Topoisomerase IIβ is required for lamina-specific targeting of retinal ganglion cell axons and dendrites

Throughout the visual system, axons and dendrites form layered arrays of connections (Huberman et al., 2010). Each layer (or lamina) is composed of synapses with similar functional specializations. In the retina, information extracted from the visual scene is transmitted via morphologically discrete layers of synapses, which correspond to parallel processing streams (Roska and Werblin, 2001). In zebrafish larvae, retinal bipolar cells (BCs), amacrine cells (ACs) and ganglion cells (GCs) form synapses across at least nine sublaminae of the inner plexiform layer (IPL) (Nevin et al., 2008). The majority of GCs project their axons to the optic tectum, another laminated structure. Each GC axon innervates one of six ‘retinorecipient’ layers in the tectal neuropil, where it forms a planar arbor (Xiao et al., 2005; Nevin et al., 2010). The postsynaptic tectal neuron dendrites also show diverse laminar selectivity, suggesting that GC axons in different laminae have molecularly and functionally distinct postsynaptic partners (Meek and Schellart, 1978; Nevin et al., 2008; Scott and Baier, 2009; Robles et al., 2011).

Although the number of cell-surface proteins implicated in laminar targeting is growing, upstream players regulating transcription of these genes have not been identified. Studies of neural differentiation and activity-dependent synaptic plasticity have begun to paint a complex picture of transcriptional regulation (Polleux et al., 2007; Flavell and Greenberg, 2008; Alberini, 2009). Transcription factors collaborate with epigenetic regulators that modify the structure and accessibility of regions of DNA to the transcriptional machinery. The accessibility of a DNA template to transcription can change following modification of a number of topological parameters: location within the nucleus (Martins and Krawetz, 2007), the types of histones present (Ju et al., 2006), acetylation of local histones (Weaver et al., 2004), methylation of cytosines in the template (Levenson et al., 2006) and local looping of the DNA strand (Sano et al., 2008).

Topoisomerase IIβ (Top2b) has recently emerged as an epigenetic regulator of transcription in later stage neural development (Tsutsui et al., 2001; Lyu et al., 2006). Like all topoisomerases, Top2b uses a strand-passing mechanism to resolve topological problems in DNA; this role in chromatin organization might influence rates of transcription at target sites. Using a forward genetic approach, we have discovered that the zebrafish top2b gene has an important role in the fine targeting of visual system neurites to their synaptic laminae. A splice-site mutation in this gene causes GC dendrites within the IPL and GC axons within the tectal neuropil to trespass between laminae, while sparing other features of cellular architecture and axon pathfinding. We demonstrate that Top2b acts within retinal clones to target GC axons to sublaminae within a single tectal lamina, stratum fibrosum et griseum superficiale (SFGS), suggesting that top2b has a GC-autonomous function in regulating the expression of target recognition molecules.

Adult zebrafish from the TL and WIK strains were maintained in our fish facility at the University of California, San Francisco (UCSF). All procedures adhered to the rules of animal use set by the National Institutes of Health and UCSF. Transgenic lines, generated and maintained in TL, were Pou4f3:mGFP [Tg(Pou4f3:gap43-GFP)s273t] (Xiao et al., 2005), Pax6:mGFP [Tg(Pax6-DF4:gap43-GFP)s220t] (Kay et al., 2001) and BGUG [Tg(Pou4f3:Gal4,UAS:mGFP)s318t] (Xiao and Baier, 2007). Larvae used live for confocal imaging were raised in phenylthiourea (0.2 mM, Sigma-Aldrich) to block pigmentation.

Detailed procedures of our n-ethyl-n-nitrosourea (ENU) mutagenesis screen have been previously described (Muto et al., 2005). Forty visually impaired mutants from this screen were investigated histologically for neurite targeting defects in the retina and tectum at 5 days postfertilization (dpf), as previously described (Nevin et al., 2008). Primary antibodies used for screening were: mouse anti-Parvalbumin (1:1200, Chemicon); rabbit anti-PKC β1 (1:800, Santa Cruz Biotechnology); and mouse Zrf3 (1:250, Oregon Monoclonal Bank). Antibodies used in follow-up studies were: goat anti-ChAT (1:50, Chemicon) rabbit anti-GFP antibody (1:4000, Molecular Probes); and mouse anti-HuC/D (1:400, Molecular Probes). TUNEL reagent (Roche) was used according to the manufacturer's protocol, with amplification of signal by secondary labeling with Alexa Fluor 488 anti-fluorescein antibody (Molecular Probes).

Established linkage mapping methods were used to locate the noto^s380 mutation in the genome (Muto et al., 2005). noto heterozygotes (in the TL strain) were crossed with fish from the highly polymorphic WIK strain, and adult TL/WIK noto heterozygotes were bred repeatedly to generate TL/WIK recombinant wild-type and mutant larvae. Recombination events at microsatellites and other sequence polymorphisms from established panels (Kwok et al., 1998; Shimoda et al., 1999; Woods et al., 2000) and additional markers of our design including marker 12 (from GenBank clone BX927411; forward primer: TGTGCATGTAACATTGACTGTAGACT; reverse primer: AGTCACACCTGCTACACTAGTTATTC), were scored in the WIK/TL recombinant mutant larvae. For candidate gene sequencing, RNA samples from wild-type and mutant larvae were isolated using TRIzol reagent (Invitrogen)/chloroform-based extraction. cDNAs from each RNA sample were reverse transcribed using the 3′ RACE cDNA Synthesis Kit (Clontech). Wild-type and mutant coding sequences of candidate genes were compared using Sequencher software (Gene Codes Corporation).

An antisense morpholino (TTTATCTGCTATAAGCTCACCTGCA) targeting the top2b exon 11 donor splice site, and a control morpholino with a 5-base sequence mismatch (TTaATgTGCTATAAcCTCAgCTcCA, lower case indicates mismatched bases) were designed and manufactured by GeneTools. These were dissolved at a concentration of 1 mM in water with Phenol Red dye (Sigma-Aldrich). Morpholino solutions were injected with a picospritzer into the yolks of wild-type BGUG embryos at the 1- to 4-cell stage. A dose of 1 pmole/embryo was identified as half-lethal (~50% survival); 0.5 pmole, 1 pmole or 1.5 pmole were used in subsequent experiments. For imaging, injected embryos were raised to 5 dpf, sorted for GFP expression, mounted in agarose and imaged. RT-PCR (described above) with 2 and 5 dpf control and morphant fish was used to confirm the morpholino targeting.

A 1.3 kb fragment (forward primer: TTGCAGCCTGGAACAAAGCTCAAG; reverse primer: AGCAAGAAGCGAAAGCAGTCAAGC) of top2b was cloned into the pCRII-TOPO vector (Invitrogen). Sense and antisense digoxigenin-labeled RNA probes were made by in vitro transcription using an RNA labeling kit (Roche). Hybridization and staining were performed on 1-6 dpf larvae, according to methods previously described (Smear et al., 2007).

Fluorescence images were collected using a Zeiss LSM 5 Pascal confocal microscope, with a 40× oil objective for slides and a 40× water immersion objective for live fish. Processing and quantification of images of retina and tectum sections have been described (Nevin et al., 2008). Live fish were mounted in agarose according to a standard protocol (Smear et al., 2007). Confocal stacks were processed to projected 3D images using ImageJ software. Several images with more than one GC axon were painted to generate one stack for each axon; individual axons were then assigned one channel in an RGB scheme to generate multicolor projected images. For the GC axon morphometry study, stacks were imported into NeuronStudio (Wearne et al., 2005) for three-dimensional tracing and length/branching quantifications.

Chimeric larvae were generated using standard methods (Gosse et al., 2008). Pou4f3:mGFP+, noto heterozygous adults were incrossed to generate donors, and non-transgenic noto heterozygotes were incrossed to generate hosts. Donor embryos were labeled by injection at the 1- to 4-cell stage with Alexa 555-dextran (MW 10,000, 1% w/v each, Molecular Probes Invitrogen). At the blastula stage, small numbers (5-15) of cells were transplanted from donor to host animal pole. At 24 hpf, transplanted hosts were sorted for red fluorescent clones in the eye. Chimeras with retinal clones, and their corresponding donors, were raised to 5 dpf and scored as either wild type or mutant; hosts with donor-derived GC axons in a donor clone-free tectum were imaged.

A primary mutagenesis screen of ~2000 genomes (Muto et al., 2005) yielded a collection of visually impaired mutants, which subsequently were sectioned and stained by immunohistochemistry with retinal and tectal cell-type markers. We became particularly interested in the blind notorious (noto) mutant owing to its specific neurite targeting defects. noto mutants had normal morphology and cytoarchitecture; DAPI staining of the retina and tectum showed the wild-type organization of cells (Fig. 1A,G, Fig. 2A,D). In addition, staining with cell type-specific markers indicated that neuron types differentiate appropriately (Fig. 1 and see Figs S1, S2 in the supplementary material) and TUNEL staining showed no striking change in cell death in the retina at 5 dpf (wild type: n=7, mean 0.9 TUNEL+ cells/central section, s.e.m. 0.6; noto: n=5, mean 0.6 TUNEL+ cells/central section, s.e.m. 0.4).

Although cell types were properly generated and localized, the local targeting of retinal neurites was perturbed (Fig. 1 and see Fig. S1 in the supplementary material). For example, Parvalbumin (Parv)+ AC neurites in the wild-type 5 dpf retina target IPL sublaminae s25, s45 and s85; these sublaminae are located at 25%, 45%, 85% of the distance from inner to outer edge of the IPL, respectively (Nevin et al., 2008). In the mutant, Parv+ AC neurites appropriately targeted strata in the two halves of the IPL, but the bands of immunoreactivity were coarse and the s25 and s45 bands were merged. Analogous errors were made by other IPL neurites, including Protein kinase C (PKC)+ bipolar cell axons, Pou4f3:mGFP+ GC dendrites, and Pax6:mGFP+ and choline acetyltransferase (ChAT)+ AC neurites. Notably, the targeting of Pax6:mGFP+ and ChAT+ AC neurites was nearly normal, but the fine sublamination of these neurites into doublet bands (e.g. s40/s45 for Pax6:mGFP) was lost.

The IPL sublamination errors in the mutant were quantified by scoring the number of distinct PKC+, Pou4f3;mGFP+ and Parv+ bands in the IPL of multiple retinas (Nevin et al., 2008). As shown in Fig. 1M, wild type and mutant each had a reproducible number of IPL sublaminae, and the observed differences were statistically significant (for all three comparisons: n=10 wild-type and 10 mutant retinas, P<0.01). These sublamination errors were more pronounced, and the IPL more disorganized, in older (7 dpf) mutants (see Fig. S2 in the supplementary material). We conclude that mutant neurites fail to target their correct sublaminae and that the mutated gene is important for sublaminar positioning of most, perhaps all, inner nuclear layer (INL) and GC layer (GCL) retinal neurons.

The noto mutant also has neurite organization defects in the tectal neuropil. In wild type, Pou4f3:mGFP+ GC axons innervate three layers, with each axon among this population arborizing in just one of these layers, either the stratum opticum (SO) or one of two sublaminae of the SFGS (Xiao and Baier, 2007). Cryostat sections of the Pou4f3:mGFP+ labeled projection in the mutant tectum revealed that the two labeled GFP+ sublaminae of SFGS are expanded and that bridges develop between them. The fine targeting of axons to the SO appeared normal in sections of the mutant tectum (Fig. 2A,D, n=10 wild-type siblings, n=15 mutants).

The postsynaptic partners of GC axons are the dendrites of tectal neurons, most of which extend from the cell bodies in the periventricular region of the tectum into strata of the neuropil. Several types of periventricular neurons extend dendrites in a lamina-specific pattern (Meek and Schellart, 1978; Scott and Baier, 2009; Robles et al., 2011). PKC and Parv antisera label two such populations (Nevin et al., 2008). Antibody staining of sections revealed that laminar targeting by these tectal dendrites is severely compromised in noto mutants (Fig. 3, n=10 wild-type and mutant larvae each). In contrast to the mutant IPL, in which the wild-type pattern was replaced with a distinct alternate pattern, mutant tectal neurites appeared to have very little or no organization; instead, the neuropil was broadly immunolabeled, suggesting a lack of neurite refinement.

GC axon targeting can be assessed by live confocal imaging of larvae expressing the paired transgenes Pou4f3:Gal4,UAS:mGFP (BGUG), where the UAS:mGFP expression is variegated (Xiao and Baier, 2007), thus labeling a small number of axons. Examination of BGUG larval tecta, such as those shown in Fig. S3 in the supplementary material, revealed a highly disorganized SFGS in mutants, suggesting multiple targeting errors, and in just one case, a crossover between SO and SFGS.

We hypothesized that individual mutant GC axons often trespass between the finer layers of SFGS, but very rarely between SO and SFGS. To test this hypothesis and learn more about the morphology of mutant GC axons, we selected larvae with BGUG expression in individually distinguishable axons. Wild-type axons in both SO and SFGS projected arbors to narrow bands within the tectum, with cross-sectional widths not exceeding 8 μm (SO: mean 4.5 μm, s.e.m. 1.3, n=4 axons from three fish; SFGS: mean 4.4 μm, s.e.m. 0.3, n=24 axons from 11 fish). Measurements from confocal images of Pou4f3c:mGFP+ wild-type larvae enabled us to estimate the width of SO (5 μm) and each sublamina of SFGS (10 μm); individual wild-type axons are therefore confined to strata as thin as, or thinner than, these identified layers. The mutant SO axons imaged were also finely targeted (mean width 3.9 μm, s.e.m. 0.7, n=6 axons from six fish); however, the population of SFGS axons was clearly distinct from wild type. First, we observed a population of GCs in which the unbranched axon extended into one lamina, then crossed into another, usually turning 180 degrees, to elaborate a branched arbor in an antiparallel conformation (see Fig. S4 in the supplementary material); these instances were scored as entry defects (6 of 24 axons; Fig. 2G). By contrast, wild-type axons always entered and elaborated arbors in the same stratum, as previously reported (Xiao and Baier, 2007).

Although entry defects contribute to the laminar disorganization seen at the population level, most of it probably results from the increased arbor width of the mutant axons. The mean SFGS axon width in the mutants was greater than in wild type (10.8 μm, s.e.m. 1.4, n=24 axons from 14 fish; t-test with unequal variance, P<0.0003); many mutant SFGS axons extended arbors wider than 10 μm and were scored as ‘multilaminar’ (10 of 24 axons; Fig. 2). Multilaminar axons were clearly different from their wild-type counterparts when viewed from the side; many branches trespassed out of the initially targeted lamina, and sometimes a secondary planar arbor formed. Morphometric analyses (Fig. 2I-K) demonstrated that these multilaminar arbors are of greater length and branch complexity than unilaminar mutant SFGS axons (including those with entry defects). Mean total arbor length, number of branchpoints and number of branchtips were increased in multilaminar axons ~50% above the mean measurement for mutant unilaminar axons (t-test, P=0.004, 0.005, 0.01, respectively; n=5 unilaminar axons and 3 multilaminar axons). However, all mutant SFGS axons measured fell within the wild-type range for these metrics. The modest decrease in size and complexity of unilaminar axons in the mutants might be the result of lamina-specific competition for limiting postsynaptic partners, as the unilaminar axons might face increased competition from multilaminar neighbors in the mutant. Alternatively, top2b might be regulating arbor complexity in a GC type-specific manner, limiting branching and extension in some and promoting these processes in others.

Genetic linkage mapping and positional cloning methods were used with a pool of ~5000 mutant larvae to localize the noto mutation to chromosome 19, between T51 marker fj94f01 to the North and our own marker 12 to the South (see Materials and methods). These and flanking markers from our study are not arranged in contiguous order in the ensEMBL zebrafish genome assemblies Zv7 or Zv8; however, we used the ensEMBL physical map in combination with our results to generate a ~2 Mb region of candidate genes. This region included the full coding sequence of top2b. Wild-type and mutant top2b were cloned in fragments; one fragment, spanning exons 9 to 20, was ~140 bp smaller in mutant cDNA. Sequencing revealed that the mutant fragment lacked exon 11. Genomic regions flanking each end of the exon were amplified and sequenced, confirming the presence of a G-to-A transition in the splice acceptor sequence of exon 11 (Fig. 4A). The alternatively spliced product is frame-shifted relative to wild type, and leads to an early stop codon in exon 12 (position 1853) of the 35-exon (4857 bp) cDNA. Top2b is a DNA-modifying enzyme that binds two double helices, cleaves one (the g-strand, for ‘gate’) and passes the other (the t-strand, for ‘transfer’) through the break, and then re-anneals the cleaved strand. Although the mutant top2b gene product retained the N-terminal ATPase region, which initially binds the t-strand, and part of the adjacent domain that transduces the power of ATP hydrolysis into a conformational change, it is predicted to lack the domains necessary to bind and cleave the g-strand, as well as the exit gate for the t-strand (Fig. 4B). Although this truncated protein is unlikely to retain topoisomerase activity, it is unclear whether the mutation results in a complete loss of function, as some wild-type splice variants might still be produced.

To confirm that the noto locus encodes the top2b gene, we phenocopied the mutant by antisense morpholino oligonucleotide (MO) gene knockdown in BGUG-transgenic fish. The top2b MO was targeted to the exon 11 splice donor site, closely simulating the top2b^s380 mutation; a 5 bp mismatched MO was used as a control. Injected embryos were raised to 5 dpf and imaged to detect possible aberrations in axon targeting. We found that injected embryos were extremely sensitive to the top2b MO (but not to the control MO); a concentration of 1 pmole/embryo was lethal to ~50% of the fish. To verify targeting and to monitor degradation of the MO in the embryos injected with this dose, top2b cDNAs were cloned and sequenced from 2 and 5 dpf control morphant and top2b morphant fish. As expected, controls had the wild-type product, whereas top2b morphant cDNAs had the aberrantly spliced product at 2 dpf (Fig. 4C). At 5 dpf, however, top2b morphant cDNA contained a small amount of wild-type top2b in addition to the truncated form. We conclude that degradation of the MO allows recovery of top2b splicing by 5 dpf. The transient nature of the knockdown might therefore confound our ability to phenocopy the neurite targeting defects seen in mutants at 5 dpf.

Nevertheless, the multilaminar arbor phenotype seen in mutants was recapitulated in a subset of the morphant SFGS axons (2 of 17 axons, from six MO-injected fish, mean width 9.5 μm, s.e.m. 0.8 μm). Although the percentage of defective tob2b morphant axons was small, the defects were pronounced; the two aberrant arbors we observed had cross-sectional widths of 25 and 38 μm, falling 20 and 33 standard deviations, respectively, outside the control mean. Injections with the control MO into BGUG+ fish did not cause any targeting defects in 23 SFGS axons imaged from eight larvae (Fig. 4D-G; mean arbor width 4.9 μm, s.e.m. 0.2 μm). Notably, the IPL sublamination defects observed in mutants were not seen in the morphants. If a small number of IPL neurites (equivalent to the fraction of multilaminar GC axons in morphants) were aberrantly targeted, their defects might be masked by the rest of the population. Moreover, BGUG+ GC dendrites have variable, dynamically changing morphologies (Mumm et al., 2006); IPL sublamination errors, if any, of individual GCs can therefore not be scored, a limitation that applies to both mutants and morphants.

To find clues for the role of top2b in larval zebrafish development, we used RNA in situ hybridization to track expression of top2b during development. At 1 dpf, top2b was expressed broadly throughout the embryo (see Fig. S5A,D in the supplementary material). Thereafter, from 2 to 6 dpf (the last time point tested), expression was restricted to the head region, and was ubiquitous in the brain (Fig. 4H-K and see Fig. S5 in the supplementary material). By 4 dpf, the gene was expressed in the retina, but appeared to be absent from the photoreceptors, weak in the outer two-thirds of the INL and strong in the innermost INL and in the GCL. This pattern indicates strongest expression in ACs and GCs. Staining was performed on progeny from incrossed noto heterozygotes; at each stage, ~25% of the stained embryos or larvae were weakly labeled. Figure 4H shows a representative collection of stained 4 dpf larvae; after imaging, the individual fish shown were genotyped for the noto mutation. As shown, the darkest staining accurately predicted a homozygous wild-type genotype, heterozygotes were stained at an intermediate level and mutants had dim staining. The reduction of top2b in noto mutants and heterozygotes indicates nonsense-mediated decay of the mutant RNA.

To explore the role of top2b in nervous system patterning, we assessed the expression of a set of early domain markers at 28-48 hours postfertilization (hpf, Fig. 5) (Warren et al., 1999; Kay et al., 2001; Thisse et al., 2001). Expression of ath5 (atoh7), encoding a basic helix-loop-helix transcription factor required for generating GCs, was prominent at 28 hpf in ventronasal retina in both wild type and noto mutants. By 48 hpf, the mutant GCL was populated with HuC+ neurons, indicating that GCs differentiate in a normal time course. Early brain development was also grossly preserved. Expression of the homeobox transcription factor pax2a in the optic stalk, choroid fissure, midbrain-hindbrain boundary, and a collection of hindbrain neurons was maintained in 28 hpf mutants. At 36 hpf, expression of the surface immunoglobulin cntn2 could be observed in the telencephalon, the trigeminal, facial and acoustic/anterior lateral line ganglia, cerebellum and hindbrain of both wild-type and mutant embryos. This survey of developmental markers indicates that the neurite targeting phenotypes observed in the mutant do not represent a failure to generate basic cell types or brain structures.

The expression of top2b might be important within the GCs to supply the axon with the correct complement of guidance receptors, or might act within tectal cells to create a proper scaffold. We addressed this question of cell autonomy using chimeric analyses. Blastula stage transplantation of undifferentiated cells from the eye fields of Pou4f3:mGFP+ donors into the same region of non-transgenic expressors was used to generate chimeric blastulae, which were raised to 5 dpf. In these larvae, we saw small populations of donor-derived axons arborized in a host tectum (Fig. 6). In chimeras in which wild-type axons innervated a mutant tectum, laminar targeting was free of errors. By contrast, mutant axons innervating the SFGS lamina of wild-type tectum appeared disorganized and trespassed between sublaminae. We quantified our findings in a subset of chimeras with individually distinguishable axons in SFGS; 13 such wild type-derived axons in mutant tecta showed the expected fine targeting, whereas multilaminar arbors were observed in 2 of the 10 distinct mutant-derived axons innervating wild-type tecta (wild type→noto: n=13 axons in four fish, mean arbor width 4.8 μm, s.e.m. 1.0; noto→wild type: n=10 axons in five fish, mean width 7.4 μm, s.e.m. 7.1 μm). These multilaminar arbor widths were 13 and 26 μm, falling 6 and 14 standard deviations from the mean wild-type arbor, respectively. We conclude that the top2b genotype of the transplanted retinal clone, rather than of the host tectum, determines the lamination phenotype of the GC axon.

Our forward genetic screen successfully uncovered a mutant with pronounced neurite laminar targeting defects despite normal cytoarchitecture, cell differentiation and formation of long-range connections. Marker analysis indicated that a wide range of neural cell types are formed at the right time and place in the retina and tectum. In particular, GC differentiation is initiated on time, as shown by ath5 expression, and a properly developed GCL, immunostained against HuC, appears by 48 hpf. Despite apparently normal cellular organization, all seven neural populations examined show some aberration in the laminar targeting of neurites in the retina and/or tectum. Defects were observed in the axons of a population of bipolar cells (PKC+), dendrites of three populations of amacrine cells (Parv+, ChAT+ and Pax6:mGFP+), and both the dendrites and axons of Pou4f3:mGFP+ GCs. Two populations of tectal neurites (Parv+ and PKC+) also lack the normal precision of laminar organization within the neuropil. In every case, neurite pathfinding to the retinal or tectal neuropil is preserved, and the wild-type laminar pattern is usually approximated, but fine targeting is disrupted. As in the mouse top2b mutant (Yang et al., 2000; Lyu and Wang, 2003), the earliest steps of development – proliferation and mitotic exit – are preserved in zebrafish noto mutants. In the mouse Top2b knockout, laminar organization of neuronal cell bodies in the cortex and pathfinding of some motor and sensory axons are also disrupted, suggesting additional defects in neural patterning not seen here in the zebrafish visual system.

Closer examination in the BGUG+ line revealed that some mutant GC axons branch exuberantly and trespass SFGS sublaminae. These defects were shown by chimeric analyses to be retina-autonomous, suggesting a role for top2b in regulating expression of targeting factors within SFGS-projecting GCs. However, because transplantation generates retinal clones containing all cell types, top2b might instead act within other retinal neurons, which in turn might instruct the synaptic choices of neighboring BGUG+ GCs. This more circuitous mechanism would probably be independent of synaptic transmission, as various disruptions and blockade of synaptic signaling between retinal neurons do not affect the laminar pattern of SFGS innervation (Nevin et al., 2008). In either case, the axon targeting defects observed are consistent with a lack of repulsion from inappropriate laminae; perhaps these GC axons have lost the ability to detect a repulsive factor. How can a broadly expressed gene like top2b contribute to a cell type-specific laminar choice? First, the Top2b protein probably cooperates with many other transcriptional regulators to coordinate transcription; these other factors probably show cell type-specific expression. Second, although top2b gene expression is broad, protein activation might not be; Top2b can be phosphorylated by numerous intracellular signals, including Casein kinase II, PKC, Protein kinase A (PKA), Extracellular signal-regulated kinase (ERK) and Mitogen-activated protein kinase (MAPK) (Ju and Rosenfeld, 2006).

Gene expression data suggest that Topoisomerase IIa (Top2a) and Top2b have divergent roles. Both gene products are broadly expressed at 1 dpf; top2a RNA is detectable in the head and neural tube (Dovey et al., 2009), whereas top2b probe labels the whole embryo. After this time point, top2a expression persists in proliferative zones (Dovey et al., 2009), whereas top2b becomes restricted to postmitotic neurons. The zebrafish data corroborate observations from mammals; top2b has been shown in developing rat cerebellum to be expressed just as top2a diminishes, as neurons exit the cell cycle and begin to differentiate morphologically (Tsutsui et al., 2001). The refinement of zebrafish top2b to brain and retina by 2 dpf coincides with the onset of local neurite targeting by ACs in the IPL and GCs in the tectal neuropil; a thin IPL plexus is visible beginning at ~42 hpf (Godinho et al., 2005), whereas the first GC axons reach rostral tectum at 48 hpf (Stuermer, 1988; Burrill and Easter, 1995). Combined, these observations suggest that the role of Top2a in neurons is predominantly in proliferation, whereas the function of Top2b continues into neurite targeting phases.

The most parsimonious explanation for the noto phenotypes stems from the known molecular function of type II topoisomerases: altering DNA topology via the passage of one double helix through another. Because of the length of chromosomes, and the association of genomic DNA with multiple proteins, local supercoiling cannot be resolved or altered by twisting the full chromosome; topoisomerases make local changes to create desirable topology. Top2a appears to be particularly important for the untangling of chromosomes prior to segregation during mitosis (Dovey et al., 2009), and perhaps has some function in transcription also (Mondal and Parvin, 2001; Mondal et al., 2003). Top2b, by contrast, appears dispensable for proliferation but involved in several facets of transcription. Top2b protein is associated with RNA polymerase (Wang, 1991), histone deacetylases (Tsai et al., 2000; Johnson et al., 2001) and human chromatin remodeling complex hACF (LeRoy et al., 2000). Possibly, each of these DNA-modifying enzymes might cooperate with topological modifications by Top2b to adjust the availability of target genes for transcription. The sites of action of Top2b are not distributed randomly throughout the genome; mammalian Top2b might preferentially associate with matrix attachment regions (MARs) (Martins and Krawetz, 2007; Sano et al., 2008). Conformational changes at or near MARs might position sections of chromatin in regions with higher or lower transcriptional activity. Alternatively, the double strand breaking activity of Top2b might act as a marker to recruit other transcriptional regulators, as has been shown for the localization of Parp1 to the promoters of several estrogen receptor target genes (Ju et al., 2006). This varied collection of mechanisms suggests that the ability of Top2b to break and passage double stranded DNA is integrated in multiple ways by distinct transcriptional cascades.

The neurite targeting phenotypes of the noto mutant support a role for Top2b in the fine coordination of transcription of a small number of targets, as a broad abrogation of transcription would result in more severe developmental defects. Gene expression analyses of rat cerebellar cultures treated with a Top2b inhibitor and of the brains of mouse Top2b mutants reveal that a small collection of genes (1-4% in one study) are up- or downregulated by Top2b. The pool of targets is enriched with developmentally important genes, including guidance factors, adhesion molecules and synapse components (Tsutsui et al., 2001; Lyu et al., 2006). The identification of analogous targets in zebrafish is an appealing area for further study. It seems likely that the tuning of expression of a number of these targets cooperatively guides visual system neurites to precise synaptic laminae.

Thanks to Joel Bordayo for technical support for the GC axon morphometry study, Jeremy Kay for discussions, and the Baier lab screening team. L.M.N. was the recipient of predoctoral fellowships from NSF and the UCSF NIH Vision training grant. This research was funded by NIH grants EY13855 and EY12406 to H.B. Deposited in PMC for release after 12 months.