Cone photoreceptor types in zebrafish are generatedby symmetric terminal divisions ofdedicated precursorsSachihiro C. Suzukia, Adam Bleckerta, Philip R. Williamsa, Masaki Takechib, Shoji Kawamurab, and Rachel O. L. Wonga,1

aDepartment of Biological Structure, University of Washington, Seattle, WA 98195-7420; and bDepartment of Integrated Biosciences, Graduate Schoolof Frontier Sciences, The University of Tokyo, Kashiwa, Chiba 277-8562, Japan

Edited* by Joshua R. Sanes, Harvard University, Cambridge, MA, and approved July 31, 2013 (received for review February 22, 2013)

Proper functioning of sensory systems requires the generation ofappropriate numbers and proportions of neuronal subtypes thatencode distinct information. Perception of color relies on signalsfrom multiple cone photoreceptor types. In cone-dominatedretinas, each cone expresses a single opsin type with peaksensitivity to UV, long (L) (red), medium (M) (green), or short (S)(blue) wavelengths. The modes of cell division generating distinctcone types are unknown. We report here a mechanism wherebyzebrafish cone photoreceptors of the same type are produced bysymmetric division of dedicated precursors. Transgenic fish inwhich the thyroid hormone receptor β2 (trβ2) promoter drivesfluorescent protein expression before L-cone precursors them-selves are produced permitted tracking of their division in vivo.Every L cone in a local region resulted from the terminal division ofan L-cone precursor, suggesting that such divisions contribute sig-nificantly to L-cone production. Analysis of the fate of isolatedpairs of cones and time-lapse observations suggest that other conetypes can also arise by symmetric terminal divisions. Such divisionsof dedicated precursors may help to rapidly attain the final num-bers and proportions of cone types (L > M, UV > S) in zebrafishlarvae. Loss- and gain-of-function experiments show that L-opsinexpression requires trβ2 activity before cone differentiation. Ec-topic expression of trβ2 after cone differentiation produces coneswith mixed opsins. Temporal differences in the onset of trβ2 ex-pression could explain why some species have mixed, and othershave pure, cone types.

vertebrate cone photoreceptors | cone genesis | zebrafish retina |in vivo time-lapse imaging

The proper functioning of neuronal circuits requires the gen-eration and wiring of a diversity of neuronal cell types. A

single neuronal cell class often comprises many subtypes thatshare similar properties, such as neurotransmitter phenotype,but differ in their precise molecular expression profile, mor-phology, and physiology (1, 2). How neuronal subtypes that shareconnectivity with the same populations of postsynaptic cells areproduced is not well understood, particularly for vertebrate cir-cuits in vivo. Specifically, are distinct presynaptic partner types ofa given postsynaptic cell generated together or produced fromseparate divisions? When during cell genesis do the presynapticcell types adopt their respective identities?Cell lineage analyses have demonstrated that many neurogenic

divisions are asymmetric, sometimes producing distinct neuronalclasses or a neuron together with a nonneuronal cell type (3, 4).Examples of progenitors that give rise to a single neuronal classhave also been reported (5–13). A single progenitor, however,can also produce two distinct neuronal subtypes (14–17). In someinstances, neurons of the same functional subtype may also sharea common progenitor (18), but their generation may involve bothsymmetric and asymmetric divisions (13, 19, 20). Recent retro-viral studies in chick retina revealed the presence of two cellclones comprising the same horizontal cell (HC) type (H1 HC)and larger clones with even numbers of H1 and H3 HCs (11). It

was suggested that HCs of the same type are generated bya symmetric terminal division of a “determined progenitor” or bytwo or more asymmetric divisions producing only one type of HC(11). To directly determine whether symmetric divisions gener-ate a specific type of neuron and to assess how common suchdivisions are, we followed cone photoreceptor genesis in vivo ina vertebrate by time-lapse imaging. We were particularly in-terested in the cone population because these cells representa good model system for investigating the modes of cell divisionthat generate a diversity of presynaptic cell types that provideconverging input onto a common set of postsynaptic cells.Cone photoreceptors in many vertebrates express a single

opsin with peak sensitivity to a specific wavelength of light (21)and are, thus, functionally distinct. Signals from these separatesensory channels recombine as they converge onto postsynapticcells in the outer retina to provide a rich palette of color in-formation underlying an animal’s visual perception (22).Zebrafish have four types of cone photoreceptors: long wave-length-sensitive (L) (red), medium wavelength-sensitive (M)(green), short wavelength-sensitive (S) (blue), and UV wave-length-sensitive (UV) cones, each of which expresses L-, M-, S-,or UV-opsin, respectively (23). Zebrafish have two L-opsin(opn1lw1, opn1lw2) and four M-opsin genes (opn1mw1, -2, -3, -4)in contrast to S- and UV-opsins, which are each encoded bya single gene (24). Our analysis revealed that L cones are producedby symmetric terminal divisions of a progenitor, revealing thepresence of a dedicated precursor for this cone type. Furthermore,

Significance

Color vision requires multiple types of cone photoreceptors,each with peak sensitivity to a specific wavelength. How dif-ferent cone types are generated in vivo is not clear. We showthat there are precursor cells individually dedicated to pro-ducing a single cone type. We tracked cone genesis in vivo intransgenic zebrafish in which red cones and their progenitorsexpress fluorescent protein driven by the thyroid hormonereceptor β2 promoter. We discovered that red cones are gen-erated by symmetric terminal divisions of a red-cone precursor.Moreover, UV, blue, and green cones also have their owndedicated precursors. Thyroid hormone receptor β2 expressionin cone precursors is required to produce pure red cones,whereas expression after cell division results in cones withmixed opsins.

Author contributions: S.C.S. and R.O.L.W. designed research; S.C.S. and P.R.W. performedresearch; S.C.S., A.B., M.T., S.K., and R.O.L.W. contributed new reagents/analytic tools; S.C.S.and A.B. analyzed data; and S.C.S. and R.O.L.W. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.

Data deposition: The DNA sequence data of the trβ2 full-length cDNA reported in thispaper has been deposited in the DNA Data Base in Japan/European Molecular BiologyLaboratory/GenBank database (accession no. AB759513).1To whom correspondence should be addressed. E-mail: wongr2@uw.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1303551110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1303551110 PNAS | September 10, 2013 | vol. 110 | no. 37 | 15109–15114

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

their subtype identity is defined before cone differentiation, mostlikely at the precursor stage. We found that other cone types canalso be produced by symmetric terminal divisions. Further-more, the transcription factor thyroid hormone receptor β2 (trβ2)is critical for L-cone fate in zebrafish, suggesting a conservedmechanism for defining cone-type fate in vertebrates (25).However, our study further raises the intriguing possibility thatspecies differences in the expression of opsins within individualcones (single opsin versus mixed opsins) may be dependent onthe temporal onset of trβ2 activity.

ResultsL-Cone Pairs Are Produced by Symmetric Division of a DedicatedPrecursor. In larval retina, all cone types are generated and dif-ferentiate within 4 d postfertilization (dpf) (26). However, howthe four cone types are generated is yet unknown. We cloned thezebrafish trβ2 promoter and found that it drives fluorescentprotein (FP) expression specifically in cone photoreceptors (Fig.1). Promoter analysis showed that the intron downstream of thefirst coding exon of trβ2 is essential and sufficient to drive geneexpression in cones, in accordance with the previous analysis oncis-regulatory elements of mouse trβ2 (27) (Fig. 1A). All trβ2:FP-expressing cones (n = 1,027; n = 3 eyes) immunolabeled exclu-sively for L-opsin (Fig. 1 B and C) and not other opsins (Fig. S1).We found that trβ2:FP expression initiated in progenitor cells

before cone photoreceptors were generated [imaging period: 24–70 h postfertilization (hpf)]. In situ hybridization showed thatendogenous message for trβ2 is present early, at the time whenFP is first detected in progenitor cells (Fig. S2). In vivo imagingof trβ2:FP retinas revealed the presence of progenitor cells thatunderwent symmetric divisions at the apical surface of the retina.The daughter cells remained at this surface and differentiatedinto cones (n = 59 divisions followed; n = 10 eyes; Fig. 1 D andE). We conclude that these divisions produced a pair of L conesbecause expression of trβ2:FP is restricted to this cone type.Thus, we identified a precursor dedicated to generating not onlycones but also a functionally distinct type of cone, the L cone. Togain a sense of how common such dedicated precursor divisionswere, we imaged the proliferative ciliary marginal zone ofa transgenic line, Tg(trβ2:tdTomato), between 51 and 60 hpf at30-min intervals (n = 3 time lapses; Fig. 1F and Movie S1). Wetracked the origin of all of the L cones within a local region (n = 32cones tracked) that were generated during the period of imagingand found that all of them were derived from symmetric division ofa tdTomato-expressing precursor, suggesting that this mode of celldivision is a major route of producing L cones.

Cell Lineages of L-Cone Precursors. We traced the lineage of trβ2:GFP-expressing progenitors by monitoring cell divisions beforethe L-cone precursors were themselves generated. Using tran-sient expression of trβ2:GFP plasmid to limit the number of FP-labeled cells, we followed the sequence of neurogenesis that ledto the production of L-cone pairs. We observed that L-coneprecursors share a lineage with HC precursors (10) (n = 5 timelapses). Both of these precursor types could arise from a priordivision that also produces ganglion cells (e.g., Fig. 2A). Wecarried out time-lapse imaging of a double-transgenic line, Tg(trβ2:tdTomato; ptf1a:GFP), in which GFP expression labelsvirtually all HC precursors and HCs (10), and examined coex-pression of tdTomato and GFP in mitotic figures in the innernuclear layer (INL). We found that about 70% of GFP-expressing mitotic figures in the INL were tdTomato-positive(30-min time interval; 8 h total; 102 of 143 mitotic figures showcolocalized signals; Fig. S3A). Because we showed previouslythat the vast majority of HCs in zebrafish are generated from HCprecursors (10), these observations suggest that L-cone pre-cursors and HC precursors are largely derived from a commonprogenitor. In some divisions (n = 3), the progenitors gave rise toganglion cells and L-cone precursors, without generating HCprecursors (Fig. 2B), in keeping with the greater number of Lcones than HCs in the differentiated retina. Finally, a division

can produce two L-cone precursors that subsequently producedfour L cones, as observed in one example (n = 1 of 9 time lap-ses). Unlike the L cones, FP expression in both HC precursorsand ganglion cells was down-regulated. Overexpression of trβ2in HC precursors, however, did not suppress the production ofHCs (Fig. S3B). In no instance did we observe trβ2:FP expres-sion in amacrine cells, bipolar cells, rod photoreceptors, orMüller cells; however, transient expression of trβ2:FP in thesecells remains a possibility. Together, our observations suggestthat L-cone precursors are produced by both symmetric (twoL-cone precursors) and asymmetric (an L-cone precursor andanother cell type) patterns of cell division involving trβ2:FP-expressing progenitors.

Fig. 1. L cones are derived from symmetric precursor divisions. (A) Intron-1 ofzebrafish trβ2 is essential for expression in cones. (B) Transient expression of FPdriven by the trβ2 promoter is restricted to cones immunopositive for L-opsin.L-opsin signal is restricted to the outer segment, where GFP labeling is dimbecause there is little cytoplasmic space in this compartment (also see Fig. S1).L-opsin signal in GFP-expressing cells were isolated by digitally removing theopsin signal outside the GFP masks. (C) Only L cones are labeled in the stabletransgenic line, Tg(trβ2:tdTomato). D, dorsal; N, nasal; T, temporal; V, ventral.(D) In vivo multiphoton time-lapse imaging reveals division of a trβ2:MYFP-labeled cell that produced two cones. (Time, hours:minutes.) AP, apical process;BP, basal process. The dotted line indicates the location of the outer limitingmembrane (also in E). Cells were labeled by transient expression of trβ2:MYFP.(E) Imaging at more frequent intervals shows that the dividing cell has a shortprocess that bifurcates at its terminal ending (BP). Cells were labeled bytransient expression of trβ2:MYFP-2A-histone2AYFP. (F ) Snapshots froma time-lapse sequence demonstrating the generation of L cones in a Tg(trβ2:tdTomato) fish. See Movie S1 for the full sequence. Postmitotic conesalready present within the patch were not tracked. Each color representsa pair of L cones produced from a terminal division during the imagingperiod. Arrowheads indicate mitotic figures. L, lateral; N, nasal.

15110 | www.pnas.org/cgi/doi/10.1073/pnas.1303551110 Suzuki et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

Other Cone Types Can Be Generated from Symmetric Terminal Divisions.Are there also dedicated precursors for the other cone types?To address this, we cloned the promoter fragment of zebrafishcone-rod homeobox (crx) that is expressed by mature photo-receptors and cells in the outer retina before cone genesis (28,29). The promoter was used to drive transient expression of FPin a small number of cells, and each cone type was labeled ge-netically or by immunofluorescence. We noticed isolated pairs ofcrx:FP-positive cones that were almost always of the same type(56 of 57 pairs; Fig. 3 A and B). These cone pairs were next toeach other, with intercone distances similar to that of their re-spective populations (Fig. S4). Although we cannot be sure thatthese cone pairs were part of a clone, the probability of obtaininggreater than 56 of 57 cone pairs comprising the same cone typeby random labeling is extremely low (Pr = 4.5 × e−30) (SIMaterials and Methods and Fig. S5).We followed the division of crx:MCFP-expressing cells in the

background of a double-transgenic line in which UV cones and Lcones are fluorescently labeled [Tg(sws1:histone2AYFP; trβ2:tdTo-mato)] (30). After tracking a division, the animals were allowed tomature until we were able to identify the progeny (Fig. 3C–G). Wefound examples of cone-precursor divisions that produced a pair ofUV cones and a pair of M cones (n = 1 each). We have not yetmanaged to directly observe a precursor for S cones (n = 4 timelapses; 20-h duration), but likely an S-cone precursor exists giventhe presence of isolated S-cone pairs in the transient transfections(Fig. 3A). Thus, all cone types can be generated by dedicatedprecursors undergoing symmetric divisions.

Trβ2 Functions Before Cone Differentiation to Specify and RestrictL-Opsin Expression. In mice, the majority of cones express bothL/M- and S-opsins, although a small fraction of cones in thedorsal retina or ventral retina expresses only M- or S-opsin, re-spectively (31). The loss of trβ2 in mice results in the absence ofM/L-opsin expression concomitant with an increase in S-opsinexpression (25). Phylogenetically, mouse S-opsin and M/L-opsincorrespond to fish UV-opsin and L-opsin, respectively (21, 23).We, thus, determined whether endogenous trβ2 is necessary forL-opsin expression in zebrafish by knocking down trβ2 using

a morpholino oligonucleotide (MO) targeting the junction re-gion of the first coding exon and the following intron of trβ2. Themorpholino sequence avoids affecting trβ1 that is transcribedfrom the same locus as an alternative splice variant. Verificationof the splicing error caused by trβ2MO is shown in Fig. S6. Therewas no significant change observed in the total density of conesin the morphant (Fig. 4 A and B) or generation of other retinalcells (Fig. S7). However, knocking down trβ2 expression severelyreduced the number of L cones and caused a correspondingincrease in UV cones (Fig. 4 A and B), consistent with thephenotype of trβ2 knockout mice (25). In contrast, the numbersof S cones and M cones were not affected (Fig. 4 A and B).We then rescued L-opsin expression in the morphant by driving

trβ2 expression in stable transgenic lines using the cone photo-receptor promoters, crx and gnat2 (guanine nucleotide bindingprotein, α transducing activity polypeptide 2). In these lines, the crxpromoter drives transgene expression early in development whencone precursors are present, and the gnat2 promoter (32) drivestransgene expression in postmitotic cones (Fig. S8). Trβ2 andMYFP were coexpressed bicistronically in fish lines, Tg(crx:MYFP-2A-trβ2) and Tg(gnat2:MYFP-2A-trβ2), by using the 2Apeptide sequence (33). We injected trβ2 morpholino into fertil-ized eggs of these transgenics and then examined cone opsin ex-pression at 5 dpf (Fig. 4A). We found that trβ2 expression undereither promoter rescued L-opsin expression (Fig. 4A). Thus, trβ2is necessary for determining L-cone fate and can induce L-opsinexpression in cones even after cone differentiation.In morpholino-treated Tg(crx:MYFP-2A-trβ2) fish, virtually all

cones expressed L-opsin, and the numbers of other cone typeswere significantly reduced (Fig. 4 A and B). The density of rodswas also reduced (Fig. S9). In contrast, in morpholino-treatedTg(gnat2:MYFP-2A-trβ2) fish, although L-opsin expression wasrestored, many cones expressed L-opsin together with anotheropsin (Fig. 4 C and D). More than half of the cones thatexpressed UV-opsin or M-opsin also contained L-opsin (Fig.4D). Interestingly, expression of trβ2 in postmitotic conesdid not prevent the increase in UV-cone density induced bythe morpholino, in contrast to overexpressing this protein underthe crx promoter in the morphant. However, although early

Fig. 2. L-cone precursors are derived from differentlineages. (A) Multiphoton time-lapse series showingsequence of cell division producing a ganglion cell(yellow), HC precursor (cyan), and L-cone precursor(red). (Time, hours:minutes.) See Movie S2 for thefull time-lapse sequence. (B) Cell divisions producinga ganglion cell and an L-cone precursor. See MovieS3 for the full sequence. (C) Division producing apair of L-cone precursors. Arrowheads indicate in-dividual L cones produced from a common pro-genitor. (D) Schematic summarizing the lineagesshared by L-cone precursors. LC, L cone; RGC, retinalganglion cell.

Suzuki et al. PNAS | September 10, 2013 | vol. 110 | no. 37 | 15111

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

overexpression of trβ2 under the crx promoter suppresses S-opsinproduction, we did not find mixed S- and L-opsin–expressingcones in Tg(gnat2:MYFP-2A-trβ2) fish. Thus, unlike UV or Mcones, S cones do not coexpress L-opsin when trβ2 is overex-pressed after cone differentiation.

DiscussionGenerating Neuronal Diversity Within a Single Neuronal Class. Ded-icated precursors producing a single neuronal class, retinal HCs,have been shown in prior works (10, 13). Retroviral lineage-tracing observations are consistent with the existence of pre-cursors that divide only to produce the same type of HC (11).Our study in zebrafish demonstrates the presence of a dedicatedcone subtype-specific precursor in vivo. We only found siblingswith identical opsin expression in our time-lapse imaging. Thus,we have directly uncovered a mode of symmetric terminal di-vision that underlies the production of not only a specific neu-ronal cell class but also, likely, each of the subtypes within thissingle cell class. Indeed, a previous in vivo-imaging study inzebrafish demonstrated that terminal division generating twophotoreceptors is not uncommon (55 of 197 terminal divisions ofall kinds tracked), although their identities were not determined(13). Our study suggests that many, if not the majority of,zebrafish cone pairs are likely to comprise the same cone type.However, like mice (6, 12), it remains possible that, in zebrafish,cones are produced by an asymmetric division, giving rise toa cone and another retinal cell type before the dedicated pre-cursors are generated (13).Our time-lapse observations demonstrated properties of

the L-cone precursors that contrast with other germinal cells.The dedicated precursor has a short basal process, ending at theouter plexiform layer rather than contacting the basal lamina.However, like fish HCs and rod photoreceptors (8, 10), divisionof the L-cone precursor occurs at the cellular layer at which theprogeny finally reside. We also found that the postmitotic cones,despite being consistently spaced, showed very little lateralmovement, possibly allowing them to rapidly integrate into theirmosaics (34).

Our analysis also indicates a clear segregation of lineagesbetween L cone and other cone types but a close lineage re-lationship between L-cone precursors and HC precursors. Re-cent studies showed that pairs of photoreceptors are generatedwithin various cell lineages, but photoreceptors are often foundtogether with HCs in the same clone (13). Our current obser-vations suggest that these photoreceptors are likely to be L conesbecause both L-cone and HC precursors are labeled in the trβ2:XFP transgenic animals. A close relationship in the lineages ofcones and HCs is also found in mice (12).The ciliary marginal zone of the teleost retina continues to add

newly generated cells throughout life and produces a crystallinecone mosaic in which the ratio of different cone types and theirspatial arrangements are stereotypic (34). As yet, we do not knowwhether symmetric terminal divisions produce specific cone typesin the adult retina, and, if so, how such a cell-generation mecha-nism can act to accurately position each cone type to form theirmosaics. It is possible that the adult ciliary marginal zone takes ona different strategy to produce cones. Future in vivo time-lapseimaging studies of the adult ciliary marginal zone are necessary todistinguish between these possibilities.It is also possible that symmetric terminal divisions giving rise

to specific cone types extend to other species. Indeed, even though,in the mouse retina, the majority of clones contain a single cone,two-cell clones comprising only cones have also been found (6,12). However, the opsin fates of the progeny were not deter-mined. Thus, it remains to be determined whether “dedicatedcone-type precursors” represent a strategy for producing distinctcone types only in cone-dominated retinas, such as zebrafish andprimate, and not in retinas comprising cones containing mixedopsins, as in mice. Alternatively or additionally, this mode of celldivision could be an efficient strategy to amplify cone numberwithin a short time frame for animals that need to developrapidly. If cone opsin expression is separately determined foreach progeny of the division, it would be necessary to coordinatethis regulation across cones to attain the final population ofeach cone type. Instead, attaining the final numbers and ap-propriate proportions of each cone type (34) via a dedicated

Fig. 3. All cone types are generated from sym-metric divisions. (A and B) Transient expression ofcrx:FP produces isolated pairs of labeled cones of thesame type. Of 57 pairs, 56 pairs comprised the samecone type (cone pairs observed: 8, L cones; 14, Mcones; 9, S cones; 25, UV cones). Only one paircomprised different cone types (L, S). Cone identitieswere obtained either using cone-specific transgeniclines [Tg(sws2:GFP) Tg(sws1:H2AYFP)] (50) or byM-opsin immunostaining. M-opsin was localized to theouter segments of a pair of crx:tdTomato-expressingcones (Inset). (C) Time lapse of Tg(sws1:H2AYFP,trβ2:tdTomato) fish injected with crx:MCFP at theone-cell stage. Shown here is a region in the dorsalretina where neighboring cell divisions gave rise topairs of UV, M, and L cones. The precursor and thedaughters for each cone type are colorized: UV (vi-olet), M (green), and L (red). Symbols at the last timepoint in C mark each cone pair that was identifiedafter fixation (D–G). See Movies S4 and S5 for thefull time lapse and orthogonal views of the divisions.L, lateral; N, nasal. (D) Single-plane view of the lasttime point of the live-imaging series in C. (E) One ofthe cell pairs was trβ2:tdTomato-positive, and, thus,the cells were L cones (arrows). (F) Immunostainingusing zpr1 revealed an M-cone pair (asterisks). (G) AUV-cone pair was identified by expression of sws1:H2AYFP (arrowheads).

15112 | www.pnas.org/cgi/doi/10.1073/pnas.1303551110 Suzuki et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

precursor would only require control of precursor generation ortheir division.

Transcriptional Regulation of Cone-Type Identity. Much work in thepast decade has focused on ascertaining the transcriptionalregulation of photoreceptor cell fate in the vertebrate retina(reviewed in ref. 35). The current model based on studies of themouse retina proposes that the photoreceptor fate is committedupon expression of the transcription factor Otx2 (36). Sub-sequently, expression of Nrl dictates rod photoreceptor fate(37), whereas, in Nrl-negative photoreceptors, Trβ2 and thyroidhormone confer M/L-opsin expression but suppress S-opsinexpression by TRβ2 forming a heterodimer with retinoid X re-ceptor γ (25, 38, 39). Our loss-of-function experiments indicate

that this role for trβ2 is conserved in zebrafish. As in mice (25),we show that, in zebrafish, trβ2 is important in the choice ofL- versus UV-cone fate. In mice, thyroid hormone becomesdistributed in a gradient across the retina as cones differentiate,thereby contributing to the differential densities of short-wave-length cones in dorsal and ventral retina (38). It is unknownwhether a gradient of thyroid hormone exists across the zebrafisheye, although this hormone is likely to be present because dei-odinase 2, its activator, is expressed in the embryonic and larvalretina [gene expression data from the Zebrafish Model OrganismDatabase (ZFIN); http://zfin.org].In contrast to mice where Trβ2 expression appears in post-

mitotic cells (27), trβ2 mRNA is observed in progenitors as earlyas 1 dpf in zebrafish. Thus, unlike mice, L-cone fate in zebrafishappears to be determined at the mitotic cell stage, rather than inpostmitotic cones (35). It is interesting to note that in the chickretina, in which each cones expresses a single opsin type (40), trβ2is also expressed in progenitors (41). Does expressing trβ2 inprecursors versus postmitotic cells determine whether an in-dividual cone expresses single or mixed opsins? Our experimentsoverexpressing trβ2 at different developmental time points sug-gest that trβ2 expression at the precursor stage only permits theproduction of cones containing solely L-opsin. However, restrict-ing expression of trβ2 to postmitotic cones allows for coexpressionof L-opsin and another opsin type (UV and M but not S). Thepresence of cones with mixed L- and M- or UV-opsin in gnat2:trβ2overexpressors suggests that UV- and M-opsin expression cannotbe suppressed by trβ2 activity after cones have differentiated. Thelack of mixed S and L cones in these overexpressors may be be-cause activation of the opn1sw2 promoter suppresses the pro-moters of L-opsins, opn1lw1 and opnlw2, which are located about 2kb downstream from the S-opsin gene (24). These observationstogether raise the possibility that cones expressing both M/L- andS-opsin in mice exist because Trβ2 acts in postmitotic cells (27),and, thus, the temporal differences in the action of this tran-scription factor may contribute to species differences in theexpression of “pure” versus “mixed” cone types.What factors regulate fate specification of cone types during

normal development? Past studies have provided insight into thetranscriptional regulation of cone versus rod photoreceptor fate(37, 42, 43), but, other than for L cones, it is not yet known whatfactors dictate cones to express UV-, S-, or M-opsin. For ex-ample, perturbation of tbx2b causes a fate switch of UV cones torods but does not alter the number of other cone types (42). Weshowed here, however, that, although not yet identified, thesefactors are likely to act at the cone-precursor stage rather thanafter cone differentiation. In trβ2 morphants, we found that theloss of L cones is accompanied by a corresponding increase inUV cones, although not S or M cones. Likewise, in mice, loss ofTrβ2 function leads to a decrease in M/L cones with a parallelincrease in S cones that is the mouse ortholog of zebrafish UVcones (25). Although the relative numbers of long- and short-wavelength cones are likely to be determined at the precursorstage in zebrafish, in rodent, this relationship appears to becontrolled in the postmitotic cones (35) and maintained by thy-roid hormone in the adult (44). What would be interesting todetermine in the future is whether, like Drosophila (45), opsinexpression of specific cone types requires interactions amongdifferent cone types to maintain their individuality and to askhow the environment causes individual cones to switch theiropsins, a feature common in salmonid fish (46).

Materials and MethodsTransient Expression of Transgenes. We injected plasmid DNAs diluted in 1×Danieu’s buffer at the final concentration of 15–25 ng/μL into the cellularpart of one-cell stage eggs using a glass capillary equipped with PicoSpitzer.Detailed information on plasmid construction is found in SI Materialsand Methods.

Imaging. We performed in vivo time-lapse imaging of zebrafish larvae eitherusing a custom-designed multiphoton microscope with a 60× 1.1 NA objective(Olympus) or using an Olympus FV1000 confocal microscope. Amira (Visage

Fig. 4. Production of pure L cones is regulated by trβ2 expression in coneprecursors. (A) L-opsin–expressing cone numbers are diminished, whereasUV-opsin–expressing cone numbers are increased in animals injected withtrβ2 morpholino (MO). L-opsin expression is rescued in the morphant whentrβ2 is constitutively active under the crx or gnat2 promoters in the stabletransgenic lines. Images were acquired at 5 dpf. (B) Quantification of thedensity of cones expressing each opsin type for the conditions in A (mean ±SEM). Numbers in parentheses indicate number of eyes. Significant differ-ences are noted by asterisks (*P < 0.001); for each cone type, comparisonswere made between trβ2 MO and the GeneTools standard control MO (StdMO) and between trβ2 MO and crx:trβ2/ trβ2 MO or gnat2:trβ2/ trβ2 MO.There was no significant difference in total photoreceptor density amongStd MO (mean ± SEM in 100 μm2, 18.6 ± 0.3; n = 9 eyes), trβ2 MO (18.9 ± 0.2;n = 7), and gnat2:trβ2/ trβ2MO (19.1 ± 0.3; n = 7). However, compared with thetrβ2 MO, the density of photoreceptors in crx:trβ2/ trβ2 MO was increased (21.6± 0.7; n = 6; P = 0.005). (C) Double immunostaining for L-opsin and M-, S-, orUV-opsin in 5 dpf Tg(gnat2:MYFP-2A-trβ2) fish injected with trβ2MO. Examplesof cones expressing L-opsin and M- or UV-opsin are indicated by arrowheads.(D) Quantification of the proportion of each cone type that also expressedL-opsin (mean ± SEM). For example, (M + L)/M is the proportion of M conescoexpressing L-opsin. Numbers in parentheses indicate number of eyes.

Suzuki et al. PNAS | September 10, 2013 | vol. 110 | no. 37 | 15113

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

Imaging), Fiji and custom-generated software were used for image analysis.Detailed information on image analysis is in SI Materials and Methods.

Antibodies. The primary antibodies used in this study were as follows: rabbitanti–L-opsin and anti–M-opsin (23); anti–S-opsin and anti–UV-opsin (23, 47);mouse monoclonal antibody (1D4) against bovine Rhodopsin, which recog-nizes L-opsin instead of Rhodopsin in zebrafish (48); zpr-1 (ZIRC) (49), whichlabels L and M cones; and rabbit anti-DsRed (Clontech). For the secondaryantibodies, DyLight (Jackson ImmunoResearch) or Alexa (Invitrogen) conju-gated antibodies were used. Detailed information about methods forimmunostaining is found in SI Materials and Methods.

Transgenic Zebrafish Lines. Using standard cloning procedures, we generatedthe following stable transgenic lines: Tg(trβ2:Tdtomato), Tg(trβ2:MYFP), Tg(sws1:H2AYFP), Tg(crx:MCFP), Tg(crx:MYFP-2A-trβ2), and Tg(gnat2:MYFP-2A-trβ2). Detail information on the DNA constructs used to generate thetransgenic lines is found in SI Materials and Methods.

Morpholino Knockdown. MO to knock down trβ2 expression (trβ2 MO: 5′-TCTAGAACTTGCAATACCTTTCTTA-3′) was purchased from GeneTools. Fornegative control, GeneTools standard control MOwas used. The verification ofthe splice error of trβ2mRNA caused by trβ2MOand its specificity are shown inFig. S6.

ACKNOWLEDGMENTS. We thank Steve Leach for providing Tg(ptf1a:Gal4-VP16) and Tg(ptf1a:GFP); Chi Bin Chien for Tg(isl2b:MGFP); Pamela Raymondfor Tg(GFAP:GFP); Jim Fadool for Tg(XOPS:GFP); David Hyde and JeremyNathans for providing opsin antibodies; Susan Brockerhoff for providingthe gnat2 promoter; members of the R.O.L.W. laboratory and Thomas Rehfor many helpful discussions and guidance; and Leanne Godinho, FeliceDunn, Mrinalini Hoon, and Haruhisa Okawa for critical reading of themanuscript. This work was supported by National Institutes of HealthGrants EY 14358 (to R.O.L.W.) and 5 T32 GM07108 (to A.B.) and a Grant-in-Aid for Scientific Research on Priority Areas (Cellular Sensor) (21026007) fromthe Ministry of Education, Culture, Sports, Science and Technology of Japan(to S.K.). S.C.S. was a recipient of the Uehara Memorial FoundationResearch Fellowship.

1. Masland RH (2001) The fundamental plan of the retina. Nat Neurosci 4(9):877–886.2. Molyneaux BJ, Arlotta P, Menezes JR, Macklis JD (2007) Neuronal subtype specifica-

tion in the cerebral cortex. Nat Rev Neurosci 8(6):427–437.3. Kriegstein A, Alvarez-Buylla A (2009) The glial nature of embryonic and adult neural

stem cells. Annu Rev Neurosci 32:149–184.4. Bertrand V, Hobert O (2010) Lineage programming: Navigating through transient

regulatory states via binary decisions. Curr Opin Genet Dev 20(4):362–368.5. Raymond PA, Rivlin PK (1987) Germinal cells in the goldfish retina that produce rod

photoreceptors. Dev Biol 122(1):120–138.6. Turner DL, Snyder EY, Cepko CL (1990) Lineage-independent determination of cell

type in the embryonic mouse retina. Neuron 4(6):833–845.7. Lumsden A, Clarke JD, Keynes R, Fraser S (1994) Early phenotypic choices by neuronal

precursors, revealed by clonal analysis of the chick embryo hindbrain. Development120(6):1581–1589.

8. Bernardos RL, Barthel LK, Meyers JR, Raymond PA (2007) Late-stage neuronal pro-genitors in the retina are radial Müller glia that function as retinal stem cells.J Neurosci 27(26):7028–7040.

9. Espinosa JS, Luo L (2008) Timing neurogenesis and differentiation: Insights fromquantitative clonal analyses of cerebellar granule cells. J Neurosci 28(10):2301–2312.

10. Godinho L, et al. (2007) Nonapical symmetric divisions underlie horizontal cell layerformation in the developing retina in vivo. Neuron 56(4):597–603.

11. Rompani SB, Cepko CL (2008) Retinal progenitor cells can produce restricted subsetsof horizontal cells. Proc Natl Acad Sci USA 105(1):192–197.

12. Hafler BP, et al. (2012) Transcription factor Olig2 defines subpopulations of retinalprogenitor cells biased toward specific cell fates. Proc Natl Acad Sci USA 109(20):7882–7887.

13. He J, et al. (2012) How variable clones build an invariant retina. Neuron 75(5):786–798.

14. Endo K, et al. (2012) Chromatin modification of Notch targets in olfactory receptorneuron diversification. Nat Neurosci 15(2):224–233.

15. Endo K, Aoki T, Yoda Y, Kimura K, Hama C (2007) Notch signal organizes the Dro-sophila olfactory circuitry by diversifying the sensory neuronal lineages. Nat Neurosci10(2):153–160.

16. Kimura Y, Satou C, Higashijima S (2008) V2a and V2b neurons are generated by thefinal divisions of pair-producing progenitors in the zebrafish spinal cord. De-velopment 135(18):3001–3005.

17. Satou C, Kimura Y, Higashijima S (2012) Generation of multiple classes of V0 neuronsin zebrafish spinal cord: Progenitor heterogeneity and temporal control of neuronaldiversity. J Neurosci 32(5):1771–1783.

18. De la Huerta I, Kim IJ, Voinescu PE, Sanes JR (2012) Direction-selective retinal ganglioncells arise from molecularly specified multipotential progenitors. Proc Natl Acad SciUSA 109(43):17663–17668.

19. Shen Q, et al. (2006) The timing of cortical neurogenesis is encoded within lineages ofindividual progenitor cells. Nat Neurosci 9(6):743–751.

20. Gomes FL, et al. (2011) Reconstruction of rat retinal progenitor cell lineages in vitroreveals a surprising degree of stochasticity in cell fate decisions. Development 138(2):227–235.

21. Ebrey T, Koutalos Y (2001) Vertebrate photoreceptors. Prog Retin Eye Res 20(1):49–94.22. Dacey DM (2000) Parallel pathways for spectral coding in primate retina. Annu Rev

Neurosci 23:743–775.23. Vihtelic TS, Doro CJ, Hyde DR (1999) Cloning and characterization of six zebrafish

photoreceptor opsin cDNAs and immunolocalization of their corresponding proteins.Vis Neurosci 16(3):571–585.

24. Chinen A, Hamaoka T, Yamada Y, Kawamura S (2003) Gene duplication and spectraldiversification of cone visual pigments of zebrafish. Genetics 163(2):663–675.

25. Ng L, et al. (2001) A thyroid hormone receptor that is required for the developmentof green cone photoreceptors. Nat Genet 27(1):94–98.

26. Schmitt EA, Dowling JE (1999) Early retinal development in the zebrafish, Danio rerio:Light and electron microscopic analyses. J Comp Neurol 404(4):515–536.

27. Jones I, Ng L, Liu H, Forrest D (2007) An intron control region differentially regulatesexpression of thyroid hormone receptor beta2 in the cochlea, pituitary, and conephotoreceptors. Mol Endocrinol 21(5):1108–1119.

28. Liu Y, Shen Y, Rest JS, Raymond PA, Zack DJ (2001) Isolation and characterization ofa zebrafish homologue of the cone rod homeobox gene. Invest Ophthalmol Vis Sci42(2):481–487.

29. Shen YC, Raymond PA (2004) Zebrafish cone-rod (crx) homeobox gene promotesretinogenesis. Dev Biol 269(1):237–251.

30. Takechi M, Hamaoka T, Kawamura S (2003) Fluorescence visualization of ultraviolet-sensitive cone photoreceptor development in living zebrafish. FEBS Lett 553(1-2):90–94.

31. Applebury ML, et al. (2000) The murine cone photoreceptor: A single cone type ex-presses both S and M opsins with retinal spatial patterning. Neuron 27(3):513–523.

32. Kennedy BN, et al. (2007) Identification of a zebrafish cone photoreceptor-specificpromoter and genetic rescue of achromatopsia in the nof mutant. Invest OphthalmolVis Sci 48(2):522–529.

33. de Felipe P, et al. (2006) E unum pluribus: Multiple proteins from a self-processingpolyprotein. Trends Biotechnol 24(2):68–75.

34. Allison WT, et al. (2010) Ontogeny of cone photoreceptor mosaics in zebrafish.J Comp Neurol 518(20):4182–4195.

35. Swaroop A, Kim D, Forrest D (2010) Transcriptional regulation of photoreceptor de-velopment and homeostasis in the mammalian retina. Nat Rev Neurosci 11(8):563–576.

36. Nishida A, et al. (2003) Otx2 homeobox gene controls retinal photoreceptor cell fateand pineal gland development. Nat Neurosci 6(12):1255–1263.

37. Mears AJ, et al. (2001) Nrl is required for rod photoreceptor development. Nat Genet29(4):447–452.

38. Roberts MR, Srinivas M, Forrest D, Morreale de Escobar G, Reh TA (2006) Making thegradient: Thyroid hormone regulates cone opsin expression in the developing mouseretina. Proc Natl Acad Sci USA 103(16):6218–6223.

39. Roberts MR, Hendrickson A, McGuire CR, Reh TA (2005) Retinoid X receptor (gamma)is necessary to establish the S-opsin gradient in cone photoreceptors of the de-veloping mouse retina. Invest Ophthalmol Vis Sci 46(8):2897–2904.

40. Okano T, Kojima D, Fukada Y, Shichida Y, Yoshizawa T (1992) Primary structures ofchicken cone visual pigments: Vertebrate rhodopsins have evolved out of cone visualpigments. Proc Natl Acad Sci USA 89(13):5932–5936.

41. Trimarchi JM, Harpavat S, Billings NA, Cepko CL (2008) Thyroid hormone componentsare expressed in three sequential waves during development of the chick retina. BMCDev Biol 8:101.

42. Alvarez-Delfin K, et al. (2009) Tbx2b is required for ultraviolet photoreceptor cellspecification during zebrafish retinal development. Proc Natl Acad Sci USA 106(6):2023–2028.

43. Stevens CB, Cameron DA, Stenkamp DL (2011) Plasticity of photoreceptor-generatingretinal progenitors revealed by prolonged retinoic acid exposure. BMC Dev Biol 11:51.

44. Glaschke A, et al. (2011) Thyroid hormone controls cone opsin expression in the retinaof adult rodents. J Neurosci 31(13):4844–4851.

45. Vasiliauskas D, et al. (2011) Feedback from rhodopsin controls rhodopsin exclusion inDrosophila photoreceptors. Nature 479(7371):108–112.

46. Cheng CL, Gan KJ, Flamarique IN (2009) Thyroid hormone induces a time-dependentopsin switch in the retina of salmonid fishes. Invest Ophthalmol Vis Sci 50(6):3024–3032.

47. Luo W, et al. (2004) Proximal and distal sequences control UV cone pigment geneexpression in transgenic zebrafish. J Biol Chem 279(18):19286–19293.

48. Yin J, et al. (2012) The 1D4 antibody labels outer segments of long double cone butnot rod photoreceptors in zebrafish. Invest Ophthalmol Vis Sci 53(8):4943–4951.

49. Larison KD, Bremiller R (1990) Early onset of phenotype and cell patterning in theembryonic zebrafish retina. Development 109(3):567–576.

50. Takechi M, Seno S, Kawamura S (2008) Identification of cis-acting elements repressingblue opsin expression in zebrafish UV cones and pineal cells. J Biol Chem 283(46):31625–31632.

15114 | www.pnas.org/cgi/doi/10.1073/pnas.1303551110 Suzuki et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0