Pathfinding in a large vertebrate axon tract: isotypic interactions guide retinotectal axons at multiple choice points

Introduction

To generate a functional nervous system, widely-separated groups of neurons must form appropriate connections. Each newborn neuron extends an axon tipped by a motile growth cone, which navigates through sequential “choice points” to eventually reach its target (Dickson, 2002; Guan and Rao, 2003; Tessier-Lavigne and Goodman, 1996). The growth cone has two potential sources of guidance information. First, it may fasciculate with “pioneer” axons—defined here as earlier-born axons that share its path—using cell-adhesion molecules. Second, it may sense secreted or cell-surface guidance ligands produced by other cells in the brain, acting at long or short range, respectively. Most recent work has focused on guidance ligands; genetic screens and biochemical purification strategies have identified many ligand/receptor pairs used by navigating growth cones (Dickson, 2002; Guan and Rao, 2003; Tessier-Lavigne and Goodman, 1996).

There is long-standing evidence, however, that pioneer-follower interactions can play necessary roles in guidance (Lopresti et al., 1973; Bate, 1976; Raper et al., 1983; Raper et al., 1984; Kuwada, 1986; Klose and Bentley, 1989; Ghosh et al., 1990; Pike et al., 1992; Hidalgo and Brand, 1997; Jhaveri and Rodrigues, 2002; Williams and Shepherd, 2002). In other cases, pioneers are not required (Keshishian and Bentley, 1983; Holt, 1984; Eisen et al., 1989; Cornel and Holt, 1992), or only facilitate followers’ guidance (Chitnis and Kuwada, 1991; Pike et al., 1992; Bak and Fraser, 2003). Nearly all of these studies were done in simple systems where a few pioneer axons could be identified, and where required roles for pioneers were tested by ablation. Furthermore, these pioneers were usually heterotypic: of a different neuronal type than the followers, with different origins and targets, so that they could only guide the followers through one leg of their journey. Thus it is not clear (1) whether the role of pioneers is generalizable to more complex systems; (2) what are the sufficient functions of pioneers, for instance if they take abnormal paths; or (3) what roles are usually played by isotypic interactions (between axons from the same neuronal type).

In vertebrates, most axon tracts are built on a large scale. They typically comprise thousands of axons, all with the same origin and target, and often develop over an extended period as new neurons are added. This raises the possibility that vertebrate axons could be guided by isotypic interactions, either between earlier and later axons (pioneer-follower interactions), or between axons that grow at the same time (community interactions). Since axons in the same tract share both origin and target, isotypic interactions could act at multiple choice points throughout their pathway. While there is often an unspoken assumption that vertebrate axons indeed use pioneers for guidance, there have been only a few direct experimental tests of this hypothesis (Holt, 1984; Eisen et al., 1989; Ghosh et al., 1990; Cornel and Holt, 1992; Bak and Fraser, 2003). Here we use the zebrafish retinotectal system to study this fundamental cellular mechanism in axon guidance. We first ask if isotypic pioneer-follower and community interactions are important in the development of a large vertebrate axon tract. We next use a pioneer replacement strategy to test whether pioneers are sufficient to affect followers, and to assess the relative importance of axon-axon interactions compared to guidance receptor signaling.

The retinotectal projection is one of the best-studied vertebrate tracts, whose formation has been studied extensively (Erskine and Herrera, 2007). Retinal ganglion cell (RGC) axons must navigate out of the eye, across the optic chiasm, and dorsally through the optic tract to reach the optic tectum. While retinal axons parallel the tract of the postoptic commissure (tPOC) after crossing the optic chiasm, they make at most fleeting contacts with tPOC axons (Burrill and Easter, 1995), and embryological manipulations that remove tPOC axons do not affect retinal axon guidance (Cornel and Holt, 1992). Thus, retinal axons do not require heterotypic pioneers. Despite the extensive retinotectal literature, there has been only one functional test of whether retinal pioneer axons might guide later retinal axons (i.e., through an isotypic interaction). In Xenopus, dorsocentral RGCs are the first to send out their axons. There was no effect on the guidance of later retinal axons when heterochronic transplants were used to delay the outgrowth of dorsocentral RGCs (Holt, 1984), suggesting that retinal pioneers are not required. Here we reexamine the role of retinal pioneers in guidance both within the eye and after exiting it. We use genetic and embryological manipulations both to remove early RGCs, and to replace them with cells lacking the Robo2 guidance receptor. We find that isotypic pioneers in fact play multiple roles during the formation of this archetypal vertebrate tract.

Materials and Methods

Results

To test the role of pioneer neurons, we devised a method to selectively remove early-born RGCs. These cells’ differentiation starts ventroanteriorly at 27–28 hours postfertilization (hpf), spreading circumferentially to fill the central retina by ~40 hpf (Hu and Easter, 1999; Masai et al., 2005). After this, new RGCs are added in successive peripheral rings (Fig. 1A). The RGC-specific transgene Tg(isl2b:GFP)^zc7 turns on by 30hpf in ventroanterior retina, reliably labeling all or the vast majority of RGCs (Fig. 1B; Pittman & Chien, in preparation). To remove early RGCs, we used a translation-blocking antisense morpholino oligonucleotide (ath5MO) to knock down function of ath5 (formally, atoh7—Zebrafish Information Network), a bHLH transcription factor expressed specifically in the eye and required cell-autonomously for RGC differentiation (Kay et al., 2005). ath5 mutants completely lack RGCs (Brown et al., 2001; Kay et al., 2001), and their RGC progenitors seem instead to take on primarily bipolar cell fates (Kay et al., 2001). We reasoned that ath5MO should have a similar, but transient, effect since morpholinos’ efficacy wanes as their concentration is diluted by embryonic growth (Nasevicius & Ekker, 2000). Indeed, we found that ath5MO causes a loss of early-born central RGCs, but allows later-born peripheral RGCs to differentiate (Fig. 1C). In isl2b:GFP embryos injected with high-dose (5 ng) ath5MO (Fig. 1C; Fig. 2E; Supplementary Figure 1), RGCs were found exclusively in peripheral retina at 6 days postfertilization (dpf)—over 4 days after central RGCs normally form—showing that differentiation is prevented rather than merely delayed. We next injected an ath5MO dose series, monitoring RGC differentiation in live embryos from 33 to 57 hpf. In control embryos, GFP+ RGCs were always present by 33hpf. With successively higher doses, the first RGCs appeared successively later, until with 5ng ath5MO, RGCs did not appear until ≥42hpf in 90% of embryos (Fig. 1D). Thus, varying the dose of ath5MO controls the time at which the first RGCs differentiate.

We then tested whether removing early-born RGCs affects the pathfinding of later axons (Fig. 2). In uninjected controls, the retinotectal projection is mature at 5 dpf, with both optic tecta strongly innervated (Fig. 2B; n=12/12). However, when central RGCs were removed by injecting 5ng ath5MO, axons of late-born RGCs failed to exit from the eye into the brain (Fig. 2C; n=19/20). We hypothesized that this failure might be related to the time of first RGC differentiation. Therefore we injected isl2b:GFP embryos with 2–5 ng ath5MO (Fig. 1D), plotting retinal exit at 5 dpf against the time of first RGC differentiation (Fig. 2D). Axons projected from eye to tectum in 100% of control larvae (uninjected, n=12; 3 ng control MO, n=12) and in 83% (45/54) of those morphants (MO-injected embryos) whose first RGCs were born by 42hpf. In contrast, axons exited the eye in only 5% (1/20) of morphants whose first RGCs were born after 42hpf. Instead, axons from peripheral RGCs grew within the RGC layer of the eye without entering the intraretinal portion of the optic nerve (Fig. 2E, Supplemental Movie 1, Supplemental Figure 1). A physical barrier is not likely the cause: ath5 morphants have normal retinal lamination, and the presumptive optic nerve can still be distinguished (Supplementary Figure 1). To test for changes in the environment of the presumptive optic nerve, we stained for Pax2, which labels the surrounding primitive glia (Macdonald et al., 1997), and for netrin1a, a marker of the optic nerve and optic nerve head (Fig. 3). Both markers appeared normal in ath5 morphants. Taken together, these results strongly suggest that an isotypic pioneer-follower interaction is required for later RGCs’ axons to exit the eye, and further that there is a critical period before 42 hpf.

We next tested whether this pioneer-follower interaction is sufficient as well as necessary, by resupplying RGCs to ath5 morphants. We targeted the presumptive eye field in blastula-stage transplants from isl2b:mCherry-CAAX donors into isl2b:GFP hosts injected with high-dose ath5MO (Fig. 4A). Since ath5 acts cell-autonomously (Kay et al., 2005), we expected that donor (non-morphant) RGCs would differentiate without affecting host cells. Indeed, while host RGCs were still restricted to peripheral retina, donor RGCs were found in central retina, and usually sent axons out of the eye to the tectum (Fig. 4B, C). Of 32 eyes with mCherry+ donor axons reaching the tectum, 24 (75%) showed rescue of retinal exit, with GFP+ host axons reaching the tectum (Fig. 4B). Retinotectal topography appeared unaffected: donor axons from central retina projected to central tectum, and host axons from peripheral retina projected to peripheral tectum (Fig. 4B). Furthermore, imaging within chimeric eyes showed that late-born host axons appeared to fasciculate with donor axons (Fig. 4C–C″). Therefore, early RGCs are both necessary and sufficient to guide later axons out of the eye.

In a second series of experiments, we tested whether, after retinal axons exit the eye, interactions between them might affect guidance to the tectum. To do this, we used cell transplants to mix retinal axons of different intrinsic pathfinding potential. In astray (ast) mutants, retinal axons make drastic pathfinding errors because they lack the Slit receptor, Robo2 (Fricke et al., 2001). Transplantation of eye primordia yields strictly eye-autonomous pathfinding defects, showing that Robo2 is required in the eye and not the brain (Fricke et al., 2001). Here, we performed cell transplants from WT or ast donors labeled with the RGC-specific transgene brn3c:GFP (Xiao et al., 2005) into nontransgenic WT or ast hosts (Fig. 5A). We then assayed the pathfinding of GFP-positive axons (which bear the donor genotype), quantitating the results by counting the number of classes of pathfinding errors made by GFP-positive donor retinal axons, yielding a score for each host from 0 for WT to 8 for strong ast (Supplementary Figure 2; see Materials and Methods).

As expected, WT donor axons in WT hosts made no pathfinding errors (Fig. 5B, B′; 0.0±0.0 errors, mean ± s.e.m.). Surprisingly, WT axons transplanted to ast hosts always made errors, which occurred at several different positions (Fig. 5C, C′; 2.71±0.40 errors, p<0.0001). Presumably, despite expressing functional Robo2, they were misguided by neighbors that lacked Robo2. Donor ast axons made many errors in ast hosts (Fig. 5D, D′; 7.93±0.07 errors), similar to axons in nontransplanted ast embryos (Supplementary Figure 2). In contrast, ast axons transplanted into WT embryos navigated far better (Fig. 5E, E′; 2.86±0.31 errors, p<0.0001). Thus, even lacking functional Robo2, their pathfinding behavior was significantly rescued by WT neighbors. These effects are not transgene-dependent: using isl2b:GFP instead of brn3c:GFP yielded qualitatively similar results (data not shown). Strikingly, then, Robo2 acts nonautonomously in cell transplants, unlike its autonomous behavior in eye transplants (Fricke et al., 2001). As a receptor, Robo2’s direct effects must take place in the growth cone that expresses it, but this does not mean that indirect effects cannot affect other axons. The strong effect of the host axons’ genotype on donor axon behavior shows that despite the clear importance of Slit-Robo signaling, its role in individual axons can be overridden by axon-axon interactions.

Is this a pioneer-follower effect? Since these chimerae contained far more host than donor cells, the host genotype could be affecting donor axon behavior because most pioneer axons are likely to be host-derived. If so, replacing pioneer RGCs with cells of a different genotype should alter pathfinding by later axons. We tested this prediction by injecting high-dose ath5MO into brn3c:GFP hosts to remove all early RGCs, replacing the pioneer RGCs using transplants from nontransgenic donor embryos, then at 5 dpf assaying the pathfinding of GFP-positive axons, which are host-derived and therefore “follower” axons (Fig. 6A). WT followers made no errors when transplanted pioneers were WT (Fig. 6B, B′; 0.0±0.0 errors). However, when pioneers were replaced with ast cells, WT followers made more errors (Fig. 6C, C′; 1.72±0.31 errors, p<.002). Thus, misguided pioneers indeed influence WT followers. As expected, ast followers made many errors when pioneers were ast (Fig. 6D, D′; 5.96±0.33 errors). When pioneers were replaced with WT cells, ast follower axons made fewer errors (Fig. 6E, E′; 4.79±0.24 errors, p<0.01). Thus, follower axons’ pathfinding is partially influenced by the pioneers’ genotype, although clearly the followers’ genotype remains the predominant influence.

Discussion

We have shown that early-born RGCs are both necessary and sufficient for retinal axons to exit the eye. This is a somewhat surprising result in light of an elegant study in Xenopus (Holt, 1984), which did not find an important role for retinal pioneers. We believe the difference derives from our ability to use ath5MO injection to completely prevent the development of all central RGCs. In contrast, the heterochronic transplants used in the Xenopus study only affected the dorsal half of central retina (the source of pioneers in Xenopus), and furthermore only delayed the development of retinal pioneers by 9–20 hours, rather than removing them entirely. A previous study in zebrafish had proposed that a small group (<10) of early retinal growth cones were pioneers based on their significant separation from the next growth cones to follow (Stuermer, 1988). It is clear from our ath5 morpholino injections that these few growth cones are not functionally pioneers, in the sense of being necessary for later axons’ guidance. Instead, we find that far more cells must be removed to prevent retinal exit: the RGCs born by 42 hpf, the critical stage (Fig. 1), number about 600 (Supplemental Movies 2, 3). This is strikingly different from invertebrate systems where ablation of a few pioneers is sufficient to perturb follower axons.

After retinal exit, our transplant experiments show that isotypic interactions between retinal axons are important for three further choice points that require Robo2 function (Fricke et al., 2001; Hutson and Chien, 2002; M. Hardy and C.-B.C., unpublished data). Misrouting in the optic chiasm leads to chiasm defasciculation (Fig. 6C); misrouting in the ventral optic tract leads to telencephalic or ventral hindbrain projections (Figs. 5C and 6C); and misrouting in the dorsal optic tract leads to aberrant crossing in the habenular or posterior commissures (Fig. 5C). At all three choice points, mutant axons can lead wildtype neighbors into error. Thus, axon-axon interactions function throughout the entire course of the retinotectal projection.

Several questions remain for future studies on axon-axon interactions. What distinguishes pioneers from followers? We know of no molecular markers that distinguish these RGC populations—for instance, they both express robo2 (Campbell et al., 2007), so the only difference may be the time and position of their birth. How do pioneers interact with followers within the eye? Most simply, later axons may fasciculate with early axons, as seen in Fig. 4C–C′; future experiments will test whether disrupting cell adhesion molecules leads to axon guidance errors within the retina, as in other vertebrates (Brittis et al., 1995; Leppert et al., 1999; Ott et al., 1998; Zelina et al., 2005). However, we cannot formally exclude other possibilities. For instance, early RGC cell bodies might secrete an attractant that draws later axons to the optic nerve head. Interestingly, WT>ath5 morphant transplants in which host axons remained trapped within the eye tended to have fewer donor axons on the tectum (data not shown), presumably reflecting fewer donor RGCs, which might provide an insufficient level of attraction. What underlies the apparent critical period before 42 hpf? Both timing and spacing are possible explanations. Ligands implicated in retinal exit (Birgbauer et al., 2000; Deiner et al., 1997; Kay et al., 2005; Kolpak et al., 2005; Li et al., 2005; Thompson et al., 2006) or their receptors, might be expressed only transiently during this period, so that RGCs born after 42 hpf would lack appropriate exit signals. Alternately, early-born RGCs may be close enough to find the optic nerve head, while later RGCs are simply too far away to sense it.

While most pioneer experiments have studied heterotypic axon interactions, there have been a few studies of isotypic interactions. In grasshopper, Myers and Bastiani (1993) found that the growth cone of the identified Q1 neuron interacts strongly with its contralateral homolog, and that ablation of one Q1 often leads to midline stalling of the other Q1’s growth cone. In the zebrafish, Bak and Fraser (2003) found that pioneer and follower growth cones in the postoptic commissure (POC) display characteristic morphology and kinetics (spread/slow and narrow/fast, respectively). After laser ablation of pioneer growth cones, followers appeared to take their place and behave like pioneers. However, POC pioneer ablation did not have any effect on the pathfinding of follower axons, and indeed we do not know of previous studies in vertebrates showing guidance by isotypic pioneers.

Here, we found that isotypic interactions after retinal exit help to guide retinal axons to the tectum. We were able to test the role of axon-axon interactions in a new way: by replacing rather than ablating RGCs, we tested sufficiency rather than necessity. Previous studies on pioneers found that their ablation prevents normal pathfinding by followers (e.g., Raper et al., 1983; Klose and Bentley, 1989; Pike et al., 1992; Whitlock & Westerfield, 1998); here, we used zebrafish transplants to test how misrouted mutant axons affect WT axons, and vice versa. We found that ast host axons can misroute WT donor axons, while WT host axons rescue ast donors to a large degree. This allowed us to compare the relative importance of different guidance mechanisms; we conclude that axon-axon interactions can be just as important as cell-autonomous Slit-Robo signaling. As well as showing a significant role for pioneer-follower interactions, our data suggest an even more important role for peer interactions between late retinal axons. An ast host axon in a WT>ast;ath5MO transplant (Fig. 6E, E′) misroutes more often than an ast donor axon in an ast>WT transplant (Fig. 5E, E′), perhaps because a larger fraction of RGCs are ast in the former case. What might distinguish pioneer-follower from peer-peer interactions? We do not necessarily expect different molecular interactions, but instead, differences in temporal and spatial proximity. Peers grow out at the same time, while peripheral followers grow out significantly later than pioneers. New retinal axons grow at the surface of the brain, so that pioneers are gradually buried underneath. Thus, while peers can interact directly, late followers will contact pioneers only indirectly, with several degrees of separation.

Our results are complemented by a recent study (Gosse et al., 2008) which showed that even a single RGC transplanted into lakritz/ath5 mutant hosts can sometimes navigate successfully to the tectum. Successful cases of tectal innervation showed a bias for central RGCs; furthermore, axons of single RGCs transplanted to peripheral retina often remained trapped within the eye. Tectal innervation appeared to be a rare event, as a large number (~5000) of transplants had to be performed; in similar experiments transplanting a few RGCs into lakritz or ath5 morphants, we also find that axons rarely reach the tectum (AJP and CBC, unpublished data). Overall, these data are consistent with our model that a large population of central RGCs are required for peripheral axons to exit the retina.

Our transplant paradigms create artificial situations in which axon-axon interactions are at war with ligand-receptor signals. For instance, a WT axon surrounded by ast axons may be shown the correct path by signals from the brain, but tugged off-course by its neighbors. During normal retinotectal development, in contrast, all axons are WT. Not only does every axon recognize the correct path to its target, but so do all its predecessors and neighbors. Therefore, signals from both outside the tract (guidance ligands) and within (axon-axon interactions) should act in coordination to produce the highly stereotyped, precise formation of this vertebrate axon tract. Since interaction (fasciculation) between isotypic axons is widely observed, we propose that this coordinated strategy is likely used throughout the development of complex vertebrate nervous systems.

Finally, we point out significant implications for the interpretation of mutant axon guidance phenotypes. For instance, imagine a tract that uses two guidance receptors, A and B, to sense partially redundant sets of guidance signals. Suppose that in the absence of fasciculation, knockout of A would cause 30% of the axons to misroute, while knockout of B would cause 10% errors. Since most axons still navigate correctly, axon-axon interactions might in fact reduce the A mutant phenotype to 10%, and the B mutant phenotype to undetectable levels. Thus, genetic redundancy may act not only at the level of single axons, but also at the level of entire tracts.

Supplementary Material