Design of an Efficient, Large-Scale Mutagenesis Screen in Zebrafish
We carried out a large-scale screen for mutants with defects in visually elicited behavior. Forty-one founder males (F0) treated with ethylnitrosourea (ENU; see Materials and Methods) were mated with wild-type (WT) females to generate more than 5,000 F1 fish. Adult F1 fish were mated with other F1 fish, or with WT fish, to generate more than 2,000 F2 families. In total, 3,171 F1 fish were used to generate the 1,896 F2 families (2,550 F1 fish for F1 × F1 crosses, and 621 F1 fish for F1 × WT) that gave at least one healthy clutch of F3 embryos in the subsequent generation. F3 embryos and larvae were obtained by random crosses between siblings from F2 families (6,468 F3 clutches in total, or 3.4 clutches per each F2 family on average). From each F3 clutch, typically 12 larvae were tested for OKR and 25 larvae for the OMR (see below). Fish were routinely scored on the seventh day postfertilization (7 dpf). Including retests, over 500,000 individual fish were screened in the course of three years. Calculations based on binomial statistics [18], taking into account the number of F1 fish used to generate the F2 families, the number of F2 families, the number of crosses for each F2 family, and the number of F3 larval fish tested, show that our screen encompassed 1,688 ENU-mutagenized genomes.
The efficiency of mutagenesis in the founder male germlines was determined by a specific-locus test, using sandy (sdy), a zebrafish tyrosinase mutant [19]. In this test, ENU-treated founder males mated with sdy heterozygous females produced six new sdy mutations in about 2,000 genomes screened. In the actual screen of F2 families, however, two new sdy mutant alleles were identified. The allele distribution of all loci, which was determined after completion of the screen and following extensive complementation tests, shows that our screen was not saturated (see Discussion). We nevertheless successfully identified new alleles of previously reported visual mutants, such as bel and nof (Table 1). Although we did not attempt to characterize mutations falling outside our screening criteria, i.e., those causing embryonic or larval lethality, we noticed (and most of the time discarded) new alleles of chk [20], bru/eby [21,22], ome, and nok [21] (unpublished data).
Two Behavioral Screening Assays, Executed in Parallel, Discovered 53 Visual Mutants
We screened for mutations disrupting behavioral responses to visual motion. A coarse grating that drifts across the fishes' visual field elicits either of two distinct responses, an OKR or an OMR. In the OMR, WT animals vigorously swim in the direction of the perceived motion (Video S1). When restrained from swimming and presented with a rotating whole-field motion stimulus, the fish show an OKR to cancel retinal slip: WT animals move their eyes to track the motion. These pursuit phases are interrupted at regular intervals by reset movements, or saccades (Video S2) [15]. To achieve high throughput, we automated both visual stimulation and analysis, as described elsewhere [16]. We found that the two screening assays were complementary: The OKR assay is slower and more labor-intensive, but has single-fish resolution; the OMR assay, on the other hand, is fast, but measures only population responses. For each assay, a behavioral index ranging from 0 (no response) to 1.0 (WT) was calculated (see Materials and Methods). Typical OMR and OKR mutant phenotypes are shown in Figure 1A and 1B.
Mutants detected by at least one of the two assays in the primary screen were kept. To select against phenotypes with general defects, we discarded mutants with overt developmental problems, as well as those that were poor swimmers, with a few exceptions. Putative F2 carriers were mated at least twice more for confirmation of the phenotype in their progeny before they were outcrossed. The OKR screen initially picked up 241 putative mutants, or “putants.” Following two retests, 46 lines (23%) were outcrossed. The OMR screen picked up 361 putants, 34 (9%) of which were confirmed and successfully propagated. In addition to high-contrast stimuli, we also routinely used a lower-contrast grating to detect subtle and/or contrast-specific visual defects. The high percentage of false positives is mostly attributable to the use of these weak test stimuli. The OKR and OMR assays were used independently within the primary screen. A considerable number of OKR mutants were later found to be OMR-deficient, and vice versa, as discussed below (Table 1).
The initial false positive rate of this behavioral screen greatly exceeded that of a morphological screen for small-eye mutants carried out in parallel [23]. However, almost all behavioral mutants were recovered in the following generation. Our strategy of extensive retesting as part of the primary screen therefore dramatically decreased the number of false positives and made this screen practical. Mutants or putative mutants with low penetrance were not kept or are not reported here. The mutants presented in this paper, therefore, were found in about 25% of the population in a clutch. To establish potential complementation groups, we systematically crossed heterozygous carriers of mutants with similar phenotypes. Noncomplementing mutations (in which the transheterozygous progeny showed a mutant phenotype) were considered to be allelic (Table 1).
Secondary Screening Assays Allowed Classification of Behavioral Phenotypes
In addition to OMR and OKR, we also assessed the larvae's visually mediated background adaptation (VBA) at 5 dpf, as a complementary strategy to enrich for visual mutants. The VBA is a neuroendocrine response that is controlled by ambient light levels and appears to depend on the function of retinal ganglion cells (RGCs) [17]. Melanophores in the skin contract their melanin granules in a bright environment, while a dark environment induces melanin dispersal [9]. We tested the VBA only in response to long (over 20 min) exposure to bright light, i.e., the mutants' ability to become pale. Figure 1C shows gradations of the VBA defect in three representative mutants. We found that, of the 89 VBA mutants discovered in the screen, 19 (21%) also had specific OMR or OKR defects. The remaining 70 “dark” mutants were either behaviorally normal or had externally visible, morphological phenotypes and were not always maintained.
To identify defects in motor functions, we systematically tested spontaneous swimming activity (SSA) (Figure 1D) in all our mutants. We also made sure that all mutants listed in Table 1, except s513, showed spontaneous, conjugate eye movements similar to WT when presented with a stationary stimulus. Finally, to identify mutants with developmental defects, we systematically examined their retinal and tectal histology and their retinotectal projections (Table 1).
Mutations May Affect Some Visual Behaviors More than Others
Because OKR and OMR are both evoked by motion of a large field grating, but differ in their motor output, our collection of mutants presented us with an opportunity to ask how well single-gene mutations can dissociate these two related behaviors. Are there mutations that impair OMR and OKR in a differential manner (weak dissociation) or even disrupt only one of the behaviors, while leaving the other unaffected (strong dissociation)? Table 1 shows that none of our mutants showed a complete absence of either OMR or OKR together with no defect at all in the other behavior. However, the two behaviors were often affected to different degrees. To reveal potential correlations, we plotted the behavioral profiles of our mutant set (Figure 2). Each data point in Figure 2 corresponds to one mutant, measured repeatedly (n > 3 clutches), and was also shaded to represent that mutant's light-exposed VBA score. Although many mutants lacked any visual responses, for those with partial OMR and OKR phenotypes, there was no clear relationship between the magnitudes of the deficit in the two behaviors (correlation coefficient r = 0.4, when mutants with OKR = 0 and OMR = 0 were excluded). Perhaps surprisingly, the severity of the VBA phenotype was not positively related to either OMR (r = −0.5) or OKR (r = −0.4) defects. The overall correlation of all OMR and OKR indices (r = 0.75) and the absence of exclusively OMR- or OKR-specific mutants suggest that these behaviors are weakly dissociable by single-gene mutation. This indicates that OMR and OKR share a major portion of the underlying neural circuitry. In contrast, the VBA appears to employ a dedicated neural pathway largely segregated, and therefore genetically separable, from motion vision.
Genes Required for Photoreceptor Differentiation and Survival
We discovered seven genes essential for photoreceptor differentiation and/or maintenance (Figures 3 and 4; Table 1). No other phenotypes could be discovered in these mutants, and at least four of them are adult viable. In two mutants (five alleles of wud and yois121), cone photoreceptors are present, but their shapes are shorter and thicker than in WT (see Figure 3). This “stumpy” morphology is not restricted to one particular cone type, as shown by labeling with zpr1, a double-cone-specific marker (Figure 3C and 3D). In five mutants (five alleles of mti, as well as goshs341, pdays351, lims382, and ssds386), all photoreceptors are lost before 6 dpf, except for a small population in the margins of the eye (Figure 4A–4J), where proliferation and differentiation of neuronal precursors continue throughout the life of the fish [23]. This suggests that some of the newborn cells select the photoreceptor fate, but die shortly after beginning differentiation. In mti mutants, degeneration spreads to the outer part of the inner nuclear layers (Figure 4F and 4H). This mutant is also the only one in this class with defective VBA (Figure 4K), as examined further below.
Six of the seven photoreceptor-defective mutants appear normal in their VBA response to light (Figure 4K). This is a curious finding, as it may suggest that classical cone/rod-mediated photoreception is not strictly required for this neuroendocrine response. It is conceivable that the pineal gland, a light-sensing organ in the dorsal forebrain, may control the VBA instead of, or together with, the retina. We therefore asked if presence of the VBA correlated with an intact pineal in our photoreceptor-degeneration mutants. Both VBA-normal and VBA-defective mutants showed a normal pineal, based on expression of shared marker zpr1 (Figure S1). This suggests that none of the mutated genes found here are necessary for the maintenance of the pineal photoreceptors. Moreover, it implies that pineal photoreceptors are not sufficient to control the VBA. This is consistent with the observation that lakritz mutants, which completely lack all RGCs due to mutation in the atonal homolog atoh7 (ath5), but which apparently have a normal pineal gland, show an extreme VBA defect (VBA = 0) [17]. Based on these combined genetic data, we propose that classical cone/rod photoreception is dispensable for this behavior and that other photosensitive cells, situated in the inner retina, signal ambient light levels to the VBA circuitry via the optic nerve.
Genes Required for General Visual Function, Including Phototransduction and Adaptation to Sudden Increases in Light
We identified 11 mutant alleles of nine genes (blds394, dadas503, dlns518, dlns393, edpos371, lajs304, mzrs130, nofs377, snevs102, zats125, and zats376) without detectable anatomical defects (unpublished data), but with complete absence of OKR and OMR (both indices 0.1 or less) (Figure 5; Table 1). The nofs377 mutation is a new allele of the alpha subunit of cone transducin [24], and the zat gene was shown by positional cloning to encode cone-specific guanylyl cyclase, Gc3 (unpublished data) [25]. Based on these findings, it is likely that some of the other seven genes in this category also encode components of the phototransduction cascade.
Other mutants were found to have variable visual impairments. We speculated that some of these mutants were unable to adjust the gain of their visual responses due to defective light adaptation. We therefore rescreened mutants with partial impairments and normal histology, using a behavioral paradigm previously developed by us to test this process in zebrafish larvae [19]. In brief, initially light-adapted fish were placed in a dark environment for a period of 45 min and then tested for OKR after return to light. The recovery of visual responsiveness following the sudden transition from dark to light served as a convenient surrogate measurement for light adaptation, although we do not know how closely this paradigm mimics adaptation. We identified five mutants (nkis136, utas301, ututs357, ymjs392, and mdrs527) in which the measured light adaptation was severely delayed (example in Figure 6). In addition, another mutant, nbks342, had a chronic impairment of both OKR and OMR, which varied with genetic background and occasionally improved with repeated stimulus presentation (unpublished data). The mutated genes may be components of light-adaptation pathways, either in photoreceptors or in the retinal network.
Genes Required for Ganglion Cell Differentiation and Axon Pathfinding
In WT animals, RGCs project to the contralateral brain and terminate in ten different arborization fields (AFs), of which AF-10, the tectum, is the largest [26]. In our collection of behavioral mutants, we found eight new mutants with specific retinofugal projection deficits (Figure 7): bojs307, darls327, walks536, exas174, misss522, michs314, drgs510, and drgs530, as well as a new allele of bel. In bels385 mutants, RGCs develop normally, but project, in variable proportions, to the ipsilateral side of the brain. The new allele was discovered in the OKR screen, because mutants showed reversed eye movements in response to a drifting grating, as is expected from a predominantly ipsilateral projection [9,27]. The reversed response is seen only when the grating rotates around the mutant, as in the OKR assay, because in this situation the direction of motion is opposite between the two eyes (e.g., temporal-to-nasal for the right eye and nasal-to-temporal for the left eye). In the OMR assay, both eyes are exposed to motion flowing in the same direction. Consequently, the OMR of bel mutants is intact.
The RGC layer of bojs307 mutants is dramatically reduced to about a third of that in WT (Figure 7A and 7B). The optic nerve is thinner, and a variable fraction (up to 50%) of the remaining RGC axons project ipsilaterally (Figure 7C and 7D). Although the axons make this abnormal choice at the midline, they nevertheless show appropriate targeting on the ipsilateral side, innervating the optic tectum as well as the other major AFs. The boj mutation complements mutations in both lakritz (encoding Atoh7/Ath5) [17] and daredevil (encoding an unknown protein) [28], two previously described genes important for RGC genesis or differentiation. The boj mutants are visually impaired to variable degrees, but most severely in the OMR. Based on our finding that the OMR is normal in bel, the OMR deficit in boj is likely due to the reduced number of RGCs, rather than the ipsilateral projection. Another possible cause could be an as-yet unknown patterning defect in the brain, which is often found in ipsilateral RGC projection mutants [29].
In darls327 mutants, the ventral branch of the optic tract is completely missing, and with it AF-2, AF-3, and AF-6; the dorsal optic tract (with AF-4, AF-5, AF-7, AF-8, and AF-9) appears intact (Figures 7F and 8). The tectum has normal size and histology, but only its dorsal half is innervated at 7 dpf; the ventral half is devoid of retinal input. We asked if the dorsal RGCs, which project their axons to the ventral branch of the optic tact in WT fish (Figure 8A), are missing in darls327 mutants. We detected differentiated RGCs throughout the retina, including the dorsal part (Figure S2). Axon tracing, following injection of 3,3′-dioctadecyloxacarbocyanine (DiO) and 1,1′-dioctadecyl-3,3,3′,3′- tetramethylindodicarbocyanine (DiD) into the nasal-dorsal and temporal-ventral quadrants of the eye, respectively, revealed that the dorsally located RGCs project into the dorsal, instead of the ventral, branch of the optic tract, sharing the same route as the ventral RGCs (Figure 8B). The absence of both the ventral optic tract and the ventral innervation of the tectum (Figure 8B and 8D) suggests that the darl gene is required for specifying dorsal RGC fate. Positional information along the temporal-nasal axis of the retina seems unaltered in the mutant. Despite the severity of the anatomical defect, this mutant's OMR and OKR scores are not substantially reduced. The VBA, however, is severely disrupted, suggesting that this neuroendocrine behavior requires input from dorsally specified RGCs.
The mutants walks536, exas174, and misss522 show specific axon targeting defects, best seen in, but not restricted to, AF-4. AF-4 is associated with the dorsal branch of the optic tract and normally has a well-ordered, compact structure (see Figure 7E). In walks536 and exas174, AF-4 is overelaborated and located at a greater distance from the optic tract (see Figure 7G and 7H). The tectum in the exas174 mutant shows an abnormal shape, particularly in the ventral-posterior region (Figure S3), and AF-9 is often missing or reduced (unpublished data). In misss522 mutants, on the other hand, AF-4 and AF-9 are reduced in size or undetectable (see Figure 7I). This mutant is completely unresponsive to motion, while the walks536 and exas174 mutants show residual OKR and OMR (Table 1). In all three mutants, AFs associated with the ventral tract appear normal. This observation, together with the finding that OMR and OKR are intact in darls327 mutants, which lack the ventral tract, suggest that one or more AFs in the dorsal tract play a key role in OMR and OKR.
In michs314 mutants, a subset of RGC axons make an abnormal turn shortly after crossing the midline and stall to form an ectopic AF (see Figure 7J). The location of this new retinorecipient area is highly consistent among individual mutants. Another OMR mutant, shirs362, has a severely retarded retinofugal projection at 5 dpf, which recovers by 7 dpf, although the dorsal optic tract remains thinner (Figure S4). Finally, in blins573 mutants, axon arbors in the tectal neuropil are disorganized and, in drg (two alleles), a subset of the RGC axons project to the incorrect layer of the tectum [28]. The axon-targeting phenotypes described here are, for the most part, so subtle and localized that they would have escaped previous lipophilic carbocyanine dye-tracing screens [30].
Genes Apparently Required for the Function of Specific Behavioral Pathways
Two mutants, ofrts373 and amjs391, show severe VBA defects with only minor OKR and OMR impairments. Strikingly, the VBA of amjs391 is reversed: The mutant turns dark in the light and light in the dark, which is the opposite of what is seen in WT. At what stage the photoresponse is inverted in this mutant will have to be elucidated. In addition, we discovered several mutants with VBA defects, but normal OMR and OKR, which are not included in Table 1.
Two other VBA mutants, dpgs128 and jakos326, showed normal OKR, but were impaired in the OMR. This selective deficit could not be explained by a locomotor problem, as both mutants show normal SSA and are adult viable. Specific deficits such as these may be either due to differential sensitivity to the stimuli presented in the two assays or due to differential effects of the mutation on the underlying neural circuits. Thus, our screen has discovered a small number of mutations that dissociate visual pathways underlying OMR and OKR.
Genes Required for Posture, Swimming, or Eye Movements
While we did not systematically keep OMR mutants with swimming defects or OKR mutants that did not move their eyes, we saved a small number of mutants whose phenotypes appeared to be informative with regard to specific neural pathways. The morphologically normal beats348, pahs374, slaks564, and flans513 mutants showed reduced OMR and/or OKR in combination with motor abnormalities. The pah gene was positionally cloned and shown to encode phenylalanine hydroxylase, an enzyme required for tyrosine and catecholamine synthesis (unpublished data). These mutations appear to primarily affect motor or other nonsensory central nervous system functions, although additional defects in visual processing may also be present.
Answers to all your Biology Questions
