into pGEM-Teasy vector (Promega), and used as a template for RNA polymerase. For bacterially expressed fusion protein, a fragment amplified using CCGAATTCGACATCGGCTACTACCTCTG and GCCTCGAGCGCACGGTTGAAGGCCTGAAC was cloned into pET23 vector (Novagen).
Chicken R-cadherin. The sequence was amplified from E17 chick retina using primers GGCCCGGCGGCCGCGGCACGGGGCGGCCGAGCGCAC and GGCCGGCTCGAGTAAGTGAGGTCAGTTCAATCCTCCT, and inserted into pCMVscript to generate riboprobes.
Histological methods are described in refs 10 and 24. Briefly, retinae were fixed in 4% paraformaldehyde in PBS, infiltrated with sucrose, frozen, and sectioned at 20 µm in a cryostat. For immunohistochemistry, sections were stained successively with primary antibodies overnight and secondary antibody for 2–3 hours. For in situ hybridization, sections were hybridized with RNA probes that had been labelled by inclusion of digoxigenin- or fluorescein isothiocyanate-conjugated UTP in the RNA polymerase reaction. Following hybridization, probes were detected immunohistochemically using a tyramide signal amplification system (TSA-Plus, Perkin Elmer). Cells in the ganglion cell layer were identified as RGCs, based on their being Thy-1-positive and choline acetyltransferase-negative31,32.
Antibodies to R-cadherin (RCD2; ref. 33), cadherin-7 (CCD7) and SV2 were obtained from the Developmental Studies Hybridoma Bank. Other antibodies were: rabbit antibodies to choline acetyltransferase34, rat anti-substance P (Fitzgerald), rabbit anti-GFP (Chemicon), mouse anti-GFP (3E6, Invitrogen) and calbindin-28K (Swant). Rabbit polyclonal antibodies to mouse Sidekick-1 cytoplasmic domain were prepared by K. Hong in this laboratory. A monoclonal antibody to chick DscamL (DS1-96, IgG1) was generated by immunizing a Balb/c mouse with a bacterially expressed extracellular domain of chick DscamL (see above). Antibodies to the extracellular domain of chick Dscam (DS2-44, IgG1, and rabbit polyclonal antibodies) and the cytoplasmic domain of chick Dscam (DS2-176, IgG1) were similarly generated by immunizing cognate fusion proteins (see above). Monoclonal antibody to Sidekick-2 (CS22, IgG1) was described previously10. Fluorescence-conjugated secondary antibodies were from Invitrogen. For arbor analysis of GFP-filled retinal neurons, sections after fluorescence in situ hybridization using appropriate probes were further stained with chicken anti-GFP antibodies (Aves Lab, Inc.), and Alexa488 conjugated anti-chicken IgY secondary antibodies (Invitrogen). Arbors were analysed as described previously10,24.
Sidekick and Dscam extracellular and transmemebrane domains were cloned into pEGFPN1. HEK293 cells (ATCC) were transfected with these clones, and stably expressing cells were selected. Cells were labelled with appropriate Cell tracker (Invitrogen), dissociated with 0.05% trypsin in the presence of EDTA for 20 min at 37 °C. The reaction was stopped by egg white trypsin inhibitor. Cells were then divided into 24 well culture dishes in Dulbecco-HamF12 medium supplemented with 1% BSA, 20 mM HEPES and 1 µg ml-1 DNase I, and rotated at room temperature for 1 h.
Human neuroblastoma cells (SH-SY 5Y; ATCC: CRL-2266) were maintained in 1:1 Dulbecco-modified Eagle’s minimal essential medium/Ham F12 supplemented with 10% fetal calf serum. 5 × 105 cells were plated on a glass cover slip in a 35 mm dish, and transfected with plasmids in OptiMEM (Invitrogen) using the DMRIE-C transfection reagent (Invitrogen) for 15 h. The medium was then replaced with fresh OptiMEM to promote neurite formation. Three days later, cultures were fixed with 4%(w/v) paraformaldehyde in PBS, and immunostained with anti-GFP and anti-SV2.
The following DNA sequences were cloned into pcDNA6.2-GW/EmGFP-miR (Invitrogen). Sdk-1-a, TGCTGAGCACTGGAAGATCCCTGCATGTTTTGGCCACTGACTGACATGCAGGGCTTCCAGTGCT; sdk-2-a, TGCTGTTGACAAAGGAGTGGGCTTTCGTTTTGGCCACTGACTGACGAAAGCCCTCCTTTGTCAA; dscam-b, TGCTGTACATGGACTTGCTGACGTCTGTTTTGGCCACTGACTGACAGACGTCAAAGTCCATGTA; dscam-c, TGCTGTTAACAGTGAGGTACATGGACGTTTTGGCCACTGACTGACGTCCATGTCTCACTGTTAA.
They were chained to generate two tandem copies as described in the manufacturer’s protocol, and amplified along with EmGFP using primers CCGGCGGCATCGATACAAGTTTGTACAAAAAAGCAGGCT and GCGGCGGCATCGATACCACTTTGTACAAGAAAGCTGGGT. The fragments were cloned into pGEM-Teasy vector, then cloned into a unique ClaI site in the retroviral vector RCAS-BP (B). For injection, RCAS plasmids were concentrated to 5 μg μl-1 in 10 mM TrisHCl, pH 7.4, 1 mM EDTA, mixed with Fast Green (final concentration 0.5%), injected into optic vesicles of SPF chick embryos (Spafas), and then electroporated as described in Methods Summary.
PiggyBac plasmids25 were obtained from M. J. Fraser (University of Notre Dame). A transposase plasmid, pCMVpiggyBACorf, was generated from p3xP3-DsRed1-orf by removing DsRed sequences with XhoI and NotI. The transposon plasmid pXL-BacII-1 was modified by introduction of the acceptor segment RfA from GATEWAY (Gateway Conversion System, Invitrogen) between SmaI and EcoRV sites, generating pXL-BacIIRfA. Full length cDNAs and IRES2-EGFP sequence were cloned into pCMVscript (Stratagene) or pEGFP-N1, then cassettes containing CMV promoter, cDNA, IRES2-EGFP and SV40 polyadenylation signal were cloned into pXL-BacII-RfA using the following attB1/B2 primers: GGGGACAAGTTTGTACAAAAAAGCAGGCTTAGTTATTAATAGTAATCAA, GGGGACCACTTTGTACAAGAAAGCTGGGTTTAAGATACATTGATGAGTT. Plasmids were concentrated and electroporated as above.
We thank R. Burgess and colleagues for sharing data on mouse Dscam before submission, K. Hong for anti-sidekick-1 antiserum and M. J. Fraser (University of Notre Dame) for piggyBac plasmids. This work was supported by a grant from the NIH to J.R.S.
Author Contributions M.Y. and J.R.S. conceived the experiments. M.Y. performed the experiments. J.R.S. and M.Y. wrote the paper.
The file contains Supplementary Figures 1-10 with Legends.
About this article
Scientific Reports (2019)