Fate mapping melanoma persister cells through regression and into recurrent disease in adult zebrafish

Melanoma, a deadly cancer of pigment producing melanocytes, ranks amongst the highest for genetic and transcriptional heterogeneity (Rambow et al., 2019; Travnickova and Patton, 2021). Therapy resistance remains a major challenge for patients with melanoma, with partial or short-term responses characteristic of targeted therapy, eventually resulting in tumour relapse (Patton et al., 2021; Shen et al., 2020b). This resistance can be intrinsic, in which pre-existing primary tumour cell states directly confer resistance, or acquired, whereby melanoma cells develop resistance upon drug exposure either by acquiring new genetic mutations or adapting their transcriptional state (Marin-Bejar et al., 2021; Marine et al., 2020; Shen et al., 2020b). These resistance mechanisms likely occur concurrently, and contribute to the high heterogeneity within melanoma and the persister cell states at the minimal residual disease (MRD) site.

Recently, using single-cell RNA sequencing, we and others have demonstrated high transcriptional heterogeneity and distinct cell states in the primary tumour and MRD, including states with low-to-no expression of pigmentation lineage markers such as melanocyte inducing transcription factor (MITF)-independent populations (Baron et al., 2020; Ennen et al., 2015; Gerber et al., 2017; Rambow et al., 2018; Tirosh et al., 2016; Travnickova et al., 2019). These studies have identified multiple melanoma cell states that have been proposed to drive tumour recurrence, while others are of unknown contribution to disease progression (Travnickova and Patton, 2021). While revealing new concepts about melanoma transcriptional cell states, these studies reflect only a single or few time points, and are thereby limited in what can be understood about the behaviour of individual cell populations over time.

This limitation is particularly critical in the context of cell plasticity in cancer models. Advances in imaging technologies coupled with lineage-tracing methods in zebrafish models now enable the fate mapping of cell populations over time (Mosimann et al., 2011; Pan et al., 2013). The power of an inducible lineage-tracing system is illustrated by advances in understanding of organ development and tissue homeostasis in zebrafish (Carney and Mosimann, 2018; Thunemann et al., 2017). For example, lineage-tracing experiments using the general neural crest marker sox10 or the recently established melanocyte stem cell marker tfap2b provided proof of the existence of multipotent melanocyte precursors during early embryonic development and their contribution to adult zebrafish pigment cell patterning (Brombin et al., 2022; Singh et al., 2016). Similarly, a conditional cre/loxP recombination system demonstrated the hierarchy of neuroepithelial progenitors and the functional heterogeneity of neural stem cells in the vertebrate adult brain using neural lineage-specific markers (Galant et al., 2016; Than-Trong et al., 2020). Furthermore, multispectral lineage tracing revealed the mechanism behind myotome generation (Nguyen et al., 2017). The regenerative capacities of zebrafish combined with an inducible lineage-tracing system have been instrumental in understanding the cell origin and lineage restrictions in regenerated organs such as lesioned heart, fins or spinal cord (Briona et al., 2015; Jopling et al., 2010; Tornini et al., 2017). Recently, multicolour tracing has been combined with mosaic mutagenesis using CRISPR-Cas9 in a novel technique called tissue editing with inducible stem cell tagging via recombination (TWISTR) to show that the fitness of mutant clones is controlled by resistance to inflammation (Avagyan et al., 2021). Despite these advances, the application of tamoxifen to control the temporal and spatial activation of Cre (Cre^ERt2) in adult zebrafish cancer models has been hampered by technical challenges, such as lack of established protocols and tamoxifen toxicity.

Here, we adapt and optimise the inducible ubi:Switch lineage-tracing system (Mosimann et al., 2011) for use in an adult zebrafish cancer model to fate map cells through melanoma growth, regression and recurrence. For the first time, we directly capture melanoma cell switching from MITF-independent persister cells to Mitfa⁺ cells to contribute to recurrent disease in vivo. Our work supports the concept that targeting persister cells will be critical to delay or prevent recurrent disease.

We have previously developed a conditional MITF-dependent zebrafish melanoma model [Tg(mitfa:BRAF^V600E);mitfa^vc7;tp53^M214K], in which melanoma regresses and recurs concurrently with the changes in MITF activity controlled by temperature (Travnickova et al., 2019). In this model, MITF activity is controlled by a temperature-sensitive splicing mutation in the mitfa gene (mitfa^vc7; zebrafish orthologue mitfa is expressed in the body melanocytes) (Johnson et al., 2011; Zeng et al., 2015). At the lower permissive temperature, MITF activity is on and promotes tumourigenesis, while at the higher temperature mitfa RNA is still expressed but is not spliced correctly, and thereby MITF protein levels and activity are abolished. Turning off MITF activity results in tumour regression with remaining persister cells at the MRD site (Travnickova et al., 2019).

To trace melanoma cells through disease states, we first set out to establish an inducible cre/loxP system on the MITF-dependent melanoma background. To this end, we expressed cre^ERt2 from the mitfa promoter to generate a transgenic Tg(mitfa:cre^ERt2) line and crossed this with a ubi:Switch reporter line (Mosimann et al., 2011). This experimental design would enable us to induce a green-to-red (GFP-to-mCherry) permanent fluorophore switch in mitfa:cre^ERt2-expressing melanoma cell populations at a defined time point in melanoma disease progression (Fig. 1A,B).

To validate the specificity of the mitfa:cre^ERt2 construct for mitfa-expressing cells, we first performed the fluorophore switch during early embryonic development using 4-hydroxytamoxifen (4-OHT) treatment (Fig. 1C). During embryogenesis and in regeneration, Mitfa is required for the generation of melanocytes, and mitfa expression marks melanoblasts and progenitors of additional pigment cells, including the yellow xanthophores (Brombin et al., 2022; Parichy et al., 2000). Upon 4-OHT treatment (three consecutive daily treatments of 20 µM), we could detect mCherry-expressing cells at 4.5 days post fertilisation (dpf) along the lateral stipe and on top of the head, in both cases colocalised with pigment (Fig. 1D,E, pink arrows). Some of the mCherry⁺ cells (Fig. 1D,E, yellow arrows) that did not colocalise with pigmented cells likely correspond to pigment cell progenitors (Brombin et al., 2022; Parichy et al., 2000). This indicates that mitfa:cre^ERt2 construct specifically labels mitfa-expressing cells.

We found that 4 µM tamoxifen treatment of adult zebrafish by immersion for three consecutive nights (11 h treatment, 13 h recovery) was successful for cre^ERt2/loxP recombination without toxicity (Fig. 1B, Fig. 2A). We chose tamoxifen over 4-OHT for its increased solubility in dimethylsulfoxide (DMSO), which is well tolerated by adult zebrafish. The treatment was completed overnight to align with the natural light–dark cycle of the fish and to prevent phototoxicity of tamoxifen (Wang et al., 2009). By 3 days post tamoxifen treatment course, we could detect mCherry in the primary tumour, but not in DMSO-treated controls (Fig. 2B; Fig. S1A). Using confocal microscopy, we validated the presence of individual clusters of mCherry⁺ melanoma cells in tamoxifen-treated tumours only (Fig. 2C,D). Immunostaining of melanoma sections with an mCherry antibody showed that mCherry⁺ cells were present both at the surface of the tumour and in the invading melanoma cells along muscle fibres, thus confirming the efficiency of this protocol to fate map cells throughout the body of the melanoma tumour (Fig. 2E,F).

Having established the system in a single colour switch reporter line, we wanted to test if our method was applicable to the multicolour system of fate mapping. To this end, we crossed the Tg(mitfa:cre^ERt2) melanoma-prone fish with the ubi:zebrabow transgenic line (Pan et al., 2013) to evaluate permanent colour changes in cre^ERt2-expressing cells from default red (RFP) to a stochastic combinatorial expression of three fluorophores upon tamoxifen treatment: cyan (CFP), yellow (YFP) and red (RFP) (Fig. 3A). Indeed, similar to ubi:Switch line, we could detect de novo fluorescent signal by 3 days post tamoxifen treatment compared to DMSO controls, mainly in the YFP channel, and at 34 days post treatment, we could detect all three fluorophores within the primary melanoma (Fig. 3B; Fig. S1B). Using confocal microscopy, we validated the presence of multicolour labelling (CFP, YFP and RFP), which allows distinction of individual cells across the labelled tissue (Fig. 3C; Fig. S1C).

Melanoma tissue often varies in the pigmentation level between individual tumours, and while non-pigmented tumours permit direct detection by fluorescence, intense pigmentation can obscure the fluorescent signal. To overcome this challenge, we performed vibratome sectioning of paraformaldehyde (PFA)-fixed melanoma tissue (Fig. 3D) followed by confocal microscopy. These imaging data show that combining thick tissue sectioning with confocal microscopy permits detection of CFP and YFP channels even in highly pigmented tissues (Fig. 3E; Fig. S1D). While we did not evaluate clonal evolution in melanoma progression (as it would require more rigorous analysis of colour switch stochasticity), this experiment demonstrated the potential for our tamoxifen treatment protocol for application in clonal analysis studies in zebrafish cancer models.

Next, we used our conditional MITF-dependent BRAF^V600E p53^M214Kubi:Switch model to trace melanoma cells from the primary tumour as it regresses and thereby determine if the cells detected at the MRD site originate from the primary tumour (Fig. 4A). Tamoxifen treatment of early-stage melanoma resulted in a GFP-to-mCherry switch in tumour lesions that was absent in DMSO-treated animals (Fig. 4B,C). Following a period of tumour growth, we transferred fish to a higher temperature to turn off MITF activity and cause tumour regression until no melanoma was detectable morphologically (5-10 weeks; see Materials and Methods). Strikingly, at the MRD site, we could detect mCherry⁺ persister cells in tamoxifen-treated fish (Fig. 4B,C). Confocal microscopy and quantification confirmed that tamoxifen-treated fish showed significantly greater mCherry signal at the MRD site compared to DMSO-treated fish (Fig. 4D,E). Whilst most DMSO-treated fish did not show any mCherry signal, we could detect it in a small proportion of the control fish (Fig. 4E). This may be due to cre expression (leakage) in those tissues where mitfa expression is very high [for example in some nodular tumours, as we have described previously (Travnickova et al., 2019)]. Immunostaining of regressed melanoma tissue sections confirmed the presence of mCherry⁺ persister cells at the MRD site (Fig. 4F). These data indicate that primary melanoma cells directly give rise to persister cells at the MRD site.

Next, we asked if the tamoxifen-induced GFP-to-mCherry switch could also be applied to melanoma persister cells in the MRD site (Fig. 5A). We transferred our melanoma fish to a higher water temperature to prevent the correct splicing of mitfa (and to thereby turn off Mitfa protein activity) to cause melanoma regression. Because the mitfa^vc7 mutation is an RNA-splicing mutation, the expression of mitfa or reporters under the control of mitfa (e.g. mitfa:GFP; mitfa:cre^ERt2) are not affected at the restrictive temperature (Fig. 5B). Once the melanoma had fully regressed, we treated the fish with tamoxifen. We detected mCherry⁺ cells 6 days after the start of tamoxifen treatment at the regression site (Fig. 5C,D). Using confocal microscopy, we were able to validate the presence of mCherry⁺ cells in the tamoxifen-treated group (Fig. 5E) that were morphologically similar to our previous observations using a Tg(mitfa:GFP) transgenic line (Travnickova et al., 2019).

We have previously demonstrated that melanoma persister cells in the MRD site do not express Mitfa protein (called MITF independent), but maintain a neural crest identity and express Sox10 (Travnickova et al., 2019). To evaluate if the ‘switched’ mCherry⁺ cells retain the same properties, we immunostained the primary and regressed tamoxifen-treated melanoma tissue sections with mCherry and Mitfa antibodies (Fig. 5F,G). No Mitfa expression was detected in ‘switched’ mCherry⁺ melanoma cells in the regressed tumour. In contrast, the primary tumour showed abundant Mitfa expression in mCherry ‘switched’ cells (Fig. 5F,G). Consecutive sections of the MRD site were stained with mCherry and Sox10 antibodies and confirmed that the mCherry⁺ cells were Sox10⁺ persister melanoma cells (Fig. S2). These data show that ‘switched’ mCherry⁺ persister cells at the MRD site are MITF independent.

Melanoma persister cells have been proposed to contribute to disease recurrence and drug resistance (Marin-Bejar et al., 2021; Rambow et al., 2018; Shen et al., 2020a; Travnickova et al., 2019; Vendramin et al., 2021), but this has not been demonstrated in an animal model using lineage tracing. Having validated the successful GFP-to-mCherry switch at the MRD site, we next sought to follow the persister cells during recurrence. Once again, we performed tamoxifen treatment and detected GFP-to-mCherry switched persister cells at the MRD site. We then restored MITF activity by lowering the water temperature and followed melanoma recurrence. Over 50 days, the mCherry⁺ cells continuously increased concomitantly with melanoma recurrence (Fig. 6A,B). Fluorescence area quantification of individual fish MRD sites show that mCherry⁺ cells drive tumour growth over time (Fig. 6C). This demonstrates that persister cells directly contribute to the growth of recurrent disease.

Confocal microscopy enabled us to visualise the tumours at cellular resolution, showing fields of mCherry⁺ melanoma cells in recurrent disease (Fig. 6D,E). Quantification of the mCherry⁺ fluorescence indicated that most recurrent disease sites expressed mCherry (Fig. 6D; 6/7 recurrence sites with an average intensity of 2×10⁵ arbitrary units). For one sample, we found that the fluorescence was hindered by strong pigmentation. Given this, we validated the presence of mCherry⁺ cells in this and other recurrent tumours using flow cytometry and could clearly detect mCherry even in highly pigmented samples (Fig. 6F).

We hypothesised that persister cells directly lead to recurrent disease by transitioning from an MITF-independent to a Mitfa⁺ state. As previously described and shown here (Fig. 5F,G), persister cells do not express Mitfa protein, but maintain their expression of Sox10 (Travnickova et al., 2019). We applied multiplex immunohistochemistry (MIHC) on sections of the recurred tumours to determine the protein expression levels of Mitfa and other melanoma markers within recurrent disease (Fig. 7A-D). MIHC allows sequential staining and stripping of several antibodies on the same slide (Pirici et al., 2009). We used an antibody against the mCherry protein to label the ‘switched’ cells in tamoxifen-treated fish compared to DMSO control, together with antibodies against the melanoma markers Sox10 and Mitfa (Fig. 7B-D). As anticipated, recurred tumours from DMSO controls lacked mCherry signal in Sox10⁺ and Mitfa⁺ tumour cells (Fig. 7B,B′). By overlaying the images of mCherry staining with Sox10 and Mitfa staining, we could demonstrate that the mCherry⁺ cells within the tamoxifen-treated recurred tumour expressed both Sox10 and Mitfa (Fig. 7D′; Fig. S3). These results indicate that melanoma persister cells undergo a cell state switch from MITF-independent to Mitfa-expressing cells, regaining features of the primary tumour and providing direct evidence of tumour cell transcriptional plasticity in vivo.

Transcriptional heterogeneity underlies the high levels of tumour cell plasticity in melanoma. Here, we apply fate mapping and MIHC to tumour cell populations through melanoma disease states to show that (1) persister cells originate from the primary tumour; (2) persister cells directly contribute to melanoma recurrence; and (3) persister cells exhibit plasticity from a MITF-independent state to Mitfa-expressing melanoma cells in recurrent disease. Thus, persister cells will be a critical drug target for delaying (or even preventing) melanoma recurrent disease.

Our system offers the opportunity to understand how individual cell states, and indeed even individual cells, contribute to drug resistance and tumour progression in zebrafish cancer models. Lessons from mouse models indicate that the efficiency of recombination is largely dependent on the expression level and pattern of the chosen promoter, and thus the tamoxifen protocol often needs to be adapted for each cre^ERt2 line (Ellisor and Zervas, 2010; Jahn et al., 2018). In particular, the Bally-Cuif group recently established a 4-OHT protocol for mosaic recombination using short-term drug exposure and long-term multiple-day immersion for maximal recombination, similar to our approach (Than-Trong et al., 2020). Hence, as inducible lineage-tracing methods become more widely applied in zebrafish adults, the zebrafish community will benefit from ensuring that tamoxifen protocols for each promoter are easily accessible.

Here, we showed the fate mapping of the mitfa:cre^ERt2-expressing melanoma cell population through disease stages, which represents a major portion of melanoma cells in our model. MITF is one of the universal diagnostic markers of cutaneous melanoma based on its expression in most melanoma cells (Compton et al., 2015). Previously, we and others have demonstrated the level and importance of transcriptional heterogeneity not only in the primary tumour but also in the persister cells at the MRD site (Marin-Bejar et al., 2021; Rambow et al., 2018; Shen et al., 2020b; Travnickova et al., 2019; Wouters et al., 2020). The persister cell states in our zebrafish models share conserved mechanisms with human melanoma cells depleted for MITF or following BRAF inhibitor drug treatment (Dilshat et al., 2021; Rambow et al., 2018; Travnickova et al., 2019). The lineage-tracing system we present here has the potential to enable fate mapping of different transcriptional cell states at the MRD site and evaluate their contribution to disease progression and response to drug treatment (Lu and Patton, 2022). This can be achieved using cell state-specific markers to recombine a subpopulation of the tumour at the desired stage.

In conclusion, we have successfully applied and validated an inducible cre^ERt2/loxP lineage-tracing system in an adult zebrafish melanoma model to follow cell states through disease stages. Combining fate mapping with live imaging, immunohistochemistry (IHC) and MIHC, we provide direct evidence for the contribution of cells at the MRD site to tumour recurrence, whilst simultaneously demonstrating their MITF-independent to Mitfa⁺ cell state shift. This system has the potential for widespread usage in the study of cancer cell plasticity over time and in the evolution of therapy resistance.

Zebrafish were maintained in accordance with UK Home Office regulations, UK Animals (Scientific Procedures) Act 1986, amended in 2013, and European Directive 2010/63/EU under project license 70/8000 and P8F7F7E52. All experiments were approved by the Home Office and Animal Welfare and Ethical Review Body (AWERB; University of Edinburgh Ethics Committee).

Fish stocks used were as follows: mitfa^vc7 (Johnson et al., 2011; Zeng et al., 2015), Tg(mitfa-BRAF^V600E), tp53^M214K(lf) (Patton et al., 2005), Tg(ubb:loxP-EGFP-loxP-mCherry) (referred to as ubi:Switch) (Mosimann et al., 2011), Tg(ubb:lox2272-loxP-Tomato-lox2272-Cerulean-loxP-YFP) [referred to as Tg(ubb:lox2272-loxP-RFP-lox2270-CFP-loxP-YFP) or ubi:zebrabow] (Pan et al., 2013), Tg(mitfa:cre^ERt2) (this study). Combined transgenic and mutant lines were generated by crossing. Adult fish were maintained either at 28.5°C, 25°C or 32°C under 14:10 h light:dark cycles.

Cre^ERt2 was amplified by PCR using pCGA_cre^ERt2 (Addgene plasmid #14797) as template. A nuclear localisation signal was added to its N-terminal. The gateway primer sequences were cre^ERt2gateF 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTGCCACCATGCCCAAGAAGAAGAGGAAGGTGTCCAATTTACTGACCGTACACC-3′ and cre^ERt2gateR: 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTTCAAGCTGTGGCAGGGAAAC-3′. The PCR product was cloned into pDONOR221, resulting in pME-cre^ERt2, a middle entry vector. Cre^ERt2 was then cloned together with the 2.1 kb zebrafish mitfa promoter into pDestTol2CG2 destination vector by the Tol2kit gateway cloning method (Kwan et al., 2007), resulting in the pEXPmitfa-cre^ERt2 expression vector. For selection purposes, the construct contains an additional GFP-coding sequence expressed from the heart-specific cmlc2 (myl7) promoter. Two nanolitres of mixed pEXPmitfa-cre^ERt2 plasmid and Tol2 mRNA (25 ng µl⁻¹ and 35 ng µl⁻¹, respectively) were injected into one-cell-stage zebrafish embryos. Zebrafish embryos with the GFP transgenic marker in the heart were selected and grown to adulthood, then bred with wild-type fish to establish stable lines.

Zebrafish were genotyped using DNA extracted from fin-clipped tissue using DirectPCR lysis reagent (Viagen) complemented with 0.1 mg ml⁻¹ proteinase K. Polymerase chain reaction (PCR) was used to establish the mutant allele status p53^M214K and to verify the presence of transgene mitfa-BRAF^V600E, as described in detail before (Wojciechowska et al., 2016).

To induce melanoma regression, selected adult melanoma-prone fish homozygous for mitfa^vc7 mutation were transferred to tanks with heaters that kept the water temperature at 32°C compared to standard system temperature of 28.5°C. Fish were monitored twice a week for tumour changes and imaged every 2 weeks to visually compare the regression over time. A tumour was considered fully regressed once no melanoma was visually detectable and stayed unchanged during two consecutive time points of monitoring (usual range of regression being 5-10 weeks). To allow melanoma recurrence, regressed fish were transferred to tanks at room temperature (24-26°C), while monitored daily for tumour progression and health condition of the fish. Fish were always imaged prior to transfer from one temperature to another, then imaged regularly to track their progression and monitored daily. Each fish was followed through recurrence until the time point the tumour reached a similar size to the primary tumour, or until reaching the humane end point criteria based on general fish fitness, swimming behaviour and location of tumour (with a maximum of 3 months of recurrent disease growth).

The emergence of melanoma ranged between 2 and 14 months of age. All fish admitted to the experiment were over 3 months old, and all experiments were finished before fish reached 18 months of age. Both females and males were included in the experiments; on average, across experiments, the ratio was 58% females (50-66%) and 42% males (33-50%).

The tamoxifen treatment protocol was adapted from Gemberling et al. (2015) and Pinzon-Olejua et al. (2017). The tamoxifen stock solution (Cayman Chemicals or Fisher Scientific) was prepared by dissolving tamoxifen powder in DMSO to reach 10 mM concentration. We prepared fresh stock for each round of treatment on day 0 for the whole 3 day treatment course and stored it at −20°C in 200-250 µl aliquots. Tamoxifen stock solution (or corresponding DMSO volume as a control) was diluted with 500 ml E3 medium (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.33 mM MgSO₄) in a fish carrier tank, and fish were immersed in the solution for 11-12 h overnight (protected from light using a black chamber box or by wrapping the tank in aluminium foil) for three consecutive nights, with a drug pause during the day. For the tamoxifen treatment of fish with regressed melanoma, the carrier tanks with fish immersed in the tamoxifen or DMSO were kept in a water bath set at 31°C to ensure the continuous suppression of MITF activity.

4-OHT (Sigma, H6278) stock was prepared by dissolving the powder in ethanol to reach 5 or 10 mM concentration and thoroughly vortexed and aliquoted into 100 µl aliquots. The 4-OHT stock solution was then diluted into 200 µM intermediate stock in E3, vortexed thoroughly while protected from light and then diluted in E3 to the final working concentration, 20 µM. Embryos were placed into six-well plates with 3 ml 20 µM 4-OHT, protected from the light and incubated at 28°C. The 4-OHT was refreshed daily for a total incubation of 3 days. Embryos were then washed three times with E3 and incubated in E3 prior to imaging. In Fig. 1D,E, embryos were incubated for 6 h after the end of 4-OHT treatment.

Fish were briefly anaesthetised in tricaine solution (0.1 g l⁻¹ MS-222), placed onto a Petri dish while immersed in the tricaine solution and imaged using a Nikon COOLPIX5400 camera attached to a brightfield microscope (Nikon SMZ1500). Each fish was imaged using three consecutive photo shots that were later merged together using Adobe Photoshop automerge function. For fluorescent imaging, anaesthetised fish were imaged under a Leica MZFIII stereomicroscope equipped with a Retiga R1 camera operated via µManager software (Edelstein et al., 2010). Images were taken using GFP and mCherry fluorescent filters. The fluorescent images were processed using Fiji software version 2.1.0 (Schindelin et al., 2012) for pseudocolouring and adjusting minimal and maximal intensity.

For zebrabow live imaging, anaesthetised fish were imaged under a Leica M205FCA stereoscope equipped with Leica K8 camera and operated via LASX software (Leica). Images were taken with a 0.63× objective at 1× or 2.5× zoom using CFP, YFP and mCherry fluorescent filters and processed as above. As shown in Fig. S1B, the CFP channel showed a certain level of autofluorescence, most prominent in the creases of jaw and gills, that was not considered as positive signal.

For confocal microscopy, zebrafish embryos were anaesthetised with 0.1 g l⁻¹ MS-222, and adult fish were culled by overdose of anaesthetic (1 g l⁻¹ MS-222, followed by death confirmation) and embedded in 1% low-melting-point (LMP) agarose (Invitrogen) with the tumour or regressed tissue oriented to the bottom of six-well glass-bottom plates (Cellvis). For dorsal imaging of zebrafish embryos, fish were incubated for 5 min with 5 mg ml⁻¹ adrenaline (epinephrine, Sigma, E4642) in order to contract the pigment in melanocytes prior to mounting in LMP agarose for imaging. Images of ubi:Switch transgenic were obtained under an Andor Dragonfly spinning disk confocal microscope equipped with an Andor Zyla sCMOS camera through a 20× air objective with 2048×2048 pixel resolution and 1 µm interval. Images of ubi:zebrabow transgenic were obtained under a Leica STELLARIS 8 confocal microscope equipped with white-light laser through 10× and 20× air objectives with 1024×1024 pixel resolution and 1 µm (20×) or 3 µm (10×) intervals. In contrast to stereoscope imaging, we did not detect CFP channel autofluorescence using confocal microscopy.

Confocal acquisitions were processed using Fiji software version 2.1.0 (Schindelin et al., 2012). Standard deviation intensity projections of all z-slices were used except for fluorescence intensity quantification, where average intensity projections were used. Fluorescence intensity has been measured as described (McCloy et al., 2014). Briefly, average intensity projections of z-stacks acquired using the Andor Dragonfly spinning disk confocal microscope were split by channel to select mCherry only. Intensity of the whole imaged region was then measured, and intensity of the small region without any tissue was used to determine background. To calculate fluorescence intensity, corrected total cell fluorescence was calculated by subtracting the background fluorescence intensity. Any negative value was replaced by 0. Fluorescence area over time was measured on the mCherry channel of Leica MZFIII epifluorescence microscope acquisitions taken before and after the tamoxifen treatment and then every 2 weeks through melanoma recurrence. Area was manually depicted using the polygon selection tool in Fiji and then measure in µm². Any sample that showed fluorescence signal prior to the tamoxifen treatment was excluded from the analysis. Graphs summarising the fluorescence intensity quantification were made using R software version 4.0.3 via R studio interface version 1.1.456 equipped with ggplot2 package (https://www.r-project.org/; https://www.rstudio.com/; https://ggplot2.tidyverse.org/).

Fish were fixed overnight in 4% PFA at 4°C with agitation. The piece of fish tissue was then rinsed twice with PBS for several minutes and mounted in 4% agarose/PBS using a well from a six-well plate as a mould. Any excess agarose above 0.5 cm from the tissue was discarded prior to sectioning. Tissue was sectioned transversely using a Leica VT1200S vibratome at a thickness of 400 µm, and the speed of vibrating blade was 0.1 mm s⁻¹. Sections were collected in PBS and mounted in 1% LMP agarose in a six-well glass-bottom plate prior to imaging using the Leica STELLARIS 8 as described above.

Adult tamoxifen-treated or DMSO control fish with recurred tumours were culled using overdose of tricaine (1 g l⁻¹ MS-222, followed by death confirmation). Tumours were individually dissected and chopped using sterile No. 9 blade scalpels. Samples were then dissociated with 0.25 mg ml⁻¹ liberase TL (Roche) for 15 min at room temperature while inverting the tube, re-suspended in FACSmax cell dissociation solution (Genlantis) and filtered through a 40 µm cell strainer. Samples were sorted by a FACSAria2 SORP instrument (BD Biosciences UK) equipped with 405 nm, 488 nm and 561 nm lasers. Green fluorescence was detected using GFP 525/50 bandpass (BP) filter and 488 nm laser, red fluorescence was detected using mCherry 610/20 BP filter and 561 nm laser, and live cells were selected as 4′,6-diamidino-2-phenylindole (DAPI) negative using DAPI 450/20 BP filter and 405 nm laser. Data were analysed using FlowJo software (BD Biosciences) version 10.8.1.

Fish samples were collected, fixed and processed as described in detail in Wojciechowska et al. (2016). Briefly, tissue was fixed by immersion in 4% PFA at 4°C with agitation for 3 days, decalcified in 0.5 M EDTA (pH 8) at 4°C with agitation for 5 days and then stored in 70% ethanol at 4°C. To obtain sections for pathology analysis, tissue was processed in 95% ethanol, absolute alcohol, xylene and paraffin, embedded in blocks, cut into 5 μm-thick sections and placed onto glass slides. Haematoxylin and Eosin staining and IHC were performed as described in detail in Wojciechowska et al. (2016). The slides were imaged using a Hamamatsu NanoZoomer SlideScanner, and the images were processed using NDP.3 software.

PFA-fixed paraffin-embedded (FFPE) sections were deparaffinised in xylene and gradually rehydrated through baths of ethanol of decreasing concentrations. Sections were bleached in 1% KOH/3% H₂O₂ solution for 15 min prior to being subjected to heat-mediated antigen retrieval in citrate buffer (0.01 M, pH 6). Sections were blocked in 10% goat serum for 1 h and incubated with the following primary antibodies: anti-mCherry (Abcam, ab125096, 1:3000 or 1:4000) (Kobayashi et al., 2014), anti-Mitfa (Sigma, SAB2702433, 1:1000) and anti-Sox10 (Abcam, ab125096, 1:4000) (Travnickova et al., 2019) overnight at 4°C. After PBS/0.1% Tween 20 washes, sections were incubated with Alexa Fluor 488 (#A11001) or 546 (#A11003) goat anti-mouse and/or Alexa Fluor 647 (#A31573) goat anti-rabbit secondary antibodies at 1:500 dilution (Invitrogen) and DAPI (Sigma, 1:500) for 30 min in the dark at room temperature and mounted in hydrophilic mounting medium prior to imaging. Sections were imaged using an Andor Dragonfly confocal microscope equipped with Andor Zyla sCMOS camera using 20× and 40× air objectives.

FFPE sections were deparaffinised with xylene before gradual rehydration and bleached in a 1% KOH/3% H₂O₂ solution for 15 min. Slides were then stained with Haematoxylin, mounted in hydrophilic mounting medium and imaged using a Hamamatsu NanoZoomer slide scanner. After coverslip removal, slides were subjected to heat-mediated antigen retrieval in citrate buffer (0.01 M, pH 6) for 7 min. The sections were then incubated in serum-free protein blocking solution (DAKO) for 30 min at room temperature and incubated in primary antibody overnight at 4°C. Following TBS/0.1% Tween 20 washes, the sections were incubated in horseradish peroxidase (HRP)-conjugated rabbit/mouse secondary antibody (DAKO Real EnVision kit) for 30 min at room temperature. Antibody staining was revealed via incubation in AEC chromogen (Abcam) for 5-30 min. Following each antibody revealing, the slides were again mounted in hydrophilic mounting medium and imaged using a Hamamatsu NanoZoomer slide scanner. The slides were then de-coverslipped and underwent chromogenic de-staining in an alcohol gradient and subsequent antibody stripping via a 75 min incubation in a solution glycine-SDS, pH 2 at 50°C, before the next blocking and antibody round (adapted from Pirici et al., 2009; Tsujikawa et al., 2017). The following antibodies were used for MIHC: anti-Sox10 (Abcam, ab229331, 1:4000), anti-mCherry (Abcam, ab125096, 1:4000) and anti-Mitfa (Sigma, SAB2702433, 1:1000).

Sample distribution was evaluated using Shapiro–Wilk test. As the samples exhibited non-Gaussian distribution, statistical evaluation was performed using nonparametric Wilcoxon test. We set up the threshold to P=0.05 to be considered as statistically significant with the following symbols: **P<0.01; *P<0.05. Statistical analysis was performed using R software version 4.0.3 through R studio interface version 1.1.456 (https://www.r-project.org/; https://www.rstudio.com/).

We are grateful to the MRC Human Genetics Unit Zebrafish Facility for zebrafish management and husbandry, Elisabeth Freyer and the Institute of Genetics and Cancer (IGC) Flow Cytometry facility, Ann Wheeler and the IGC Imaging Facility for supporting the imaging experiments.