Formation of retinal direction-selective circuitry initiated by starburst amacrine cell homotypic contact

To explore how the DS circuit creates its IPL sublayers, we began by determining when the sublayers first emerge in mouse. This analysis focused on SACs and ooDSGCs because bipolar cells develop later (Morgan et al., 2006). Previous estimates of layer emergence vary widely (Stacy and Wong, 2003; Sun et al., 2013) due to the lack of adequate markers to study dendrite development in neonatal SACs. We therefore assembled a suite of mouse lines and antibody markers for this purpose, enabling anatomical studies of the full SAC population as well as individual cells (Figure 1B–C; Figure 1—figure supplement 1; Figure 2—figure supplement 1). These markers revealed that SAC dendrites form two continuous well-defined laminae by P1 (Figure 1B,E). Some dendrites were stratified already at P0, even though the P0 IPL neuropil is less than one-cell diameter wide (Figure 1B; Figure 1—figure supplement 1). Further supporting this timeline, individual P1 SACs made lamina-specific projections (Figure 1C): 96% of OFF SACs in the inner nuclear layer (INL), and 99% of ON SACs in the ganglion cell layer (GCL), stratified within the expected IPL sublayer (n = 49/51 OFF; 78/79 ON; four mice). By contrast, ooDSGCs projected rudimentary and unstratified dendrites at P1 (n = 18 cells, three mice, none were stratified; Figure 1E; Figure 1—figure supplement 2; also see Peng et al., 2017). Even at P2, only 30% of ooDSGCs co-fasciculated with SAC arbors; the rest projected diffusely within the IPL (n = 23 cells, two mice; Figure 1D,E; Figure 1—figure supplement 2). These results indicate that SACs form IPL sublayers at P0-P1, and are joined later by their synaptic partners.

To gain insight into how SACs form their sublayers, we next investigated the cell-cell interactions that immediately precede SAC dendrite stratification. Because SACs stratify early – before any other cell type investigated to date (Figure 1; Kay and Sanes, 2013; Stacy and Wong, 2003) – they are unlikely to form strata by following pre-existing laminar cues. Instead, we hypothesized that SACs create their sublayers by engaging in homotypic interactions. To test this idea, we examined embryonic retina to determine if and when SACs establish homotypic contact. SACs exit the cell cycle at the apical retinal surface and migrate radially through the outer neuroblast layer (ONBL). They next arrive at the inner neuroblast layer (INBL), where postmitotic neurons reside (Hinds and Hinds, 1978); Figure 2A,B). Then they begin to innervate the nascent IPL, which begins to appear in some retinal regions at E16 (Figure 2A). To reveal SAC morphology throughout these steps, the early SAC marker Isl1 (Galli-Resta et al., 1997) was used to drive Cre-dependent expression a membrane-targeted GFP (mGFP) reporter (Isl1^mG mice). We also examined the orientation of SAC dendrite projections using antibodies to internexin, a marker of SAC primary dendrites (Figure 2—figure supplement 1). Staining was performed at E16, when SACs at all stages of their early development could be discerned (Figure 2A–D).

Figure 2 with 3 supplements see all

Download asset Open asset

Since mature SACs contact each other in the IPL, we expected that the onset of SAC homotypic contact would occur around the time of their earliest IPL projections. Surprisingly, however, this analysis revealed that SACs begin to contact each other within the INBL cell body layer upon the conclusion of their radial migration. Migrating SACs rarely interacted, but on arrival at the INBL, SAC arbors were observed touching the soma or primary dendrite of neighboring SACs (Figure 2A–D). The majority of INBL SACs engaged in these soma-layer contacts, such that a GFP⁺ arbor network connected them (Figure 2G). Analysis of primary dendrite orientation indicated that soma-layer contacts likely arose due to projections targeted within this layer: Unlike mature SACs, which exclusively project their primary dendrites toward the IPL, many E16 SACs projected tangentially through the INBL – that is, toward neighboring somata (Figure 2E,F). We even noted cases where SACs appeared to project directly towards each other (Figure 2E). These observations suggest that post-migratory SACs initiate contact with each other by generating an arbor network in the INBL cell body layer.

Many E16 SACs also innervate the nascent IPL, raising the question of whether the soma- or IPL-layer projection establishes the first homotypic contact. We concluded that soma-layer SAC contact precedes IPL innervation, for three reasons. First, soma contacts were found in retinal regions where the IPL had not yet emerged (Figure 2—figure supplement 2). Second, soma contacts were observed among cells that still showed migratory morphological features, such as apical and/or basal processes (Deans et al., 2011; Hinds and Hinds, 1978), and did not yet project into the IPL (Figure 2D; Figure 2—figure supplement 2). Third, SAC dendrite polarization in the tangential plane was highly transient: By P1, the vast majority of SAC primary dendrites were oriented toward the IPL (Figure 2E,F). These three observations suggest that INBL SACs transiently seek out homotypic soma contact before shifting to target the IPL.

We next sought to determine how long the soma-layer SAC arbor network persists. To this end, we examined SAC anatomy at early postnatal ages using Isl1^mG and Chat^mG (Figure 1—figure supplement 1) mice. At P0-1, although SAC arbors within the soma layers no longer express internexin (Figure 2—figure supplement 1), the arbor network remained remarkably prominent (Figure 2G). Most OFF SACs assumed a bi-laminar morphology, with one set of arbors in the IPL and another set targeting neighboring SACs in the INL (Figure 2H–J,L; Figure 2—figure supplements 2–3). INL contacts were highly SAC-selective: 88.8% of branches terminated homotypically (n = 122 arbor tips from 22 cells), significantly greater than the contact rate expected by chance (Figure 2—figure supplement 3). By P2-3, however, this dense INL network was mostly gone (Figure 2G,L; Figure 2—figure supplement 2). ON SACs also made soma layer projections between P0 and P3 that contacted neighboring SAC somata (Figure 2K,L; Figure 2—figure supplement 3). Together, these observations demonstrate that both ON and OFF SACs make transient soma-layer homotypic contacts that arise prior to IPL dendrite elaboration, and are disassembled at P2-3 after SAC sublayers have formed (Figure 2M).

SAC homotypic contacts arise at a time when they could serve as a cue for IPL innervation and sublayer formation. To test this idea, we developed a genetic strategy to prevent SACs from contacting each other in vivo. Ptf1a encodes a transcription factor required for progenitor cells to assume an amacrine fate (Fujitani et al., 2006; Nakhai et al., 2007); Figure 3—figure supplement 1). We crossed conditional Ptf1a^flox mutant mice (Krah et al., 2015) to a Cre line (Six3-Cre; Furuta et al., 2000) that drives widespread recombination in central retina but spares some progenitors from Cre activity in peripheral retina (Figure 3A; Figure 3—figure supplement 1). In Six3-Cre; Ptf1a^flox/flox mice (abbreviated Ptf1a-cKO), only these spared Cre^– progenitors were capable of giving rise to SACs, indicating that any SACs produced in these mutants are wild-type at the Ptf1a locus (Figure 3C). Therefore, the Ptf1a-cKO mutant creates a situation where otherwise-normal SACs are present at significantly lower density than in wild-type retina (Figure 3B,C). In P1-2 mutants, some SACs were effectively segregated from their neighbors – these were termed ‘solitary’ SACs – while others had neighbors sufficiently nearby that they touched (Figure 3B–F; Figure 3—figure supplement 2).

Figure 3 with 2 supplements see all

Download asset Open asset

Comparing solitary to touching SACs in Ptf1a-cKO retinas revealed a role for homotypic contacts in promoting IPL innervation and sublayer formation. At P1-2, touching SACs projected normally to the IPL, similar to SACs from Ptf1a⁺ littermates (Figure 3D,E,G). This suggests that any changes in retinal cell type composition caused by loss of Ptf1a (Figure 3—figure supplement 1) are not by themselves sufficient to perturb SAC sublayer formation. By contrast, solitary SACs largely failed to innervate the IPL (Figure 3F,G). This was not caused by abnormal migration: Solitary SACs were properly positioned at the IPL border, but sent only rudimentary arbors into it (Figure 3F; Figure 3—figure supplement 2). Solitary SACs were also more likely to project processes into the soma layers (Figure 3G), and when they did so, the projections were typically more elaborate than those observed in wild-type retina (Figure 3D,F; Figure 3—figure supplement 2). Thus, solitary SACs overgrew arbors directed toward neighboring somata instead of growing IPL dendrites. Both types of projection errors were also seen at P15, indicating that early errors persist to retinal maturity (Figure 3—figure supplement 2). Misprojecting SACs were still closely apposed to numerous other amacrine cells, and their arbors were intermingled in the IPL, strongly suggesting that generic amacrine interactions are not sufficient to ensure normal dendrite targeting (Figure 3—figure supplement 2). Instead, homotypic interactions are specifically required for IPL innervation and sublayer formation.

To understand how SACs initiate IPL innervation upon homotypic contact, we next sought to identify the molecular cues that SACs use to recognize that contact has occurred. The cell-surface protein MEGF10 (Figure 4A) is a strong candidate to mediate homotypic recognition in this context, for four reasons. First, it is selectively expressed by SACs during the perinatal period (Figure 1B; Figure 1—figure supplement 1). Second, the onset of its expression coincides with onset of SAC homotypic contact at the conclusion of radial migration (Figure 4B). Third, MEGF10 protein is present on soma-layer SAC arbors, making it available to transduce signals arising on these arbors (Figure 4C). Finally, MEGF10 mediates SAC-SAC interactions in a separate context – during formation of the orderly ‘mosaic’ among SAC cell bodies across the retina (Kay et al., 2012). Thus, we tested whether MEGF10 also mediates SAC-SAC recognition to initiate IPL innervation. If so, SACs from mice lacking Megf10 gene function should have phenotypes similar to solitary Ptf1a-cKO SACs – that is, reduced IPL innervation and increased arborization in cell body layers.

Figure 4

Download asset Open asset

To test this prediction, we examined SAC anatomy in Megf10 null mutants (Kay et al., 2012) and littermate controls at P0-1, when sublayers are first forming. We found a striking effect on sublayer formation: Both ON and OFF strata were absent or severely disrupted in mutants (Figure 5A). The cause of sublayer absence was investigated using pan-SAC labeling (Figure 5A,B) and single-cell analysis (Figure 5C; Figure 6D). These studies revealed a severe deficit in IPL dendrite arborization: Most Megf10^–/– SACs made only rudimentary, unstratified IPL projections at P0-1 (n = 1/15 OFF SACs were stratified). Other amacrine cell types showed normal dendritic morphology in Megf10 mutants (Figure 5—figure supplement 1), indicating that the phenotype was specific to SACs. Loss of IPL innervation was not due to aberrant SAC radial migration, because, at P0, mutant SACs had reached the inner retina in normal numbers (wild-type, 2600 ± 287 SACs/mm²; mutant, 3153 ± 145 SACs/mm²; p=0.144, 2-tailed t-test; n = 3 each group), and were positioned adjacent to the IPL, similar to littermate controls (Figure 5A). Furthermore, most mutant SACs sent at least some arbors into the IPL at P0-1 (Figure 5A,C; Figure 6D), suggesting that they migrated to a location from which IPL innervation was feasible. However, the mutant SAC arbors that reached the IPL appeared undifferentiated, with a lack of space-filling branches (Figure 5A,C). As a result, not only did their arbors enclose a significantly smaller IPL territory, but they also failed to sample as much of their enclosed territory as control SACs (Figure 5C; also compare to control cell in Figure 2H). By P3 some ON SAC IPL innervation was evident, but OFF SAC arbors remained largely confined to the soma layer; those that did reach the IPL remained undifferentiated (Figure 5B; Figure 6A,D). These observations indicate that deletion of MEGF10 causes an IPL innervation phenotype strongly reminiscent of Ptf1a-cKO solitary SACs: Both manipulations profoundly impair SAC dendrite arborization within the IPL, preventing timely sublayer formation.

Figure 5 with 1 supplement see all

Download asset Open asset

Figure 6 with 1 supplement see all

Download asset Open asset

In contrast to their underinnervation of the IPL, Megf10 mutant SACs arborized exuberantly in the soma layers (Figure 6A). Both ON and OFF SACs were affected (Figure 6D,E; Figure 6—figure supplement 1), but the OFF SAC phenotype was particularly striking: Starting at P1, the mutant INL network became much more elaborate than the control network of any age (Figure 6A,C). INL arbor density increased in mutants from P0 to P1 and remained high at P3; by contrast, control SACs largely eliminated their INL projections over the same period (Figure 5A,B; Figure 6A,E). To understand how mutant SACs generate a denser and more persistent soma-layer network, we assessed single SAC morphology (Figure 6A,D). From this analysis, we determined that one reason for the denser mutant network, particularly at P2-3, was that a larger number of mutant cells projected to the soma layers (Figure 6E). However, this reason was not sufficient to explain the denser mutant INL network at P1 (Figure 6C), because at that age the number of cells projecting to the INL was similar in mutants and littermate controls (Figure 6E). Therefore, to account for this increase in INL arbor density, we surmised that individual mutant SAC must, on average, overinnervate INL. Supporting this conclusion, we found that mutant SACs frequently had more extensive INL arbors than littermate control SACs (Figure 6B). Further, mutant SACs continued to grow primary dendrites tangentially within soma layers at P1, when the vast majority of control SACs only targeted the IPL (Figure 6F; also see Figure 2F). These observations indicate that mutant SACs continue to expand their soma layer arbor network at P1. Thus, as with solitary Ptf1a-cKO SACs, soma layer projections were both more frequent and more exuberant for Megf10^–/– SACs.

Together, these data suggest that MEGF10 governs a developmental transition from soma-layer to IPL-layer dendrite growth (Figure 6G): Whereas control SACs have only a brief period of soma-layer growth, switching to IPL ramification around P0, Megf10 mutant SACs do not make this transition and instead persist in soma-layer innervation. As a result of this failed transition, many individual mutant SACs ramify extensively in the INL but underinnervate the IPL, causing the dendrite targeting phenotypes that were observed at the population level (Figure 6A–C). We conclude that, because MEGF10 regulates IPL innervation in this way, MEGF10 is required for initial formation of SAC IPL sublayers.

Given the similar phenotypes of Megf10 mutant and solitary Ptf1a-cKO SACs, we hypothesized that MEGF10 is the molecular cue that triggers IPL innervation upon SAC-SAC contact. A key prediction of this model is that SACs should require MEGF10 signals from their neighbors to target their dendrites properly. To test this prediction, we generated a conditional Megf10^flox allele and used it to create a situation where Megf10⁺ SACs were surrounded by Megf10^– mutant cells. This was accomplished via the same Six3-Cre strategy that we employed in our Ptf1a-cKO studies (Figure 3A–C). In central retina of Six3-Cre; Megf10^flox/lacZ (Six3-Megf10-cKO) animals, the vast majority of cells expressed a Cre-dependent GFP reporter, indicating that they lacked Megf10 function (Figure 7A). Accordingly, SACs projected exuberantly to the INL and sublayer formation was disrupted, as in null mutants (Figure 7B; Figure 7—figure supplement 1).

Figure 7 with 2 supplements see all

Download asset Open asset

In peripheral retina, some SACs escaped Cre activity, leading to absence of the GFP reporter and continued MEGF10 protein expression (Figure 7A,B; Figure 7—figure supplement 1). Our model predicts that these cells should have mutant dendrite phenotypes despite retaining MEGF10. To test this prediction, we imaged βgal-stained OFF SACs from Six3-Megf10-cKO and littermate control mice at P2. This age was chosen because wild-type and null mutant mice showed a large difference in SAC INL projection frequency (Figure 6E). In littermate controls, we found that βgal⁺ SACs rarely projected to the INL (Figure 7C,D); therefore, they behaved like control SACs from earlier experiments (Figure 6E). By contrast, Megf10⁺ SACs surrounded by mutant SACs in Six3-Megf10-cKO retina showed a high rate of INL projections, nearly identical to their Megf10^– neighbors (Figure 7B,D; Figure 7—figure supplement 1). Thus, when Megf10⁺ SACs are deprived of MEGF10 signal from adjacent SACs, they make exuberant soma-layer projections. This finding implicates MEGF10 as a transcellular signal that controls SAC dendrite targeting (Figure 7K).

Next, we investigated how SACs receive this MEGF10 signal from their neighbors. Given that MEGF10 can function as a receptor in other contexts (Chung et al., 2013; Kay et al., 2012), we speculated that MEGF10 might act as its own receptor. In support of this idea, co-immunoprecipitation experiments using intracellularly truncated Megf10 constructs showed that MEGF10 can interact with itself through its extracellular domain (Figure 7I,J; Figure 7—figure supplement 2). Thus, MEGF10 appears biochemically capable of acting as both ligand and receptor.

If MEGF10 is indeed a receptor in this context, SACs should require it to detect contact with MEGF10-expressing homotypic neighbors. To test this prediction, we asked whether removal of Megf10 from a single SAC, during the period of soma-layer homotypic contact, would impair its IPL innervation despite normal MEGF10 expression by surrounding cells. We used Chat^Cre to achieve sparse recombination in SACs of neonatal mice, as in the anatomy experiments described above (Figure 2H–K; Figure 6D). In Chat-Megf10-cKO animals, MEGF10 immunostaining was used to identify SACs that lost MEGF10 protein prior to P3 – that is, during the period when soma-layer arbors are present (Figure 7F,G). MEGF10^– cells constituted a small minority of SACs at P3, meaning that they were generally surrounded by MEGF10⁺ neighbors (Figure 7—figure supplement 1). In this context, MEGF10^– SACs produced more exuberant soma-layer arbors than neighboring MEGF10⁺ cells, while sending only minimal arbors into the IPL (Figure 7E–H). Thus, single MEGF10^– SACs had phenotypes similar to SACs from mice entirely lacking Megf10 (Figure 7G,H; compare to Figure 6D). By contrast, adjacent MEGF10⁺ cells in the same Chat-Megf10-cKO retinas were indistinguishable from littermate control SACs (Figure 7E,F,H). Therefore, when Megf10 is lost during dendro-somatic contact (but not after; see below), SACs make projection errors typical of neurons deprived of homotypic interactions, and they do so even if their neighbors express MEGF10 and are developing normally. Together, these experiments support the conclusion that MEGF10 is a receptor through which SACs detect each other to terminate soma-layer growth and initiate IPL innervation (Figure 7K).

We next asked whether neonatal MEGF10-mediated interactions influence the anatomy of SAC IPL sublayers at maturity. We found that SAC sublayers eventually formed (by P5; Figure 8H), and were present in the mature Megf10^–/– retina, but they were marred by numerous errors. Sporadically, and at apparently arbitrary retinal locations, two kinds of local laminar disruptions were apparent. First, there were discontinuities in the ON and OFF strata, such that mutant SACs did not completely innervate their sublaminae (Figure 8A–C). These discontinuities diminished retinal coverage within each mutant sublayer by ~15% (OFF decrease, 15.0 ± 0.9%; ON decrease, 13.7 ± 4.0%; mean ±SD; n = 9 fields of view/2 mice per genotype). Innervation gaps were not observed for other amacrine cells, indicating that SACs were selectively affected (Figure 8—figure supplement 1). Examination of single SACs revealed that while dendritic patterning substantially recovered between P1 and adulthood, SAC arbor territories remained significantly smaller in mutants (Figure 8D). These phenotypes suggest that mutant SACs never fully made up for their initial IPL innervation deficit, thereby contributing to gaps in the dendritic plexus.

Figure 8 with 1 supplement see all

Download asset Open asset

The second type of SAC error in mature Megf10^–/– IPL was dendrite mistargeting to ectopic IPL strata (Figure 8A,B,E). Both ON and OFF SACs were affected; in each case, ectopic arbors were mostly found in IPL regions inappropriately close to the soma layers (Figure 8A,B). En-face images of mutant IPL revealed that ectopic OFF arbors formed a patchy but extensive fascicle network connecting many of the cells (Figure 8E,F; 78.5 ± 3.5% of SACs participated in the network, mean ±95% CI). This IPL network was morphologically similar to the ectopic INL network observed in mutants at earlier ages (Figure 6C), raising the possibility that the early network gives rise to the adult network by shifting location from the INL to the IPL. Supporting this view, we found that a soma layer-to-IPL transition occurs at P5, when mutant SACs began projecting to ectopic IPL locations in addition to the soma layers (Figure 8G,H; Figure 8—figure supplement 1). This transition occurred without a significant change in the number of mutant SACs projecting into the ectopic network (Figure 8F; Figure 8—figure supplement 1), suggesting that the same cells continued to participate in the network but simply altered their anatomy to target the IPL. Thus, early exuberant soma-layer projections appear to give rise to adult IPL ectopias, starting between P3 and P5.

Together, these two adult mutant phenotypes demonstrate that DS circuit sublayer formation is delayed and imperfect in the absence of MEGF10. While other mechanisms appear to partially compensate for MEGF10 in generating the sublayers, such mechanisms are not sufficient to prevent persistence of innervation gaps and laminar targeting errors. Thus, MEGF10 is essential for normal formation of the mature SAC IPL projection.

Next, we sought to directly test the idea that MEGF10 is required early – at the time of initial SAC homotypic contact – to ensure normal SAC IPL lamination at maturity. To this end, we used Megf10^flox mice to delete MEGF10 at different times. Deletion prior to the onset of homotypic contact, using the Six3-Cre line, fully phenocopied Megf10^–/– adult IPL errors (Figure 9A), suggesting a requirement for MEGF10 at the time of contact. To remove MEGF10 from SACs that had already established homotypic contact, we used Chat^Cre. In this line, the number of SACs expressing Cre gradually increases over the first postnatal days to encompass the full SAC population (Xu et al., 2016). Therefore, Chat-Megf10-cKO mice can be used both for early, sparse MEGF10 deletion (Figure 7F–H) and for later, broad MEGF10 deletion. MEGF10 immunostaining revealed that this late, broad deletion occurs between P3 and P5 (Figure 7—figure supplement 1), such that MEGF10 expression is largely preserved during the period when homotypic soma-layer contacts exist (Figure 2L), but is eliminated shortly thereafter. In this Chat^Cre-mediated deletion regime, SAC laminar targeting and gap errors were exceedingly rare (Figure 9A). These experiments therefore define a time window for MEGF10 function (Figure 9C): Adult IPL targeting phenotypes require absence of MEGF10 during the soma-layer projection phase of SAC development – that is, prior to P3. Any additional activity of MEGF10 after P3 is dispensable for the adult IPL phenotype. These findings strongly support a model whereby the functions of MEGF10 during early homotypic contact – that is, promoting IPL innervation and terminating soma-layer arbor growth – are necessary for development of normal SAC IPL innervation at maturity.

Figure 9 with 1 supplement see all

Download asset Open asset

In addition to laminar targeting errors, Megf10 mutants also show disruptions in the mosaic spacing of SAC cell bodies across the retina: Instead of a regular, uniform distribution, mutant SAC positioning is random (Kay et al., 2012). We considered the possibility that SAC IPL errors might arise due to MEGF10 effects on soma spacing. Two lines of evidence suggest that this is not the case. First, the two phenotypes were not well correlated at the individual SAC level: Regardless of the severity of their mosaic spacing defects, SACs made IPL targeting errors at a constant rate (Figure 9—figure supplement 1). This finding suggests that disturbed cell positioning does not influence the probability of making an IPL error. Second, using our Megf10^flox allele, we were able to dissociate the IPL and mosaic phenotypes: Deletion of MEGF10 after P3 in Chat-Megf10-cKO mice caused mosaic patterning deficits, but IPL projections were largely normal (Figure 9A,B). This finding demonstrates that IPL laminar perturbations are not an inevitable consequence of altered soma positioning. Altogether, these experiments support the notion that altered SAC position makes at best a minor contribution to IPL phenotypes; instead, delayed IPL innervation and exuberant soma-layer arborization are likely the major sources of perturbed SAC projections at maturity.

Next, we asked whether MEGF10, and its effects on SAC sublayer formation, are important for assembly of the broader DS circuit. To this end, we tested the impact of SAC IPL stratification errors on laminar targeting by their circuit partners. First, we examined ooDSGC IPL projections using the Hblx9-GFP (referred to as Hb9-GFP; Figure 10) and Drd4-GFP (Figure 10—figure supplement 1) transgenic lines, which label ooDSGC subtypes with different preferred directions (Trenholm et al., 2011; Huberman et al., 2009). In littermate control mice (n = 9), ooDSGC dendrites were tightly and selectively associated with SAC arbors, as shown previously (Vaney and Pow, 2000). This association was maintained in Megf10 mutants: Both normal and ectopic SAC IPL arbors reliably recruited ectopic ooDSGC projections (Figure 10A–C; Figure 10—figure supplement 1; n = 240 ectopias from five mutants,>97% contained ooDSGC arbors). Further, when SAC gaps were present in the mutant IPL, ooDSGC dendrites typically grew around the gap edges and failed to enter them (Figure 10D; Figure 10—figure supplement 1; n = 325 gaps from five mutants, >95% devoid of ooDSGC arbors). Thus, SACs provide both permissive cues required for ooDSGC IPL innervation, and also attractive cues sufficient to recruit ooDSGCs to the wrong IPL sublayer.

Figure 10 with 1 supplement see all

Download asset Open asset

Next we determined the impact of altered SAC lamination on the axons of bipolar cells that participate in the DS circuit (Figure 11A). We examined the four cell types (BC2, BC3a, BC5, and BC7) that make extensive monosynaptic connections with SACs and ooDSGCs (Duan et al., 2014; Ding et al., 2016; GreeneGreene et al., 2016; KimKim et al., 2014; Chen et al., 2014). Bipolar axons were marked with type-specific antibodies and mouse lines reported previously (Wässle et al., 2009; Duan et al., 2014), as well as a novel transgenic marker of BC5 (Gjd2-GFP; Figure 11—figure supplement 1). In wild-type retina, DS-circuit bipolar cells arborized in close contact with SAC dendrites; however, unlike ooDSGCs, they remained adjacent to SACs rather than overlapping them (Figure 11A–D; Figure 11—figure supplement 1). This arrangement was preserved in Megf10 mutants: Axons of all four bipolar cell types were recruited to ectopic IPL locations by mistargeted SAC arbors, where they stratified adjacent to SACs (Figure 11B–D,F; Figure 11—figure supplement 1). For example, BC5 and BC7 terminals always sandwiched SAC arbors, regardless of their IPL location – even when doing so required formation of a supernumerary BC axon field between the normal and ectopic SAC sublayers (Figure 11C,D). To quantify the mistargeting effect, we measured the position of BC5 and BC7 terminals adjacent to ON SAC ectopias. Their arbors were pushed farther apart by SAC arbor clumps (Figure 11C–E), which shifted BC7 terminals significantly toward the GCL by ~4 µm (69 ± 0.8% of IPL depth in control regions to 74 ± 1.9% in affected regions; mean ± S.E.M.; n = 21 control, n = 6 affected; 2-tailed t-test, p=0.0024). These observations indicate that DS-circuit bipolar cells, like ooDSGCs, respond to SAC attractive cues. However, in contrast to ooDSGCs, bipolar cell projections were minimally affected by SAC IPL gaps. While BC5 and BC7 terminals were slightly mispositioned in the absence of SAC arbors – they were closer together – innervation of gap regions was otherwise normal (Figure 11C–F). Thus, DS-circuit bipolar axons either do not require SAC-derived signals for IPL innervation, or the relevant signals are capable of acting over larger distances than the typical SAC IPL gap size (35–45 µm maximum diameter). Altogether, these analyses of DS circuit anatomy in Megf10 mutants support the notion that early-stratifying SACs form a scaffold that directs IPL laminar targeting of their circuit partners using multiple guidance strategies.

Figure 11 with 1 supplement see all

Download asset Open asset

Finally, we investigated the extent to which developmental events controlled by MEGF10 affect DS circuit function. We sought to determine whether the anatomical perturbations caused by loss of MEGF10 – that is, SAC laminar targeting and mosaic spacing errors – alter direction coding by ooDSGCs. To do this, we recorded from wild-type and Megf10^–/– retinas on a large-scale multielectrode array (Field et al., 2007; Yu et al., 2017). ooDSGCs were identified based on their responses to drifting gratings and moving bars (see Materials and methods), which unambiguously distinguished them from other recorded RGCs (Figure 12A). Because MEGF10 is not expressed in the adult DS circuit (Kay et al., 2012), we could be confident that any mutant physiological phenotypes reflect anatomical changes that arose during development.

Figure 12 with 1 supplement see all

Download asset Open asset

These experiments revealed that ooDSGCs with robust direction selectivity were present in both wild-type and Megf10^–/– retinas (Figure 12A,B), and constituted a similar fraction of the RGC population in both strains (wild-type: 80/609, 13.1%; mutant: 74/551, 13.4%). Furthermore, loss of Megf10 did not alter the organization of ooDSGC preferred directions along cardinal axes (Oyster and Barlow, 1967), or the fraction of ooDSGCs preferring each direction (Figure 12—figure supplement 1). These results are consistent with the observation that mutant SACs remain paired with ooDSGC dendrites and bipolar cell axons even when normal lamination and arbor spacing are disrupted. They indicate that the qualitative functional properties of the circuit are still present.

However, a more careful examination of DS tuning properties in Megf10^–/– retinas revealed clear quantitative differences in ooDSGC responses. Moving bars were used to measure the width and strength of direction tuning for each identified ooDSGC across the populations recorded on the electrode array (Figure 12C). Tuning width was measured as the circular standard deviation of the tuning curve, while tuning strength was measured as the normalized response difference to motion in the preferred and null directions (see Materials and methods). These experiments revealed systematic shifts toward broader (Figure 12D) and weaker (Figure 12E) direction tuning across the population of ooDSGCs in Megf10 mutant retinas. This was mainly due to higher null direction spiking among ooDSGCs in mutants (Figure 12B,C,E). Furthermore, these effects on tuning width and strength persisted across a broad range of stimulus contrasts (Figure 12—figure supplement 1). These results demonstrate that disruption of MEGF10-dependent developmental patterning degrades the precision and strength of ooDSGC direction tuning. They further suggest that perturbations to the anatomical regularly of the circuit across space (e.g. laminar uniformity and SAC spacing) may effectively introduce noise in the DS circuit that broadens and weakens direction tuning (see Discussion).

This idea led us to consider additional functional properties of ooDSGCs that might depend on the spatial regularity of the DS circuit, and therefore might be perturbed in Megf10 mutants. One such property is the generation of symmetric DS responses to stimuli that are darker or brighter than the background (Figure 12F,G). This ON-OFF symmetry allows the DS response to be largely insensitive to contrast reversals (Amthor and Grzywacz, 1993); it arises because ooDSGCs receive highly symmetric SAC inputs in both ON and OFF sublayers (Figure 1A). In Megf10 mutants, ON-OFF anatomical symmetry is disturbed, because ON and OFF SAC errors are not spatially correlated (Figure 8A–C). We hypothesized that this might lead to disparities in the direction tuning of individual cells’ ON and OFF responses. Indeed, Megf10^–/– ooDSGCs exhibited greater separation (i.e. less coherence) between their ON and OFF preferred directions than wild-type ooDSGCs, across a broad range of contrasts (Figure 12H; Figure 12—figure supplement 1). These results support the idea that MEGF10 serves to establish a highly uniform and regular network of SAC dendrites (via controlling both the precise timing of INL lamination and through regularizing inter-SAC spacing), the net effect of which is to allow greater precision and coherence in the direction tuning of ooDSGCs.

During radial migration, newborn central nervous system neurons have a multipolar morphology, but on arrival at their final position within the tissue they become highly polarized (Nadarajah et al., 2001; Tabata and Nakajima, 2003; Cooper, 2014; Chow et al., 2015; Krol et al., 2016; Hinds and Hinds, 1978). This morphological change enables elaboration of dendrites and integration into local circuitry. If dendrite differentiation begins early, migration is impaired (Hoshiba et al., 2016), suggesting that the transition from migratory to mature morphology must be highly regulated to ensure that neurons only differentiate once they arrive at their final position. The extracellular cues that signal arrival are poorly understood in most nervous system regions.

Here, we show that SACs use homotypic recognition, mediated by MEGF10, to initiate IPL-directed dendrite morphogenesis. When deprived of homotypic neighbors or MEGF10, SACs at the IPL retain a multipolar morphology (compare Figure 2C to Figures 3F and 6A) instead of polarizing arbors toward the IPL. This indicates that the transition from migratory to mature morphology is impaired in the absence of SAC homotypic recognition. We show that migrating SACs first establish homotypic contact upon arrival at the inner retina. At this stage, they are still multipolar (Figure 2D), but they orient primary dendrites tangentially within the INBL to ultimately contact their SAC neighbors. These contacts occur prior to IPL innervation, and are required for it to occur in a timely manner. SACs lacking neighbors or the molecular means to detect them (i.e. MEGF10) appear to persist in this multipolar soma-layer-targeting phase, causing over-innervation of the INL/GCL and delaying IPL innervation (Figure 6G). Thus, establishment of homotypic contact is a key checkpoint for the progression of SAC dendrite differentiation and IPL sublayer morphogenesis.

We propose that the function of this checkpoint is to ensure that SACs elaborate dendrites only when they have arrived adjacent to the IPL. The presence of other SACs that have already completed their migration is a reliable indicator of arrival in the proper location. Because soma-layer SAC contacts appear earliest, and because MEGF10 selectively influences IPL innervation during the period when they exist, we favor the notion that the key homotypic interactions occur through these arbors. However, we cannot exclude that IPL-based interactions also play a role. INL-directed arbors resembling those we describe can be discerned in many developing zebrafish amacrine cells (Godinho et al., 2005; Chow et al., 2015), raising the possibility that this mechanism applies across species and across other amacrine cell types. Because most neurons require a way to control when and where they differentiate, we anticipate that this homotypic contact strategy, or variations upon it, may have important roles in the differentiation of other CNS neurons at the completion of their radial migration.

We conclude that MEGF10 is the molecule responsible for homotypic recognition during SAC IPL innervation. Four key results support this conclusion. First, MEGF10 is expressed at the right time and place to assume this role: It is expressed selectively in SACs (Figure 1), upon conclusion of their radial migration, and in the soma-layer arbors that we propose mediate recognition (Figure 4). Second, Megf10 null mutant SACs phenocopy the dendrite polarization errors seen in solitary Ptf1a-cKO SACs, suggesting that homotypic recognition requires Megf10. Third, co-immunoprecipitation experiments indicate that MEGF10 interacts with itself via its extracellular domain, suggesting it could act as both ligand and receptor. While this biochemical interaction may take place in the cis configuration, the fourth line of evidence indicates that MEGF10 interacts in trans as well: Using a conditional-null Megf10 allele in vivo, we show that MEGF10 is required on the cell that sends homotypic signals as well as the cell receiving those signals. Loss of MEGF10 on either side leads to dendritic phenotypes resembling solitary SACs and Megf10 null mutants. Together, these data are consistent with a model whereby SAC-SAC contact initiates a transcellular MEGF10 homophilic interaction, in which MEGF10 serves as both receptor and ligand to trigger the switch from migratory to mature morphology (see model, Figure 7K).

This homophilic model of MEGF10 function is consistent with its role during establishment of mosaic cell body patterning (Kay et al., 2012). In that context, MEGF10 acts as ligand and receptor to mediate cell-cell repulsion, thereby spacing SAC somata evenly across the retina. Here we discover a second MEGF10 function in SAC IPL innervation. Because the two SAC phenotypes have different underlying cell biology (soma movement vs. dendrite dynamics), and separable temporal requirements for MEGF10 function (Figure 9), it seems unlikely that they reflect disruption of a single biological event. Instead, MEGF10 appears to act at distinct, albeit partially overlapping times, to control different aspects of SAC development, each of which are regulated by contact with homotypic neighbors (see model, Figure 9C).

Our results shed light on the mechanisms controlling SAC dendrite lamination. While repulsion mediated by Sema6a and PlexinA2 prevents OFF SACs from straying to the ON sublayer (Sun et al., 2013), molecules required for formation of the SAC sublayers have not been identified. We show that SACs deprived of homotypic neighbors or MEGF10 initially fail to form IPL sublayers, and when they eventually do so, their strata are riddled with errors. Both the lack of sublayers at early stages and the dendritic mistargeting to inappropriate sublayers at maturity are novel SAC phenotypes; they implicate MEGF10 as a key player in forming SAC IPL sublayer-specific projections.

It is generally assumed that sublayer formation has two basic molecular requirements: 1) Attractive/adhesive molecules that mediate co-fasciculation of stratified arbors; and 2) repulsive cues that prevent straying of arbors into other sublayers (Lefebvre et al., 2015; Sanes and Yamagata, 2009). Our MEGF10 studies suggest an additional, earlier requirement for cell-cell interactions that occur prior to neuropil innervation. The purpose of this surprisingly early SAC-SAC interaction, we propose, is to ensure that SACs grow dendrites at the right time and place to co-fasciculate with their SAC neighbors. The molecular basis of this homotypic co-fasciculation – clearly another essential player in sublayer formation – remains to be determined. MEGF10 is probably not involved; the co-fasciculation system appears intact in Megf10 mutants given that sublayers do eventually form. Perhaps this system is part of the mechanism that compensates for loss of MEGF10 to ultimately generate the sublayers.

When IPL arborization is delayed by loss of Megf10, two SAC errors ensue. First, SACs generate mistargeted dendritic material that appears to persist as ectopic IPL sublayers. Second, SACs never completely innervate their sublayers, resulting in fragmented IPL strata. These two errors are caused by delays rather than an ongoing requirement for MEGF10 during later stages of arbor growth, as shown by conditional mutant experiments (Figure 9). Thus, our findings support the idea that timing is critical to the sequential lamination of the IPL: When SAC dendrites arrive in the IPL too late, they encounter a different cellular and molecular milieu that may not support the proper development of their arbors. In this view, the normal role of MEGF10 in DS circuit assembly is to instigate SAC dendrite outgrowth at the crucial time when laminar self-assembly can occur.

SACs may face an additional obstacle to overcoming their delayed IPL innervation in Megf10 mutants: abnormal soma positioning. While mosaic spacing errors do not account for the Megf10 mutant ectopic IPL phenotype, we cannot exclude the possibility that the placement of IPL arbor gaps might be at least partly explained by soma position. If SACs are struggling to make up for their delayed IPL innervation, it is plausible that increasing the distance between SACs (as happens sporadically due to random positioning) might further hinder the development of complete retinal coverage.

Because of their early stratification, SAC dendrites have been proposed to act as a scaffold that guides assembly of the DS circuit (Stacy and Wong, 2003). A key prediction of this model is that laminar targeting of later-stratifying cell types should depend on the existence of this scaffold. We show using a SAC-specific manipulation – removal of Megf10 – that disruption of SAC stratification causes their bipolar and ooDSGC circuit partners to make corresponding projection errors. Based on the kinds of errors we observed, SACs appear to provide attractive, permissive, and possibly even repulsive arbor sorting cues to influence the laminar positioning of their circuit partners. This work thus constitutes the first critical test of the scaffolding model, and provides strong support for it. We find that SACs use homotypic interactions to initiate formation of their circuit sublayers, and then heterotypic interactions to recruit circuit partners to join them. SACs might achieve their scaffolding functions directly, by providing guidance cues to their partners; or they may do so indirectly, by patterning the IPL projections of an intermediary cell type that in turn guides later-arriving projections. Direct scaffolding may be mediated in part by Cadherins 8 and 9, which regulate interactions between SAC dendrites and DS circuit bipolar cell axons (Duan et al., 2014). Molecular mediators of ooDSGC-SAC dendrite interactions remain to be identified.

Evidence that the SAC scaffold can be repulsive – or at least can exclude bipolar arbors from certain IPL regions – came from our observations of BC axon anatomy. In wild-type retina, we were surprised to note how completely the BC3a, BC5, and BC7 axon terminals were excluded from the SAC territory – they contacted it but did not enter (Figure 11B–D; Figure 11—figure supplement 1). This behavior stands in stark contrast to the behavior of ooDSGC dendrites, which completely overlapped SACs (Figure 10A–B; Vaney and Pow, 2000). Moreover, in Megf10 mutants, the laminar distance between BC5 and BC7 terminals was reduced in the absence of SAC arbors, and increased in the presence of SAC ectopias, further suggesting the existence of local SAC-BC repulsion. The finding that SACs exclude bipolar circuit partners from their sublayers appears at first counterintuitive. But given that no bipolar cell type is exclusively devoted to the DS circuit (Wässle et al., 2009; Greene et al., 2016; Kim et al., 2014), a mechanism must exist to ensure that they can also contact non-DS partners. We speculate that SACs initially recruit their bipolar partners using long-range attractive cues, and then use contact-repulsion (or an equivalent arbor sorting mechanism) to displace bipolar arbors such that they remain in contact with the SAC layers but also innervate adjacent layers. This model is consistent with bipolar arbor phenotypes in Megf10 mutants, but will require further study.

We found that impairment of SAC interactions in the perinatal retina causes permanent functional DS circuit deficits. In Megf10 mutants, direction tuning of ooDSGCs becomes broader and weaker, and their ON/OFF preferred directions are less aligned. Direction tuning is degraded in large part because mutant ooDSGCs have aberrant spiking responses to null-direction stimuli. This suggests that impaired null-direction inhibition – which arises from SACs – is a key contributor to the phenotype. Broader ooDSGC tuning curves have been shown, in modeling studies, to degrade population-level coding of directional information, and the ability of downstream neurons to extract such information (Fiscella et al., 2015). Thus, the physiological phenotypes we identified are likely sufficient to impair the ability of mutant retina to appropriately relay visual information.

Dysfunctional DS circuit physiology in Megf10 mutants is almost certainly a consequence of its effects on development, because neurons do not express MEGF10 beyond the second postnatal week (Kay et al., 2012). Further, even though MEGF10 is expressed by Müller glia in adulthood, we have been unable to detect any changes in Müller glia anatomy or interactions with DS circuit synapses upon loss of Megf10 function (Wang et al., 2017); J.W. and J.N.K., unpublished observations). We therefore conclude that anatomical changes to the DS circuit arising during development are responsible for circuit dysfunction.

The fundamental change to DS circuit anatomy in Megf10 mutants is altered distribution of arbors and synapses, unlike other manipulations which simply serve to destroy SAC radial morphology or disrupt synaptic partnering among DS circuit cells (Sun et al., 2013; Duan et al., 2014; Kostadinov and Sanes, 2015; Peng et al., 2017). In Megf10 mutants, the combined effect of mosaic spacing defects and IPL laminar targeting errors is to disturb the regularity of SAC IPL innervation. As a result, some parts of the visual map become over-innervated (e.g. Figure 10C) while others are uninnervated (Figure 10D). In turn, ooDSGCs are recruited to the over-innervated regions and excluded from uninnervated gaps, likely causing sporadic local inhomogeneity in synapse density across visual space. According to some models of DS, which posit that the total amount of SAC inhibition is the key factor underlying DS responsiveness, these relatively small-scale changes would be considered unlikely to change circuit function (Taylor and Vaney, 2002; Demb, 2007). A more recent alternate view is that the fine spatial arrangement of glutamatergic inputs to SACs, and the synaptic balance of SAC and bipolar input onto ooDSGC dendrites, are both important for DS responses (Ding et al., 2016; Vlasits et al., 2016; Poleg-Polsky and Diamond, 2016; Sethuramanujam et al., 2016; Sethuramanujam et al., 2017). The finding that Megf10 mutants have DS tuning phenotypes suggests that local synaptic arrangements are indeed important for the DS computation. More broadly, this finding shows that the developmental mechanisms we describe here are important for enabling circuit function, raising the possibility that other circuits throughout the retina and CNS may use similar developmental mechanisms to establish their functional connectivity.

All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee of Duke University. The animals were maintained under a 12 hr light-dark cycle with ad lib access to food and water. Retinas from adult (4–8 weeks old) Megf10^–/– mutant mice and wild-type control mice with same genetic background were used for experiments performed on the multielectrode array (MEA). Animals were dark-adapted overnight prior to the experiment.

For this study, the following transgenic and mutant mouse lines were used: (1) Megf10^{tm1b(KOMP)Jrs} (Kay et al., 2012), referred to as Megf10^– or Megf10^lacZ; (2) Ptf1a^tm3Cvw (Krah et al., 2015), referred to as Ptf1a^flox or (when crossed to Cre mice) Ptf1a-cKO; (3) Isl1^tm(cre)Sev (Yang et al., 2006), referred to as Isl1^Cre; (4) Tg(Hlxb9-GFP)1Tmj/J (Trenholm et al., 2011), referred to as Hb9-GFP; (5) Chat^tm2(cre)Lowl (Rossi et al., 2011), referred to as Chat^Cre; (6) Tg(Six3-cre)69Frty (Furuta et al., 2000) referred to as Six3-Cre; (7) Kcng4^{tm1.1(cre)Jrs} (Duan et al., 2014) referred to as Kcng4^Cre; (8) Tg(Drd4-EGFP)W18Gsat (Huberman et al., 2009), referred to as Drd4-GFP; (9) Tg(Gjd2-EGFP)JM16Gsat, referred to as Gjd2-GFP; (10) Tg(Gad1-EGFP)G42Zjh, referred to as Gad1-GFP. Two Cre reporter strains were used that express membrane-targeted green fluorescent protein (mGFP) upon Cre recombination: (1) Gt(ROSA)26Sor^{tm4(ACTB-tdTomato,-EGFP)Luo}, also known as mT/mG (Muzumdar et al., 2007); (2) Rosa26^fGFP (Rawlins et al., 2009). An additional Cre reporter strain was used that expresses tdTomato fluorescent protein upon Cre recombination: Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze} (Madisen et al., 2010). See Key Resources table for repository stock numbers where applicable.

To produce Megf10^flox mice, Megf10^{tm1a(KOMP)Jrs} mice (Kay et al., 2012) were crossed to germline Cre strain B6;SJL-Tg(ACTFLPe)9205Dym/J, thereby generating a functional allele (also known as Megf10^tm1c) in which exon four was flanked by loxP sites.

HEK293T cells were obtained from, validated by, and mycoplasma tested by ATCC. The cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) with 10% bovine growth serum, 4.5 g/L D-glucose, 2.0 mM L-glutamine, 1% Penicillin/Streptomycin in 10 cm cell culture dishes. Cells were passaged every 2–3 days to reach confluence. Before splitting, culture media were removed and Dulbecco’s phosphate-buffered saline (D-PBS) was used to rinse cell layers as well as removing residual serum. Cells were detached from dish with 4 ml of 0.05% Trypsin and incubated at 37°C until cell layer is dispersed (about 5 min). Equal volume of complete culture media was added to the dish to inhibit protease activity. The suspension was centrifuged at 200 x g for 5 min. Supernatant was aspirated and the cells were suspended with appropriate amount of media and plated (1:4-1:8). Cells used for experiments were passaged no more than 10 times. Cell stocks were stored as 2 million cells per vial in complete culture media with 10% DMSO in liquid nitrogen.

To study SAC anatomy during embryonic stages, Isl1^Cre was crossed to lox-stop-lox-mGFP Cre reporter mice (mT/mG or Rosa26^GFPf; see Key Reagents) to generate Isl1^mG animals. Timed-pregnant dams were sacrificed at E16 and eyes collected from embryos (n = 11 mice from three litters). Tissue was processed as described for postnatal eyes, except fixation time was 60 min. Cross-sections were stained with anti-GFP to reveal the morphology of Isl1^mG-expressing neurons, as well as Sox2 to distinguish Isl1^mG-positive SACs from RGCs. (All cells shown in Figure 2B–G were confirmed to be SACs by Sox2 co-labeling.) In combination with these markers, anti-internexin staining was used to assess orientation of primary dendrites. Location and/or presence of the IPL was determined using Hoechst nuclear staining, which revealed cell body-free neuropil regions, and/or by Isl1^mG labeling of neuronal processes, which filled these neuropil regions (Figure 2—figure supplement 2). We assessed anatomy of mGFP⁺ migrating SACs in the ONBL, as well as SACs in the INBL that were concluding their migration. Morphology of ON SACs in the GCL could not be discerned due to Isl1 expression by RGCs (Figure 2A,B), but because displaced amacrine cells pause at the INL-IPL border before crossing to the GCL (Chow et al., 2015), the population of cells available to analyze might have included both ON and OFF SACs.

To measure the orientation of primary dendrites at E16 and P1, the angle ROI function in ImageJ was used. This function outputs an angle degree measurement (absolute value) between two line segments. The first line segment of the angle was drawn to follow the trajectory of the internexin⁺ primary dendrite; the endpoint was at the cell body. The second line segment of the angle was a plumb line to the IPL (i.e. it was drawn to intersect the IPL at ~90˚). As such, dendrites oriented exactly toward the IPL were assigned an angle of 0˚. At E16 the IPL was occasionally not present yet; in this case the second line segment was a plumb line to the inner limiting membrane. In cases where the internexin⁺ dendrite curved, we traced the initial trajectory of the dendrite as it emerged from the cell body. Dendrites were classified as projecting (1) towards the IPL; (2) toward the ONBL; or 3) tangentially, according to the angle scheme delineated in Figure 2—figure supplement 1E. Image stacks were randomly selected for analysis from a larger library of images; within each selected stack every SAC was traced. Isl1^mG and Sox2 were used to confirm the SAC identity of each measured cell, as well as the trajectory of the internexin⁺ dendrite.

The homotypic nature of SAC soma-layer contacts was investigated by imaging single Chat^mG-labeled OFF SACs in mice also carrying a single copy of the Megf10^lacZ allele (Figure 2I,J). Anti-βgal staining was used to reveal the full SAC population, including arbors. En-face images were captured in Z-stacks spanning the INL and IPL; slices corresponding to each layer were separately Z-projected for display in Figure 2 and Figure 2—figure supplement 3. To quantify the frequency of SAC-SAC contacts, we used Z-stacks from P1 tissue to examine the trajectory and termination site of each dendritic tip in three dimensions. The fraction of Chat^mG-labeled dendrites terminating on the βgal-positive soma or arbor of a neighboring SAC was quantified. To be counted, the putative contact needed to be confirmed in a single Z-stack slice; where necessary, 3D reconstructions and orthogonal views were used to confirm contact.

We also performed the same analysis on Z-stacks in which one channel had been flipped about the horizontal and vertical axes. This served as a negative control to measure the frequency with which GFP and βgal arbors interact by chance, given their density and geometry in the P1 retina. Sample sizes are given in main text and in Figure 2—figure supplement 3.

Single SACs labeled in Chat^mG and Chat^mG;Megf10^–/– mice were morphologically assessed in cross-sections. GFP signal was amplified with anti-GFP antibody staining. All GFP⁺ SACs on any given slide were imaged and analyzed, to avoid cell selection bias, with the exceptions of: 1) cells severed by the cryosectioning process; 2) cells with arbors that could not clearly be distinguished from those of their neighbors; 3) cells in the far retinal periphery, where sections were oblique to retinal layers, obscuring IPL strata. In experiments analyzing Megf10 mutants, littermates were always used as controls to avoid complications arising from the fact that the precise state of retinal development at the time of birth might vary from litter to litter.

A cell was scored as innervating the IPL if it ramified branched dendrites within the neuropil. Dendrites that entered the neuropil but did not branch or stratify (e.g. Figure 6D) were not sufficient. A cell was scored as projecting to the soma layer if arbors emanating from the cell soma or primary dendrite terminated or arborized in the INL (for OFF SACs) or GCL (for ON SACs). The arbor was required to be ~≥1 cell radius in length (i.e. small fine arbors were not counted). One other important exception that was not counted: We observed that many SACs at young ages had single unbranched arbors extending ~180˚ away from the IPL (e.g. Figure 2J,K – all four cells have such arbors, even the ones that do not project towards neighboring SAC somata). These processes were not counted for two reasons. First, their trajectory was such that they were unlikely to join the soma-layer dendrite network or contact neighboring somata. Second, these 180˚ arbors were sometimes still present in P5 SACs (Figure 2—figure supplement 2) and therefore they did not appear to be subject to the same developmental regulation as tangentially-directed arbors (Figure 2L). This observation suggests they are fundamentally different, and likely serve a different (as yet uncharacterized) purpose. No obvious difference in their frequency was observed between wild-type and Megf10 mutants.

To produce graphs in Figures 2L, 6E and 8F, the fraction of cells making ectopic projections – either to the soma layer or to inappropriate IPL sublayers – was calculated for each genotype and each time point. To determine whether a GFP⁺ IPL arbor was located in normal or abnormal IPL strata, Megf10:βgal was used as a counterstain. Chat^Cre was rarely expressed in OFF SACs at P0, making it difficult to obtain large sample sizes at this age. For this reason, and because soma-layer projection frequency did not appear to differ much between P0 and P1, the data from each time point was pooled for analysis of Megf10 litters.

Sample sizes for Figure 2L: P0, n = 25 OFF, 63 ON; P1, n = 51 OFF, 79 ON; P2, n = 46 OFF, 55 ON; P3, n = 33 OFF, 49 ON; P5, n = 15 OFF, 26 ON; P7, n = 23 OFF, 34 ON. Data were from four litters of mice, each of which was assessed at no less than two of these time points.

Sample sizes for Megf10; Chat^mG experiments (Figure 6E; Figure 8F): Megf10 heterozygous littermate controls: P0/1, n = 11 OFF, 25 ON; P2, n = 25 OFF, 23 ON; P3, n = 17 OFF, 22, ON; P5, n = 16 OFF, 16 ON. Megf10 mutants: P0/1, n = 6 OFF, 25 ON; P2, n = 14 OFF, 20 ON; P3, n = 34 OFF, 41 ON; P5, n = 48 OFF, 54 ON. Data were from two litters of mice.

For the adult data reported in Figure 8F,a different procedure was used; see ‘Quantification of Mosaic Spacing Phenotypes’ section below.

P1-P2 retinas carrying Megf10^lacZ and Hb9-GFP were co-stained for βgal and GFP. RGCs with dendrites that co-fasciculated with βgal-positive IPL strata were counted. Cells that projected to βgal-positive regions, but also filled non-SAC-projecting IPL regions, were not counted as co-fasciculated. To judge co-fasciculation, we used two criteria: 1) inspection of dendrite anatomy across the confocal stack; 2) fluorescence profiles of GFP and βgal channels across IPL (see next section below). Examples of cells falling into each category are provided in Figure 1 and Figure 1—figure supplement 2. See Results for sample sizes.

Images of retinal cross sections were processed in ImageJ. A vertical ROI (12.5 µm wide) was drawn to perpendicularly bisect the IPL strata, from the edge of the INL to the edge of the GCL. IPL stratification levels were reported as percentage of IPL width. Intensity was calculated for each pixel along the length of the ROI as an average across its width. Background (minimum pixel value) was subtracted; then, all pixel intensity values were normalized to the maximum value of that ROI. Location of fluorescent peaks was calculated as the pixel with maximum intensity; if multiple pixels had the same intensity the peak was defined as the center of the plateau. The procedure was typically performed on single confocal optical sections, but for some P1-2 cells, which have much smaller arbors, it was necessary to use a maximum-intensity projection of a small number of slices in order to fully capture dendrite morphology.

For BC5-BC7 arbor distance measurements (Figure 9F), distances as percentage of total IPL width were compared by one-way ANOVA/Tukey’s post-hoc test. n = 14 measurements from two control mice; n = 7 normal IPLs, 11 SAC clumps, 11 SAC gaps from 3 Megf10^–/– mice.

The MEGF10-ΔICD-GFP construct was reported previously (Kay et al., 2012), which was originally made from pUbC-MEGF10-GFP (Addgene #40207). It encodes a version of MEGF10 in which the cytoplasmic domain is truncated after the 9^th amino acid and replaced by GFP. Inclusion of those nine amino acids was necessary to achieve plasma membrane localization. For this study, it was subcloned into the pEGFPN3 plasmid, containing the CMV promoter, to make pCMV-MEGF10-ΔICD-GFP.

To make the MEGF10-ΔICD-Flag construct, Megf10 (truncated after the 9^th intracellular domain amino acid as above) was PCR amplified from pUbC-MEGF10-GFP vector using M10flag_Fwd forward primer and Cyto9_flag_Rev1 reverse primer. Resulting PCR products were digested with NotI and AscI restriction enzymes and ligation cloned into pEGFPN3 vector linearized with corresponding restriction enzymes.

Statistical analysis was performed using GraphPad Prism software (anatomy/development studies) or using custom JAVA-based software and MATLAB software (physiology studies). This software is available, together with the primary data it was written to analyze, at a public repository (Roy and Field, 2017; https://github.com/Field-Lab/megf10-dstuning; copy archived at https://github.com/elifesciences-publications/megf10-dstuning). Statistical tests used for each experiment are given in the appropriate Materials and methods section above, and/or in the figure legends. Sample sizes for each experiment are given in the appropriate Methods section above or else in the Results. p-Values (α = 0.05) are given in figure legends, or in the Results if no figure is shown. Error bars are defined in figure legends. Exact p-values are reported unless the value was less than 1.0 × 10⁻⁷.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

For financial support we thank the National Eye Institute (EY024694 to JNK; EY026344 to TAR; EY024567 to GDF; EY5722 to Duke University); Pew Charitable Trusts, E Matilda Ziegler Foundation, McKnight Endowment Fund for Neuroscience, Alfred P Sloan foundation (JNK); the Whitehall Foundation (GDF); and Research to Prevent Blindness Unrestricted Grant (Duke University). We thank Sidney Kuo and Greg Schwartz for suggesting use of the Gjd2-GFP mouse line; members of the Kay lab and Cagla Eroglu for their comments on the manuscript; Megan Stogsdill and Ari Pereira for mouse colony support; and X Duan and J Sanes (Harvard) for Kcng4^Cre mice.

Animal experimentation: All animal experimental procedures were reviewed and approved by the Institutional Animal Care and Use Committee of Duke University (protocol A005-16-01).

© 2018, Ray et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.