To investigate the guidance cues and growth cone behaviors involved in the initial outgrowth of axons from an early developing brain nucleus, we studied the formation of the zebrafish MLF, one of the first axon tracts to develop in the zebrafish central nervous system (CNS; Figure 1a) 32. The MLF arises from bilateral clusters of neuronal cell bodies, called the nucMLF, that lie in the ventral midbrain. At the onset of MLF formation (16 hours post fertilization (hpf)), the nucMLF consists of 6–8 neurons and by 24 hpf the cluster has grown to approximately 30–35 neurons 3334. During the initial wave of axonogenesis (16–30 hpf), unipolar nucMLF cells each extend an axon caudally along the ipsilateral ventral neural tube. The caudal-most nucMLF cell extends the leading MLF axon into the hindbrain and then the more rostrally positioned cells extend axons that fasciculate with the leading axon 35. We have previously reported that Semaphorin3D, which is expressed both rostral and medial to the nucMLFs, plays a role in repelling nucMLF axons in the caudal direction 3637; however, nothing is known about other extracellular cues involved in guiding the initial pathfinding decisions of nucMLF axons. Here, we further investigate guidance mechanisms that control how nucMLF axons initially extend in the correct direction and fasciculate with the leading nucMLF axon.

We first characterized the pattern of nucMLF axon outgrowth by examining fixed embryos at stages between 16 and 24 hpf. To specifically visualize nucMLF neurons, we used a transgenic line in which a pitx2 promoter construct drives green fluorescent protein (GFP) expression (Tg(pitx2c:gfp)) in nucMLF neurons (Figure 1b–e; see Materials and methods). We verified that all nucMLF neurons are labeled by double labeling embryos with antibodies against ZN-12, a marker of nucMLF neurons, and GFP (Figure 1b). At the onset of nucMLF axonogenesis, cell bodies each extended one axon in the direction of the caudal-most nucMLF cell body (Figure 1c–e, asterisk), which extended its own axon into the hindbrain, parallel to the ventral midline. In their initial trajectory, the earliest nucMLF axons typically did not contact neighboring axons since their cell bodies were positioned at some distance from one another, but later as these axons approached the caudal-most nucMLF cell and as the number of nucMLF cells increased, axon-axon contact was prevalent. Interestingly, the axons did not appear to extend into the tissue surrounding the nucMLF. Rather, they appeared to extend directly toward the caudal-most nucMLF cell body and eventually converged on a point just rostral to this cell (Figure 1c–e, arrow). In our subsequent analysis, we will refer to this point as the convergence point. Beyond the convergence point, MLF axons extend into the hindbrain as a tight fascicle.

To better understand nucMLF growth cone behaviors and gain insight into potential guidance factors that might control initial axon direction, convergence, and fasciculation, we performed in vivo time-lapse imaging of these neurons. We reasoned that specific growth cone behaviors (that is, retraction, pausing, branching, and so on) would reflect underlying molecular mechanisms responsible for these behaviors. To visualize growing nucMLF axons in vivo, we removed the yolk cell from Tg(pitx2c:gfp) embryos and imaged the nucMLF from the ventral side of the embryo proper (see Materials and methods for details), beginning at approximately 16 hpf. Consistent with the fixed staging series, each nucMLF neuron extended one axon from its caudal side, which grew towards the convergence point. Once growth cones contacted neighboring MLF axons, they maintained these contacts and extended along the axon en route to the convergence point. This behavior was highly conserved and these interactions appeared to expedite growth towards the convergence point, suggesting that axon-axon contacts may be important for MLF formation (n = 11 embryos, 32 growth cones; Figure 2a and Additional file 1). In some cases, axons extended from cell bodies that were positioned up to 25 μm from their nearest neighbor, and, therefore, the majority of their path towards the convergence point was not along a neighboring axon. This observation suggested that axon-axon interactions were not the only factor required for proper convergence. Indeed, we found that if a wild-type growth cone extended aberrantly into the surrounding tissue, away from the convergence point, it was able to retract and redirect its growth in the appropriate direction without contact with other axons (Figure 2b and Additional file 2). This result suggests that nucMLF axons are guided by surrounding tissues, perhaps by repulsive cues that funnel them towards the convergence point.

Given the significant axon-axon interactions we observed in the live imaging experiments, we asked whether TAG-1, a GPI-linked Ig superfamily CAM known to mediate axon-axon interactions 19, was critical for the initial growth and convergence of MLF axons. tag-1 mRNA is expressed by a subset of neurons in the zebrafish brain, including the nucMLF (Figure 3a) 38. We examined the expression of TAG-1 protein during initial stages of MLF development (16–24 hpf), and found that it is expressed on the nucMLF axons and weakly on the nucMLF cell bodies throughout these stages (Figure 3b–d, and data not shown). In addition, TAG-1 is expressed on axons from the nucleus of the tract of the postoptic commissure (nucTPOC) that begin to extend past the lateral side of the nucMLF around 19–20 hpf, and on a couple of cells lateral to the nucMLF, but not otherwise in the surrounding neuroepithelium. We injected Tg(pitx2c:gfp) embryos at the 1–4 cell stage with a translation blocking TAG-1 morpholino antisense oligonucleotide (TAG1MO) 31, allowed them to develop to 24 hpf, and examined the pattern of nucMLF axon outgrowth in fixed embryos by immunolabeling with an anti-GFP antibody. We have previously shown that injection of TAG1MO, but not a standard control morpholino (STDCONMO), eliminates TAG-1 immunolabeling, indicating the loss of TAG-1 protein 31. TAG-1 knockdown caused several defects in nucMLF axon initial growth and convergence, but did not appear to alter the number of nucMLF neurons (Figure 3i–k, m). Normally, nucMLF axons initially extend only within the area containing the nucMLF cell bodies (called the 'nucMLF zone'; outlined in Figure 3i–l) prior to reaching the convergence point. However, after TAG-1 knockdown, nucMLF axons aberrantly extended into the tissue medial, lateral, and rostral to the nucMLF zone (Figure 3k, open arrowhead). This phenotype was observed in 67.6% (n = 173) of TAG1MO injected embryos, compared with only 8.4% (n = 107) of STDCONMO injected embryos (Figure 3i–k, m). In addition, the axons converged at a point more caudal than the typical convergence point in 73.2% (n = 168) of embryos injected with TAG1MO, while injection of STDCONMO caused this defect in only 8.4% (n = 107) of embryos (Figure 3i–k, arrowheads, and Figure 3m).

Knockdown of TAG-1 also significantly stunted nucMLF axon outgrowth. Normally, the leading edge of the MLF has reached the anterior spinal cord by 24 hpf 34; however after TAG-1 knockdown, MLF axons had not extended beyond the anterior hindbrain in 62.2% (n = 172) of embryos, compared with only 2.8% (n = 107) of STDCONMO injected embryos (Figure 3j, m). To determine whether these defects were specifically due to TAG-1 knockdown and not to nonspecific morpholino effects, we attempted to rescue the defects by co-injecting TAG1MO with a TAG-1 mRNA construct that did not include the majority of the TAG1MO binding region. TAG-1 overexpression did not cause any non-specific gross morphological changes, but did significantly reduce the percentage of embryos with nucMLF defects, suggesting that these defects are specific to TAG-1 knockdown (Figure 3m). Furthermore, knockdown of L1, another Ig superfamily CAM that is expressed by the nucMLF 39, did not cause the above defects, although it did cause MLF defasciculation in the hindbrain 35, suggesting that not all CAMs expressed by these neurons are required for MLF axon convergence. Together, these data suggest that TAG-1 is required for the efficient convergence of nucMLF axons and that it potentially functions to promote positive axon-axon interactions and/or allow nucMLF axons to interpret signals from the surrounding tissue.

Interestingly, the nucMLF axon defects induced by TAG-1 knockdown appeared very similar to defects we previously observed in bashful (bal) embryos, which have a mutation in the laminin-α1 subunit (Figure 3l) 1540. Laminin is expressed broadly in the basal lamina along the ventral surface of the neuroepithelium, directly apposed to the nucMLF and the MLF axons during all stages examined (16–24 hpf; Figure 3e–h, and data not shown). In bal mutants, the nucMLF axons also appear to extend aberrantly outside the nucMLF zone before eventually converging in the anterior hindbrain. Moreover, these axons appear defasciculated and stalled in the hindbrain. The similar defects suggest that TAG-1 and laminin-α1 may function together to guide the initial convergence of nucMLF axons or that they may have similar influences on growth cone behavior to produce comparable pathfinding defects. However, the fixed tissue analysis of nucMLF axon defects was insufficient to address these questions.

To determine how TAG-1 and laminin-α1 loss of function influences nucMLF growth cone behaviors, we performed in vivo time-lapse imaging in TAG1MO injected Tg(pitx2c:gfp) embryos and bal;Tg(pitx2c:gfp) embryos. Interestingly, live imaging results indicate these cues play both overlapping and unique roles in regulating nucMLF growth cone behaviors to cause the apparently identical defects observed in fixed tissue. In TAG1MO injected or bal embryos, axons extended in aberrant directions and into the tissue surrounding the nucMLF zone, without retracting or immediately correcting their growth direction (Figure 4a, b and Additional files 3, 4). Some of these misdirected axons eventually converged with other nucMLF axons that had extended properly (Figure 4a, arrowhead), while others approached or crossed the ventral midline (Figure 4a, b, arrows) or grew laterally after exiting the nucMLF zone. Interestingly, axons extended into medial (30%, n = 49 axons in 10 embryos) and lateral (23%, n = 49 axons in 10 embryos) tissues with comparable frequency after TAG-1 knockdown. However, in bal embryos, the majority of axons extending into surrounding tissue grew towards the ventral midline (53%, n = 80 axons in 8 embryos; 16% extended laterally), and in some cases, these axons accelerated as they neared the midline. To quantify the directional outgrowth errors, we determined the average percentage of time that axons extended in the correct direction ('on-track') and whether each MLF axon initially emerged from its cell body in an 'on-track' or 'off-track' position. We defined 'on-track' as growth within a region defined by a line extending from the cell body at 90 degrees to the nucMLF centerline and another line connecting the cell body to the convergence point (Figure 4c). The nucMLF centerline was defined as an imaginary line drawn along the MLF and extending rostrally through the nucMLF (Figure 4a, c, dashed line). Growth outside of this region or across the nucMLF centerline was considered 'off-track'. On average, axons spent significantly more time growing off-track after TAG-1 or laminin-α1 loss of function (Figure 4d). Axons from uninjected embryos extended off-track only 3% of the time (n = 40 axons in 11 embryos). However, axons in TAG1MO injected and bal embryos grew off-track 35% (n = 44 axons in 10 embryos) and 75% (n = 31 axons in 8 embryos) of the time, respectively. In bal embryos, a significant number of axons (40%, n = 83 axons in 8 embryos; Figure 4e) initially emerged from their cell bodies in an incorrect direction, which may, at least in part, explain why these axons grew off-track the majority of the time measured. In contrast, the majority of axons in TAG1MO injected embryos emerged from their cell bodies in an on-track position and later extended off-track. Taken together, these data suggested that TAG-1 and laminin-α1 may both be required for nucMLF growth cones to be guided by the surrounding tissue and that laminin-α1 is important for establishing the initial polarity of nucMLF neurons.

The live imaging experiments also revealed a role for TAG-1, but not laminin-α1, in regulating axon-axon interactions among converging nucMLF axons. In TAG1MO injected embryos, we commonly observed growth cones contacting neighboring axons and then subsequently retracting (Figure 5a and Additional file 5). In many cases, growth cones would sample multiple other axons, but fail to maintain any of the interactions. None of these behaviors were observed in bal embryos where aberrantly guided axons maintained their interactions once they contacted other axons (Figures 4b and 5b, and Additional files 4 and 6). To quantify the maintenance of axon-axon interactions, we calculated the average percentage of time in which growth cones were in contact with neighboring axons once contact was initiated, the average duration of each contact, and the average number of axons contacted per growth cone. On average, axons in TAG1MO injected embryos were in contact with neighboring axons only 47% (n = 22 axons in 10 embryos) of the time compared with 95% (n = 32 axons in 11 embryos) and 96% (n = 11 axons in 8 embryos) of the time in uninjected and bal embryos, respectively (compare Figures 5c and 2a). Moreover, the average duration of these contacts was 70 minutes (n = 35 axon contacts in 10 embryos) in TAG1MO injected embryos versus 176 minutes (n = 34 axon contacts in 11 embryos) and 177 minutes (n = 12 axon contacts in 8 embryos) in uninjected and bal embryos, respectively (Figure 5d). Finally, prior to converging, each growth cone contacted an average of 1.06 axons (n = 32 growth cones in 11 embryos) in uninjected and 1.09 axons (n = 11 growth cones in 8 embryos) in bal embryos (Figure 5e). However, after TAG-1 knockdown, each growth cone sampled 1.6 axons (n = 22 growth cones in 10 embryos; Figure 5e), suggesting that growth cones repeatedly sample neighboring axons despite their inability to maintain these interactions. The failed axon interactions in TAG1MO injected embryos may contribute to their increased off-track extension since these axons often changed their trajectory and extended away from the nucMLF centerline after breaking contact with a neighboring axon. Moreover, these axon-axon defects may have potential additional consequences in terms of neuronal polarity. In one case, we observed the growth cone of a newly emerged MLF axon contact a neighboring axon and then collapse (Figure 5a, arrow). After collapsing, the axon retracted to its cell body and then what appeared to be a new axon emerged from the opposite side of the cell body, consequently changing the neuron's polarity. We never observed such polarity changes in uninjected embryos, nor did we ever observe a nucMLF neuron extend multiple axons and then prune all but one axon in any of the groups. Collectively, these observations suggested that TAG-1 knockdown reduced positive axon-axon interactions, which can cause nucMLF axons to grow off-track and possibly influence nucMLF neuronal polarity.

Typically, nucMLF axons do not branch as they initially converge in the midbrain and extend through the hindbrain. Only much later in development do they normally develop collateral branches that innervate the spinal cord 41. However, in bal, we commonly observed axons that extended collateral branches or growth cones that split and produced divergent axons (Figure 6a and Additional file 7). nucMLF axons did not branch significantly in uninjected or TAG1MO injected embryos. Branches were counted if they lasted for at least ten minutes and many of the branches lasted for the duration of the imaging session (up to six hours). On average, axons in bal embryos each extended 0.71 branches (n = 49 axons in 8 embryos), compared to 0 branches in uninjected embryos (n = 51 axons in 11 embryos) and 0.17 branches in TAG1MO injected embryos (n = 42 axons in 10 embryos) (Figure 6b). In bal embryos, approximately 54% (n = 49 axons in 8 embryos) of axons extended at least one branch, and 19% had more than one branch. The majority of these branches extended from the axon shaft. These observations suggest that the presence of laminin-α1 normally suppresses axon branching, and that TAG-1 does not influence axon branching during nucMLF axon convergence.

The stunted MLF outgrowth phenotype we observed in the fixed tissue analysis of TAG1MO injected and bal embryos (Figure 3i, j, m) 15 may be due, in part, to their initial extension in incorrect directions, but a reduced growth rate may also contribute to this phenotype. Therefore, we quantified the growth rate of axons in uninjected, TAG1MO injected, and bal embryos from their initial extension until they reached the convergence point or an equivalent position in the rostral-caudal axis (Figure 6c). Axons in uninjected embryos grew at an average of 14.11 ± 1.08 μm/h (Figure 6c). However, axons in TAG1MO injected or bal embryos grew significantly slower at an average of 10.63 ± 1.13 μm/h and 8.27 ± 0.96 μm/h, respectively (Figure 6c). Thus, in addition to their inefficient pathfinding towards the convergence point, reduced growth rates likely contribute to the retarded MLF axons observed after TAG-1 or laminin-α1 loss of function.

Zebrafish (Danio rerio) were maintained in a laboratory breeding colony on a 14/10 h light/dark cycle. Embryos were maintained at 28.5°C and staged as described previously 49. bashful mutant embryos were obtained from fish carrying the bal^uw-1 allele 15, which contains a mutation that results in a predicted protein truncation at amino acid 1,424 of laminin-α1 50. Animals were handled in accordance with guidelines set forth by NIH and IACUC, and the animal use protocol was approved by the University of Wisconsin Animal Care and Use Committee (assurance number A3368-01).

The Tg(pitx2c:gfp) transgenic line was generated using an internal promoter of the pitx2 gene that specifically produces the pitx2c isoform 51. This promoter was obtained by amplifying a region from exon 2 to exon 4 of the pitx2 gene from a PAC clone. This fragment was fused to the enhanced GFP (EGFP) coding sequence followed by the SV40p(A) signal sequence from pEGFP-N1 (Clontech, Mountain View, CA, USA) and cloned into pBSIIKS- (Stratagene, La Jolla, CA, USA). The primers used for PCR amplification of the pitx2c promoter were: 5'-AGGGTACCGGACTCCCACTGCCGCAAACT-3' and 5'-TCTAGAGCTCTGTAATGCAAAAGGAAACAC-3'. The primers used to amplify the EGFP-SV40p(A) fragment were: 5'-TAACGAGCTCATGGTGAGCAAGGGCGAGGA and 5'-CAGAATTCTGAGTTTGGACAAACCACAAC-3'. To generate transgenic zebrafish, the vector sequence was removed from the pitx2c:egfp construct prior to injection into embryos at the one-cell stage. Three independent lines were isolated that all showed expression of EGFP in the trigeminal ganglia and the nucMLF.

For whole-mount immunohistochemistry, embryos were fixed in 4% paraformaldehyde overnight, blocked in 5% sheep serum and 2 mg/ml bovine serum albumin in phosphate buffered saline with 0.1% Tween, and incubated overnight (4°C) in anti-GFP (1:1,000; Sigma, St Louis, MO, USA), ZN-12 (1:250; Zebrafish International Zebrafish Center, Eugene, OR, USA) or anti-TAG1 (1:500; 52). Antibody labeling was performed with the Vectastain IgG ABC immunoperoxidase labeling kit (Vector Laboratories, Burlingame, CA, USA). For fluorescent double labeling, Alexa-conjugated secondary antibodies (4 μg/ml; Invitrogen-Molecular Probes, Carlsbad, CA, USA) were used to bind primary antibodies.

Morpholino oligonucleotides were synthesized by Gene Tools (Corvallis, OR, USA). To knockdown TAG-1, we used a morpholino sequence (TAG1MO; 5'-CCACACCCA-GACCAGACACTTATTT-3') that has been previously reported and shown to block TAG-1 protein 31. As a control, we injected a standard control morpholino (STDCONMO; 5'-CCTCTTACCTCAGTTACAATTTATA-3'). Morpholino oligos were injected into newly fertilized embryos at the one- to four-cell stage as described previously 53. Optimal concentrations for the TAG1MO were determined by titrating the dose from 1–10 ng per embryo; 10 ng injections appeared to cause non-specific toxicity, 2.5–5 ng injections appeared to cause robust defects without non-specific cell death, and 1 ng injections caused partial, but insignificant, defects. Therefore, we injected TAG1MO at approximately 2.5–5 ng to elicit full knockdown and equivalent doses of STDCONMO were injected for each experiment.

To create a TAG-1 construct potentially capable of rescuing TAG-1 knockdown, we subcloned full-length TAG-1 into the pCS2⁺ vector and then removed the first 20 bases of the sequence complementary to the 25-mer TAG1MO sequence, located in the 5' untranslated region. The mMessage mMachine Kit (Ambion, Austin, TX, USA) was used to transcribe 5' capped mRNA from pCS2⁺-TAG1. Approximately 100 pg of RNA was injected into one-cell stage embryos. Overexpression was verified by immunolabeling for TAG-1 at 24 hpf.

All brightfield images were captured on a Nikon TE300 inverted microscope equipped with a Spot RT camera (Diagnostic Instruments, Sterling Heights, MI, USA) and processed with Metamorph software (Universal Imaging Corp., West Chester, PA, USA). Fluorescent images of fixed preparations are confocal images captured on an Olympus Fluoview 1000 Laser Scanning Confocal Microscope.

Preparation of embryos for imaging was adapted from Langenberg and colleagues 54. Briefly, the yolk cell was paralyzed by adenosine 5' (β,γ-imido)triphosphate (Calbiochem, San Diego, CA, USA) injection at 15.5 hpf, and then removed. Each embryo was mounted between two coverslips separated by high vacuum silicon grease in 67% L-15. Embryos ranged in age from 16–18 hpf at the start of imaging and were imaged for 2–12 hours in a temperature controlled environment set to 29°C. Images were captured using a 60× dipping objective (NA = 1.00) on a Nikon (Tokyo, Japan) E-600FN equipped with standard epifluorescence, a filter wheel, and a CoolSnap HQ camera (PhotoMetrics, Tucson, AZ, USA). Images were captured every one minute, and exposure times were typically 300–600 ms. Images and movies were captured, processed, and analyzed with MetaMorph software (Molecular Devices, Sunnyvale, CA, USA.