Main

The 58 genes of the mouse Pcdh locus are tandemly arranged in α-, β- and γ-subclusters, called Pcdha, Pcdhb and Pcdhg, which encode 14, 22 and 22 cadherin-like proteins, respectively8 (Fig. 1a). In the Pcdha and Pcdhg subclusters, single variable exons encoding extracellular, transmembrane and juxtamembrane domains are spliced to three constant exons, generating proteins with unique extracellular but common intracellular domains8. The complexity of this locus is reminiscent of that of Dscam1, which mediates self-avoidance in Drosophila4,5,6,7,15. Moreover, Pcdh genes, like Dscam1, exhibit stochastic expression, and both Pcdhg and Dscam proteins exhibit isoform-specific homotypic recognition13,14. In contrast, the two vertebrate Dscams are not complex genes, so although they mediate both repulsive and attractive interactions among neurons16,17,18,19, they are unlikely to underlie self/non-self discrimination. We therefore investigated roles of Pcdh genes in these processes.

Figure 1: Pcdhgs are required for self-avoidance of SAC dendrites.
figure 1

a, Pcdh locus comprises Pcdha, Pcdhb and Pcdhg subclusters. Pcdha and Pcdhg isoforms are assembled by splicing of one variable exon encoding extracellular (ECD) and transmembrane domains to three constant exons encoding the intracellular domain (ICD). b, SACs are present in both the inner nuclear layer (INL) and the ganglion cell layer (GCL) and extend dendrites that form radially symmetrical arborizations confined to thin sublaminae in the inner plexiform layer (IPL). c, SAC dendrites avoid isoneuronal dendrites (red) but interact heteroneuronally with other SACs (blue), forming synapses on their dendrites. d–i, Morphology of a single SAC, labelled with membrane-Cherry, in the GCL in control and Pcdhg mutant retinas. Wild-type SAC dendrites self-avoid. In Pcdhg mutants, self-avoidance defects include self-crossing and bundling of dendrites. Crossings are detected at 0.2 µm xy resolution in single 0.8-µm optical sections (e, g, i show magnified views of boxed areas in d, f, h). Images with 0.2 µm z resolution are shown in Supplementary Fig. 1. j, SAC dendritic self-crossings in first–fifth order branches per SAC. Graph underestimates difference between genotypes because the most severely affected mutant SACs could not be scored. **P < 0.01. k, l, Number of terminal branches (k) and dendritic field diameter (l) do not differ between wild-type and mutant SACs. Panels j–l show means ± s.e.m.; n = 8 cells from 5–6 animals per genotype. Scale bars, 50 μm (d, f, h) and 10 μm (e, g, i).

PowerPoint slide

Previous studies of mouse mutants lacking all 22 Pcdhg genes revealed that they are required for survival of multiple neuronal types20,21,22,23. To seek roles of Pcdhgs in self-avoidance, we focused on a retinal interneuron, the SAC, which expresses Pcdhg genes22 and exhibits marked dendritic self-avoidance24. Radially symmetric SAC dendritic arborizations are confined to narrow planes within the inner plexiform (synaptic) layer; SACs have no axons. Dendrites of a single SAC seldom cross one another, yet dendrites of neighbouring SACs cross freely (Fig. 1b, c; Supplementary Fig. 1) and even form synapses with each other24,25, suggesting that they can distinguish ‘self’ from ‘non-self’.

We used a conditional mutant (Pcdhgfcon3)22 to bypass the neonatal lethality of constitutive Pcdhg mutants and employed Cre drivers that delete Pcdhg genes from all or subsets of retinal cells. We visualized individual neurons by infection with recombinant adeno-associated virus (rAAV) expressing a fluorescent protein (XFP; Fig. 1d, e), biolistic delivery of DNA encoding XFP, or intracellular injection of a fluorescent dye. We identified SACs, the sole cholinergic neurons in retina, with antibodies to choline acetytransferase (ChAT), which also demonstrated the association of XFP-positive SAC dendrites with dendrites from other (XFP-negative) SACs (Supplementary Figs 1 and 2).

SAC morphology was profoundly altered in Pcdhg mutant retinas (Pcdhgfcon3/fcon3; retina-cre, called Pcdhgrko/rko here; see Methods for genotypes). Dendrites arising from a single SAC frequently crossed each other and sometimes formed loose bundles (Fig. 1f–i and Supplementary Fig. 1). Crossing frequency was increased several-fold in both proximal and distal regions of the arborization (Fig. 1j). These defects were highly specific, in that the diameter of SAC arborizations, the number of dendritic termini, the laminar targeting of SAC dendrites, and the mosaic arrangement of SAC bodies were all unaffected in Pcdhgrko/rko mutants (Fig. 1k, l and Supplementary Figs 1 and 2). Thus, Pcdhgs are dispensable for many aspects of SAC morphogenesis but are required for their self-avoidance.

In the absence of Pcdhg genes, neurons of many types die in elevated numbers during the period of naturally occurring cell death20,21,22,23. Although SACs are largely spared in Pcdhg mutants22, their dendritic defects might be secondary to loss of other neurites with which they ordinarily interact. To test this possibility, we blocked apoptosis by deleting the Bax gene, which is required for naturally occurring and Pcdhg-dependent neuronal death22,23,26. SAC morphology was normal in Bax−/− mice, but self-avoidance defects persisted in Bax−/−; Pcdhgrko/rko double mutants (Supplementary Fig. 3).

We next asked whether Pcdhgs are required for the development of SAC arborizations, or only for their maintenance. In wild-type neonates, SACs extended dendrites that branched profusely and contacted each other (Fig. 2a–c). By postnatal day (P)12, however, excess neurites and isoneuronal contacts were eliminated, resulting in a radial arborization with evenly spaced branches (Fig. 2d, and see ref. 24). Thus, self-avoidance arises rapidly following a short period of isoneuronal ‘sampling’. In Pcdhgrko/rko mice, SACs were clearly aberrant by P3, exhibiting excessive crossing and tangling of neurites (Fig. 2e–g). Excess branches were subsequently eliminated, but whereas most crossing branches were eliminated in controls, many persisted in mutants (Fig. 2h). Thus, Pcdhgs may lead to self-avoidance by mediating repulsive interactions that bias the rearrangement process to selectively eliminate contacts among isoneuronal branches.

Figure 2: Pcdhgs pattern developing SAC dendrites in a cell-autonomous manner.
figure 2

a–h, SACs in developing wild-type and Pcdhg mutant retinas. Wild-type SACs extend fine, exuberant branches (P3, P5) that make transient intradendritic contacts (P5, P8); by P12, excess branches and isoneuronal contacts are eliminated. Dendrites of mutant SACs display excessive self-crossing and bundling by P3; by P12, excess branches are eliminated, but crossing dendrites remain. i, j, Cultured Pcdhg mutant SACs exhibit loss of symmetric growth and uneven distribution of neurites. DIV, days in vitro. k, Histogram of fractal dimensions (Df, metric for space-filling) for 47 wild-type (black) and 47 mutant (grey) SACs. Wild-type SAC in i has Df of 1.61 and mutant SAC in j has Df of 1.53. l, Mean Df for cultured SACs (n = 47 cells), SACs in vivo at P5 (n = 6) and adult (n = 9). ***P < 0.001. Error bars, s.e.m. Scale bars, 50 µm (ad, also apply respectively to e–h) and 20 μm (i, j).

PowerPoint slide

To initiate analysis of the mechanism by which Pcdhgs mediate self-avoidance, we next asked whether they act cell-autonomously. We selectively removed Pcdhg genes from SACs using a ChAT-Cre line. In this case, Pcdhg-negative SACs were surrounded by Pcdhg-positive neurons of other types. We also deleted Pcdhg genes from individual SACs using a transgenic line that expressed tamoxifen-activated Cre recombinase in SACs; we activated Cre with a low dose of tamoxifen and introduced a Cre-dependent reporter to mark mutant SACs. In this case, Pcdhg-negative SACs were surrounded by Pcdhg-positive SACs. In both cases, SACs lacking Pcdhg genes exhibited striking self-avoidance defects (Supplementary Fig. 4). To test whether Pcdhgs can act in completely isolated SACs, we used fluorescence-activated cell sorting to purify SACs from a transgenic line in which they are selectively labelled by an orange fluorescent protein (Thy1-OFP3) and cultured them at low density. Isolated SACs extended dendrites that formed radial, web-like arborizations (Fig. 2i), reminiscent of those observed at P5 in vivo (Fig. 2b). In contrast, SACs from Pcdhgrko/rko; Thy1-OFP3 mice exhibited less symmetrical and unevenly spaced arborizations, reminiscent of those observed in Pcdhgrko/rko retinas at P5 (Fig. 2j and Supplementary Fig. 5). Analysis of the space-filling capacity of dendritic arborizations2,27 (see Methods) revealed that defects in vitro were similar in magnitude to those in vivo (Fig. 2k, l). Thus, Pcdhgs do not depend on intercellular interactions to promote self-avoidance.

We next assessed the requirement for isoform diversity in Pcdhg-dependent self-avoidance. We used RT–PCR (PCR with reverse transcription) to survey expression of Pcdhg isoforms in whole retina, in amacrines generally and in SACs specifically. All 22 Pcdhg variants were expressed in each preparation, with no indication of decreased diversity in purified subpopulations (Supplementary Fig. 6). We then analysed a targeted mouse mutant, Pcdhgtcko, in which three contiguous Pcdhg variable exons, C3–C5, had been deleted. Expression of the remaining 19 Pcdhg isoforms is unperturbed in this allele28. Because Pcdhgtcko homozygous mice die at birth28, we generated transheterozygous animals (Pcdhgtcko/fcon3;retina-cre) so that only retina lacks both copies of Pcdhgc3-c5. In these retinas, neuronal death was as prevalent as in those of Pcdhgrko/rko mice22,28, yet SACs exhibited normal self-avoidance (Fig. 3a, e).

Figure 3: No single Pcdhg isoform is necessary and any isoform is sufficient for dendrite self-avoidance.
figure 3

a, SACs lacking Pcdhgc3–c5 (pcdhgtcko/fcon3;retina-cre) exhibit self-avoidance. b, Replacement of all 22 Pcdhgs by the PcdhgC3 isoform, using the RC::cC3-cherry transgene, rescues SAC dendrite self-avoidance. c, SACs lacking Pcdhga1-a3 (pcdhgtako/tako) exhibit self-avoidance. d, Replacement of all 22 Pcdhgs by the PcdhgA1 isoform, using the RC::cA1-cherry transgene, rescues SAC dendrite self-avoidance. e, Compared to mutants lacking all 22 isoforms, self-crossings of SACs in retinas expressing 19 or 1 isoforms are restored to control levels. ***P < 0.001; n.s., not significant. Data are mean ± s.e.m., from 7 SACs from pcdhgtcko/fcon3;retina-cre retinas, 3 SACs from pcdhgtako/tako, and 9 from remaining genotypes. Scale bar in a, 50 μm; applies to bd also.

PowerPoint slide

In a complementary approach, we generated a line in which the single PcdhgC3 isoform, fused to a fluorescent protein (mCherry), could be expressed in any cell in a Cre-dependent manner (ROSA26-CAG::lox-Stop-lox-Pcdhgc3-mCherry or cC3-mCherry). Thus, in cC3-mCherry;Pcdhgrko/rko mice, Cre both deletes all 22 endogenous Pcdhg genes and activates the single PcdhgC3-mCherry isoform throughout the retina. Analysis of mCherry fluorescence confirmed Cre-dependent expression of the transgene in all retinal cells and appropriate localization of the fusion protein to cell membranes and synaptic layers (Supplementary Fig. 7). Expression of Pcdhgc3 alone rescued self-avoidance defects of Pcdhg mutants (Fig. 3b, e).

To test the possibility that only some isoforms are dispensable for self-avoidance, we analysed a second set of isoforms. We generated Pcdhgtako, which lacks the Pcdhga1-a3 variable exons28, and a line that expresses Pcdhga1-mCherry in a Cre-dependent manner (cA1-mCherry). Results were similar to those for the C3–C5 group: self-avoidance persisted in the absence of PcdhgA1–A3 and was rescued by replacement of all Pcdhg isoforms with PcdhgA1 alone (Fig. 3c–e and Supplementary Fig. 7). From these results, we conclude that no single Pcdhg isoform is necessary but any single isoform is sufficient for dendritic self-avoidance.

Although Pcdhg isoform diversity is not required for isoneuronal self-avoidance, it may be required to ensure that dendrites of adjacent SACs do not avoid each other, which would prevent them from interacting. The ability to generate a SAC population expressing a single Pcdhg isoform (Pcdhga1 or Pcdhgc3) enabled us to test this idea. We injected closely spaced pairs of SACs with different fluorophores (Fig. 4a) and measured the extent to which their dendrites overlapped. To determine whether this method reliably revealed interactions among SACs, we rotated, flipped or rotated and flipped the image of one of the cells, and recalculated overlap. Only the real image showed an overlap greater than that of the manipulated images (Fig. 4b). We then measured overlap for pairs of SACs from wild-type, mutant and single isoform-expressing mice, normalizing for intercellular distance by comparing overlap to the value calculated from the flipped image (Fig. 4c–e and Supplementary Fig. 8). Overlap was equivalent in wild-type and mutant retina, but significantly decreased in retinas expressing a single isoform (Fig. 4f); values for Pcdhga1 and Pcdhgc3 were similar (1.01 and 1.08). Likewise, the mean length of overlapping segments was greater than expected for random overlap in wild-type and mutant but not in single isoform-expressing pairs (Fig. 4g). Thus, when all SACs express the same Pcdhg isoform, heteroneuronal dendrites avoid each other, just as isoneuronal dendrites do in control SACs. We conclude that isoform diversity enables SACs to distinguish isoneuronal from heteroneuronal dendrites.

Figure 4: Reducing Pcdhg diversity disrupts heteroneuronal SAC interactions.
figure 4

a, Two nearby SACs from a wild-type mouse injected with contrasting fluorescent dyes. Right panel shows image of the green SAC flipped vertically. b, Overlap between red and green cells in a. First two bars are derived from the two panels in a. The green cell was rotated in 45° steps or flipped and then rotated (manipulations indicated by symbols beneath graph); third and fourth bars show mean overlap ± s.e.m. derived from these images (n = 7). All inversions and rotations decrease overlap, indicating that overlap in the real image is non-random. c–e, Tracings of SAC pairs, and versions flipped as in a, from wild-type (c), Pcdhgrko/rko (d) and cA1;Pcdhgrko/rko (e) mice. Overlap shown in black. f, Overlap between neighbouring cells, expressed as ratio between overlap measured in real and flipped images. Bars show mean ± s.e.m. for 11, 9 and 8 pairs from wild-type, Pcdhgrko/rko and single isoform-expressing (cA1;Pcdhgrko/rko and cC3;Pcdhgrko/rko) animals. Expression of a single isoform in neighbouring SACs decreases their interaction. g, Mean length of overlapping segments between SAC pairs. R, real image; F, flipped image. *P = 0.05; **P < 0.05; ***P < 0.01. Error bars, s.e.m.; n as in f. Scale bar in a, 50 μm; applies to ce also.

PowerPoint slide

Finally, we asked whether Pcdhgs mediate self-avoidance in areas other than retina. We examined cerebellar Purkinje cells, which have elaborate, planar dendritic arborizations known to exhibit self-avoidance3 (Fig. 5a–c). Importantly, stochastic and combinatorial expression, which underlies the ability of Drosophila Dscam1 to mediate self-avoidance4,5,6,12,14,15,29, has been documented for Pcdhg genes in Purkinje cells10. We selectively deleted Pcdhg genes from Purkinje neurons using an L7-cre transgene, marked cells with a vector that expresses fluorescent proteins in a Cre-dependent manner, and examined them at P15, P21 and at P35, after arborizations have matured30. Deletion of Pcdhg genes from Purkinje cells had no detectable effect on their survival, shape, size or branching pattern (Fig. 5d, e, h, i and Supplementary Fig. 9), but their arborizations were disorganized and dendrites often crossed over each other (Fig. 5f, g). Use of a Cre-dependent reporter revealed that deletion remained incomplete at P8, at which time Purkinje dendrite growth was already advanced (Supplementary Fig. 9). It is therefore possible that earlier deletion of Pcdhg genes would lead to a more dramatic effect. Nonetheless, these results demonstrate a role for Pcdhg genes in Purkinje cell self-avoidance.

Figure 5: Purkinje cell dendrite self-avoidance requires Pcdhg s.
figure 5

a–c, Control Purkinje cells (a, b) labelled with Cre-dependent AAV-XFP in L7-cre transgenic mouse. Self-avoidance is clear in high-magnification view in c (shows area boxed in b). d–f, Purkinje cells lacking Pcdhgs and labelled as in a–c have disorganized arborizations marked by frequent self-crossing defects. Panel f shows area boxed in e. g, Self-crossings detected in single confocal z-sections of 7,225 μm2 unit area from controls and mutants. **P < 0.01; ***P < 0.001; n = 8, 15 and 15 cells at P15, P21 and P35 respectively from ≥3 mice per genotype. h, i, Area of dendritic arborizations (n = 20 cells) and cell density (>40 regions) do not differ between control and mutant Purkinje cells. Data show mean ± s.e.m. Scale bars, 50 μm (a, b, d, e) and 10 μm (c, f).

PowerPoint slide

In summary, although vertebrate Pcdh genes and Drosophila Dscam1 are structurally unrelated, they have remarkable parallels: both encode numerous isoforms from a single locus, the isoforms are expressed stochastically and combinatorially, and the encoded proteins interact homophilically7,8,10,11,12,13,14. We have now shown that in mammalian neurons, Pcdhgs, like Dscam1 (refs 4, 5, 6, 12), promote self-avoidance during development by a cell-autonomous mechanism. In addition, for both Dscam1 and Pcdhg genes, diversity appears to underlie self/non-self discrimination, presumably because neighbouring neurons are unlikely to express the same isoforms and are therefore free to interact7,12,14,15,29. Thus, two phyla appear to have recruited different molecules to mediate similar, complex strategies for self-recognition during formation of neuronal arborizations. These parallels raise the question of why vertebrate and invertebrate nervous systems have invested heavily in mechanisms that promote self-avoidance. In principle, self-avoidance allows neurons to cover their receptive or projective fields maximally while retaining the ability to overlap those of neighbouring neurons1,2,3. However, to our knowledge, the effect of perturbing self-avoidance on circuit function has yet to be assessed in any system. We can now address this issue by electrophysiological analysis of SACs, Purkinje cells, and their synaptic targets in Pcdhg mutant mice.

Methods Summary

Transgenic, knockout and knock-in mouse lines used for this study, as well as methods for marking cells are described in Methods. Identity of labelled SACs was confirmed by immunolabelling retinas for the cholinergic neuron-specific marker. Dendrite self-crossings were quantified by number of branch overlaps detected in single confocal planes.

Online Methods

Mouse strains

The Pcdhgfcon3 conditional mutant allele, in which the third constant exon is flanked by loxP sequences and which generates a functionally null allele following Cre recombination, was described previously21,22. Retina-specific Chx10-cre31 and Six3-cre transgenic mice32 were provided by C. Cepko (Harvard) and W. Klein (M.D. Anderson Cancer Center), respectively. Bax−/− mutants33, Purkinje-specific L7Bac-cre transgenic mice34, Chat-cre, in which the Cre recombinase gene was targeted to the endogenous ChAT gene35, and Rosa-CAG-LoxP-STOP-LoxP-tdTomato-WPRE reporter mice36 were obtained from Jackson Laboratories. A line of BAC transgenic mice in which regulatory elements from the fstl4 gene drive expression of CreER was generated as described in ref. 37. In this line, called line 1 to distinguish it from the line called ‘BD’ in ref. 37, CreER was expressed in SACs, as well as sparse other amacrine cells. We believe that expression reflects influences at the site of transgene integration rather than expression of fstl4. Thy1-OFP3 transgenic mice, in which Thy1 promoter and regulatory elements direct expression of Kusabira Orange (OFP) in SACs and subsets of retinal ganglion cells (RGCs), were described previously37. Pcdhgtcko and Pcdhgtako mice were generated using standard gene targeting techniques28. Mice were maintained on a C57/B6J background. All experiments were carried out in accordance with protocols approved by the Harvard University Standing Committee on the Use of Animals in Research and Teaching.

Generation of single Pcdhg isoform conditional knock-in mice

Pcdhga1 and pcdhgc3 full-length cDNAs were amplified from RNA isolated from P21 C57/BL6 mouse brain, and cloned in frame into pCMV-mCherry-N1 (Clontech). Linker sequence residing between the third constant exon and gfp in Pcdhgfusg knock-in mice and shown to produce functional Pcdhg-GFP fusion proteins in vivo20 was subcloned into pCMV-pcdhga1/c3-mCherry-N1. Targeting vector pRosa26-PAS38 was modified as described in ref. 39 to include a CAG cassette (chicken β-actin promoter and CMV immediate-early enhancer), a Gateway RfA destination site (Invitrogen), a WPRE fragment (woodchuck hepatitis virus posttranscriptional element), and a STOP sequence was cloned from pBS302 (Addgene plasmid 11925). LoxP-STOP-loxP-Pcdhga1/c3-mCherry was recombined into pROSA26-CAG-Rfa-WPRE-FNF-iSceI, creating pROSA26-CAG-loxP-STOP-loxP-Pcdhga1/c3-mCherry-WPRE-FNF-iSceI targeting vectors. The iSceI-linearized vectors were electroporated into 129/B6 F1 hybrid ES cell line V6.5. G418-resistant, targeted ES clones were identified by PCR: 1.7 kb fragment amplified by 5′-Rosa-F: GGCGGACTGGCGGGACTA and 5′-CAG-R: CCAGGCGGGCCATTTACCGTAAG; and 8.2 kb fragment amplified by 3′-CherryF: CTCCCACAACGAGGACTACACCATC and 3′-RosaR: GCATTTTAAAAGCATGAAACTACAAC. ES cell transfections and blastocyst injections were performed by the Genome Modification Facility, Harvard University. Following germ-line transmission, the FRT-neo-FRT cassette was excised by crossing to mice that express Flp recombinase ubiquitously40. Gt(ROSA)26Sor::CAG-loxP-STOP-loxP-Pcdhga1/c3-mCherry conditional knock-in mice are called cA1-mCherry and cC3-mCherry.

Labelling of neurons

Plasmid encoding pAAV2/2-CAG-palmitoylation tag-mCherry-WPRE was used to generate recombinant AAV2/2 expressing membrane-tagged Cherry. To label SACs in retina expressing cC3-mCherry or cA1-mCherry, we used rAAV2/2-CBA-YC3.6-WPRE expressing a calcium sensor that includes cytosolic YFP and used here for visualization of neuronal morphology41. Recombinant AAV2/2-CAG-memb-mCherry and rAAV2/2-YC3.6 were prepared at the Harvard Gene Therapy Institute ((1–2) × 1012 genome copies per ml). Optimal titres of (1–2) × 109 viral genome particles per ml for AAV2/2-CAG-memb-mCherry and 2 × 1010 viral genome particles per ml for rAAV2/2- YC3.6 were prepared in phosphate-buffered saline (PBS, pH = 7.4). rAAV2/9 expressing GFP and mCherry were generated and provided by D. Cai and K. Cohen in our laboratory; high titre virus was produced at the University of Pennsylvania Vector Core.

To inject virus into eyes, adult mice were anaesthetized with ketamine/xylazine by intraperitoneal injection. A 30½G needle was used to make a small hole in the temporal eye, below the cornea, and 1.5 μl of rAAV virus was injected into the vitreous humour with a Hamilton syringe and 33G blunt-ended needle. Animals were killed and retinas were dissected 4–6 weeks following injection. For cerebellar virus infection, P1–P2 mice were anaesthetized with ice and a small puncture was made into the caudal-medial position of one cortical lobe; 1.5 μl of rAAV2/9-GFP; mCherry virus was injected with a Hamilton syringe and 33G blunt-ended needle. Mice were analysed 12–35 days after infection.

For biolistic transfection of SACs, gold particles (1.0 μm diameter, Bio-Rad) were coated with plasmids encoding tdTomato driven by CMV promoter24. Live retinas were dissected, transected with four radial incisions, flattened with photoreceptor side down, and mounted onto a nitrocellulose filter (Millipore). Gold particles were delivered using a Biolistics Helios Gene gun device (Bio-Rad), and retinas were cultured in Ames medium (Sigma) in an oxygenated incubator heated to 37 °C for 12–16 h.

To assess interactions between dendrites of neighbouring SACs, we injected pairs of cells with fluorescent dyes. Retinas from mice expressing OFP in SACs (Thy1-OFP3) were mounted RGC side up and perfused with Ames medium bubbled with 95%O2/5%CO2 at 25 °C. OFP+ SACs were visualized with epifluorescence, and impaled with high resistance electrodes (50 MΩ) filled with a K+ based intracellular recording solution supplemented with 50 μM Alexa Fluor 568 (for targeting) and 200 μM of Alexa Fluor 488 or 647 (for filling, Invitrogen). Square voltage pulses of 3 V were applied to SACs at 50 Hz using a BK Precision Model 3011B function generator. After filling one SAC, the electrode was replaced with a second containing the contrasting dye and the second cell was filled. Images of labelled SAC pairs in live retinas were acquired at 40× on a Zeiss LSM 510 confocal microscope.

Tissue preparation and immunohistochemistry

Mice were killed with intraperitoneal injection of Nembutal, and either enucleated immediately or transcardially perfused with Ringer’s solution followed by 4% paraformaldehyde (PFA) in PBS. Eye cups were removed and fixed in 4% PFA on ice for 1 h, followed by dissection and post-fixation of retinas for an additional 30 min, then rinsed with PBS. Brains were post-fixed in 4% PFA at 4 °C overnight. Animal procedures were in compliance with the US National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Animal and Care and Use Program at Harvard University.

Whole-mount preparations and cryosections of retinas were performed as described22,42. Briefly, whole retinas were incubated for 1–2 h in blocking buffer (0.4% Triton-X, 4% normal donkey serum in PBS), then incubated for 6 days at 4 °C with primary antibodies. Sagittal 80 μm sections of cerebellum were obtained with a vibratome (Leica), incubated in blocking buffer, and with primary antibodies for 2 days at 4 °C. Following washing, retinas and brain sections were incubated for 3 h at room temperature with Alexa-conjugated secondary antibodies (Invitrogen or Jackson ImmunoResearch). Whole retinas were flattened with photoreceptor side down onto nitrocellulose filters. Retina flat-mounts and brain sections were mounted onto glass slides, covered with Vectashield (Vector) or Fluoromount G (Southern Biotech), and imaged on an Olympus FV1000 scanning confocal microscope. Antibodies used were as follows: chick and rabbit anti-GFP (Aves and Millipore); rabbit anti-DsRed (Clontech); goat anti-choline acetyltransferase (Millipore); guinea pig anti-vGluT3 (Millipore); rabbit anti-Calbindin (Swant); mouse anti-syntaxin HPC1 clone (Sigma); rabbit anti-cleaved caspase3 (Cell Signaling Technology). Nuclei were labelled with DAPI, Po-pro1, or NeuroTrace Nissl 435/455 (Invitrogen).

SAC purification and culture

To isolate and culture wild-type and Pcdhg mutant SACs in vitro, we crossed the Thy1-OFP3 transgene, which selectively directs expression of Kusabira Orange (OFP) in SACs and subset of RGCs37, into Pcdhgfcon3; Six3-cre mice. Retinas from genotyped Pcdhgfcon3/fcon3; Six3-cre; Thy1-OFP3 mutant and control P2 mice were dissociated using papain22. OFP+ SACs were isolated by fluorescence activated cell sorting (FACS, MoFlo), plated onto poly-l-lysine-coated glass coverslips (Warner) and cultured for 7–9 days in RGC growth media modified from Meyer-Franke43 in the following ways: (1) substitution of NS2144 for B27, (2) substitution of N2 (Invitrogen) for Sato stock, (3) addition of TGF-β1 and TGF-β2 (2.5 ng ml−1; Peprotech), and (4) addition of mouse glia-conditioned medium (15%). One-third of media was exchanged with fresh media every three days. Cells were fixed with cold 4%PFA/4% sucrose for 15 min, and immunostained for syntaxin and calbindin to confirm SAC identity, and for GFP to confirm Pcdhg−/−; GFP SACs from unrecombined Pcdhg-GFP+ SACs due to variegated Six3-Cre activity in retina.

Image analysis

For best reproduction and clarity of SAC arborizations, maximized projections of confocal images were inverted and contrast-enhanced using Photoshop (Adobe Systems). For morphometric analysis of SACs, we used Fiji software and selected confocal image series of wild-type and Pcdhg mutant SACs situated in comparable retinal eccentricities. Self-crossings per dendritic branch order were quantified as number of branch overlaps detected in single confocal planes; crossings occurring distal to fifth branch order could not be quantified accurately owing to severity of defects in mutants. Dendritic field diameter was measured as the longest axis of arborization. In some cases, arborizations were re-imaged by oversampling using a 60× 1.45NA objective at x,y,z resolution of 47 × 47 × 131 µm and then subjected to deconvolution using Huygens software (http://www.svi.nl/HuygensProfessional).

For analysis of SAC density and mosaic regularity, confocal z-stacks of ChAT-labelled SACs through the GCL and INL were acquired at similar locations in central retina. Sample sizes were 4–5 areas (0.099 mm2) per animal, 2–4 animals per genotype. For each field, xy coordinates of SAC arrays were obtained by manually marking centres of cells using Fiji and used to compute SAC density (number per mm2), packing factor45, and density recovery profiles (DRP)46 with WinDRP software (http://www.mpimf-heidelberg.mpg.de/~teuler/WinDRP/ReadMe.htm).

To compare the space-filling and complexity of control and mutant SAC arborizations, we computed fractal dimensions, Df, which provide a measure of how completely dendrites fill its area2,27,47,48. To calculate Df, we applied the box-counting method as implemented in the FracLac 2.5 plug-in for ImageJ software (http://rsb.info.nih.gov/ij/plugins/fraclac/FLHelp/Introduction.htm; NIH). Confocal images of cultured mutant and control SACs were obtained at equivalent laser scanning parameters with a 60× oil immersion lens, and maximum projections and thresholded, binary images were processed using Image J. Box counts using a series of progressively smaller box sizes (d) were scanned in a region of interest covering the SAC arborization, and the number of boxes intersected by pixels [k(d)] were analysed; this computes Df, which represents an inverse linear regression between log[k(d)] and log(d). Df ranges from 1.0 (straight line with a dimension of 1) to 2.0 (plane with a dimension of 2); a difference of 0.1 represents a doubling of complexity27.

For analysis of dendrite overlap between arborizations of neighbouring SACs, pairs with somata separated by 80–160 µm were selected because their dendrites are known to interact25. Images were processed using Fiji or Photoshop software. To estimate the amount of dendritic overlap that would occur by chance if two SAC arborizations occupy the same territory, we flipped or rotated the image of one SAC, realigned cell body position, and merged images. This method was inspired by work on tiling of RGC dendrites49. We measured total overlapping pixels in real and flipped images, interpreting ratios of >1 (real/flipped) as indicating non-random interactions between SACs.

Purkinje cell dendrite self-crossings detected in single confocal planes were counted in a 7,225 μm2 region of interest assigned to middle of arborization. Purkinje arborization areas were measured using the convex-hull selection in Fiji. Calbindin-labelled Purkinje somata residing along a 635 μm segment in lobules III–VI in single confocal planes were counted to measure Purkinje cell density.

Means were compared using the two-tailed Student’s t test on condition of equivalent variances determined by F-test, or with the Mann–Whitney non-parametric test. Means of multiple samples were compared using ANOVA and posthoc Tukey test.

RT–PCR of dissociated retina cells

We used FACS to sort live cells from dissociated P7 whole retina, VC1.1+ amacrine cells, and OFP+;Thy1.2 SACs cells, as described previously37,50. Amacrine cells were sorted from a live cell suspension of dissociated retinal cells using monoclonal VC1.1 antibody (200 μg ml−1, Sigma) and an anti-IgM secondary conjugated to phycoerythrin-Cy7 (Southern). OFP+ SACs were sorted from OFP+ RGCs by negative selection of Thy1.2-PE-Cy7 labelled RGCs. In each condition, 2,000 cells were sorted directly into RNA lysis buffer (Qiagen); RNA was purified and first strand cDNAs were generated with Superscript RT III (Invitrogen). Primers that uniquely detect the 22 Pcdhg variable exon-constant exon spliced transcripts were adapted from ref. 21, with modifications to avoid cross-hybridization. These primers, and others used to assess purity of the sorted population, are listed in Supplementary Table 1. PCR program used is: 94 °C for 2 min; 30 cycles of 94 °C for 20 s, 56 °C for 30 s, 72 °C for 1 min; 72 °C for 7 min.