Zebrabow: multispectral cell labeling for cell tracing and lineage analysis in zebrafish

A deeper understanding of developmental and neurobiological processes requires high-resolution visualization of cell lineages and assemblies. The accessibility and translucency of zebrafish make it an ideal system for dissecting the cellular basis of vertebrate development. Indeed, substantial progress has been made in labeling zebrafish cells and following their trajectory through development. Initial advances based on cell labeling with organic dyes resulted in the establishment of fate maps, lineage diagrams and neural circuits (Kimmel and Law, 1985). The discovery of genetically encoded fluorescent proteins in conjunction with novel transgenic and microscopy technologies allowed for visualization of different cell types and subcellular compartments (Distel et al., 2006; Kawakami, 2004; Keller et al., 2008; Megason, 2009).

Despite these significant developments, current technologies still have several limitations. When a large number of cells are labeled, individual cells are often difficult to distinguish. In the nervous system, overlapping axons and dendrites cannot be resolved with conventional fluorescence microscopy, rendering it difficult to trace the precise connectivity of individual neurons. Similar problems arise during time-lapse visualization owing to the low speed and resolution of current confocal and multiphoton technologies. To circumvent this challenge, two or three fluorescent colors (Distel et al., 2006; Hatta et al., 2006; Megason, 2009; Teddy et al., 2005) or faster imaging techniques (Dunsby, 2008; Huisken and Stainier, 2007; Keller et al., 2008) have been used. In these studies, however, each cell is labeled with the same set of colors (for example membrane in red and nuclei in green), which provides no distinction between adjacent cells.

One potential solution to this problem involves labeling adjacent cells with many different colors, which was achieved by the development of Brainbow (Lichtman et al., 2008; Livet et al., 2007). The Brainbow construct contains a promoter followed by three fluorescent proteins: RFP, CFP and YFP (Fig. 1A). Expression of one and only one of these three proteins (per one copy of the construct) is achieved by the use of Lox sites, the recognition sites for Cre recombinase. Remarkably, transgenic mice that carried multiple reporter insertions showed a large variety of colors owing to stochastic recombination and combinatorial expression of fluorescent proteins in each cell (Fig. 1B). The unique combination provides each cell a distinct color, allowing resolution of individual cell boundaries. In addition to enhancing visual resolution, Brainbow can also be used as a multi-lineage marker (Buckingham and Meilhac, 2011; Kretzschmar and Watt, 2012). The stochastic recombination events in individual progenitor cells are inherited by their progeny, resulting in clones marked by different colors (Gupta and Poss, 2012; Snippert et al., 2010; Tabansky et al., 2012).

Previous studies have applied the Brainbow technology to zebrafish and have shown that Cre induction can generate many distinct colors from microinjected Brainbow plasmid DNA (Pan et al., 2011) or β-actin-2:Brainbow transgenes (Gupta and Poss, 2012). These studies were restricted to early embryogenesis and heart tissue, respectively, and have not addressed several key questions: can Brainbow be used in a wide variety of tissues? What are the optimal strategies to achieve broad or tissue-specific labeling? How can color diversity be maximized? How stable are colors over time and through cell divisions? Can Brainbow be used for clonal analysis in multiple organs? Here, we address these issues and describe tools and methods that will allow for the broad application of the Brainbow technology in zebrafish (Zebrabow): we describe new transgenic lines for ubiquitous or tissue-specific multicolor labeling; we show that multicolor labeling facilitates axonal tracing; we maximize color diversity by optimizing Cre activity; we show that colors are stable over time and after cell divisions; and we demonstrate that Zebrabow can be used for long-term clonal analysis in a wide variety of tissues. Long-term time-lapse imaging shows that the zebrafish corneal epithelium undergoes dramatic changes in clonal structures during development and suggests that late-born corneal clones are formed by centripetal expansion from the periphery. The Zebrabow tools and methods reported here provide an important resource to facilitate multicolor fluorescent labeling in a wide variety of applications.

Zebrafish from the TL/AB strain were maintained using standard procedures (Westerfield, 2000). Embryos were raised at 28.5°C in embryo water containing 0.1% Methylene Blue hydrate (Sigma, St Louis, MO, USA) and 0.03% Instant Ocean sea salt (United Pet Group, Cincinnati, OH, USA). At 24 hours post-fertilization (hpf), embryos were transferred to water with 0.003% 1-phenyl-2-thiourea (PTU; Sigma) to inhibit pigment formation. Developmental stages were determined as described previously (Kimmel et al., 1995). The pax2a:CreER¹⁹ transgenic was obtained from the Brand laboratory (Biotechnologisches Zentrum, Dresden, Germany) (Hans et al., 2009). The cmlc2:CreER transgenic was obtained from the Poss laboratory (Duke University, Durham, NC, USA) (Kikuchi et al., 2010). The ubi:CreER transgenic was obtained from the Zon laboratory (Boston Children’s Hospital, Boston, MA, USA) (Mosimann et al., 2011). The s1102t Gal4 driver line was obtained from the Zebrafish International Resource Center (Eugene, OR, USA). The s1020t Gal4 driver line was obtained from the Baier (Max Planck Institute of Neurobiology, Martinsried, Germany) and Isacoff (University of California, Berkeley, CA, USA) laboratories (Scott and Baier, 2009; Scott et al., 2007).

Genomic DNA was prepared with a modified HotSHOT method (Meeker et al., 2007). Embryos or fin clips were digested with 100 μl 50 mM sodium hydroxide (Sigma), incubated at 95°C for 20 minutes and then chilled to 4°C. 10 μl 1 M Tris-HCl (pH 7.5) (Sigma) was then added to each sample. Genomic DNA was mixed with primers and GoTaq qPCR Master Mix (Promega) and reactions were run on an Opticon 2 Real-Time Cycler (Bio-Rad). Primers for genomic DNA control are: 5′-TCAGTCAACCATTCAGTGGCCCAT-3′ and 5′-CAGGAAAGGGAATGCAGGGTTTGT-3′. Primers for Zebrabow are: 5′-ACCTGGTGGAGTTCAAGACCATCT-3′ and 5′-TCGTACTGTTCCACGATGGTGTAGTC-3′. Copy number was determined with the comparative Ct method (Tesson et al., 2010). Genomic DNA from ubi:Zebrabow-S heterozygote was used as single-copy reference. Δ(Ct)=(Ct for Zebrabow PCR) - (Ct for control PCR); ΔΔCt=Δ(Ct, sample) - Δ(Ct, reference). Copy number in the sample=2^-ΔΔCt.

Zebrabow transgenesis constructs were generated with an upstream Gateway selection cassette (Life Technologies) (Zebrabow-GateDest; supplementary material Fig. S1A). The Gateway selection cassette allows insertion of promoter elements upstream of Zebrabow via homologous recombination (Kwan et al., 2007; Villefranc et al., 2007). The Zebrabow vectors have flanking Tol2 transposase and I-SceI meganuclease recognition sites to enhance transgenesis efficiency (Kikuta and Kawakami, 2009; Thermes et al., 2002). Tol2-mediated transgenesis mediates high efficiency single insertions into the genome, whereas meganuclease-mediated transgenesis mediates lower efficiency, but multicopy, insertions into a single genomic locus. Completed constructs were tested by plasmid DNA injection into embryos (Pan et al., 2011).

The ubi:Zebrabow transgenic construct was generated by inserting the ubiquitin b promoter (Mosimann et al., 2011) into the Zebrabow-GateDest vector by homologous recombination. Plasmid DNA was isolated with QIAprep Spin Miniprep Kit (Qiagen) and repurified with QIAquick PCR purification Kit (Qiagen) to remove residual RNase. To generate transgenic lines, 35 pg Tol2 mRNA and 20 pg plasmid DNA were co-injected into one-cell stage embryos. Eighty percent (18/22) of crosses between injected breeders gave rise to Zebrabow-expressing F1s. The Tg(ubi:Zebrabow-S^a132) line (referred to as ubi:Zebrabow-S or line S) shows Mendelian inheritance consistent with a single transgene. Using inverse PCR, the insertion site was mapped to chromosome 7 (chr7:27,259,360), in the first intron of an uncharacterized gene.

The Tg(ubi:Zebrabow-M^a131) line (referred to as ubi:Zebrabow-M or line M) had higher expression than ubi:Zebrabow-S. Using quantitative PCR of genomic DNA with ubi:Zebrabow-S heterozygote as a single copy control, 16-32 copies of ubi:Zebrabow transgene were found in this line. Expression starts between 6 and 9 hpf, and was present in all cell types examined. As ubi:Zebrabow insertions segregate in each generation, it is necessary to raise larvae with the brightest RFP expression. Quantitative PCR is then used to select for adult animals that harbor high numbers of transgenic insertions. Despite these measures, some insertions are lost in each generation. Sperm from F2 ubi:Zebrabow-M males was cryogenically preserved, allowing replenishment of the repertoire of Zebrabow transgenes by in vitro fertilization. Additionally, the high rate of transgenesis with Tol2 transposase allows individual investigators to generate ubi:Zebrabow multiple insertion lines easily.

The Tg(UAS:Zebrabow-V^a130) line (referred to as UAS:Zebrabow-V or line V) was generated using the UAS:Brainbow construct described previously (Pan et al., 2011). The 14×UAS-E1B promoter (Köster and Fraser, 2001) was used, and transgenesis was performed with I-SceI meganuclease (New England BioLabs, MA, USA) (Thermes et al., 2002). F1 progeny from potential founders were injected with Gal4 mRNA and screened by RFP expression. Quantitative PCR of genomic DNA showed that there are two insertions (likely in the same locus) in this line.

The Tg(UAS:Zebrabow-B^a133) line (referred to as UAS:Zebrabow-B or line B) was generated using the non-repetitive 4×UAS promoter (Akitake et al., 2011). Plasmid DNA was co-injected with Tol2 mRNA into a Gal4 transgenic line with strong expression in the brain, skin and notochord (D.S. and A.F.S., unpublished). Potential founders were screened by RFP expression and a single line was identified. Quantitative PCR suggests 9-31 inserts per fish in this line. The UAS:Zebrabow-B line was maintained in the background of the s1102t Gal4 line; embryos with strong expression in the trigeminal ganglion were selected in each generation. Quantitative PCR was performed to select for adult animals harboring high numbers of transgenic insertions.

The Tg(hsp:Cre^a134) line (referred to as hsp:Cre) was generated by inserting the zebrafish hsp70 promoter and the Cre gene into a plasmid containing flanking I-SceI meganuclease sites (a gift from Y. Rong, NIH, Bethesda, MD, USA). Plasmid DNA was injected into one-cell stage embryos and potential founders were screened by fin-clip PCR with transgene specific primers. hsp:Cre lines were maintained separately from Zebrabow lines to avoid sporadic recombination in the germline. Some heat-induced recombination in superficial structures can occur if embryos are exposed to larger volumes of low-melting point agarose. This can be avoided by using small volumes of low-melting point agarose (that cool quickly) or by using alternative mounting methods, e.g. methylcellulose.

Fish were anesthetized with ethyl 3-aminobenzoate methane sulfonate salt solution (Mesab, Sigma) and mounted in 1.0-1.5% low-melting point agarose on glass-bottomed culture dishes (P35G-0-14-C, MatTek). Confocal images were acquired with a 20× objective (Olympus XLUMPFL, NA=0.95) on an upright Olympus FV1000 microscope (Olympus, USA). The following laser lines and emission filters were used. RFP (dTomato): 561 nm laser and 585-685 nm band-pass filter. YFP: 515 nm laser and 535-565 nm band-pass filter. CFP (Cerulean): 440 nm laser and 465-495 nm band-pass filter.

For time-lapse imaging, settings were kept identical, except for the confocal pinhole. In the ubi:Zebrabow-M line, fluorescence intensity gradually increases during development. To prevent image saturation, the confocal pinhole was reduced to decrease the amount of captured photons without changing the ratio of different fluorophores. For Hue and saturation analysis, tif image files were processed with Photoshop software (Adobe, USA). In Photoshop, color display mode was set to Hue-Saturation-Brightness (HSB) by selecting the ‘HSB sliders’ option in the ‘Color’ window. Using the eye dropper tool with sample size set at 5×5 pixels, an area at the center of a cell was selected. The average color values of the selected area was displayed in the ‘Color’ window and recorded manually. Single z-sections were used for color measurement. In cases where cells are less than 5 pixels wide in a single z-section, maximal intensity projection of two to three z-sections was used. Outlines of clones were determined by manually marking boundaries of cells in optical sections. For two-photon imaging, a Zeiss 710 NLO two-photon system with a 20× objective (W Plan-Apochromat, NA=1.0) was used (Zeiss, Germany). A Chameleon Ti:Sapphire laser (Coherent, USA) was tuned to 850 nm for CFP excitation and 930 nm for RFP and YFP excitation. Axonal tracing was carried out manually using the NeuronJ plugin in the ImageJ software (Schneider et al., 2012).

Purified Cre protein was obtained from the Liu lab (Harvard University, Cambridge, MA, USA). Wild-type Cre was fused with a C-terminal 6×His tag and purified from bacterial lysate with Ni-NTA agarose resin, eluted and dialyzed (Cronican et al., 2010). For injection, 50 μM Cre protein was diluted 1:20 or 1:40 in 0.1 M potassium chloride (Sigma) and 1 nl of solution was injected into a single blastomere between the one- and eight-cell stages.

12 hpf embryos were transferred with 1 ml of embryo water into 1.5 ml Eppendorf tubes and put into a 37°C heat block for 2.5, 5, 10, 15 or 20 minutes. Tubes were then left in 28.5°C incubator to cool down for 20 minutes before being transferred back into a Petri dish.

Trangenics containing ubi:CreER were treated with tamoxifen (25 μg/ml, Sigma) or 4-hydroxy-tamoxifen (5-10 μM) at late gastrulation stage (between 6 and 12 hpf). After treatment, tamoxifen or 4-hydroxy-tamoxifen was washed out with embryo water three times. Tamoxifen treatments for pax2a:CreER and cmlc2:CreER were performed as previously described (Hans et al., 2009; Kikuchi et al., 2010).

For Zebrabow transgenesis, we used the Brainbow1.0L DNA cassette (Fig. 1A) (Livet et al., 2007). The default expression is RFP, but Cre-mediated recombination switches expression to either CFP or YFP. The arrangement of two pairs of mutually incompatible Lox sites ensures that recombination occurs only once and is irreversible. These three fluorescent proteins have non-overlapping excitation/emission profiles and their colors can be separated with commonly used optical filters (Livet et al., 2007; Pan et al., 2011; Weissman et al., 2011). We generated two types of Zebrabow transgenic lines: ubi:Zebrabow and UAS:Zebrabow (Table 1).

To create Zebrabow lines with ubiquitous and sustained expression from embryonic to adult stages, we used the promoter from the zebrafish ubiquitin B (ubi) gene (Fig. 2A) (Mosimann et al., 2011). The ubi promoter drives ubiquitous and persistent transgenic expression and labels many cell types that are not labeled by other broad activity promoters, including the β-actin-2 promoter used in the priZm Brainbow line (Gupta and Poss, 2012). We characterized two high-expressing lines: line S (ubi:Zebrabow-S) with a single mapped ubi:Zebrabow insertion; and line M (ubi:Zebrabow-M) with multiple independent insertions (Fig. 2). Without Cre recombination, both lines express only RFP (Fig. 2B). In line S, Cre recombination generates three colors (in heterozygotes) or six colors (in homozygotes) (Fig. 2C,D). In line M, Cre recombination results in stochastic, combinatorial expression of RFP, CFP and YFP (Fig. 2E). Most cells express all three fluorophores, and more than 30 colors were commonly observed in a single optical section (see below). Expression before 24 hours post-fertilization (hpf) is relatively weak but quickly increases afterwards. Strong fluorescence was observed both in juvenile and adult animals (Fig. 2F; supplementary material Fig. S1).

We tested ubi:Zebrabow-M with different Cre expression methods to achieve either broad or tissue-specific multicolor labeling (Tables 2 and 3). For broad labeling, we found three approaches that generated consistent multicolor labeling: (1) Cre protein microinjection at the blastula stage (Fig. 2E); (2) crossing to a heat shock-inducible Cre line (hsp:Cre; Fig. 5C); and (3) crossing to a tamoxifen-inducible Cre line (ubi:CreER; Fig. 5F) (Mosimann et al., 2011). To test whether ubi:Zebrabow-M could be used for tissue-specific labeling, we crossed ubi:Zebrabow-M to two tissue-specific Cre lines: (1) cmlc2:CreER, which is expressed in cardiomyocytes (Kikuchi et al., 2010); and (2) pax2a:CreER¹⁹, which is expressed in the forebrain diencephalon (Hans et al., 2009). Multicolor labeling was observed in Cre-expressing tissues, whereas other tissues only expressed RFP (default expression) (Fig. 2G,H). Therefore, multiple methods of Cre-mediated recombination can be used to generate ubiquitous or tissue-specific labeling.

To render Zebrabow expression Gal4-inducible, we generated two UAS:Zebrabow reporter lines, one with standard 14×UAS repeats and more variegated expression (line V, UAS:Zebrabow-V), and the other with 4× non-repetitive UAS and broader expression (line B, UAS:Zebrabow-B) (Table 1 and Fig. 3) (Akitake et al., 2011; Köster and Fraser, 2001). UAS:Zebrabow-V has two non-segregating copies of Zebrabow (determined by quantitative PCR of genomic DNA, see Materials and methods) and shows robust expression. In the presence of Gal4 but without Cre recombination, the UAS line behaves like a RFP reporter line (Fig. 3B). After Cre recombination, we observed six colors in UAS:Zebrabow-V heterozygous animals (consistent with expression of two copies of UAS:Zebrabow) (Fig. 3C) and more than ten colors in the progeny of heterozygous parents (Fig. 3D). The UAS:Zebrabow-B line has more than ten independent inserts (determined by quantitative PCR), which results in a wide variety of fluorescent colors after Cre injection (Fig. 3F). Both lines are compatible with multiple Gal4 lines (Fig. 3; data not shown), but UAS:Zebrabow-B labels many more cells than the UAS:Zebrabow-V line for a given Gal4 driver (Fig. 3E,F). Some Gal4 lines, e.g. atoh7:Gal4 (expressed in retinal ganglion cells) (Del Bene et al., 2010), are able to drive Zebrabow expression only in line B and not line V (C. Riegler and F. Engert, personal communication). These results suggest that line B is more permissive to Gal4-induced expression and might have a lower threshold for Gal4 activation. The sparse labeling and strong expression in UAS:Zebrabow-V is therefore best suited for tracking a few cells (see below), whereas the broader expression and more diverse colors in UAS:Zebrabow-B may be better suited for analyses of large cell populations.

Different Gal4 and Cre expression methods were tested in UAS:Zebrabow-B for broad and tissue-specific labeling (Tables 2 and 3). Broad Zebrabow expression can be induced by Gal4 mRNA and Cre protein co-injection at the blastula stage (Fig. 3G,H). Labeling density can be controlled by injecting Gal4 mRNA at the one-cell stage (ubiquitous labeling) or into one of the blastomeres at the four- to eight-cell stage (sparse labeling). Expression starts between 5 and 8 hpf and is very robust by 14 hpf, in contrast to the relatively weak embryonic expression of ubi:Zebrabow at this stage. However, fluorescence gradually declines as Gal4 is degraded. For tissue-specific labeling with Gal4 driver lines, both hsp:Cre transgenic or Cre protein injection were suitable. Somatosensory axons labeled with UAS:Zebrabow-V showed normal development, suggesting that strong Zebrabow expression does not affect axonal outgrowth or formation of presynaptic terminals (Fig. 4) (Pan et al., 2012; Sagasti et al., 2005). Fluorescent colors are consistent at different points along the axon and in the cell body (supplementary material Fig. S2). The strong expression and relatively sparse labeling in UAS:Zebrabow-V is well suited for time-lapse imaging, and color diversity facilitates tracing of overlapping axonal arbors (Fig. 4C-F; supplementary material Movies 1 and 2).

To compare color diversity under different experimental conditions, we measured the hue and saturation of individual cells. Hue describes the visible light spectrum, from red to violet, and is measured in degrees, with red at 0°, green at 120° and blue at 240° (Fig. 5A). Saturation is the purity of color, describing the percentage of a given hue that is mixed with neutral color (white or gray). Pure red has 100% saturation, pink has intermediate saturation and white has 0% saturation (Fig. 5A). Fluorescence brightness also contributes to color, but brightness is significantly influenced by differences in expression level, cellular topography and light scattering at different depth of imaging. We therefore focused our analysis on hue and saturation.

Using the heat-shock inducible hsp:Cre line, we found that the extent of recombination could be controlled by the duration of heat shock. With a brief heat shock (2.5 minutes), very few cells underwent recombination. These cells were mostly magenta (red+blue) and yellow (red+green), with very similar hue and saturation values (Fig. 5B). With an intermediate heat shock (5 minutes), color diversity increased both visually and on the Hue-Saturation graph (Fig. 5C). Based on hue and saturation we estimated that there were ∼30 distinct colors in a single optical section. A longer heat shock (20 minutes) resulted in a reduction in color variety, as most cells were different shades of blue and green (Fig. 5D).

Optimization of color was also possible with pharmacological induction of Cre using the ubi:CreER transgenic line (Fig. 5E,F) (Mosimann et al., 2011). A low dose of tamoxifen (10 μg/ml) produced limited color diversity, whereas a high dose of tamoxifen (25 μg/ml) produced a similarly high level of color variety, as generated by intermediate length of heat shock. These results demonstrate that Zebrabow color diversity can be optimized by varying Cre levels and activity.

Stably maintained and inherited colors allow lineage tracing and fate mapping. Characterization of color profile of cells at an early time point allows re-identification of the cells with the same color profile at a later time point. This approach is ideally suited to determining the lineages of different progenitor populations or the clonal origins of organs or tumors. Zebrabow and Brainbow techniques rely on genomic rearrangements that are stably passed on from progenitor to progeny. It has been thus assumed that color would also be stable and identical among cells with the same clonal origin (Gupta and Poss, 2012; Tabansky et al., 2012). However, this assumption has not been directly validated. Changes in gene expression could conceivably occur without further Cre-mediated genomic modification, e.g. by gene silencing or other epigenetic mechanisms (Shilatifard, 2006; Vastenhouw and Schier, 2012). Color changes within a lineage could erroneously split single lineages into several distinct lineages (Walsh and Cepko, 1992). To address this issue, we followed a small cluster of cells from embryonic to post-embryonic stages and asked whether color remained stable over time and was faithfully inherited by daughter cells. The Hue-Saturation plot is a convenient way of evaluating color inheritance and stability. If colors remained the same, measurements from different time points would form a cluster on the Hue-Saturation graph; if not, data points would be dispersed.

We focused on the epithelial cells on the surface of the eye (the cornea), which are large and superficial. The corneal epithelium is also an important model system for stem cell biology and lineage analysis (see next section). Ubi:Zebrabow-M;ubi:CreER embryos were treated with tamoxifen at gastrulation stage (10-11 hpf) and corneal epithelial cells were followed over time (Fig. 6A). At 2 days post-fertilization (dpf), epithelial cells form single-color clusters that tile the corneal surface (Fig. 6B). Cells within a cluster have very similar Hue-Saturation profiles, suggesting that cells within a cluster share common lineage (Fig. 6E). Movement of cells between clusters was negligible, which allowed us to use relative location in addition to color to track clusters from embryonic to larval stages (Fig. 6B-D). Between 5 and 10 dpf, the number of cells per cluster increased almost threefold (3.4 cells/cluster to 9.4 cells/cluster, Fig. 6C,D,G), but the Hue-Saturation profile of individual cells stayed remarkably consistent (Fig. 6E,F). In very rare cases, the color profile did change slightly between 2 and 5 dpf (asterisks, Fig. 6E,F), probably owing to the time needed for mature fluorescent protein to accumulate. These results strongly suggest that color is faithfully inherited among the large majority of daughter cells and is stable over time. The combination of time-lapse imaging and color analyses suggest that single-color clusters in the cornea are likely to be formed by clonally derived cells and we will refer to them as clones in the following sections.

To test whether Zebrabow would be suitable for long-term clonal analyses, we investigated the developmental changes in the corneal epithelium. In the mammalian cornea, epithelial cells with strong proliferative capacity and stem cell-like properties are initially distributed across the entire cornea (Chung et al., 1992; Collinson et al., 2002; Davies et al., 2009). As the embryo develops, corneal stem cells become restricted to the peripheral edge of the cornea, the limbus (Davies and Di Girolamo, 2010). This shift in corneal stem cell location is reflected in the clonal composition of the cornea during the course of development. Random cell labeling in transgenic mice using the lineage marker lacZ results initially in small scattered clones that are then gradually replaced by large wedge-shaped clones (Collinson et al., 2002). The zebrafish cornea is structurally similar to mammalian cornea, but developmental changes in clonal structures have not been investigated (Soules and Link, 2005; Zhao et al., 2006). The cornea is therefore an excellent system to test the extent to which Zebrabow can reveal complex clonal relationships and dynamics.

To label clones, we induced Cre at 6-12 hpf in ubi:Zebrabow-M;ubi:CreER embryos. Confocal z-stacks of the cornea were acquired in a time series. At embryonic and early larval stages (2-10 dpf), the corneal surface was covered by the same cohort of embryonic epithelial clones (Fig. 6B-D). From 10 to 15 dpf, several notable changes began to reshape clone structure (Fig. 7A-D). In the peripheral cornea, clones were scattered and interspersed by smaller cells (Fig. 7C,C′); in the central cornea, rosette-like structures were frequently observed, accompanied by the presence of extruded epithelial cells (Fig. 7D-D′). This is reminiscent of the extrusion and death of surface epithelial cells in other systems (Eisenhoffer et al., 2012; Eisenhoffer and Rosenblatt, 2011). Consistent with the idea of cell extrusion, the number of cells per clone decreased significantly between 10 and 15 dpf (Fig. 6G).

At juvenile stage (3-4 weeks), clone structure becomes markedly different. Large wedge-shaped clones are seen, interspersed with scattered embryonic epithelial cells (Fig. 7E,F). This clonal structure is maintained in adults, similar to the clones seen in the adult mammalian cornea (data not shown). Indeed, the peripheral borders of the clones overlay the limbus and expand to the neighboring conjunctiva (Fig. 7G,H). To explore how clones are formed over time, we followed individual animals over the span of several weeks (5-23 dpf or 10-30 dpf) (Fig. 7I-L; supplementary material Fig. S3). We found that nascent clones were often found in the peripheral cornea by 10 dpf, underneath embryonic epithelial cells (Fig. 7J). From 10-23 dpf, clones expand peripherally into the conjunctiva and centrally towards the central cornea. At 30 dpf, some clones have reached the center of the cornea. These results are consistent with the model that corneal epithelial cells are derived from the limbus (Collinson et al., 2002; Nagasaki and Zhao, 2003; Sun et al., 2010). A working model of clonal development in the corneal epithelium is shown in Fig. 7M-P. These results demonstrate that dynamic clonal structures can be observed via long-term Zebrabow clonal analysis.

Last, we asked whether single-color cell clusters such as those seen in the cornea can be observed in other organ systems. As cells within these clusters are likely to be clonally related, the size and shape of single-color clusters provide clues about the extent of progenitor cell proliferation and cell migration during organogenesis. We used ubi:Zebrabow-M transgenic fish and generated embryonic clones by pulsed tamoxifen-induced Cre recombination or Cre protein injection. We first examined the nervous system, specifically in regions of the dorsal telencephalon and optic tectum that undergo robust post-embryonic neurogenesis (Grandel and Brand, 2012; Ito et al., 2010; Kizil et al., 2012; Mueller and Wullimann, 2003; Wullimann, 2009; Zupanc et al., 2005). In the dorsal telencephalon, single-color clusters of various shapes can be seen near the ventricular (dorsal) surface starting at 14 dpf (Fig. 8A). This is consistent with active neurogenesis in the ventricular surface and suggests that large areas of the forebrain are formed by clonally related cells (Zupanc et al., 2005). By contrast, the dorsal optic tectum is mostly composed of medial-lateral clusters at 14 dpf (Fig. 8C). This structure is consistent with recent evidence showing that tectal stem cells are located medially and their progeny migrate laterally (Ito et al., 2010). These clone-like clusters were also observed at 28 dpf, despite the significant increase in brain size, suggesting that the observed structures are stable (Fig. 8B,D). Overall, these results suggest that clustering of clonally related cells might be a prominent feature of the mature dorsal telencephalon and optic tectum.

We also observed single-color clusters in other organ systems. In the pectoral fin, chondrocytes (cartilage cells) form small and scattered clusters, suggesting that local and dispersed proliferation is the predominant mode of growth (Fig. 8E) (Dewit et al., 2011). In the caudal fin, proximal-distal arrays of lateral line organs were seen, consistent with previous descriptions of clones of lateral line cells (Fig. 8F,G) (Tu and Johnson, 2011). These results demonstrate the potential of Zebrabow for long-term lineage analyses in different tissues.

The results presented here demonstrate the strength and versatility of multicolor genetic labeling in zebrafish and provide well-characterized tools and validated methods that can be readily adopted. The ubi:Zebrabow and UAS:Zebrabow lines are suitable for both broad and tissue-specific multicolor labeling: color is faithfully inherited between daughter cells and is stable over time, and clone-like structures can be identified in tissues ranging from the cornea and forebrain to caudal fin and cartilage. As different Zebrabow and Cre combinations have distinct advantages (Tables 2 and 3), investigators can select from a wide array of approaches to suit their imaging needs. Thus, Zebrabow provides the resources necessary for systematic anatomical and lineage studies during zebrafish development.

Our results show that Zebrabow colors remain stable over time. However, colors only become stable after color establishment and maturation. First, Cre needs to be expressed and then recombine the Lox sites. Second, fluorescent protein concentration needs to reach a steady state. The non-default fluorophores (CFP and YFP) need time to accumulate, whereas the default fluorophore (RFP) needs time to degrade. Thus, the color maturation rate depends on the timing of Cre recombination, the amount of residual RFP, and accumulation of mature CFP and YFP. One way to minimize maturation time is to induce Cre-mediated recombination prior to the significant accumulation of RFP. To maintain color stability, it is also essential that Cre activity is transient so that RFP expression is not continuously changed to CFP or YFP expression.

Imaging conditions also impact color stability. Different fluorescent proteins have different photostability profiles, excitation and emission spectra, and abilities to withstand fixation (Shaner et al., 2008; Shaner et al., 2005; Weissman et al., 2011). These factors can impact color after sample preparation and imaging. For example, in deeper tissues, longer wavelength light (emitted by RFP) will be scattered less than shorter wavelength light. We have focused on more superficial structures because imaging beyond the depth of 200-300 μm remains challenging. It will be interesting to test whether recent advances in microscopy or tissue clearing techniques will improve Zebrabow imaging in deeper structures (Hama et al., 2011; Kaufmann et al., 2012; Keller et al., 2008; Kuwajima et al., 2013). Intense laser excitation may also skew the relative ratio of the three fluorophores because RFP and CFP are less photostable than YFP (Weissman et al., 2011). Our results show the ability of Zebrabow to generate stable colors, but color establishment and stability need to be tested empirically.

Color diversity is another important variable in multicolor imaging. High color diversity makes each cell more traceable and reduces the chance that different clones have identical colors. Our results show that Zebrabow color profiles change in a predictable trajectory in response to increasing Cre levels and activity. As long as Cre activity is tunable (with heat shock or tamoxifen), it is possible to generate optimal color diversity. We also found that the number of cells that have undergone recombination increases with increasing Cre activity. Recombination takes place in fewer cells under low color diversity conditions and in more cells under high diversity conditions. In applications in which both sparse labeling and high color diversity are required, transplantation approaches (e.g. ubi:Zebrabow into wild type) can be used to reduce labeling density (data not shown).

Our study demonstrates that Zebrabow has the potential for clonal analysis in a wide variety of tissues, including the cornea, brain, muscle, cartilage and vasculature. The diversity of clone-like clusters in different organs suggests different modes of progenitor expansion during organogenesis. For example, the presence of large cohesive clusters suggests not only rapid proliferation of a small pool of progenitor cells but also limited dispersal of daughter cells from their site of origin. Such aspects of cell behavior can be studied readily with the techniques used here.

Putative clones can be identified not only by the shared colors of cohesive clusters, but color may also be used to determine whether dispersed cells might be clonally related. Compared with single or double labeling, the wide diversity of color in Zebrabow reduces the likelihood that unrelated cells have the same color. However, extensive dispersal of cells with similar color can hinder the unambiguous assignment of clonal relationships. One approach to help identify clones would be to compare the number of cells per single-color clone (clone size) at different labeling density: if cells with the same color are clonally related, clone size will be the same regardless of labeling density. By contrast, if many unrelated cells share the same color, clone size would increase with labeling density. Such calculations have been performed in retroviral clonal analysis and could be applied to color analysis (Galileo et al., 1990).

In the cornea, it has long been believed that epithelial stem cells are located exclusively in the limbus and that corneal clones are formed by centripetal growth from the limbus (Davies and Di Girolamo, 2010; Lavker et al., 2004). Interestingly, new evidence suggest that corneal stem cells might be scattered over the entire cornea and that corneal clones may be formed by centrifugal, rather than centripetal, growth (Majo et al., 2008). We find that many zebrafish corneal clones are derived from the peripheral cornea, a region analogous to the limbus. Although our results do not exclude the possibility that a subset of clones might originate from the central cornea, time-lapse imaging of single clones suggests that corneal clones form by centripetal expansion of limbus-derived clones. These results demonstrate that the Zebrabow resource described here is ideally suited to address fundamental questions in organogenesis and tissue homeostasis.

We thank D. Cai for helpful suggestions; C. Mosimann and L. Zon for suggestions and the ubi:CreER transgenic; K. Cerveny for help with inverse PCR and adult zebrafish histology; J. Cronican and D. Liu for providing purified Cre protein; M. Brand for the pax2a:CreER transgenic; K. Poss for the cmlc2:CreER transgenic; H. Baier and E. Isacoff for the s1020t Gal4 driver line; Y. Rong for the pBSII-KS-I-SceI plasmid; Z. Tobias for analyses on the onset of Zebrabow expression; J. Miller and K. Hurley for fish care; J. Gagnon, K. Rogers and A. Pauli for comments on the manuscript; and members of the Schier laboratory for discussion.