Transducin Duplicates in the Zebrafish Retina and Pineal Complex: Differential Specialisation after the Teleost Tetraploidisation

Gene duplications provide raw materials that can be selected for functional adaptations by evolutionary mechanisms. We describe here the results of 350 million years of evolution of three functionally related gene families: the alpha, beta and gamma subunits of transducins, the G protein involved in vision. Early vertebrate tetraploidisations resulted in separate transducin heterotrimers: gnat1/gnb1/gngt1 for rods, and gnat2/gnb3/gngt2 for cones. The teleost-specific tetraploidisation generated additional duplicates for gnb1, gnb3 and gngt2. We report here that the duplicates have undergone several types of subfunctionalisation or neofunctionalisation in the zebrafish. We have found that gnb1a and gnb1b are co-expressed at different levels in rods; gnb3a and gnb3b have undergone compartmentalisation restricting gnb3b to the dorsal and medial retina, however, gnb3a expression was detected only at very low levels in both larvae and adult retina; gngt2b expression is restricted to the dorsal and medial retina, whereas gngt2a is expressed ventrally. This dorsoventral distinction could be an adaptation to protect the lower part of the retina from intense light damage. The ontogenetic analysis shows earlier onset of expression in the pineal complex than in the retina, in accordance with its earlier maturation. Additionally, gnb1a but not gnb1b is expressed in the pineal complex, and gnb3b and gngt2b are transiently expressed in the pineal during ontogeny, thus showing partial temporal subfunctionalisation. These retina-pineal distinctions presumably reflect their distinct functional roles in vision and circadian rhythmicity. In summary, this study describes several functional differences between transducin gene duplicates resulting from the teleost-specific tetraploidisation.


Introduction
Phototransduction is the process whereby light is converted into electrochemical signals. It predominantly takes place in photoreceptor cells, but also in other non-specialised cell types.
The present study was designed to provide an evolutionary perspective on the specialisation of transducin paralogs, focusing on rods versus cones in the retina and possible specialisations in the pineal complex. The expression patterns of the transducin genes in zebrafish were analysed during the ontogeny as well as in the adult stage, giving special attention to the gene duplicates retained after 3R: gnb1a/gnb1b, gnb3a/gnb3b and gngt2a/gngt2b.

Materials and Methods
This study was carried out in strict accordance with the recommendations of the Federation of Laboratory Animal Science Associations. The project was approved by the Uppsala Ethical Committee on Animal Experiments (Uppsala djurförsöksetiska nämnd), permit numbers: C33/10, C294/12 and C315/12. Prior to any procedure, all animals were anaesthetised with benzocaine (0.5 ml/L). Phylogenetic cladogram summarising the evolution of the transducin subunit genes within vertebrates. The three transducin subunit genes present in the vertebrate ancestor cone-like photoreceptors ("proto-cones") (GNAT, GNB and GNGT) gave rise to two paralogs for each gene represented in extant tetrapods after 1R and 2R: GNAT1-GNAT2, GNB1-GNB3 and GNGT1-GNGT2. This facilitated the appearance of rods. In teleost fishes the 3R tetraploidisation resulted in three pairs of paralogs that have been retained in zebrafish: gnb1a/gnb1b, gnb3a/gnb3b and gngt2a/gngt2b.

Zebrafish embryos, larvae and adults
The embryos and larvae used in this study were AB wild type and albino strains. All embryos and larvae were collected and prepared for gene expression analyses as described previously [33]. The adult animals were 6 months to 1 year old AB zebrafish (n = 58), purchased from the Science for Life Laboratory Zebrafish Technology Platform (Uppsala University), or conespecific transgenic zebrafish, Tg(gnat2:EGFP) [34], provided by Dr. B. Kennedy (University College Dublin) (n = 6).
Standard cloning procedures were applied to synthesise the riboprobes. In summary, specific primers against the 3´untranslated region (3´UTR) of each gene were used to amplify a 202-550bp sequence by PCR. The obtained amplicon was inserted into a pCRII-TOPO vector and cloned in one shot chemically competent TOP 10 E. coli. The primer pairs for each transducin subunit gene are shown in Table 1. The use of the 3´UTRs ensures higher specificity, as they hold higher nucleotide variability than the coding regions. Subsequently, all riboprobes (sense and antisense) were synthesised by either T7 or SP6 RNA polymerases using a DIG RNA Labelling Kit (Roche: cat. no. 11175025910), labelling them either with digoxigenin (DIG) or fluorescein (FITC).
Primer pair sequences used to amplify all zebrafish-specific transducin genes. Primer pairs 1-9 were used to synthesise the antisense and sense riboprobes used in ISH experiments, Table 1. Primer sequences used to amplify the transducin genes.
To prepare the sections for ISH, adult zebrafish heads were dissected immediately after anaesthesia, fixed by immersion in 4% paraformaldehyde diluted in phosphate buffer (0.1M, pH 7.4) for 6 to 8 hours, and washed in phosphate buffer saline (0.1M, pH 7.4) overnight. The tissues were cryoprotected by immersion in 30% sucrose diluted in phosphate buffer and sectioned in a cryostat (Microm Cryo-Star HM 560). Transversal or sagittal sections of the whole head (12-20 μm thick) were obtained and stuck to positive-charged slides.
ISH (n = 20) and WISH (n = 4) were performed according to Hauptmann and Gerster, 2000 [33], with minor adaptations in the case of ISH. The final staining reaction was carried out using different substrates for the AP enzyme bounded to the Fab fragments: NBT/BCIP or Fast Red tablets (Roche: cat. no. 11681451001 and 11496549001, respectively). Double ISH according to Hauptmann, 2001 [35] was also performed in specific cases.
Specific expression was tested in the eye for each gene by RT-PCR using eye cDNA (see description below). To test the specificity of the antisense probes, sense probes were synthesised and incubated in parallel with the antisense probes, obtaining no staining for any of them. In addition, non-labelled antisense probes were used in a competitive reaction with the labelled antisense probes at different concentrations.
To confirm the cell type assignment for each gene, ISH experiments (n = 6) were performed on Tg(gnat2:EGFP) zebrafish (Figs. 2B and C). The cone-specific EGFP fluorescence was enhanced by incubation with a mouse anti-GFP antibody (1:400, Invitrogen: cat. no. A-11120) and a secondary donkey anti-mouse coupled Alexa 488 (1:1000, Invitrogen: cat. no. A-21202).  Raymond and Barthel, 2004 [36] and a Nomarski contrast photomicrograph from a radial section of an adult AB zebrafish retina (A). All photoreceptors have a terminal in the outer plexiform layer (OPL). The outer nuclear layer (ONL) is subdivided by the outer limiting membrane (olm) into a sublayer where the rods have their nuclei (black asterisk: rONL) and a more external sublayer where the cones have their nuclei (empty asterisks: cONL). Note that the short single cones (SSC) have their nuclei in the rONL while their oil droplets (black arrows) and outer segments (OS) are in the cONL. The rods' myoids (empty arrows) are embedded in the cONL. The long cone inner segment sublayer (lCIS) includes the oil droplets and myoids of the double cones (DC) and the long single cones (LSC) and part of the LSC OS. The outermost part of the retina (top) is covered by the pigmentary epithelium where the DC OS (dCOS) and the rods' ellipsoid (arrowhead) and outer segments (ROS) are embedded. B and C are confocal photomicrographs of in situ hybridisation experiments performed using Tg (gnat2:EGFP) zebrafish, which expresses EGFP in all cones. Specific Fast Red staining in rods for gnb1a (B) and cones for gnat2 (C) can be observed. Arrows in B point to the rod´s myoids that in some cases could be mistaken as SSC. In C, the arrowhead points to a DC, the asterisk to a LSC and the star to a SSC. Scale bar in A represents 25 μm. In order to prevent the EGFP decay, the hybridisation time was reduced from overnight to 6 hours. Antibodies against GNAT1, GNB1 and opsins were also used as markers but not included in the study.

Microscopy and photography
The general anatomy of the zebrafish retina has been described previously in detail [37], and it is known to follow the general organisation pattern common to vertebrates [38]. Our results were analysed based on overall cell morphology, topological location of the nuclei and oil droplets of the different photoreceptor cell types, as well as cone-specific EGFP fluorescence of the Tg(gnat2:EGFP) zebrafish (Fig. 2). We have used the nomenclature proposed by Raymond and Barthel, 2004 [36] for the different cone types present in the zebrafish retina: double cones (DC, middle and long wavelength), long single cones (LSC, short wavelength) and short single cones (SSC, ultraviolet).
Bright-field, fluorescence and Nomarski contrast photomicrographs, as well as combinations, were taken for the ISH experiments on slides, using a Zeiss Axioplan 2 microscope equipped with a Zeiss AxioCam camera. In addition, an inverted LSM510 Zeiss confocal microscope was used for detailed pictures. The images of WISH embryos and larvae were acquired using a stereomicroscope Nikon SMZ1500 with a Nikon DS-Vi1 camera All images were processed and the figures merged using CorelDRAW Graphics Suite X6.

Expression level analyses
Gene expression levels were analysed by reverse transcriptase quantitative PCR (RT-qPCR) for gnat2, gnb1a, gnb1b, gnb3a and gnb3b, using β-actin I (bactin1) and α-tubulin 1 (tubaI) as internal control (housekeeping) genes. Adult animals (n = 30) were kept in a 14-10 light-dark cycle and sacrificed at different Zeitgeber (ZT) time points during 24h (ZT0, ZT4, ZT8, ZT12, ZT16 and ZT20). This enables the study of rhythmic oscillations on gene expression without any endogenous interference derived from a dark or light adaptation processes. Total mRNA was extracted from adult eyes (n = 5 per time point) or from whole 1 day post-fertilisation (dpf), 2 dpf and 3 dpf embryos (n = 30) with an RNeasy Mini Kit (Qiagen: cat. no. 74104) precipitated for higher purity, treated with DNase I (Fermentas, cat. no. EN0525) and used for downstream cDNA synthesis in RT and NRT reactions using M-MLV RT Reverse Transcriptase (Invitrogen: cat. no. 28025-013) according to the manufacturer´s instructions.
RT-qPCRs were carried out in a CFX96 RT-PCR detection system (Bio-Rad), using IQ SYBR Green Supermix (Bio-Rad: cat. no. 1708880) and specific primers ( Table 1). All samples were analysed in triplicates. No template controls (NTCs) were also performed for each primer pair. A reference pool of cDNA and NRT was also used in triplicate for each plate to control inter-plate variation between different runs. After each RT-qPCR run, a melt curve was made to control for primer dimers and the correct amplicon. The results were compared to the expression of β-actin, as internal housekeeping, using the 2 -ΔCt method and analysed using Lin-RegPCR 2014.6 [39]. The statistical differences were calculated using StatPlus:mac V5 two-way ANOVA and Tukey HSD post-hoc test.

Results
The present study focuses on the specialisations of the zebrafish transducin genes. For all positive cases, the staining was found in the photoreceptors, not in the inner retina, and most of them also in the pineal complex. No specialised expression within photoreceptor types (rods or cones) could be found for any of the transducin gene 3R-generated duplicates. Additionally, stronger intensity was observed in the dorsal retina for the genes with expression in the entire retina, which could be related to the asymmetrical cell distribution and organisation between the dorsal and the ventral retina previously described in zebrafish [40].

Expression patterns of the transducin genes in the adult zebrafish retina
The presence in the zebrafish genome of all the transducin genes: gnat1, gnat2, gnb1a, gnb1b, gnb3a, gnb3b, gngt1, gngt2a and gngt2b was verified by PCR using genomic DNA and the 3 UTR primers designed for each gene (see Table 1). The same primers were also used to investigate the transducin gene expression in the retina by RT-PCR using retina cDNA and confirmed that all transducin genes, except gnb3a, are transcribed in the retina. Subsequently, subtype-specific antisense DIG/FITC-labelled riboprobes were used in ISH. Positive staining was observed in the outer retina for all the genes, with the exception of gnb3a (Fig. 3). Some of the mRNAs were also detected in the pineal complex (see description below), but not in any other part of the head.
1. The expression of gnat1 and gnat2 genes is restricted to rods and cones, respectively. There are two genes coding for zebrafish transducin alpha subunits: gnat1 and gnat2. These have previously been shown to be rod-and cone-specific, respectively [31]. No duplicate of any of these genes was retained after 3R [14], indicating no further specialisation in teleost fishes. Nevertheless, we analysed their expression patterns to provide a complete picture of the transducin gene expression in the zebrafish retina. Gnat1 mRNA was restricted to the rod inner segments (Fig. 3A). The staining can be observed in the thin layer of cytoplasm surrounding the nuclei, and more intensely in the rod's myoid, embedded in the cONL (Fig. 3A). Gnat2 mRNA was found in the inner segments of all cone types: DC, LSC and SSC (Figs. 2C, 3B).
2. The gnb1 paralogs retained after 3R are co-expressed in adult rods. Contrary to the alpha subunits, the transducin beta subunit genes have retained duplicates after 3R ( [14], Although ISH is a qualitative method, the analysis of these results suggested different expression levels for the two genes (Figs. 3C, D). This was quantified by RT-qPCR and we found that there is, indeed, a statistically significant higher expression for gnb1a at all time-points (p<0,05; Fig. 4F). The expression levels of gnb1b are continuous throughout the day, while gnb1a shows a slight rhythmic oscillation with two peaks of expression around ZT8 and ZT16 (p<0,05; Fig. 4F).
3. The gnb3 paralogs show different expression patterns after 3R. Gene expression analysis of gnb3a and gnb3b showed spatial subfunctionalisation (compartmentalisation). Expression of gnb3b could be observed in cones of the dorsal and medial retina exclusively (Figs. 3F, 5B). However, expression of gnb3a could not be detected in adult retinae (Figs. 3E, 5A). Further RT-qPCR experiments revealed comparable expression levels of gnb3a in adult eyes and whole 3 dpf embryos (Fig. 5C), while the expression levels for cone-specific gnat2 and gnb3b showed significant expression increase from 3 dpf embryos to adult eyes.
4. The gngt2 paralogs retained after 3R show compartmentalisation in adult cones. This study shows rod-specific expression of the gngt1 gene in zebrafish (Fig. 3G), in agreement with results from other vertebrate species [26]. The cone-specific transducin gamma gene gngt2 was duplicated in 3R and zebrafish, unlike other investigated teleosts, has retained both copies: gngt2a and gngt2b [14]. Their expression patterns revealed cone-specific restriction

Expression of five transducin genes was detected in the adult pineal complex
The pineal complex in zebrafish consists of the pineal and parapineal organs, which are located dorsal to the diencephalic roof and connected to it via the pineal stalk. The pineal organ displays a flat T-or Y-shaped structure with a pineal vesicle opened to the third ventricle. It consists of ependymal cells, melanocyte-like cells, fibrous astrocytes and two types of photoreceptor-like cells that represent the morphofunctional unit of the gland [41]. The parapineal organ is an unpaired, left-sided accessory organ that has different names depending on and gngt1, respectively. Their expression is observed in the rod nuclei of the rONL, but the strongest staining is observed in their myoids, which are embedded in the cONL sublayer (thin arrows). Panels B, F, H and I show the expression of gnat2, gnb3b, gngt2a and gngt2b, respectively, in all cones: DC (arrowheads), LSC (asterisks) and SSC (stars). The brown retinal pigment epithelium is shown the uppermost part for each picture. The stratification of the outer retina is evident in all panels except H, due to the exclusive ventral expression of gngt2a, where the stratification becomes unclear. Panel E shows the lack of staining in the adult retina for the gnb3a gene. Note that the weaker stained band into the ONLc for the cone-specific transducin subunits (thick arrows) corresponds to the rods' myoid position (thin arrows in rods photomicrographs). Scale bar in panel A: 50 μm. For abbreviations see Fig. 2  the subclass of animals: parapineal in fish, parietal eye in reptiles and frontal organ in amphibians [42]. In this study we did not use any pineal versus parapineal cell markers, apart from the structural and morphological evidence, so we will refer to the pineal complex unless specified otherwise.
The present work shows the presence of mRNA for gnat1 (Fig. 7D), gnat2 (Fig. 7H), gnb1a  expression was detected for gnb1b, gnb3a, gnb3b or gngt2b. Expression of gnb3a (Fig. 7P) could not be observed in the adult pineal complex; however, it was detected in larval stages (Figs. 7M-O). The gnb1 expression pattern in the pineal complex provides another example of specialisation, i.e., while the two paralogs are co-expressed in the rods of the retina (Fig. 4), only gnb1a is expressed in the pineal complex (Figs. 7I-L). Transducin expression starts early during ontogeny Expression of all transducin genes was analysed by WISH at 19 hours post-fertilisation (hpf), 26 hpf, 36 hpf, 48 hpf, 52 hpf, 3 dpf, 4 dpf and 5 dpf (Figs. 7,8,9), with the same probes used for the ISH on adult sections. At 19 hpf the retina and the pineal complex are poorly developed, therefore we used this stage as a control of non-expression but were not incorporated.
1. Expression onset of the transducin genes in the retina is around 2 dpf. For all transducin genes, expression starts in the ventral retina (Fig. 9), around 48 hpf (Fig. 8), in agreement with the known cell differentiation process in the retina [36,43]. Transcripts start to be de-  (Fig. 8) and the photoreceptor mosaic pattern develops in the following days as previously described [21,43]. No correlation between onset of expression and photoreceptor type was observed.   (Figs. 9G-H). The only difference between the two 3R-generated gnb1 paralogs was the intensity of the staining, which was stronger for gnb1a (Fig. 9F) than for gnb1b (Fig. 9H), like we have described in adults (Figs. 3C-D). A wide background-like staining for both genes was observed in the whole body, as previously described from 8-16-cell stage [30]. Similar wide expression was also observed for gnb3a (Figs. 7M-O) and gngt2a (Figs. 7W-Y). In all four cases, this background-like staining becomes progressively restricted to the retina and/or the pineal complex (Figs. 7, 9), while the stronger intensity of the staining in the eye is due to the higher amount of mRNA.
In contrast to adults, where no expression of gnb3a was detected, embryonic and larval expres-  addition, gnb3a, whose expression was not detected in adults, was found to be expressed in the pineal complex from 26 hpf (Figs. 7M-P). As in the adult stage, no expression of gnb1b was detected. However, it was remarkable to find a transient expression of the cone-specific transducin paralogs gnb3b (Figs. 7Q, R) and gngt2b (Figs. 7A´, B´), from 26 to 48 hpf and 36 hpf to 3 dpf, respectively. Their expression ceases at hatching, when the transducin expression profile is completed in both the retina and the pineal complex.
Despite the background-like staining throughout the body described above for gnb1a, gnb1b, gnb3a and gngt2a, a clear but pale staining in the pineal complex could be distinguished from 26 hpf onwards for all of them, except gnb1b.

Discussion
All three families of transducin subunit genes were duplicated in the early vertebrate 1R and 2R tetraploidisations [14], resulting in separate sets of subunits for rods and cones [17,18]. Subsequently, two transducin gene families were duplicated in the teleost-specific 3R tetraploidisation; gnb and gngt, generating duplicates for rod gnb1 and cone gnb3 and gngt2 [14]. Since the 3R event took place more than 300 million years ago, it can be expected that these duplicates have undergone subfunctionalisation or neofunctionalisation. The alpha subunit genes have not retained duplicates after 3R, presumably due to their central roles as primary effectors of the phototransduction cascade [14]. For those gene duplicates retained after 3R, our analyses have revealed diverse subfunctionalisations, including diverging expression patterns and expression levels. These subfunctionalisations are most likely due to differential regulation of the 3R duplicates. To identify the regulatory elements, we searched for conserved non-coding elements (CNEs) with mVISTA, using the gene nucleotide sequence plus an additional 10kb upstream sequence, for the zebrafish transducin 3R duplicates and their spotted gar orthologs. However, we did not find any significant differences. More extensive and sophisticated analyses are needed to identify any such regulatory elements.
The amino acid sequences are well conserved for each pair of duplicates in the zebrafish, with identities of 99% for the GNB1 proteins and 75% for the GNB3 and GNGT2 pairs [14]. Only three amino acid changes were found between the GNB1 duplicates, and those are not in the regions that facilitate binding to the alpha or gamma subunits. Therefore, it may be assumed that any functional differences (interactions with other components) would be minor. For the other two duplicated genes, any consequences for protein interactions will have to be investigated experimentally.
Expression of transducin paralogs in the adult zebrafish is limited to photosensitive organs: the retina and the pineal complex ISH experiments using transducin-specific riboprobes on adult zebrafish heads revealed positive staining limited to the primary photosensitive structures, the retina and/or the pineal complex, for all genes except gnb3a. In contrast, gnb3a expression was observed during development, suggesting temporal subfunctionalisation. However, the expression levels in 3 dpf larvae and adult eye are very similar and extremely low, which offers an explanation to the lack of staining in adults. Adult eyes have much lower levels of gnb3a mRNA in relation to their cell number and size. Only one expressed sequence tag for gnb3a (NCBI accession number: EH460287.1) was identified through BLAST searches in the NCBI EST database for zebrafish adult brain and related tissues using our probe as the query sequence. This is consistent with low expression levels in these tissues.
1. Expression of transducin genes in the adult zebrafish retina is restricted to photoreceptor cells. The expression patterns of the transducin genes in the retina are conserved across vertebrate species (for references see specific cases below). We did not observe expression for any of the transducin genes in the zebrafish inner retina, in contrast to what has been found in some other vertebrates. Expression of gnb1 and gnb3 has been found in amacrine and bipolar cells, respectively, in different species of mammals [18,44]. GNB3 expression in both cones and bipolar cells appears to be highly conserved as it has also been described in frogs, chicken, mice, guinea pigs, dogs and non-human primates [20]. We have previously identified 3R duplicates of the gnb3 gene in all investigated teleost species [14]. In goldfish, GNB3 expression has been reported in bipolar cells [20], but the authors pointed out that there may have been some cross-reactivity. Furthermore, the goldfish has undergone a fourth tetraploidisation. Studies in other teleosts will determine if this is a common feature, and investigations in representatives from more basally radiating vertebrate lineages will be required to determine whether GNB3 expression in bipolar cells was lost in teleosts or gained in tetrapods.
2. Zebrafish gnb1 paralogs are co-expressed in rods. We found that the zebrafish gnb1 paralogs gnb1a and gnb1b are expressed in the vast majority of rods throughout the retina, yet we cannot say with certainty that all rods express both genes. GNB1 was initially described in rats as a rod-specific protein [18], and subsequently identified in many different vertebrate species. Although the retention of two almost identical zebrafish gnb1 paralogs suggests specialisation by diverging morphological or temporal expression, we did not observe such differences.
The continuous expression of gnb1b and the rhythmic oscillation of gnb1a expression suggest a dosage effect, or that one of the genes is expressed in an organ we have not investigated, probably the one with the lowest expression, gnb1b. Higher expression of gnb1a at ZT8 could be related to the higher visual sensitivity demonstrated for zebrafish in the afternoon [45]. However, no clear explanation was found for a peak of expression at ZT16.
The co-expression of this pair of paralogs is peculiar, but not unique within the phototransduction cascade. We have recently found a similar case between the pde6ga and pde6gb paralogs (work in progress) and it was also suggested for the arrestin genes, where both arrSa and arrSb paralogs are expressed (and very likely co-expressed) in rods [46] and the opsin GPCR kinases grk1a and grk1b [47]. We postulate that there might be two parallel phototransduction pathways within rods, at least for a few steps of the pathway.
3. Cone transducin paralogs retained after 3R show compartmentalisation. Two conespecific transducin paralog pairs have been retained after 3R: gnb3a/gnb3b and gngt2a/gngt2b. We found that both pairs display compartmentalisation, but in quite different ways. The gngt2 paralogs display a beautiful case of spatial subfunctionalisation, where gngt2a is expressed exclusively in the ventral retina and gngt2b is expressed in the dorsal and medial retina (Fig. 6). For the gnb3 paralogs, gnb3b is expressed in the dorsal and medial retina, while very little expression of gnb3a was revealed by RT-qPCR in adult eye, which was undetectable by ISH. Consequently, the functionality of the cones in the ventral retina could be questioned since they seem to have none or very little beta transducin subunit. However, a recent publication demonstrated that gnb3 knock-out mice produce stable responses with normal kinetics and saturating amplitudes, although with an approximately fourfold reduction in sensitivity [48]. We propose that a reduced sensitivity of the ventral retina may have evolved as a protective mechanism. Zebrafish live in shallow water, in subequatorial rivers, so the direct light incidence from above is high and could be harmful. The ventral retina has been shown to have a higher number of photoreceptors, as well as shorter photoreceptor outer segments and thicker RPE [40], specialisations that may also be protective. Furthermore, light-induced photoreceptor degeneration experiments always damage the dorsal retina more severely [49], suggesting a reduced sensitivity of the ventral part.
These observations provide evidence not only for morphological but also molecular specialisations of the ventral retina. In addition to the gnb3 paralogs, other phototransduction genes have been found to have specialised paralogs that are exclusively expressed in the ventral retina: the red opsin lws-1 [50], the green opsin rh2-4 [51], and the gngt2a (present results, see Fig. 10).
Aside from opsins and transducins, there are other gene families involved in zebrafish vision that show subfunctionalisations after 3R or due to species/lineage-specific (local) duplications: guanylyl cyclase [52], guanylyl cyclase activating protein [53,54], arrestin [46], G proteincoupled receptor kinase [47,55] and retinoid binding protein [56]. However, none of these show compartmentalisation in the ventral retina, or have not yet been investigated in this regard. Impact of 3R on transducin genes in the pineal complex: partial compartmentalisation of gnb1 paralogs Comparative studies of the pineal complex are complicated due to its variability across species. In zebrafish, the pineal organ contains two types of light-sensitive photoreceptor cells that share extensive similarities with retinal rods and cones, such as cell morphology, responses to light stimuli, and expression of components involved in phototransduction [8,55,[57][58][59][60][61].
In adults, only gnat2 expression has been reported [61]. In addition to this, we describe here the expression of gnat1, gnat2, gnb1a, gngt1 and gngt2a, but not gnb1b, gnb3a, gnb3b and gngt2b in the pineal complex. The identified subunits would be able to form a rod-like transducin heterotrimer (gnat1-gnb1a-gngt1; Fig. 11), while only alpha and gamma subunits for the cone-like transducin are present (gnat2-gngt2a; Fig. 11). The lack of detectable gnb3a in the adult pineal complex could be due to temporal subfunctionalisation or to very low expression levels, like in the retina, as its expression was found in both organs early in development. Nevertheless, studies analysing the amount of GNB3a protein and its co-localisation with opsins must be carried out in order to investigate the functionality of a possible cone-like transducin.
The lack of expression of gnb1b in the pineal complex suggests a partial compartmentalisation for the gnb1 paralogs: gnb1a is expressed in both the retina and the pineal complex, while gnb1b is restricted to the retina. No other case of partial compartmentalisation has been reported for the zebrafish phototransduction proteins.
It is not clear whether the pineal complex represents a primitive stage during eye evolution, or if it has evolved largely independently from the vertebrate eye [7]. In contrast to the retina, the structure of the pineal organ has changed dramatically in mammals relative to non-mammalian vertebrates as a result of the progressive replacement of direct to indirect photosensitivity [41]. The presence of the components needed to form functional rod-and cone-specific phototransduction pathways has been demonstrated previously in pineal cells of neonatal rats, although expression of many of these genes decline rapidly during development. These findings strongly support loss of photosensitivity in the mammalian pineal organ during ontogeny [62]. In contrast, lampreys [11], zebrafish ( [26], present results), other teleost fishes [9], as well as amphibians and reptiles (see [10,41]) seem to maintain functionality for the two photoreceptor cell types during their entire lifespan. Thus, a complete analysis of all the phototransduction proteins expressed in the zebrafish pineal organ and their cell-type specificity would offer important insights into this interesting possibility.
The ontogenetic analysis of the transducin genes reveals early establishment of the adult pattern and two partial subfunctionalisations This comparative ontogenetic analysis of the transducin genes between the zebrafish retina and pineal complex shows an earlier onset of the expression in the pineal complex than in the retina. Similar results have previously been reported for the GNAT [27,59] and GNGT subunits [26,32]. Earlier expression in the pineal organ compared to the retina has been also reported for opsins in lampreys [63] and teleost fishes [64][65][66]. These results are not surprising, as the two organs have different embryonic origins and their maturation is differentially regulated [67]. The early onset of genes involved in phototransduction in the pineal has been suggested previously to relate to regulation of hatching [66].
1. The early establishment of the adult pattern highlights the importance of the ventral retina in the larvae. Retinal differentiation in zebrafish starts in the ventral region, different from most other vertebrates, in which it is initiated in central locations [21]. The sequential events leading to photoreceptor differentiation can be summarised as follows: rhodopsin expression starts at 50 hpf, colour opsins at 52 hpf [43], outer segment development starts at 54 hpf, synaptic ribbons are discernible at 62 hpf and around 74 hpf the eyes respond to light [21].
A general overview of the ontogeny of the transducin genes reveals onset of expression in the retina around 48 hpf (Fig. 8), which is similar to the opsins [51] and PDE6 (unpublished results). All three cases are in accordance with the development of the visual system, but are not consistent with the idea that rods do not become functional until around 10 dpf [68,69]. A difference between rods and cones in onset of functionality has been demonstrated previously and attributed to different retinoic acid levels [21]. However, to our knowledge, the reason for this long delay in acquiring functionality in putative "fully-developed" rods is not yet known. In addition, ontogenetic analyses of the transducin genes show early establishment of the adult expression patterns, i.e., the exclusive ventral expression of gngt2a and absence of a cone beta subunit in the ventral retina. This suggests an important role of the ventral retina from an early point in development, a pattern that has been demonstrated previously for the opsins [50,51]. Taken together, this supports the protective hypothesis of the ventral retina since these specialisations would provide exclusive expression of a few genes related to light detection in the newly hatched larvae, during a time when they predominately swim at the bottom of a stream or tank with light incidence mainly from above [21].
Developmental analyses, including the present study, have revealed ubiquitous expression of gnb1a, gnb1b, gnb3a and gngt2a in the whole body from very early in development ( [70], present results), which gradually becomes restricted to the retina and/or the pineal complex during the larval stage ( [71,30], present results). Detailed ontogenetic analysis of the GNB subunits have revealed a similar spatiotemporal plan for both gnb1a and gnb1b: both mRNAs are first observed at the 8-16 cell stage [30] and both genes are involved in the migration and polarisation, but not differentiation, of the primordial germ cells [30]. These data, together with the testis expression of gngt1 [26], suggest a crucial role of the "rod-specific" βγ dimer in the development of the gonads, a process that is not complete until around 30 dpf [72]. This coincides with the time-point when the adult eye has reached its mature size [68]. By this timepoint, gnb1a, gnb1b and gngt1 are completely restricted to the photosensitive structures in the head. In the retina, gnb1a and gnb1b follow an identical developmental pattern.
2. Two partial subfunctionalisations between the retina and the pineal complex from the larval stage. The ontogenetic analysis of the pineal complex reveals expression of gnat1, gnat2, gnb1a, gnb3a, gngt1 and gngt2a from 26 hpf. This would facilitate the formation of functional rod-like (gnat1-gnb1a-gngt1; Fig. 11) and cone-like transducin heterotrimers (gnat2-gnb3a-gngt2a; Fig. 11). This is different from the adult pineal complex where expression of gnb3a could not be demonstrated. Moreover, we found an interesting transient expression of gnb3b and gngt2b in the pineal complex prior to hatching. This finding reveals a partial temporal subfunctionalisation, as both genes are expressed in the retina from 48 hpf. Their functional consequences are not clear, but the timing implies a role in hatching regulation [66]. Partial compartmentalisation was also observed between the retina and the pineal complex for gnb1a and gnb1b.

Conclusions
The evolutionary consequences of 3R for the zebrafish transducin subunits are manifold. Three paralogous pairs have been retained after 3R: gnb1a/gnb1b, gnb3a/gnb3b and gngt2a/gngt2b, showing a very interesting range of specialisations. Gnb1a and gnb1b genes are co-expressed in the same rods and gnb1b is absent from the pineal complex, which indicates a partial compartmentalisation. The cone-specific gnb3b gene is expressed in the dorsal and medial retina, while the expression of gnb3a in adults is so low that it cannot be detected by ISH. Similar expression levels of gnb3a were found in 3 dpf larvae, where ISH staining can be observed in both the retina and the pineal complex. Additionally, gnb3b mRNA is only transiently detected in the pineal complex prior to hatching, indicating a partial temporal subfunctionalisation. The cone-specific gngt2b gene is expressed in the dorsal and medial retina, while gngt2a is exclusively expressed in the ventral retina. This is a very clear case of compartmentalisation that highlights an important specialisation of the ventral retina. In addition, gngt2b has undergone a process of partial temporal subfunctionalisation similar to that of gnb3b. Taken together, these results show multiple examples of altered gene expression reflecting anatomical and developmental specialisations of 3R paralogs.