FRaeppli: a multispectral imaging toolbox for cell tracing and dense tissue analysis in zebrafish

A prerequisite for uncovering the cellular architecture of organs and entire embryos, as well as the intricate cellular processes that guide tissue formation, such as cell interaction, movement and proliferation, is the capability to visualize the resident cell types at single-cell resolution (Megason and Fraser, 2007). The transparency and rapid growth of zebrafish are ideal for dissecting the cellular processes that establish tissue architectures in vertebrate development and disease. Uniform membrane labelling within a tissue generates an approximate view of cellular shapes, suitable for uniformly shaped cells, while elaborate cellular morphologies and intricate contacts, e.g. via protrusions (Caviglia and Ober, 2018), remain elusive. Instead, sparse labelling by expression of a single membrane-targeted fluorescent protein (FP) (e.g. Cayuso et al., 2016; Sanders et al., 2013) outlines individual cells and has significantly advanced our understanding of dynamic cell behaviours. Nevertheless, information concerning the cellular topologies, the nature and morphology of contacts between adjacent cells is frequently lacking, because the cellular microenvironment remains unlabelled.

The need to distinguish neighbouring cells has been met partially by the Brainbow system (Livet et al., 2007), which randomly labels neighbouring cells by the stochastic expression of three or four FPs (reviewed by Richier and Salecker, 2015; Weissman and Pan, 2015). Stochastic combinatorial expression of spectrally distinct FPs is achieved by their sequential arrangement in a DNA cassette, with each FP flanked by recombinase recognition sites, permitting stochastic recombinase-mediated DNA excision or inversion. This approach, pioneered in mice, has been adapted in a variety of species, including zebrafish (Gupta and Poss, 2012; Livet et al., 2007; Pan et al., 2013; Richier and Salecker, 2015; Weissman and Pan, 2015). Key advances range from uncovering distinct neural networks (e.g. Fontenas and Kucenas, 2021; Forster et al., 2017) to clonal dominance of stem cell-like populations in differentiating tissues (e.g. (Gupta and Poss, 2012; Gurevich et al., 2016). Brainbow-tools targeting FPs to specific cellular compartments, such as the membrane, have been successfully used in mice and flies, but rarely in zebrafish (Chan et al., 2022; Forster et al., 2017). Moreover, comprehensively resolving the complex organization of tissues requires information about specific cell types, subcellular structures (e.g. cytoskeleton) or signalling pathway activity as landmarks, which could be achieved by combining multicolour cell labelling systems with spectrally distinct signals, such as infrared proteins (Cook et al., 2019). However, the majority of currently available reporter lines express GFP, which overlaps with the spectra of existing Brainbow-like systems and thereby precludes their combined use.

To address those experimental needs, we introduce the FRaeppli (Fish-Raeppli) tools: transgenic lines inspired by Raeppli, a whole-tissue labelling system developed in Drosophila (Kanca et al., 2014). FRaeppli expresses four bright monomeric FPs – mOrange2, mKate2, mTFP1 and TagBFP – that are spectrally distinct from GFP and infrared fluorophores, allowing simultaneous acquisition of six distinct labels. We demonstrate the versatility of the membrane-tethered fraeppli-caax for large-scale visualization of cell topologies and the nuclear fraeppli-nls for lineage tracing analysis in dense tissues. We show its versatility with examples of larval and adult tissues, genetic strategies to control clone size, live imaging and a hyperspectral protocol for time-efficient data acquisition.

For a versatile next-generation multicolour cell labelling system in zebrafish, we adapted Drosophila Raeppli (Kanca et al., 2014) and produced F(ish)Raeppli. The colour-cassette consists of four bright FPs: mOrange2 (E2-Orange in fraeppli-nls), mKate2, mTFP1 and TagBFP (Fig. 1A, Fig. S1). Based on their excitation and emission spectra, imaging of these FPs can be performed with fluorophores in the green/GFP and far-red range. FRaeppli is therefore compatible with many transgenic lines and diverse cell and organelle labels. We generated two different versions for application-specific use: in fraeppli-nls, a nuclear localization signal targets the FPs to the nucleus (NLS; Fig. 1B,C) enabling cell lineage and growth analyses; in fraeppli-caax, a Ras farnesylation sequence (CAAX) directs the FPs to the cell membrane for cell shape analyses (Fig. 1E). Recombination of the colour-cassette is mediated by the bacteriophage PhiC31 integrase (Bischof et al., 2007; Lister, 2011; Thorpe et al., 2000), only thus far used for transgenesis in zebrafish (Hu et al., 2011; Mosimann et al., 2013). PhiC31 catalyses recombination of unidirectional attB and attP sites, excising DNA fragments flanked by these sites. In FRaeppli, a single attB site is placed directly after the promoter and attP sites before each of the FP open reading frames (Fig. 1A). Recombination events generate attL sites, incompatible with further PhiC31-mediated recombination and thereby initiating stable expression of one FP per cell (Fig. S1). PhiC31 is built into the FRaeppli cassette and controlled by an upstream 1.5 kb-STOP-cassette flanked by lox2272 sites. Cre-mediated recombination of these sites excises the STOP-cassette, placing phiC31 expression under direct control of an upstream activating sequence (UAS). Expression of the FRaeppli construct is driven by the binary Gal4/UAS system (Brand and Perrimon, 1993). Binding of the transcriptional activator Gal4 or optimized versions (Distel et al., 2009) to regulatory UAS elements drives transcription of both PhiC31 and the stochastically selected FP, following PhiC31-mediated attB/attP recombination (Fig. S1D).

We tested FRaeppli recombination in the forming liver first by injecting cre mRNA into stable transgenic fraeppli-nls or fraeppli-caax embryos at the one-cell stage, labelling cells in four colours in most embryos (Fig. 1B, Fig. S2A). This indicates the correct expression of the FRaeppli-inbuilt phiC31 and subsequent induction of all four recombination events. Next, we tested recombination after conditional Cre expression by crossing fabp10a:kalTA4;fraeppli-caax to hsp70l:cre and subjecting embryos to a heat-shock at 33 h post-fertilisation (hpf). They expressed all four FPs in the liver (Fig. S2B,C), indicating that all elements of FRaeppli work correctly. Three independent lines each of fraeppli-nls and fraeppli-caax were identified, fraeppli-nls^cph1, fraeppli-nls^cph2, fraeppli-nls^cph3, fraeppli-caax^cph4, fraeppli-caax^cph5 and fraeppli-caax^cph6, with different levels of transgene activation (Fig. S2A-E). Most experiments were performed with the most penetrant lines: fraeppli-nls^cph1, fraeppli-nls^cph2, fraeppli-nls^cph3 and fraeppli-caax^cph4.

We expanded the conditional strategies for FRaeppli labelling by generating an inducible transgenic PhiC31 line: hsp70l:phiC31^cph7. Activation is faster, as it bypasses Cre-induced excision of the STOP-cassette by directly triggering colour-cassette recombination and is therefore more suitable for early embryogenesis (Fig. S1C). Similar to the FRaeppli-inbuilt PhiC31, exogenous PhiC31 expression in prox1a:kalTA4;hsp70l:phi;fraeppli-nls or fraeppli-caax larvae efficiently induces all recombination events (Fig. 1C,E). Examining mRNA and protein expression over time, by in situ hybridization chain reaction (HCR) and hsp70l:phiC31-sfGFP-NLS expression, respectively, showed increasing expression of phiC31 mRNA between 2 and 4 h and of PhiC31 protein 4 h post heat-shock (Fig. S3A-C; Movie 1). Direct induction of FRaeppli colour recombination by PhiC31 allows subsequent independent Cre-mediated gene inactivation, which was impossible with previous Cre-controlled multicolour tools.

Unbiased colour selection ensures differential labelling of neighbouring cells. We therefore assessed colour representation using different transgene activation strategies and determined no significant colour bias in fraeppli-nls or fraeppli-caax (Fig. 1D, Fig. S2E). Nonetheless, we noticed that the timely appearance of the FPs can vary, mostly starting with tagBFP-CAAX or tagBFP-NLS (Fig. S3D,H), followed by the other FPs and mKate2-CAAX or E2-Orange-NLS appearing last. Depending on the experimental context, the lag time between the first and last FP can take between 4 and 30 h. To exclude asynchronous PhiC31-mediated recombination as the underlying reason, we assessed recombination of the colour-cassette at the genomic level (Fig. S3E-G) by performing PCR amplification at 10 and 24 hpf in single hsp70l:gal4;fraeppli-caax embryos injected with phiC31 mRNA. This showed successful recombination of all four FP loci in individual embryos at 10 hpf (Fig. S3F-F″′), indicating synchronized recombination during blastula stages. Likewise, cre mRNA-injected embryos of the same genotype show efficient amplification of all recombined FP loci 24 h after heat-shock, indicating that FRaeppli-inbuilt PhiC31 can also drive synchronous recombination of all FP loci (Fig. S3G). Delayed FP appearance is likely due to a combination of individual FP maturation times, e.g. 13 min for tagBFP versus 78 min for E2-Orange (www.fpbase.org) and mRNA stability issues given the colour-cassette contains a single polyA, after tagBFP. The farther the respective FP is from the polyA, the longer it takes to build up detectable FP protein levels. Furthermore, the strength of a given Gal4 driver influences the temporal gap between recombination and colour detection. In the liver, it can take 11 or 30 h for all FRaeppli FPs to appear, after stimulating recombination using fabp10a:kalTA4 or prox1a:kalTA4, respectively. Conversely, using a strong ubiquitous Gal4-driver during gastrulation stages can drive the appearance of all colours between 9 and 13 hpf, making it suitable for the study of early developmental processes (Fig. S3H). The strength of the Gal4 driver is therefore an important consideration for experimental design. These results corroborate that FRaeppli lines stochastically label cells by simultaneous recombination events, which is pivotal for tissue morphology and lineage-tracing studies.

A key feature of multicolour cell labelling is the versatile control of recombination in space and time, to mark progenitor cells before the onset of differentiation for lineage tracing, or to label individual cells for high-resolution cell morphology and cell interaction studies. Different PhiC31, Cre and Gal4 expression methods were assessed for labelling density by temporally controlling FRaeppli recombination in the liver (Fig. 2G). Dense labelling can be achieved by phiC31 mRNA injection mediating recombination during blastula stages and prox1a:kalTA4-induced FP expression in liver progenitors (Fig. 2A). Labelling density is slightly reduced when cre mRNA injection removes the STOP cassette and prox1a:kalTA4 controls recombination by subsequent expression of the FRaeppli-inbuilt PhiC31 (Fig. 2B). Late induction of recombination by hsp70l:cre at 70 hpf results in lower labelling density (Fig. 2C). Similar activation strategies lead to progressively reduced labelling using the hepatocyte driver fabp10a:kalTA4, which is activated 1-1.5 days later, led to progressively reduced labelling (Fig. 2D-F), demonstrating that diverse tissue-specific drivers can control both FRaeppli recombination and expression at different developmental stages.

To validate FRaeppli for broad embryo labelling, we employed the ubiquitous driver hsp70l:gal4 and phiC31 or cre mRNA injection in fraeppli-caax embryos. Initiating recombination at different time-points between 26 and 72 hpf, labels cells at different densities (Fig. 2H-K, Fig. S2G-I), without apparent tissue bias.

Altogether, the combination of the Gal4/UAS and Cre/lox systems opens up the possibility of selecting a subset of cells within a population by using two different promoters. For example, this is achieved by intersecting a tissue-wide Gal4 driver with a cell type-specific Cre- or CreERT2 line, or vice versa. This dual activation strategy enables the analysis of new subpopulations arising from genome-wide transcriptome studies by visualizing cell shapes, behaviours and lineages at single-cell resolution. Thus, FRaeppli is a highly versatile cell-labelling system, as multiple approaches allow to control temporal and broad or tissue-specific multicolour expression, achieved by strategic combination of Gal4/UAS, Cre/lox and PhiC31 systems.

One challenge for imaging multispectral cell labelling is the frequently overlapping excitation and emission spectra of the used fluorophores (Cutrale et al., 2017), which is usually addressed by sequential imaging (Kanca et al., 2014; Fig. S4A). However, imaging time increases concomitantly with colour number (Fig. S4B). Alternatively, simultaneous excitation and acquisition of all channels by spectral imaging reduces imaging time (Dickinson et al., 2001). To test the feasibility, we first acquired all spectral data of larval prox1a:kalTA4;fraeppli-nls livers by lambda scanning (Fig. 3A), using 32 channels collecting signal from the 410-694 nm range, effectively reducing imaging time by 50-75%. Second, we employed the Hyper-Spectral Phasor software (HySP), which transforms the acquired spectral information by Fourier transformation into a 2D PhasorPlot, (Fig. 3A′) for subsequent label-unmixing (Fig. 3A″; Cutrale et al., 2017). This powerful method allows fast unmixing of large spectral images without calibration using reference samples for each spectral signature. Directly comparing the same prox1a:kalTA4;fraeppli-nls liver acquired sequentially with band pass filters (Fig. S4D) or spectrally (Fig. S4E) showed that hyperspectral imaging combined with HySP processing enables additional spectral separation of autofluorescence-derived noise (Cutrale et al., 2017), which is advantageous for metabolically active tissues like the liver. This method is suitable for live and fixed tissues of fraeppli larvae and adults (Fig. 3D, Fig. S4C,E).

Adding cellular and subcellular markers to FRaeppli-labelled populations would significantly extend the experimental possibilities for morphology and lineage studies. Hence, we tested the compatibility of FRaeppli with subpopulation-specific cell labels. First, we combined fraeppli-nls with the endothelial kdrl:GFP line, showing live how labelled hepatic cells arrange along the blood vessel network and the clear spectral separation of all labels, including GFP (Fig. 3C). Furthermore, inducing hsp70l:phiC31-mediated fraeppli-nls recombination in liver progenitors before lineage decision produced clones of diverse fates, containing tp1:GFP-positive biliary cells and tp1:GFP-negative hepatocytes, showcasing the advantage of FRaeppli for lineage-tracing studies involving GFP reporters (Fig. 3D).

Next, we visualized the liver architecture (Fig. 3B) by staining fraeppli-caax and fraeppli-nls adult liver sections and larval livers for actin and Anxa4 or TO-PRO, highlighting the hepatocyte canaliculi, blood vessel network and biliary network or nuclei, respectively (Fig. 3E-E′, Fig. S5A-B′). These applications demonstrate the compatibility of FRaeppli with at least two spectrally distinct labels, highlighting features on a tissue, cellular and subcellular scale. Notably, FRaeppli expression, depending on the Gal4-driver, is not only stably maintained into adulthood, but is also compatible with SeeDB2 clearing (Ke et al., 2016; Fig. S5A; Movie 2), permitting long-term clonal analysis.

Moreover, inducing colour recombination in embryos carrying multiple transgene copies would increase the number of distinct colour hues for differential cell labelling. Crossing two fraeppli-nls alleles lead to up to ten colours in the developing liver (Fig. 3F,F′), indicating successful recombination of all possible combinations, with comparable cell numbers expressing two distinct FPs (Fig. 3G). Altogether, we established the compatibility of FRaeppli with hyperspectral imaging and analysis strategies, with tissue-clearing approaches, and with common GFP and infrared tools that significantly expand experimental possibilities for high-resolution multicolour morphological and functional studies.

A powerful advantage of FRaeppli-CAAX is the ability to differentially label the morphologies of adjacent cells in order to untangle complex cellular interactions. Using this approach, we investigated zebrafish liver architecture and its similarity with that of mammals. Previous studies have reported that canaliculi, the apical membrane of hepatocytes, exhibit an intracellular topology by forming tube-like membrane invaginations within single hepatocytes (Figs 3B and 4A; Lorent et al., 2004; Sakaguchi et al., 2008). In mammals, the apical tubules are lined by membranes of adjacent cells, thus displaying intercellular shared topologies (Fig. 4A; Treyer and Musch, 2013). However, canalicular topologies in the zebrafish liver have not been assessed at single-cell resolution. We therefore stained fabp10a:kalTA4;fraeppli-caax larvae injected with cre mRNA using a multidrug resistance protein 1 (Mdr1) antibody, which marks mature canalicular membrane at 5 days post fertilisation (dpf). Combining commercial and custom-made image analysis tools (see Materials and Methods), we developed an analysis pipeline to quantify canalicular topologies (Movie 3). This revealed, in addition to the known intracellular canaliculi (Fig. 4A,B), a substantial proportion of intercellular canaliculi, shared between the membranes of neighbouring hepatocytes (Fig. 4A,C,D; Movie 3), demonstrating plasticity of canalicular topology in zebrafish. Altered canalicular topologies are also observed upon chronic liver injury in mammals (Clerbaux et al., 2019; Kamimoto et al., 2020), further suggesting that canaliculi topology is generally more plastic than previously known and that common mechanisms of canaliculi formation exist between zebrafish and mammals.

Live imaging is instrumental for elucidating morphodynamic cell behaviours driving embryonic development. Following several independent clones in the same organ enables investigation of cell rearrangement and interactions during tissue formation. Sparse labelling with a single membrane-localised FP has shown that directional progenitor migration determines liver position (Cayuso et al., 2016). Yet it is unknown to what extent cells rearrange as the organ differentiates. We monitored hepatocyte behaviours as the liver develops its functional organisation by time-lapse imaging in prox1a:kalTA4;fraeppli-caax embryos injected with phiC31 mRNA (Movie 4). Simultaneous tracking of individual cell clusters over time showed continuous changes of cluster morphologies and interactions, including protrusive activity evident at cluster edges, that indicate dynamic cell rearrangement during hepatic differentiation (Fig. 4E-F′). Tracing cluster dynamics showed diverse behaviours, including intermingling of adjacent clusters that led to fragmentation (Fig. 4E′; Movie 4). Cluster merging can occur in parallel, when detached cells contact other clusters and fuse (Fig. 4E″,F′). Combining real-time imaging with multicolour FRaeppli labelling revealed that differentiating hepatocytes are unexpectedly dynamic, which is essential information for understanding the establishment of tissue architecture and organ growth (Rulands and Simons, 2016). It further highlights the suitability of FRaeppli for collecting time series of dense tissues at subcellular resolution, including those of internal organs. The presented FRaeppli applications not only generate valuable insights into how complex tissues like the liver form but also allude to the extended experimental possibilities that the tools offer.

In summary, the FRaeppli toolbox offers a new cell labelling system that is: (1) targeting FPs to cell membranes and nuclei required for deciphering intricate cell behaviours and efficient lineage tracing; (2) spectrally compatible with many existing transgenic reporter lines and cell labels; and (3) amenable to sequential PhiC31-based cell labelling and conditional Cre/lox gene inactivation. Together with the growing number of conditional Cre-based gene inactivation tools (Carney and Mosimann, 2018) FRaeppli will enable connecting cell morphologies with gene function, which is the essential next step for understanding how tissues are built. The experimental versatility of FRaeppli thereby opens up powerful new applications for analyses of cellular behaviours in dense multicellular tissues.

Zebrafish (Danio rerio) embryos and adults were kept according to standard laboratory conditions (Westerfield, 2000). Research was performed in accordance with ethical guidelines approved by the Danish Animal Experiments Inspectorate (Dyreforsøgstilsynet). The following transgenic lines were used: tgBAC(prox1a:kalTA4)^uq3bh (Koltowska et al., 2015), tg(hsp70l:gal4)^fci1 (Cayuso et al., 2016), tg(kdrl:EGFP)^s843 (Jin et al., 2005), tg(hsp70l:cre)^zdf13 (Feng et al., 2007), tg(-2.8fabp10a:Cre,cryaa:Venus)^s955 (Ni et al., 2012) and tg(tp1-MmHbb:EGFP)^um14 (Parsons et al., 2009).

NLS- and CAAX-tagged fluorescent proteins from Drosophila Raeppli were combined into a final pBlueScript (pBS) vector following the original strategy described by Kanca et al. (2014). The order of FPs differs in the two FRaeppli cassettes: the pBS-FP-CAAX cassette contains mKate2-mTFP1-mOrange2-TagBFP, whereas pBS-FP-NLS cassette contains E2-Orange-mKate2-mTFP1-TagBFP. A fragment containing the SV40pA was amplified with primers 334 and 335 from the original pCS2+ vector (Rupp et al., 1994). The amplicon was cut with XbaI/SpeI and ligated into pBS-FP-CAAX cassette and pBS-FP-NLS cassette, both linearized with SpeI. A fragment containing the UAS promoter was amplified using primers 332 and 333 from UAS:ephrinb1 (Cayuso et al., 2016) and cloned into pJET1.2 (Thermo Fisher Scientific). A cassette containing a minimal attB site, an H2B-NeonGreen-STOP coding sequence flanked by two Lox2272 sites and several unique restriction sites for subsequent cloning steps (called ALNG) was synthesized as GeneBlock (IDT) and cloned using XbaI/NcoI into pJET-UAS cut with the same enzymes.

The H2B-NeonGreen contained in the Lox2272-flanked cassette, which was supposed to be expressed in unrecombined cells, and was subsequently replaced with standard GFP amplified with primers 371 and 372, and cloned in the ALNG cassette by AscI/NdeI, given the suspected toxicity of NeonGreen in zebrafish. Unfortunately, neither version was expressed, probably owing to the distance from the UAS.

The phiC31 integrase sequence followed by the BGH polyadenylation (pA) sequence was amplified with primers #51-340 from pCDNA-phiC31-BGHpA (Bischof et al., 2007), where a small sequence in the multiple cloning site was removed by XbaI/XhoI restriction and filled by Klenow DNA pol I (NEB). The full phiC31 integrase CDS was cloned after the ALGN cassette of pJET-UAS-ALNG using NsiI/BsrGI. The cmlc2:mTurquoise marker cassette was amplified from UAS-self-cmlc2:mTurquoise (derived from pGEMT-cmlc2:mTurquoise, a kind gift from J. Goedhart, University of Amsterdam, The Netherlands) with primers 267 and 353, and inserted in pJET. This plasmid was linearized with NheI and the cassette containing UAS-ALGN-phiC31 integrase was cloned into it after release with AvrII/SpeI.

The resulting pJET-UAS-ALGN-phiC31-cmlc2:mTurquoise was used in subsequent steps to derive the final FRaeppli 1.0, which unfortunately could not be employed for the generation of transgenic lines, as the plasmid had already recombined in bacterial cells due to bacterial activation of the UAS promoter, followed by the unanticipated expression of the integrase.

To generate an intact FRaeppli 2.0 plasmid, refractory to recombination in bacterial cells, an intron from SV40 (Reddy et al., 1979) was placed at position 546 of the phiC31 integrase-coding sequence, leading to a premature stop codon, which blocks the translation of a functional integrase only in bacterial cells. In eukaryotic cells, the intron is spliced out; therefore, a functional integrase can be generated upon UAS activation. A cassette containing the second lox2272 site, the first part of phiC31 integrase, the SV40 intron and linkers needed for subsequent cloning steps were synthesized as a GeneBlock (IDT) and subcloned into pJET-fraeppli2.0. Subsequently, the NdeI/BstEII fragment of pJET-fraeppli2.0 cloned into pJET-UAS-ALNG-phiC31-cmlc2:mTurquoise also cut with the same enzymes was introduced into the SV40 intron in the integrase CDS. The resulting plasmid was then cut with NotI and the fragment (UAS-ALGN-phiC31-intron-cmlc2:mTurquoise) cloned into pBS-FP-CAAX cassette and pBS-FP-NLS cassette. The miniTol2 sequences (Urasaki et al., 2006), taken from UAS-self- cmlc2:mTurquoise were inserted in the resulting pBS vectors by substitution of the XhoI-bound fragment, generating the final FRaeppli2.0-NLS and FRaeppli2.0-CAAX, which were used for transgenesis. All primers used in this study are listed in Table S1.

The optimized Gal4 activator kalTA4-coding sequence (Distel et al., 2009) was inserted after the fabp10a promoter (Her et al., 2003) into the I-SceI-fabp10a:cre-cryaa:Venus plasmid (Feng et al., 2007), where cre was removed with ClaI/NotI. An unnecessary polyA sequence was then removed by cutting with XbaI. The final plasmid was sequenced with primers 327 and 331.

The EcoRI/NheI phiC31-integrase-BGHpA fragment was sub-cloned from the UAS:phiC31-cryaa:Citrine plasmid (this work) into the I-SceI-containing backbone of the fabp10a:kalTA4-cryaa:Citrine plasmid, previously linearized with EcoRI/XbaI. The hsp70l promoter was amplified from hsp70l:mCherry-2A-wnt2 (Poulain and Ober, 2011) using primers 613 and 614, cut with ClaI/NsiI, and subsequently cloned into I-SceI-phiC31 integrase cut with the same enzymes. Primers 86, 382, 347 were used for sequencing the I-SceI-hsp70l:phiC31 integrase plasmid. The he1a:palm-Citrine marker cassette was generated by replacing the promoter sequence in the UASself-cryaa:Citrine plasmid (Cayuso et al., 2016) with the hatching enzyme 1a (he1a) promoter sequence (a kind gift from the R. Koester laboratory, Technical University of Braunschweig, Germany). The he1a:palm-Citrine-SV40polyA cassette was amplified with primers 617 and 618, and cloned into I-SceI-phiC31 cut with Acc65I/SwaI, using blunt-end cloning.

The previously described I-SceI-hsp70l:phiC31-he1a:palm-Citrine was modified by replacing the stop codon after the phiC31 sequence with a linker sequence (5′-acttggttccag-3′). The resulting plasmid was called I-SceI-hsp70l:phiC31-linker-he1a:palm-Citrine. The sfGFP sequence (a kind gift from H. Knaut's laboratory, Skirball Institute of Biomolecular Medicine, NYU, USA) was amplified with primers 628 and 629, and cloned into I-SceI-hsp70l:phiC31-linker-he1a:palm-Citrine with PspOMI. Finally, the NLS sequence was added in frame at the C-terminal of the CDS by an intermediate subcloning step into pBSII-BFP-NLS. The final plasmid I-SceI-hsp70l:phiC31-sfGFP-NLS-he1p:palm-Citrine was sequenced with primers 382, 355, 581 and 615.

fraeppli2.0-nls and fraeppli2.0-caax plasmids were injected into one-cell stage embryos together with tol2 mRNA (30 pg mRNA and 20 pg DNA/embryo) as previously described (Kawakami et al., 2000). Individual F1 fraeppli fish were tested for recombination efficiency, using cre or phiC31 integrase mRNA injection. The three or four best lines are maintained as the stable lines fraeppli-nls^cph1-3,cph9 and fraeppli-caax^cph4-6. The I-SceI-hsp70l:phiC31-he1a:palm-Citrine and I-SceI-fabp10a:kalTA4-cryaa:Venus plasmids were injected in one cell-stage embryos together with I-SceI enzyme (NEB) as previously described (Soroldoni et al., 2009). Stable transgenic insertions are maintained as hsp70l:phiC31^cph7 and fabp10a:kalTA4^cph8.

To either prime or activate FRaeppli recombination, cre mRNA (20-25 pg) or phiC31 integrase mRNA (30-40 pg) was injected into one-cell stage embryos, as described previously (Felker and Mosimann, 2016; Mosimann et al., 2013). For temporally controlled activation, fraeppli and desired Gal4/KalTA4 driver lines were crossed to either tg(hsp70l:cre)^zdf13 or tg(hsp70l:phiC31-he1a:palm-Citrine)^cph7. Embryos were subjected to a 30-60 min heat-shock at 39°C and selected for recombination at desired stages. The recombination in Cre-primed embryos follows the sequential onset of transgenic Gal4/KalTA4 and subsequent FRaeppli-inbuilt PhiC31 integrase expression.

mRNA expression of hsp70l:phiC31 was analysed by in situ Hybridization Chain Reaction (HCR) RNA FISH version 3.0 (Choi et al., 2018). HCR probes against PhiC31 were purchased from Molecular Instruments. Fixation, detection and amplification were performed according to the HCR v3.0 protocol for whole-mount zebrafish larvae (Choi et al., 2016), apart from the following modifications: 1 pmol of the probe set were used for detection and only 15 pmol hairpin solutions were used for signal amplification. Samples were counterstained with 1:500 DAPI overnight at 4°C. Embryos were mounted in Vectashield and imaged with a Zeiss 880 confocal microscope using a 40× oil objective. Maximum intensity projections of z-stacks were generated in Imaris and representative images are shown in the figures.

Embryos and larvae intended for imaging experiments were raised in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂ and 0.33 mM MgSO₄) supplemented with 0.2 mM 1-phenyl 2-thiourea (PTU, Sigma-Aldrich) to inhibit pigmentation. For live imaging, embryos and larvae were embedded in 0.8% low melting agarose and anesthetized with Tricaine (164 mg/l; MS-222, Sigma-Aldrich) dissolved in E3/PTU. For the time series in Fig. 4, four embryos were selected for three or four colour clones and mounted in a single plate for concurrent imaging every 12-15 min for 8-12 h using sequential acquisition parameters (Fig. S4A). Stained larvae were embedded in VectaShield (Vector labs) for whole-mount imaging. Whole liver lobes or vibratome sections of adult livers were cleared and imaged in SeeDB2S or embedded and imaged in VectaShield. Imaging was performed on Zeiss LSM 780, 880 and 980 confocal microscopes equipped with PMT detectors for sequential imaging and a spectral detector (GaAsP-PMT, 32 channels, 410-694 nm range, 8.9 nm bandwidth) for spectral acquisition. Five-colour sequential imaging was performed with a Leica Stellaris confocal microscope equipped with a tunable white-light laser and four PMT/HyD detectors. In spectral acquisitions, all four FRaeppli FPs were simultaneously excited with 405 nm, 456 nm and 561 nm lasers. In sequential acquisition experiments, 515 nm and 594 nm excitation lasers were needed to discriminate between mOrange2/E2-Orange and mKate2. Additional FPs or fluorophores were excited with a 488 nm and a 633 nm laser. Details of imaging parameters for different microscope set-ups are described in Fig. S4A.

Embryos were fixed with 4% PFA at 4°C overnight and stained as previously described (Thestrup et al., 2019). Adult fish were fixed overnight in 4% PFA at 4°C, using a rotator. Livers of adult fish were dissected, embedded in 4% agarose/water, cut into 130 µm sections using a vibratome (Leica) and stained using the desired antibodies. Alternatively, dissected liver lobes were stained as whole-mount and subsequently cleared following the SeeDB2S protocol (Ke et al., 2016). The following primary and secondary antibodies were used: α-2F11 (mouse monoclonal, 1:10,000, a gift from Julian Lewis, CRUK London Research Institute, UK), anti-MDR1 (rabbit polyclonal antibody, 1:2000, Santa Cruz Biotechnology, sc-8313), goat anti-mouse 488 (1:500, Jackson Immunoresearch) and goat anti-rabbit 647 (1:500, Jackson Immunoresearch). Fixed embryos and adult liver lobes were incubated in 0.1% PBS-Triton X, Alexa Fluor 488 phalloidin (1:1000, Molecular Probes, A12379) and TO-PRO-3 stain (1:50,000, Thermo Fisher Scientific) for 1 h at room temperature or overnight at 4°C. Fresh samples were used for staining and imaging, as FRaeppli endogenous fluorescence is stable in fixed tissue for at least 1-2 weeks.

Images were processed with Bitplane Imaris and Zeiss ZEN software. Maximum intensity projections of z-plane stacks are shown, unless indicated differently in the figure legends. For FRaeppli CAAX images, the gamma value of each FRaeppli channel was increased to 1.2-1.5 to better visualize dark regions of the cell membranes. The gamma value was not altered for analysis purposes.

Lambda scans were spectrally unmixed using Hyper-Spectral Phasor software (HySP) (version 0.9.10) (Cutrale et al., 2017) (https://bioimaging.usc.edu/software.html). 16-bit spectral images stored in lsm5 format were opened in HySP with harmonic number 1. Images were de-noised in the phasor window to reduce spectral scattering by 10-fold filtering. The minimum intensity values were manually adjusted to subtract dark background, and the phasor was displayed in logarithmic scale. Different clusters in the phasor plot corresponding to the spectra of the fluorescent proteins were manually gated using the polygon ROI selector (see Fig. 3A′) following the Detailed Guide to HySP (https://bioimaging.usc.edu/software.html#instructions). The overlap of the ROI in the image window was used to verify the ROI positions.

Of note, for fraeppli-caax we recommend comparing the sequential and spectral acquisition for any given experiment, as faithful colour assignment at cell interfaces of touching cells using the hyperspectral phasor plot is more complex. This also depends on the labelling density obtained. The most reliable, though slower, option is to use sequential acquisition settings. This does not apply to fraeppli-nls acquisitions as nuclear signals never spatially intersect.

To quantify colour choice in fraeppli-caax, embryos of the genotype fraeppli-caax^cph4; prox1a:kalTA4^uq3bh; hsp70l:cre^zdf13 were either injected with cre mRNA at the one-cell stage or hsp70l:cre expression was induced at 33 hpf. Stacks of the first 30 µm of liver tissue were acquired by live imaging with spectral acquisition at 5 dpf, followed by manual annotation of the recombined cells in the Zeiss ZEN software. Embryos of genotype fraeppli-nls^cph1; prox1a:KalTA4^uq3bh; hsp:phiC31^cph7 were used to quantify colour choice in fraeppli-nls. Recombination was induced by heat shock at 36 hpf as described above. The left liver lobe of 5 dpf larvae was acquired live by spectral imaging. Sequential HySP unmixing was followed by segmentation in Imaris and determination of the number of nuclei using the Spots function.

DNA was extracted from single embryos of the indicated genotypes (Fig. S3F,G). PCRs amplifying recombined transgene species or un-recombined controls were performed (strategy in Fig. S3E). The following primers were used: 666 and 667 (a; detecting a common region in the transgene), 170-662 (b; detecting TagBFP in recombined transgene), 170-346 (c; detecting mTFP1 in recombined transgene), 170-663 (d; detecting mOrange2 in recombined transgene) and 170-664 (e; detecting mKate2 in recombined transgene). The same PCR reactions were performed on 10 ng of FRaeppli 1.0 and 2.0 plasmid as control templates, as FRaeppli 1.0 recombines randomly during E. coli growth and thereby was used as positive control of recombination (see ‘Constructs’ section). Pictures of agarose gels were taken with different exposure times depending on the PCR, to avoid overexposing bands. Band quantification was performed with GelAnalyzer 19.1 (www.gelanalyzer.com; by Istvan Lazar, Jr and Istvan Lazar, Sr), using a morphometric algorithm to automatically define the background. Each band for the recombined versions (b,c) was first normalized to the intensity of the common band (a) from the same sample to correct for differences in DNA content. A second normalization factor for correcting the differences in picture exposure time was then calculated by measuring the 1000 bp band of the ladder (GeneRuler 100 bp plus, Thermo Fisher Scientific) in each picture and applied to each band. Finally, the PCR efficiency, for detecting each recombined colour, was calculated by comparing the normalized intensities of the bands (b-e) of the positive control (FRaeppli 1.0 construct). The PCR efficiency normalization factor for each colour was then applied to every experimental band of the corresponding colour.

For comparison of recombination efficiency in independent fraeppli-caax and fraeppli-nls lines (Fig. S2D,E), stacks of the first 100 µm of liver tissue were imaged live with spectral acquisition at 5 dpf, followed by manual cluster annotation. In this context, adjacent labelled cells were considered as clusters.

In 5 dpf stained larvae, FRaeppli FPs were imaged with spectral acquisition and the MDR-stained canaliculi (Alexa 488) in a separate acquisition with a PMT detector, as only a weak signal was detected in the spectral image. The spectral image was processed with the hyperspectral phasor software (see above) and seven channels were identified: four FRaeppli colours (B, C, Y and R) and the intersecting channels between Tag-BFP and mTFP1 (B/C), between mTFP1 and MDR (C/MDR), and between mOrange2 and MDR (Y/MDR). The intersecting channels were used in the analysis to correctly assign pixels that were positive for two adjacent colours in the spectrum, indicating spatial colocalization. The optimized unmixing gates were applied to all images and minimally adjusted to include all significant pixels in each image.

Segmentation was performed in Imaris. First, the seven unmixed channels (exported from the Phasor software) and the MDR channel of each image were combined in a single file in Imaris, following the application of a 5×5×5 median filter to the MDR channel. Subsequently, all the eight channels (cell membranes and canaliculi) were segmented in 3D with the ‘Surface’ function. An additional surface was created by merging the surfaces of the seven FRaeppli channels (cell membranes). The merged surface was used to select a subset of canaliculi overlapping FRaeppli-labelled cells, by intersecting the two surfaces (FRaeppli membranes and canaliculi). The individual surfaces of the segmented membranes were subsequently used to create masked channels, whereby each colour-positive pixel was set to the same maximal value by thresholding (each included pixel=1; each excluded pixel=0). By doing this, intensity quantifications for each masked channel would also translate in relative volume estimation. For each FRaeppli-positive segmented canaliculus, several parameters were exported for further analysis: the Intensity Mean of each masked channel, indicating the relative volume of each canaliculus that was positive for that colour, the Shortest distance to Surface, indicating all the reciprocal distances between the canaliculi, and the Sphericity.

All parameters were imported into a custom-made script coded in R (RStudio Team, 2019; R Core Team, 2020) to extract the topological information for all analysed canaliculi. First, each canaliculus was associated with one or multiple colours, using a 5% threshold of pixel occupancy for each channel, and the intersecting channels as quality controls. Second, the sphericity (<0.75) and distance (<0.5 μm) between each canaliculus were used to classify canaliculi with an ‘acinar’ configuration, namely those shared between neighbouring hepatocytes that also contain an intracellular region (Fig. 4A). Acinar canaliculi with >30% of pixels occupied by one or multiple colours were considered in the final classification.

A threshold of total colour occupancy of >70% was applied for all the remaining canaliculi, which were subsequently divided into ‘intracellular’ or ‘shared’ according to the presence of one or multiple colours, respectively. An additional distance filter was applied to exclude intracellular canaliculi of neighbouring cells of the same colour in a 14 µm radius, which is the typical size of a hepatocyte, as shared or intracellular configurations cannot be discriminated in those cases. Thereby, only canaliculi included in isolated cells or cells surrounded by neighbours of different colours were included in the final classification.

Statistical analysis was performed with GraphPad Prism (version 9.3.1). Statistical significance was calculated using the two-tailed Student's t-test; n.s. P≥0.05. All data are presented as mean±s.d. Unless indicated otherwise, n refers to sample size (e.g. individual embryos) and N refers to biological replicates. For adult tissue stainings, at least three sections were analysed for each biological replicate.

We apologise to colleagues whose primary work was not included due to space constraints. We thank E. Caussinus and M. Affolter for sharing the Raeppli constructs; X. Guo for key cloning advice; R. Heyne for cloning of fabp10a:kalTA4; E. Ambrosio Galindo for help with processing of adult tissue; O. Andersson, W. Herzog, K. Koltowska, B. Hogan, C. Mosimann, D. Stainier and D. Wilkinson for additional constructs and zebrafish lines; and M. Walther and K. Kristiansen in Zürich and the department of experimental medicine (AEM) in Copenhagen for expert fish care. We gratefully acknowledge the DanStem Imaging Platform (University of Copenhagen), the Translational Imaging Center (University of Southern California), the ScopeM (ETH-Zurich), and the Center for Microscopy and Image Analysis (University of Zurich) for their support and assistance in this work.