The cytoplasmic tail of rhodopsin triggers rapid rod degeneration in kinesin-2 mutants

Photoreceptor degeneration can lead to blindness and represents the most common form of neural degenerative disease worldwide. Although many genes involved in photoreceptor degeneration have been identified, the underlying mechanisms remain to be elucidated. Here we examined photoreceptor development in zebrafish kif3a and kif3b mutants, which affect two subunits of the kinesin-2 complex. In both mutants, rods degenerated quickly, whereas cones underwent a slow degeneration process. Notably, the photoreceptor defects were considerably more severe in kif3a mutants than in kif3b mutants. In the cone photoreceptors of kif3a mutants, opsin proteins accumulated in the apical region and formed abnormal membrane structures. In contrast, rhodopsins were enriched in the rod cell body membrane and represented the primary reason for rapid rod degeneration in these mutants. Moreover, removal of the cytoplasmic tail of rhodopsin to reduce its function, but not decreasing rhodopsin expression levels, prevented rod degeneration in both kif3a and kif3b mutants. Of note, overexpression of full-length rhodopsin or its cytoplasmic tail domain, but not of rhodopsin lacking the cytoplasmic tail, exacerbated rod degeneration in kif3a mutants, implying an important role of the cytoplasmic tail in rod degeneration. Finally, we showed that the cytoplasmic tail of rhodopsin might trigger rod degeneration through activating the downstream calcium signaling pathway, as drug treatment with inhibitors of intracellular calcium release prevented rod degeneration in kif3a mutants. Our results demonstrate a previously unknown function of the rhodopsin cytoplasmic domain during opsin-triggered photoreceptor degeneration and may open up new avenues for managing this disease.

Photoreceptor cells consist of highly polarized neurons with a distinct photosensitive compartment, termed the outer segment (OS). 2 The OS comprises a modified cilium with a large stack of disk membranes containing opsins, the most abundant protein of the cell. The retina contains two types of classic photoreceptors, rods and cones, which are responsible for vision at low and high levels, respectively. Because of a lack of protein translation machinery, all of the OS protein components must be first synthesized in the inner segment (IS) and transported to the OS through a narrow channel called the connecting cilium (1).
Similar to cilia in other cells, the connecting cilium contains the microtubule-based axoneme with a "9 ϩ 0" configuration. The initiation and maintenance of connecting cilia require an elaborate intraflagellar transport (IFT) process that is driven by two classic motors, kinesin-2 (anterograde) and cytoplasmic dynein 2 (retrograde). The heterotrimeric kinesin-2 motor is composed of three subunits, including two kinesin-2 motor subunits (KIF3A and KIF3B) and one nonmotor subunit (KAP3) (2,3). In vitro studies suggest that KIF3C, another kinesin-2 protein, may substitute for KIF3B in the kinesin-2 motor (4,5). In addition, KIF17, another member of the kinesin-2 family, functions as a dimer during ciliary transport (3).
Kinesin-2 family members mediate protein transport in photoreceptor cells. Loss of function of KIF3A in the mouse retina interrupts the transport machinery and leads to rapid photoreceptor cell death (6 -8). In zebrafish, mutation in kif3a gene leads to mispositioning of the basal bodies and photoreceptor cell death (9). Dysfunction of Kif3b, however, affects only a subset of cilia, and photoreceptor connecting cilium is initially absent but then partially recovered latterly (10). Although inhibition of Kif17 by dominant-negative expression or morpholino-based knockdown may affect OS development, zebrafish kif17 mutants are able to grow to adulthood with normal photoreceptor functions (10 -12). Furthermore, another study has suggested that KIF17 is dispensable for photoreceptor development (13).
Photoreceptor degeneration occurs in many human genetic disorders, including retinitis pigmentosa (RP), the most common form of inherited retinal degeneration (14). Previous studies suggest that ectopic opsin accumulation leads to photoreceptor cell death and that diminishing expression of the rho gene can delay the degeneration of rod photoreceptors (7,15). However, the requirement of kinesin-2 for opsin transportation remains controversial. Mutation of either the IFT or KIF3A gene results in rod degeneration together with mislocalization of rhodopsin, indicating that IFT mediates opsin transport (6,16,17). Further analysis using the fluorescence recovery after photobleaching method in live cells also shows that KIF3A plays a role during opsin transport (18). Conversely, recent studies suggest that opsin transport may still occur without the function of both KIF3 and KIF17 motors (13,19,20).
The structure and function of photoreceptor cells are similar in zebrafish and humans. The zebrafish retina contains five types of photoreceptor cells, including one rod photoreceptor cell type and four cone photoreceptor cell types. According to their absorption spectra and morphology, cone photoreceptors are classified as short single cones sensitive to ultraviolet light, long single cones sensitive to blue light, and double cones comprised of a red-sensitive and a green-sensitive cell (21,22). Cone and rod photoreceptors form a highly ordered mosaic in the adult retina (23). Previously, we reported that rod photoreceptors degenerated in zebrafish kif3b mutants, although the underlying mechanisms are yet to be understood (10). Here we further generated zebrafish kif3a mutants and investigated rod and cone development therein. As for kif3b mutants, rod photoreceptors in kif3a mutants exhibited a faster degeneration process than cones. Further experiments demonstrated that opsins were accumulated in the apical part of cone photoreceptors, whereas most rod opsins localized to the cell membrane. We next generated a zebrafish mutant allele lacking the carboxyl-terminal cytoplasmic tail of rhodopsin and found that rods were maintained in kif3a/rho double mutants. Furthermore, we showed that the cytoplasmic domain plays a key role during rod degeneration, as overexpression of full-length opsin or the cytoplasmic domain alone, but not the form lacking the cytoplasmic domain, accelerated rod degeneration in kif3a mutants. Finally, drug treatment with calcium channel blockers increased the rod survival ratio in kif3a mutants, suggesting that the calcium signaling pathway is involved in photoreceptor degeneration.

Rapid rod degeneration in kif3a mutants
To elucidate the mechanisms underlying the degeneration of rod photoreceptors observed following kif3b dysfunction in early zebrafish embryos (10), in this study, we further generated kif3a mutants using CRISPR/Cas9 methods (supplemental Fig.  S1, A-C). Similar to other ciliary mutants, zebrafish kif3a mutants displayed body curvature at 3 days post fertilization (dpf). Whole-mount immunostaining with an anti-acetylated ␣-tubulin antibody demonstrated that all cilia were absent in kif3a mutant larvae (supplemental Fig. S1, D-I), which differs from the consequences of mutation of the other two kinesin-2 genes, kif3b and kif17 (9 -11). The complete absence of cilia has also been described in another kif3a mutant line derived from N-ethyl-N-nitrosourea mutagenesis (data not shown and Ref. 9). These results indicate that Kif3a is indispensable for ciliogenesis in zebrafish.
We next examined photoreceptor development in kif3a mutants by crossing these animals to Tg(Xla.Rho:GFP), a rod-specific transgenic line driven by the Xenopus rhodopsin promoter (referred to here as XOPS-GFP) (23). At 3 dpf, the GFP signals in the retina of wild-type and kif3a mutants were indistinguishable (supplemental Fig. S2). However, kif3a mutants displayed a rapid decrease of GFP signals in the retina in the following 48 h of development, and only scattered GFP signals could be observed in the sections of mutant retina at 5 dpf ( Fig.  1, A-C, and supplemental Fig. S2). We further analyzed the expression of rod-specific genes via quantitative PCR (qPCR) and whole-mount in situ hybridization (WISH) and found that these genes were all down-regulated in kif3a mutants at 5 dpf ( Fig. 1, D-F). These results suggest that mutation of kif3a leads to fast rod degeneration from 3 to 5 dpf, which is similar to that observed in kif3b mutants (Fig. 1, D and F) (10).

Cone photoreceptor development in kif3a mutants
In kif3b mutants, cone photoreceptors survive longer than 5 dpf, as shown by both WISH and qPCR analysis (Fig. 2, A and  B). Similarly, WISH results for two cone opsin genes, uv and green, demonstrated similar staining levels at 5 dpf in the retinae of kif3a and wild-type larvae ( Fig. 2A), whereas qPCR analysis indicated that the expression of cone-specific genes decreased slightly (Fig. 2C). Next, we examined cone development via whole-mount immunostaining using the zpr-1 antibody, which recognizes Arr3a and selectively labels cell body of the red-and green-sensitive double cones (22,24). In contrast to the fast degeneration of rods, the majority of double cones were still present at 4 dpf in the retinae of kif3a mutants, and only a small number of cones degenerated at 5 dpf, suggesting that cones degenerate much slower than rods in kif3a mutants ( Fig. 2D and supplemental Fig. S2). Similarly, the double cones of kif3b mutants showed a grossly normal staining pattern comparable with that of wild-type siblings at 5 dpf. In contrast, the cone photoreceptors in another ciliary mutant, ovl, which harbors a mutation that affects one of the IFT proteins, Ift88, displayed a much more severe degeneration phenotype at the same developmental stage ( Fig. 2D and supplemental Fig. S2).
Next, we crossed kif3a mutants to a stable transgenic line, Tg(hsp:GFP-CT44), which expresses GFP fused to the 44 C-terminal amino acids of rhodopsin driven by a heat shock promoter (25,26). In wild-type embryos, this fusion protein could be easily transported to the outer segment within 4 h after heat shock (25). We reasoned that such GFP signals might localize to the inner cell membrane of photoreceptors because of the dysfunction of IFT in kif3a mutants. However, we noticed that the majority of GFP signals accumulated at the apical region of photoreceptor cells in kif3a mutants after heat induction at 5 dpf (Fig. 2E). Further immunostaining results with Tom20, a mitochondrial marker, suggested that these GFP fusion proteins localized apically to the mitochondria (Fig. 2E). Similarly, such protein accumulation also occurs in UV light-sensitive photoreceptor cells, as shown using another stable transgenic line expressing tdTomato fused to the same outer segment targeting signals driven by the sws1 promoter (27,28). Noticeably, a small amount of fusion proteins also localized to the cell membrane and synapse in the mutants (Fig. 2E, arrows). These results suggest that opsin proteins may be sequestered apically in the mutant photoreceptors. Interestingly, most of these Photoreceptor degeneration in kinesin-2 mutants ectopic opsins did not colocalize with Rom1b, another outer segment membrane protein (supplemental Fig. S3). Next, we performed electron microscopy analysis of the retina of 5-dpf larvae. As shown in Fig. 2, F-I, irregular membrane structures were consistently formed on the exteriors of mitochondria in the photoreceptors of kif3a mutants. Considering that rods already degenerate at this stage, these results suggest that the majority of opsins localized to the apical region of cone photoreceptors in kif3a mutants.

Redundancy between Kif3b and Kif3c
In contrast to the complete absence of cilia in kif3a mutants, loss of function of Kif3b only results in ciliogenesis defects in a subset of cilia ( Fig. 3F) (10). Robust cilia were still present in the central nervous system (CNS) of kif3b mutant embryos on a Tg(bactin2:Arl13b-GFP) transgenic strain background (data not shown). Similarly, although rods degenerated quickly in kif3b mutants, the cones were relatively normal compared with those of kif3a mutants (Fig. 2, A-D, and supplemental Fig. S2). The mild phenotype of kif3b mutants may be related to the presence of the Kif3c protein, which can also bind to Kif3a in vitro (4). Zebrafish carry two KIF3C homologues, kif3ca and kif3cb, both of which are mainly expressed in the CNS, although kif3ca is expressed at a relatively higher level (Fig. 3, A-D). To test in vivo functional redundancy between Kif3c and Kif3b, we further generated zebrafish kif3ca mutants (Fig. 3E). Although cilia developed normally in kif3ca mutants, kif3b/kif3ca double mutants failed to develop cilia in ear cristae and photoreceptor cells, suggesting that Kif3ca is involved in the formation of these cilia in kif3b mutants (Fig. 3F). In addition, the photoreceptor layer was also severely disorganized in the double mutants (Fig.  3F). Next, we generated a stable transgenic line expressing GFP fused to the N terminus of the Kif3ca protein and driven by a heat shock promoter (Fig. 3G). We found that all examined cilia that were absent in kif3b mutants could be fully rescued upon heat induction, including motile, non-motile, single, and multiple cilia (Fig. 3, H-P). These results demonstrate that Kif3b and Kif3ca are interchangeable during ciliary transport.
Next, we focused on photoreceptor development in kif3b mutants. The expression of rod-specific genes was maintained following Kif3ca protein overexpression (Fig. 4, A and I). Rod outer segments and photoreceptor-connecting cilia were recovered as well (Fig. 4, B-H). In contrast, overexpression of another member of the kinesin-2 protein family, Kif17, could not prevent rod degeneration, although the cilia could be rescued in several other tissues, including the spinal cord ( Fig. 4A) (10). Finally, overexpression of neither kif3ca nor kif3cb can

Photoreceptor degeneration in kinesin-2 mutants
rescue cilia or photoreceptor defects in kif3a mutants, although both can rescue ciliary defects in kif3b mutants, suggesting the essential function of Kif3a during ciliogenesis (supplemental Figs. S4 and S5). Together, these results suggest that function of Kif3ca and Kif3b are redundant in ciliogenesis, which is the reason underlying the mild phenotype in kif3b mutants.

Excess opsin accumulation contributes to the rapid degeneration of rod photoreceptors
Considering that opsin mislocalization is one of the causes of photoreceptor cell death, we further investigated opsin distribution using the Tg(hsp:GFP-CT44) transgenic line. Specifically, the transgene was heat-induced at 3.5 dpf when most of the rods were still present in the kif3a mutants. In the photoreceptors of wild-type embryos, the GFP signals exclusively accumulated in the outer segments 12 h after heat shock, whereas robust signals were still present in the cell membrane of the photoreceptors in kif3a mutants (data not shown). To better distinguish cones and rods, we isolated single photoreceptors and simultaneously labeled them with zpr-1 and zpr-3 antibodies. Zpr-3 is a monoclonal antibody that can recognize outer segments of both rods and short double cones (29,30). In the isolated mutant photoreceptors, the short double cones would be labeled with both antibodies (cell membrane, zpr-3; cell body, zpr-1), whereas only the cell membrane would be labeled in rod cells (Fig. 5A). Comparison of the average fluorescence intensity in the cell membrane between double cones and rods identified that the GFP signals were significantly higher in the membrane of rods than those of cones (Fig. 5, A and B). We further investigated the distribution of endogenous proteins by staining with an anti-bovine rhodopsin antibody, 4D2. Because the 4D2 antibody recognizes both rod and double cone opsins in zebrafish (supplemental Fig. S6), we performed this analysis using the XOPS-GFP transgenic line to identify rod photoreceptors. Similarly, rod cell membrane was found to contain a higher level of opsins than those of cones (Fig. 5, E and F).
Next, we repeated the opsin transport experiments in ift88 mutants, which exhibit rapid degeneration of both rods and cones (supplemental Fig. S2). As shown in Fig. 5C, both rod and cone cell membranes contained robust but similar fluorescence intensities, indicating that opsin transport was similarly affected in cones and rods lacking Ift88 proteins (Fig. 5, C and  D). The high GFP intensity in the cell membrane was consistent with the rapid photoreceptor degeneration in kif3a and ift88 mutants. These results suggest that excess opsin accumulation in the cell membrane of rods contributes to their fast degeneration in kif3a mutants.

Disruption of rhodopsin function prevents rod degeneration in kif3 mutants
As the rod cell membrane contains a higher level of opsin proteins, we reasoned that elimination of rhodopsin might prevent rod degeneration in kif3 (kif3a and kif3b) mutants. To test this hypothesis, we generated a mutant line harboring a 5-bp

Photoreceptor degeneration in kinesin-2 mutants
deletion in the rho gene, which encodes a truncated protein lacking the entire cytoplasmic carboxyl terminus (amino acids 315-354) (Fig. 6A). Whole-mount immunostaining and ultrastructural results suggested that the photoreceptor cell layer appeared to be grossly normal in the mutants at the larval stage (Fig. 6, B-E, and data not shown). Notably, the truncated proteins still localized to the outer segments at the larval stage, as shown by zpr-3 and 4D2 antibody staining (Fig. 6, D and E, and data not shown). In contrast, the majority of rod outer segments were missing in the adult retina (Fig. 6, F-I). The rod nuclear layer was also significantly thinner than that of wild-type siblings (Fig. 6, F-I). In the remaining rod photoreceptors, the mutant rhodopsins ectopically localized to the cell membrane (Fig. 6I, arrows). These results suggest that the rod cells had degenerated during development in this mutant line.
We further generated kif3a/rho and kif3b/rho double mutants and investigated the development of rod photoreceptors at 5 dpf. WISH and qPCR results with rod-specific genes showed that the expression level was largely maintained in the double mutants (Fig. 7, A-D and H, and supplemental Fig. S7). Although outer segments did not form, the photoreceptor cell layers were grossly intact in the double mutants (Fig. 7, E-G). Notably, rod degeneration was also prevented in kif3a/rho ϩ/Ϫ mutants (Fig. 7H). Strikingly, the expression level of the mutant rho gene remained unchanged, and the truncated proteins were still enriched in the cell membrane in the double mutants (Fig.  7, D, G, and H). These results suggest that the cytoplasmic tail plays a role during rod degeneration.

The rhodopsin cytoplasmic tail mediates photoreceptor degeneration in kif3 mutants
To further test the role of the rhodopsin cytoplasmic tail during photoreceptor degeneration, we overexpressed wildtype or mutant rhodopsin in kif3a mutants and investigated rod development therein. In wild-type embryos, overexpression of either full-length or mutant rhodopsin lacking the cytoplasmic tail did not affect rod development (Fig. 7, I and Q). In contrast, overexpression of full-length rhodopsin accelerated rod degeneration in kif3a mutants, as suggested by the number of GFPpositive rod photoreceptors (Fig. 7, J, L, and Q). The number of rod cells appeared to be indistinguishable between mutant retinae injected with the cytoplasmic tail-deficient rhodopsin and uninjected mutant controls (Fig. 7, J, K, and Q). Strikingly, overexpression of the cytoplasmic tail alone by heat shock induction significantly affected the survival of rod photoreceptors (Fig. 7, M-P and R). These results strongly suggest that the rhodopsin cytoplasmic tail participates in photoreceptor degeneration in kif3a mutants.
In the outer segments, rhodopsin activates its G-protein, transducin, and initiates the phototransduction signaling pathway. We speculated that rhodopsin might activate the apoptosis signaling pathway via its cytoplasmic tail when localized to the cell body membrane ectopically. As calcium is a key element in apoptotic signaling, we tested the effects of modulating calcium signaling on rod development in kif3 mutants. First, we treated the kif3a mutants and siblings with thapsigargin, which helps raise intracellular free calcium by inhibiting the endoplas-

Photoreceptor degeneration in kinesin-2 mutants
mic reticulum Ca-ATPase (31). As shown in Fig. 8, thapsigargin treatment exaggerated rod degeneration in both kif3a and kif3b mutants (Fig. 8, A-E). In contrast, drug treatment with two intracellular calcium release inhibitors, dantrolene or 2-aminoethoxydiphenyl borate, increased the number of surviving rod cells significantly (Fig. 8, F and G). These results suggest that ectopic rhodopsin may trigger a calcium-dependent apoptosis pathway via its cytoplasmic tail.

Discussion
Members of the kinesin-2 family play diverse roles during vertebrate ciliogenesis. In zebrafish, mutation in kif3b gene results in defects in a subset of cilia, whereas almost all cilium formation is affected in kif3a mutants (Refs. 9, 10 and this study). In contrast, Kif3c and Kif17 appear to be dispensable for zebrafish development (9,10). During photoreceptor development, kinesin-2 proteins drive the movement of IFT complexes and facilitate the transport of cargo proteins through the connecting cilium. Our results show that both Kif3a and Kif3b are required for photoreceptor development. Although not necessary, Kif3ca could substitute for Kif3b to maintain protein transport in the photoreceptors as well as in the other types of cilia examined. Notably, dysfunction of either Kif3a or Kif3b led to rapid degeneration in rod photoreceptors, whereas cones survived longer than rods in the mutants. In kif3b mutants, cones degenerated slower than in kif3a mutants, possibly because of the presence of small amounts of Kif3ca in kif3b mutants that maintain the IFT mechanisms.
The apical accumulation of opsin proteins in cone photoreceptors represents a notable phenomenon in kif3a mutants. We consider that such accumulation occurs mainly in cone photoreceptors, as robust opsin fusion proteins were still present on the apical regions when the larvae were heat induced as late as 7 dpf, at which time the majority of rods had degenerated (data not shown). Our results demonstrated that rods and cones differed from each other upon removal of the kif3a gene. In photoreceptor cells of zebrafish larva, the outer segments start to form at around 2.5 dpf, concomitant with the initial synthesis of opsin and other outer segment proteins in the inner segments. These outer segment proteins are first transported to the base of the connecting cilium, driven by cytoplasmic dynein 1 motors (Fig. 8H), and then transported to their destination through the connecting cilium. Subsequently, the outer segments start to expand with the delivery of additional opsins, membrane proteins, and other scaffold proteins. In kif3a mutants, these proteins can still be transported to the basal body region; however, no connecting cilium is formed because of the lack of IFT, which is necessary for tubulin delivery, a key component of the ciliary axoneme (32,33). In rod photoreceptors, membrane proteins, including rhodopsin, are transported to the base of the outer segments to form a new membrane disk. However, consequent to the lack of a connecting cilium, these proteins will be initially concentrated at the cell membrane in the apical region and then quickly dispersed to the entire plasma membrane by diffusion (Fig. 8H). Such ectopic rhodopsin proteins may trigger the apoptosis signaling pathway via their cytoplasmic tails, which would further activate calcium signaling and lead to eventual photoreceptor degeneration. In kif3/rho double mutants, lack of the cytoplasmic tail prevents the activation of cell apoptosis signaling, and rod degeneration is blocked (Fig. 8H). In contrast, cones appear to have different transport mechanisms because of the continuity of the membrane disks, which can quickly deliver opsins to the entire outer   In wild-type photoreceptors, opsins were synthesized in the inner segment and transported to the outer segment in both rods and cones. In kif3 mutants, ectopic opsins accumulate in the cell membrane of rods and activate the calcium signaling pathway, leading to rapid rod degeneration. In contrast, abnormal membrane structures develop in the apical region of cone photoreceptors, which leads to a lower concentration of opsin protein in the cell membrane and a longer survival time. In kif3/rho double mutants, opsin proteins are still enriched in the cell membrane, but lack of a cytoplasmic tail prevents activation of the downstream cell apoptosis pathway. In ift88(ovl) mutants, both rods and cones contain higher amounts of ectopic opsins in the cell membrane and degenerate quickly (See "Discussion" for more details). **, p Ͻ 0.01; ***, p Ͻ 0.001. Scale bar ϭ 25 m.

Photoreceptor degeneration in kinesin-2 mutants
segments over a short time period (34,35). Although the connecting cilium fails to form in the cones of kif3a mutants, opsins can still be transported, and abnormal membrane structures start to form because of the lack of the ciliary axoneme. Under these conditions, a substantial portion of cone opsins become sequestered in the apical region, and only small quantities of opsins are diffused to the plasma membrane, which improves the survival ratio of cone photoreceptors (Fig. 8H).
In comparison, both cones and rods degenerate rapidly in ift88(ovl) mutants, indicating the different roles of Ift88 and Kif3 motors in opsin transport. IFT88 has been reported previously to localize in both the ciliary axoneme and the periciliary cytoplasm in photoreceptor cells and may participate in the transport of post-Golgi vesicles to the ciliary membrane during early development stages (36,37). When Ift88 is absent, the first step of opsin transport will be interrupted, and these membrane proteins will consequently be misdirected to the cytoplasmic membrane, resulting in rod and cone degeneration (Fig. 8H).
Opsin mislocalization constitutes one of the major causes of photoreceptor cell degeneration (7,15). We further confirmed this by showing that interference with the function of rhodopsin could inhibit the degeneration of rod photoreceptors in kif3 mutants. Notably, the rhodopsin present in the rho mutants lacked only the last cytoplasmic domain, and its expression level was similar to that of control siblings. Because the truncated proteins could be transported to the outer segments, it is possible that mutant rhodopsin will activate phototransduction, as it contains all three cytoplasmic loops that are necessary for the binding of G-protein (38). Of note, the cytoplasmic tail plays an essential role during deactivation of rhodopsin and, hence, the termination of the visual signals in that it contains several phosphorylation sites that are necessary for the binding of G-protein-coupled receptor kinase and arrestin (39 -41). Thus, the absence of the cytoplasmic tail may lead to excessive signaling because of overactive rhodopsin, constituting the main reason of progressive photoreceptor degeneration in humans (42,43) and zebrafish mutants.
The recovery of rod degeneration in kif3/rho double mutants suggested that the carboxyl-terminal cytoplasmic domain participated in the cell death process. In wild-type embryos, overexpression of either full-length or truncated proteins did not affect rod development, suggesting that ectopic localization of rhodopsin in the cell body membrane is a prerequisite for inducing rod degeneration. Similarly, thapsigargin treatment did not lead to significant rod degeneration in Tg(hsp:GFP-CT44) transgenic embryos, where the overexpressed CT44 mainly localizes to the outer segments (supplemental Fig. S8). Conversely, the means by which the cytoplasmic tail triggers cell apoptosis remains unknown. The signaling transduction pathways in the cell body differ considerably from those of the outer segment. It is possible that this ectopic G-proteincoupled receptor binds to certain cytoplasmic G-proteins and activates downstream signaling pathways, as suggested by previous study in Ambystoma tigrinum (44). One of the phenotypes associated with such signaling pathways is the increase of internal Ca 2ϩ signals, as suggested by the drug treatment results. Further experiments are necessary to decipher the mechanisms underlying this apoptotic pathway.
Finally, the mutation site in rho zebrafish mutants (H315) is close to the human Q312ter rhodopsin mutation, which might also result in complete loss of the entire cytoplasmic tail (45). Interestingly, human patients harboring the Q312ter mutation displayed mild retinal disease compared with other rhodopsin mutations (45), possibly suggesting that the cytoplasmic tail may also function during rods degeneration in human RP patients. In conclusion, we found a novel function of the cytoplasmic tail during ectopic rhodopsin-triggered rod degeneration. Future studies of this signaling pathway may benefit the treatment of human RP.

Whole-mount in situ hybridization and immunohistochemistry
For in situ hybridization, rod-and cone-specific genes were either purchased from Open Biosystems as reported previously (10) or cloned from a wild-type cDNA library into pGEM-T vectors. Probe preparation and hybridization were carried out using standard protocols (47). For immunohistochemistry, the following antibodies were used: anti-acetylated ␣-tubulin (1:500, Sigma), zpr-3 (1:500, Zebrafish International Resource Center), zpr-1 (1:500, Zebrafish International Resource Center), Tom20 (1:200, Abcam), and anti-rhodopsin 4D2 (1:500, Abcam). For larva fish, cryosectioning and antibody staining were performed as described previously (48). To isolate single adult photoreceptors, eyeballs were dissected from euthanatized adult fish and washed in L15 medium. After removal of the cornea and lens, the retina was put into 100 l of L15 medium and briefly rinsed several times. The supernatant, which contains single rod photoreceptors, was transferred to poly-L-lysinecoated slides, incubated for 1.5 h at 28°C, and then fixed in 4% paraformaldehyde (PFA) overnight at 4°C. The leftovers, which contained cone photoreceptors and other neurons, were also fixed in 4% PFA. The following immunostaining steps were the same as for larva photoreceptors. After staining, the confocal images were collected on a Leica TCS Sp8 upright confocal microscope.

Quantitative PCR
The primers used for qPCR are summarized in supplemental  Table S1. The PCR reaction was set up using Eva-Green Master Mix (ABM) and performed on an ABI 7500 real-time PCR system (Applied Biosystems). The amplification condition parameters were 95°C for 15 s, followed by 40 cycles of 95°C for 5 s, 60°C for 15 s, and 72°C for 35 s, and each sample had triplicate reactions. Relative gene expression levels were quantified using Photoreceptor degeneration in kinesin-2 mutants the comparative Ct method (2 Ϫ⌬⌬Ct method) based on Ct values for target genes and zebrafish ␤-actin.

Opsin accumulation analysis
To better control the expression level of the GFP-CT44 fusion protein, we first generated a stable transgenic line expressing GFP-CT44 driven by a heat shock promoter and then crossed to the kif3a ϩ/Ϫ or ift88 ϩ/Ϫ heterozygotic fish. The mutant larvae containing the transgene were heat-induced at 42°C for 1 h at 3.5 dpf and then collected at 4 dpf. After euthanasia with Tricaine, the eyeballs were collected and digested with papain (30 units/ml, Roche) in L15 medium at 37°C for 5 min and then fixed in 4% PFA overnight at 4°C. After fixation, the samples were ground, seeded onto a poly-lysine-coated slide, and dried at 37°C. Immunohistochemistry was performed using zpr-1 and zpr-3 antibodies simultaneously as described previously (48). For endogenous experiments, the kif3a ϩ/Ϫ heterozygotic fish was crossed to the XOPS-GFP transgene, and anti-bovine rhodopsin 4D2 antibody was used to label endogenous opsin protein. For quantitation of opsin accumulation, the fluorescence intensity was calculated by drawing a line through each side of the cell membrane using ImageJ software (National Institutes of Health). The average fluorescence intensity per cell was used to evaluate opsin accumulation.

Electron microscopy
Wild-type and kif3a mutant larvae were fixed overnight in 2.5% glutaraldehyde in 0.1 M PBS. The samples were washed three times in PBS and fixed again in 1% osmium tetroxide. After that, embryos were washed, gradually dehydrated to 100% acetone, and then embedded in Epon812 resin. Ultrathin sections were collected and stained with uranyl acetate and lead citrate. The sections were observed on a transmission electron microscope (JEM-1200EX) and imaged with Olympus Soft Imaging Solutions.

Overexpression and pharmaceutical treatments
For rescue experiments, the full-length kif3ca gene was fused to the C terminus of GFP and cloned into the Tol2 Gateway vector driven by a heat shock promoter. The stable transgenic line was identified and crossed to kif3b heterozygotic mutants for phenotypic analysis. For mRNA overexpression, full-length kif3cb or kif17 genes were PCR-amplified and cloned into the PXT7 vector. The mRNA was synthesized using the Ambion T7 transcription kit and injected into zebrafish embryos at a concentration of 300 ng/l. For rho overexpression experiments, full-length or mutant (‚C) rho genes were PCR-amplified and cloned into the PXT7 vector with the following primers: forward, 5Ј-CGGAATTCACCATGAACGGTACAGAGGG-ACC-3Ј; reverse (full-length), 5Ј-CCGCTCGAGTTACGCCG-GAGACACGGAGC-3Ј; reverse (‚C), 5Ј-CCGCTCGAG-TCAACGGAACTGCTTGTTCATGC-3Ј. The synthesized mRNA was injected into zebrafish embryos carrying the XOPS-GFP transgene at the one-cell stage with a concentration of 200 ng/l. The injected embryos containing the XOPS-GFP transgene were collected at 4 dpf and cryosectioned with a Leica CM1850 cryostat. Sections from the center of the retina were captured for evaluation of rod development. For overexpression of the cytoplasmic tail of rhodopsin, we generated zebrafish kif3a mutants carrying Tg(hsp:GFP-CT44) and XOPS-GFP double transgenes. Both the mutant larvae and their siblings were heat-induced at 42°C for 2 h at 3.5 dpf, transferred to 28°C, and collected at 4 dpf for rod degeneration analysis.
For drug treatment, transgenic embryos were treated either with dantrolene (50 M, Sigma) or 2-aminoethoxydiphenyl borate (0.2 M, Sigma) from 3 dpf to 5 dpf and then fixed in 4% PFA for rod degeneration analysis. For thapsigargin treatment, transgenic embryos were incubated at 1 M for 10 h at 3.5 dpf and then fixed in 4% PFA for evaluation of rod degeneration. For XOPS-GFP/Tg (hsp:GFP-CT44) double-transgenic embryos, the larvae were first heat-induced at 42°C for 2 h at 3.5 dpf and then treated with 1 M thapsigargin for 10 h either immediately after heat induction or at 4.5 dpf. After that, the treated embryos were fixed in 4% PFA for evaluation of rod degeneration.

Statistical analysis
Scatterplot data are presented as mean Ϯ S.D. as indicated in the figure legends. Statistical significance was evaluated using a t test, and p Ͻ 0.05 was taken to be statistically significant.
Author contributions-C. Z. designed all experiments. D. F. performed most of the experiments, including immunohistochemistry, q-PCR, in situ hybridization, and electron microscopic analysis. Z. C., Y. K., and H. X. performed the mutant and transgenic analyses. K. Y., B. X., and S. M. performed the mutant screen for kif3a and rho fish. D. F. and C. Z. analyzed the data and wrote the manuscript.