To visualize dynamic TGFb response across melanoma development, we designed a TGFb Inducible Enhancer (TIE) reporter using human melanoma cell (A375) ChIP-seq data following a 2-hr TGFB1 treatment. RNA-sequencing indicated that following a 2-hr treatment of A375 cells with 10 ng/mL human recombinant TGFB1, 223 genes were significantly up-regulated (q<0.05) including typical TGFb target genes SMAD7, JUNB, and PMEPA1, while 94 genes were down-regulated (Figure 1—figure supplement 2A, left). Hallmark gene set enrichment analysis (GSEA) of genes ranked by log2 fold-change (log2fc) confirmed that TGFb was the top up-regulated pathway following 2-hr treatment (q=0) (Figure 1—figure supplement 2A, right). This indicates that a 2-hr treatment with human recombinant TGFB1 ligand is sufficient to activate the TGFb pathway.

To identify a TGFB1 inducible enhancer region, we selected a region of chromatin that was exclusively open upon stimulation, based on H3K27ac ChIP-seq, with unique SMAD2/3 binding following treatment (Figure 1A). The enhancer region under the SMAD2/3 peak was cloned upstream of a beta-globin minimal promoter and EGFP in a Tol2 vector backbone. TIE:EGFP was tested for inducibility in the presence of ubiquitously expressed constitutively active SMAD2 and SMAD3 (ubi:caSMAD2/3) by electroporation of adult zebrafish skin, with ubiquitous BFP (ubi:BFP) as a control for electroporation efficiency. TIE:EGFP reporter activity was significantly increased in the presence of ubi:caSMAD2 and ubi:caSMAD3, confirming that the reporter is inducible (Figure 1—figure supplement 1A).

The TIE:EGFP reporter construct was microinjected into Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- zebrafish in order to generate melanomas. TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- stable embryos have EGFP expression in the anterior brain at 24 hr post fertilization (hpf) extending along the brain and spinal cord beginning at 48 hpf (Figure 1B). This expression pattern is consistent with that previously found in a Smad3 zebrafish reporter, although in our hands this Smad3 reporter was not active in adults (Casari et al., 2014). TIE:EGFP embryos were treated at 24 hpf with 50 or 100 µM of SB431542, a TGFb type I receptor kinase inhibitor, and imaged at 48 hpf (Kim et al., 2021). Treatment with this TGFb inhibitor for 24 hr drastically reduced TIE:EGFP signal (Figure 1—figure supplement 1B, C). This indicates that the TIE:EGFP reporter is specific to TGFb signaling.

To visualize TGFb response across melanoma development, single cell Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- embryos stably expressing TIE:EGFP were injected with an empty MiniCoopR vector (MCR:MCS) containing the mitfa minigene to generate melanomas. Tyrosinase gRNA and Cas9 protein were also injected to knock-out melanocyte pigment as well as mitfa:mCherry to allow for melanocyte visualization. Normal melanocytes rarely had TIE:EGFP expression, with the exception of occasional fin melanocytes (Figure 1—figure supplement 1D). TIE:EGFP has yet to be observed in mitfa:mCherry high early melanomas (Figure 1—figure supplement 1E). However, of the fish with melanoma formation (n=56), 55% turned on TIE:EGFP in advanced melanomas defined by protrusion from the body plane (Figure 1C). TIE:EGFP⁺ cells often occur in clusters throughout the tumor and these cells are TIE:EGFP⁺ for many weeks at a time. This data indicates that advanced zebrafish melanomas develop TGFb responsive zones.

To understand the transcriptional differences between TIE:EGFP⁺ and TIE:EGFP^- melanoma cells, single TIE:EGFP⁺ and mitfa:mCherry⁺ cells from TIE:EGFP expressing melanomas were processed for single cell RNA-sequencing at 23 and 42 weeks post fertilization (wpf), respectively (melanomas used are depicted in Figure 1C and Figure 1—figure supplement 2B). We used SORT-seq to sequence flow cytometry-sorted single cells and in silico linked transcriptomes to fluorescence intensities as recorded by FACS indexing (Muraro et al., 2016). A UMAP of the sequencing results from our combined replicates is shown (Figure 2A; UMAP with no batch correction in Figure 2—figure supplement 1A). Most cells were identified as melanoma cells expressing mitfa and/or sox10, but we identified an mpeg1.1-expressing TIE:EGFP⁺ macrophage cluster as well (Figure 2A).

The melanoma cell population was separated into TIE:EGFP^-, TIE:EGFP^low, and TIE:EGFP^high based on FACS intensities. We found that immediate-early TGFb target genes, identified following an acute 2 hr TGFB1 treatment in zebrafish melanoma cells, ZMEL1, were down-regulated in TIE:EGFP^high cells (Figure 2B). However, we identified 29 genes that were up-regulated in TIE:EGFP^high cells. We termed these genes, related to the extracellular matrix, ‘chronic’ TGFb targets (Figure 2C; Katsuno et al., 2019; Verrecchia et al., 2001). At the time of dissociation, these melanomas had TIE:EGFP⁺ cells for several weeks, therefore they were likely not in an acute TGFb response phase (Figure 1C and Figure 1—figure supplement 2B). Using GSEA analysis we found that the top down-regulated pathways in TIE:EGFP^high cells were interferon alpha and gamma (Figure 2—figure supplement 1B). This suggests that TGFb has an immune suppressive effect in these melanomas.

Although the vast majority of literature focuses on acute TGFb signaling, our TIE:EGFP reporter remains on in melanoma, reading out chronic TGFb signal. We next asked what transcription factors are necessary for induction of our novel chronic TGFb reporter and therefore required for chronic TGFb signaling. We performed HOMER motif analysis of ~12,000 SMAD2/3 responsive regulatory elements. These regions were identified in A375 melanoma cells stimulated with TGFB1 and were bound by SMAD2/3 upon stimulus. We found that these SMAD2/3 bound regions are highly enriched for AP-1 motifs (Figure 2D). ChIP-seq in A375 cells in the presence or absence of TGFB1 treatment confirmed the binding of AP-1 transcription factors JUNB and ATF3 (Figure 2E). ATF3 and JUNB were bound at 19% and 48% of SMAD2/3 responsive elements, respectively, before TGFB1 treatment was administered. This data indicates that AP-1 factors, in particular JUNB, may be important for SMAD binding to chromatin.

We hypothesized that AP-1 factors occupy SMAD responsive elements prior to stimulation, stabilizing open chromatin to allow the TGFb response to occur quickly upon stimulation. To test this hypothesis, we deleted AP-1 motifs in our TIE enhancer driving luciferase. We identified AP-1 motifs (highlighted in orange) in our TGFb-induced enhancer using MoLoTool and HOCOMOCO motifs (Figure 2F, top; Kulakovskiy et al., 2018). AP-1 motif #1 was the most significant (p=8.4e^–6), and AP-1 motif #2 had a lesser p-value of 8.2e^–4. We deleted either AP-1 site #1 alone, AP-1 site #2 alone, or both AP-1 sites, as well as the most significant SMAD2/3 motif (p=6.7e^–6), shown in red (Figure 2—figure supplement 2). In A375 cells, loss of AP-1 site #1 rendered the enhancer region unresponsive to TGFb signaling, however loss of AP-1 site #2 did not alter inducibility (Figure 2F, bottom). This result indicates that AP-1 site #1 is necessary for TGFb inducibility at this enhancer. Deletion of the SMAD site also destroyed TGFB1 responsiveness. This data suggests that AP-1 binding is required for responsiveness of this chronic TGFb inducible enhancer in melanoma.

We identified TIE:EGFP⁺ macrophages in TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/-;MCR:MCS melanomas that express mitfa and sox10, suggesting recent phagocytosis of melanoma cells (Figure 3A). There are two potential models for why macrophages are TIE:EGFP⁺: (1) macrophages express TIE:EGFP themselves, or (2) macrophages are engulfing TIE:EGFP⁺ cells. To test these hypotheses, we crossed TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- zebrafish to a mpeg:mCherry reporter line that labels macrophages with mCherry (Ellett et al., 2011). These embryos were injected with MCR:BRAF^V600E, 2x U6:p53/Tyr gRNA mitfa:Cas9, and mitfa:BFP to generate pigmentless, BFP⁺ melanomas. As the TIE:EGFP reporter is cytoplasmic, if mpeg:mCherry⁺ macrophages are endogenously expressing TIE:EGFP, we would expect the entire macrophage to appear yellow via imaging. Confocal imaging of these tumors showed that the majority of a TIE:EGFP⁺ region of the tumor is composed of clustered TIE:EGFP⁺ macrophages. Within these regions, we observed that a subset of macrophages contain only puncta of TIE:EGFP signal, suggesting phagocytosis of TIE:EGFP⁺ cells, while a separate subset are entirely yellow (suggesting endogenous reporter expression; Figure 3B and Figure 3—figure supplement 1A). In 10 out of 13 tumors analyzed by confocal microscopy, about 60–80% of TIE:EGFP⁺ macrophages contained TIE:EGFP fragments, while the remaining TIE:EGFP⁺ macrophages were diffusely TIE:EGFP⁺. In the remaining three tumors, about 30% of TIE:EGFP⁺ macrophages contained TIE:EGFP fragments, while the remaining were diffusely TIE:EGFP⁺. Therefore, in most tumors (with rare exceptions), the majority of TIE:EGFP⁺ macrophages are EGFP⁺ because they phagocytosed a TIE:EGFP⁺ cell. Indeed, we observed macrophages actively phagocytosing TIE:EGFP⁺ cells (Figure 3—figure supplement 1C). In three out of 13 tumors imaged by confocal microscopy, we observed that in TIE:EGFP⁺ patches, macrophages were engaged closely with a cluster of TIE:EGFP⁺;mitfa:BFP⁺ tumor cells, likely in the process of phagocytosis (Figure 3B and Figure 3—figure supplement 1B). Together, this data reveals that a large proportion of macrophages engulf a TIE:EGFP⁺ cell population that includes melanoma cells.

To determine if macrophages are preferentially phagocytosing TIE:EGFP⁺ melanoma cells, three TIE:EGFP⁺;mpeg:mCherry⁺;mitfa:BFP⁺ melanoma were excised, digested, and processed for flow cytometry analysis (Figure 3C–D). Viable cells were separated into TIE:EGFP^- and TIE:EGFP⁺ (Figure 3D, far left). In all three tumors, less than 1% of sorted cells were TIE:EGFP⁺, indicating this TGFb responsive population is rare. TIE:EGFP^- and TIE:EGFP⁺ (Figure 3D, middle) cells were plotted relative to mpeg:mCherry and mitfa:BFP. Q2 represents the total percentage of TIE:EGFP^- or TIE:EGFP⁺ cells that were melanoma cells phagocytosed by macrophages. The percentage given in Q2 is plotted for several tumors in Figure 3D (far right). Macrophages phagocytose a significantly higher percentage of TIE:EGFP⁺ melanoma cells compared to TIE:EGFP^- melanoma cells. This was confirmed by qPCR, which showed that mitfa and sox10 expression is enriched in TIE:EGFP⁺ macrophages compared to TIE:EGFP^- macrophages (Figure 3—figure supplement 2). Altogether, scRNA-seq, confocal imaging and flow cytometry analyses demonstrate that macrophages preferentially phagocytose TIE:EGFP⁺ melanoma cells compared to TIE:EGFP^- melanoma cells.

The chromatin remodeler SATB2 is amplified in 4–8% of melanoma patients and its expression correlates with patient outcome. Our lab has previously shown that SATB2 overexpression leads to accelerated melanoma onset, produces more aggressive, metastatic tumors, and induces a TGFb signature (Fazio et al., 2021). SATB2 was overexpressed under the mitfa promoter in the MiniCoopR backbone (MCR:SATB2) in TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- zebrafish. Of the fish with SATB2 melanomas (n=27), 68% turned on TIE:EGFP in tumors, and its signal was expressed throughout the tumor volume (Figure 4A). To understand the transcriptional differences between TIE:EGFP⁺ and TIE:EGFP^- melanoma cells, we sequenced the transcriptome of single cells from an MCR:SATB2 expressing melanoma at 30 wpf using SORT-seq (tumor used shown in Figure 4A, top). We identified most cells as melanoma cells with mitfa and/or sox10 expression, and again an mpeg1.1 expressing TIE:EGFP⁺ macrophage population (Figure 4B and Figure 4—figure supplement 1A).

To determine if there is a change in state of the macrophages that express TIE:EGFP (either endogenously or via phagocytosis), we subset macrophages from the SATB2 expressing melanoma and separated macrophages that were TIE:EGFP⁺ or TIE:EGFP^- based on flow cytometry intensity. TGFb is known to convert macrophages from a pro-inflammatory M1-like to an anti-inflammatory M2-like state which can be characterized by gene expression. In the zebrafish, these include M1 markers acod1, tnfa, csf3a/b, and socs3b, as well as M2 markers mrc1b, vegfaa, alox5ap, and marco (Viola et al., 2019; Wu et al., 2015; Jeannin et al., 2018; Wilson, 2014; Martinez and Gordon, 2014; Hwang et al., 2020; Ye et al., 2021; Georgoudaki et al., 2016). We found that although TIE:EGFP⁺ macrophages are not clearly polarized, they express a combination of M1-like and M2-like marker genes (Figure 4—figure supplement 1B).

To understand the transcriptional differences between TIE:EGFP⁺ and TIE:EGFP^- melanoma cells, SATB2 expressing melanoma cells were divided into TIE:EGFP^-, TIE:EGFP^low, and TIE:EGFP^high cells based on flow cytometry intensity. These MCR:SATB2 results confirmed what was observed in the TIE:EGFP⁺ MCR:MCS tumors. We found that in TIE:EGFP^high melanoma cells, chronic extracellular matrix TGFb target genes were up-regulated and the top down-regulated pathways by GSEA again were interferon alpha and gamma (Figure 4C and Figure 4—figure supplement 1C). Down-regulation of interferon suggests that the TGFb responsive melanoma cells likely evade adaptive immune responses, such as interferon-mediated antigen presentation to CD8⁺ T cells. Finally, 40% of the MCR:SATB2 fish with melanomas (n=27) had TIE:EGFP expressed in early melanomas. These lesions are mitfa:mCherry high, indicating a proliferation of melanoma cells, but do not yet extend off the body plane. Meanwhile, 0% of MCR:MCS fish with melanomas (n=56) had TIE:EGFP⁺ early melanomas (Figure 4D). This shows that overexpression of SATB2 leads to acceleration of TGFb response in melanoma and suggests patients with this amplification may be more susceptible to TGFb inhibitors.

Tgfb-Induced-Enhancer:Beta-globin-minimal-promoter-EGFP:pA pDestTol2pA2 was cloned by PCR amplifying the enhancer (chr1:22747452–22747734) in Figure 1A using A375 human melanoma gDNA (Forward primer: TTCTTTGTCATCCTGGTAGAGCAAATCGAG, Reverse primer: GACAGGTCGCACCTGAGTCC) (Advantage 2 PCR Kit, Clontech #639207, Kyoto, Japan). PCR product was gel purified (Qiagen #28604, Hilden, Germany) and cloned into a pENTR-5’TOPO vector (Invitrogen #45–0711, Waltham, MA, USA). This 5 ’entry vector was Gateway recombined upstream of a mouse beta-globin-minimal-promoter-EGFP middle entry vector, a 3 ’entry polyA, and the pDestTol2pA2 backbone (ThermoFisher #12538120, Waltham, MA, USA). Mitfa:mCherry was Gateway recombined using the mitfa zebrafish promoter and a mCherry middle entry plasmid (ThermoFisher #12538120). Ubi:caSMAD2 was cloned by PCR amplifying human SMAD2 pDONR221 (DNASU HsCD00045549, Tempe, AZ, USA) using the following primers to insert constitutively active mutations (Forward: CACCATGTCGTCCATCT TGCCATTCACGCCGCC;Reverse: CTATTCCATTTCTGAGCAACGCACTGAAGGGGATC) (Sigma 11732641001, St. Louis, MO, USA), gel purifying (Qiagen #28604) and inserting into a pENTR/D-TOPO vector (Invitrogen #45–0218). This middle entry vector was Gateway recombined using a 5’ entry zebrafish ubiquitious promoter (Addgene #2732, Watertown, MA, USA), a 3’ entry pA vector, and a pDestTol2pA2 backbone (ThermoFisher #12538120). Ubi:caSMAD3 was cloned by PCR amplifying human constitutively active SMAD3 from Smad3 pCMV-SPORT6 (Harvard Plasmid Repository #HSCD00339271, Boston, MA, USA) (Sigma 11732641001), gel purifying (Qiagen #28604) and inserting into a pENTR/D- TOPO vector (Invitrogen #45–0218). This middle entry vector was Gateway recombined using a 5’ entry zebrafish ubiquitious promoter, a 3’ entry pA vector, and a pDestTol2pA2 backbone (ThermoFisher #12538120). Ubi:BFP was cloned via Gateway recombination using a 5’ entry zebrafish ubiquitious promoter, a 3’ entry pA vector, and a pDestTol2pA2 backbone (ThermoFisher #12538120). Mitfa:BFP was cloned via Gateway recombination using a 5’ entry zebrafish mitfa promoter, a 3’ entry pA vector, and a pDestTol2pA2 backbone (ThermoFisher #12538120). MCR:SATB2 and MCR:MCS are from Fazio et al., 2021. MCR:BRAF^V600E (Addgene #118846) and 2xU6:p53/tyr gRNA mitfa:Cas9 (Addgene #118844) are published (tyr gRNA: GGACTGGAGGACTTCTGGGG; p53 gRNA = GGTGGGAGAGTGGATGG CTG) (Ablain et al., 2018).

TIE:EGFP reporter fish were created by injecting TIE:Beta-globin-minimal-promoter- EGFP:pA pDestTol2pA2 at 6 ng/µL into single cell Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- zebrafish embryos (referred to as TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/-) (Patton et al., 2005). F0s were screened at 5 dpf for EGFP. At approximately 2 months post fertilization, F0s were crossed to Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- fish and F1s were screened as embryos for EGFP. EGFP⁺ fish were raised to adulthood.

Electroporation protocols were adapted from Callahan et al., 2018. Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- zebrafish were anesthetized using 4% MS-222 (Pentair, TRS1, Minneapolis, MN, USA). Zebrafish were subcutaneously injected with 2 µL mix using a Hamilton syringe (Hamilton #80300, Reno, NV, USA) anterior to the dorsal fin. The following amounts of vectors were injected into each fish: 200 ng TIE:Beta-globin-minimal-promoter-EGFP:pA pDestTol2pA2, 400 ng ubi:caSMAD2, 400 ng ubi:caSMAD3, 666 ng ubi:BFP, and 266 ng Tol2 pCS2FA-transposase (Kwan et al., 2007). DNA was prepped using Qiagen HiSpeed Plasmid Maxi Kit (QIAGEN #12662). Zebrafish were then electroporated with a BTX ECM 830 machine (BTX #45–0662 Holliston, MA, USA) using the following parameters: LV Mode; 45 V; 60ms pulse length; 5 pulses, 1 second interval. The cathode paddle was placed on the side of the fish that was injected. Electroporated fish were imaged approximately one week post electroporation using a Nikon SMZ18 Stereomicroscope (Nikon, Tokyo, Japan). To quantify TIE activity, we imaged each electroporated fish, quantified GFP intensity, and divided by the area of BFP to account for variation in electroporation efficiency.

For stereoscope imaging, zebrafish were anesthetized using 4% MS-222 (Pentair, TRS1) and imaged using a Nikon SMZ18 Stereomicroscope using consistent imaging parameters. For confocal imaging, tumors represented in Figure 3B and Figure 3—figure supplement 1C tumors were imaged using Nikon C2si Laser Scanning confocal using a 10 X objective at 1 X magnification. Z-stacks were aligned, and images minimally processed using Imaris (RRID:SCR_007370) or ImageJ software (RRID:SCR_003070). TIE:EGFP⁺ tumors depicted in Figure 3—figure supplement 1A, B were imaged using Zeiss LSM 980 NLO Multi-photon microscope using confocal detection at 15 X (Figure 3—figure supplement 1A, left), 45 X (Figure 3—figure supplement 1A, right), 20 X (Figure 3—figure supplement 1B). Z-stacks were aligned, and images were minimally processed using Imaris.

TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- F1s were crossed and F2 embryos were screened for EGFP, dechorionated at 24 hpf, and placed in a 24 well plate. Embryos were treated with either E3 zebrafish water, DMSO vehicle control (Sigma #D2650), 50 µM or 100 µM SB 431542 hydrate (Sigma # S4317-5MG). Twenty-four hours post treatment embryos were imaged using a Nikon SMZ18 Stereomicroscope (1 x objective; 3 x zoom; 7ms white light and 300ms GFP exposure used for all conditions).

SB-treated 48 hpf embryos (TIE:EGFP) were fixed with 4% PFA overnight at 4°C. The embryos were then dehydrated by washes of 1:2 MeOH:PBST, then 2:1 MeOH:PBST, then 100% MeOH, and stored in 100% MeOH. Embryos were re-hydrated using washes of 2:1 MeOH:PBT, then 1:2 MeOH:PBT, then 100% PBT. Samples were then digested with Proteinase K (Sigma Aldrich, #03115828001) at 10 µg/mL and washed with PBT. Samples were then re-fixed in 4% PFA, followed by several PBT washes. Samples were incubated in ‘PreHyb’ solution (50% formamide, 5 X SSC, 0.1% Tween 20, citric acid to pH6, 50 µg/mL heparin, 500 µg/mL tRNA) for 4.5 hr at 70 °C. They were then incubated in ‘ProbeHyb’ solution (same as PreHyb with addition of GFP1 probe at final concentration of 200 ng/500 µL) overnight at 70 °C. Samples were then consecutively washed in ‘WashHyb’ solution (50% formamide, 5 X SSC, 0.1% Tween 20, citric acid to pH6) at concentrations of 75%, then 50%, then 25% (in 2 X SSC) at 70 °C. Samples were then washed in 2 X SSC, then 0.2 X SSC (75%, then 50%, then 25%) in PBT. They were then incubated in blocking solution (PBT, 2% sheep serum, 2 mg/mL BSA) for 4 hr, then incubated in antibody solution (1:4000 of anti-DIG-AP (Sigma Aldrich, #11093274910) in blocking solution) overnight at 4 °C. Samples then underwent multiple PBT washes, then washing with staining wash solution (100 mM Tris HCL pH9.5, 50 mM MgCl2, 100 mM NaCl, 0.1% Tween 20). Samples were then stained (NBT 50 mg/mL, BCIP 50 mg/mL) for 30 min, then washed with PBT. Samples were post-fixed using 4% PFA at 4 °C overnight. Samples were then cleared using a series of MeOH washes: 1:2 MeOH:PBT, 2:1 MeOH:PBT, 100% MeOH. Samples were then washed with a series of glycerol washes: 30% glycerol (in PBT), 50% glycerol, 70% glycerol. They were then imaged in 70% glycerol using a stereomicroscope. Intensity of probe staining was analyzed, and embryos were binned into categories representing ‘low’, ‘medium’, or ‘high’, GFP intensity.

Melanocyte development is conserved between zebrafish and mammals (Mort et al., 2015). The master regulator of the melanocyte lineage, MITF, is conserved in zebrafish (mitfa) and required for melanocyte development (Lister et al., 1999). Expression of mitfa:BRAF^V600E together with a homozygous p53 missense mutation, leads to development of zebrafish melanomas (Patton et al., 2005). A mitfa^−/− mutation in Tg(mitfa:BRAF^V600E);p53^−/− fish prevents melanocyte development and spontaneous melanoma formation. Melanocyte development can then be mosaically rescued via injection of MiniCoopR (MCR), a transposon-based vector that contains a mitfa minigene, often alongside a candidate oncogene driven by the mitfa promoter (Ceol et al., 2011). In the absence of a candidate oncogene, the mitfa promoter is followed by an empty multiple cloning site (MCR:MCS), which is sufficient to generate melanomas in a genetic background already containing the BRAF^V600E and p53^-/- mutations. Tyrosinase, an enzyme required for melanocyte pigmentation, is also knocked out in our model to generate unpigmented melanomas, allowing for easy imaging of fluorescent reporters. TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/- tumors were generated by injecting single cell TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa^-/- zebrafish embryos with MiniCoopR Multiple Cloning Site (MCR:MCS) at 20 ng/µL, mitfa:mCherry at 10 ng/µL, tyr gRNA (GGACTGGAGGACTTCTGGGG) at 10 ng/µL, Cas9 protein (PNA Bio CP02) at 50 ng/µL, and Tol2 mRNA at 20 ng/µL (White et al., 2008). Experiments involving the mpeg:mCherry reporter line required us to cross the TIE:EGFP;Tg(mitfa:BRAF^V600E);p53^-/- zebrafish line with a Casper (roy^-/-;mitfa^-/-) mpeg:mCherry line, so progeny would be heterozygous for the BRAF^V600E and p53^-/- mutations. We therefore generated TIE:EGFP;mpeg:mCherry;Tg(mitfa:BRAF^V600E);p53^-/- tumors by injecting into the one-cell stage MCR:BRAF^V600E at 10 ng/µL, 2x U6:p53/Tyr gRNA mitfa:Cas9 at 10 ng/µL, mitfa:BFP at 10 ng/µL, tyr gRNA at 10 ng/µL, Cas9 protein at 50 ng/µL, and Tol2 mRNA at 20 ng/µL. Tyrosinase gRNA was created by annealing a tyr oligo template (CCTCCATACGATTTAGGTGACACTATAGGACTGGAGGACTTCTGGGGGTTTTAGAG CTAGAAATAGCAAG) to a constant oligonucleotide (AAAAGCACCGACTCGGTGCCACTT TTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC). The annealed oligo was filled in using T4 DNA polymerase (New England BioLabs, M0203S, Ipswich, MA, USA), PCR amplified, gel purified, transcribed (MEGAscript T7/SP6 Thermo Fisher Scientific, AM1333), and cleaned up (Zymo Research, R2051, Irvine, CA, USA). Embryos were grown to adulthood and were monitored for tumor development around 8–12 wpf.

All zebrafish (Danio rerio) work was performed in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Animal research protocols were approved by the Institutional Animal Care and Use Committee of Boston Children’s Hospital, Protocol #20-10-4254R. All zebrafish work operated according to the guidelines of the Institutional Animal Care and Use Committee of Boston Children’s Hospital.

Tumors were excised and dissociated for 30 min with occasional chopping using 0.075 mg/mL Liberase (Sigma #5401119001) in DMEM (Gibco #11965–092, Waltham, MA, USA) with 1% Penstrep (Corning #30–002 CI, Corning, NY, USA). Casper zebrafish skin was used to exclude autofluorescence when setting gates (White et al., 2008). A mitfa:mCherry⁺/EGFP^- tumor from a TIE:EGFP;mpeg:mCherry;Tg(mitfa:BRAF^V600E);p53^-/- fish or mitfa:mCherry;crestin:EGFP;tyr^-/- zebrafish skin was used to set the gates for mCherry intensity. Ubi:EGFP zebrafish skin was used to set the gate for EGFP intensity. Dissociated samples were filtered through a 40 µm filter and resuspended in FACS buffer (PBS/10%FBS/1% Penstrep) before filtering through a FACS tube (Corning #352235). Single cells were sorted into 384-well-cell-capture plates containing barcoded primers from Single Cell Discoveries (https://www.scdiscoveries.com/) using a BD FACS ARIA II sorter. SYTOX was used as a live/dead marker (ThermoFisher #S34857). Library preparation and Illumina sequencing was performed by Single Cell Discoveries (Muraro et al., 2016). SORT-seq data are demultiplexed and aligned to zebrafish Ensembl GRCz11 annotation using scruff R packages with the following parameter (bcStart = 7, bcStop = 14, bcEdit = 1, umiStart = 1, umiStop = 6, keep = 60) (Hao et al., 2021, Wang et al., 2019). The 384 SORT-seq barcodes are downloaded from https://github.com/anna-alemany/transcriptomics/blob/master/mapandgo/bwamap/bc_celseq2.tsv ( Alemany, 2018). Analysis was completed in R Studio using Seurat (RRID:SCR_016341) (min.features=200; 600<nFeature RNA <10000; percent.mt <10; obj.resolution=0.2; GFPhigh>4000) (Hao et al., 2021; Stuart et al., 2019; Butler et al., 2018; Satija et al., 2015). Batch correction was performed using FindVariableFeatures and FindIntegrationAnchors, nfeatures = 15,000. Pathway analysis was conducted using Gene Set Enrichment Analysis (GSEA; RRID:SCR_003199) version 4.1.0 (Mootha et al., 2003; Subramanian et al., 2005). Zebrafish genes were converted to human using DIOPT Ortholog Finder version 8.5 (RRID:SCR_021963) and the best match was used for GSEA analysis (Hu et al., 2011).

Tumors were excised and dissociated for 30 min with occasional chopping using 0.075 mg/mL Liberase (Sigma #5401119001) in DMEM (Gibco #11965–092) with 1% Penstrep (Corning #30–002 CI). Casper zebrafish skin was used to exclude autofluorescent cells (White et al., 2008). Mpeg:mCherry⁺ and ubi:EGFP⁺ zebrafish skin was used to set the gates for mCherry and EGFP intensity, respectively. A mitfa:BFP⁺ patch of skin as well as flk:BFP⁺ skin were used to set the gates for BFP intensity. Dissociated samples were filtered through a 40 µm filter and resuspended in FACS buffer (PBS/10%FBS/1% Penstrep) before filtering through a FACS tube (Corning #352235). DRAQ7 was used as a live/dead marker (Abcam #ab109202, Cambridge, UK). Cells were analyzed using a BD FACS Aria II 5 Laser System. Two technical replicates of 1 million cells were sorted from each tumor, for a total of 2 million cells per tumor. Data was processed using FlowJo version 10.8.1 (RRID:SCR_008520). Debris, doublets, and autofluorescent cells were removed from the analysis and viable cells were separated into TIE:EGFP^- and TIE:EGFP⁺.

Human melanoma A375 and zebrafish melanoma ZMEL1 cells were grown in filter sterilized DMEM (Gibco #11965–092) supplemented with 10% heat-inactivated FBS (Gemini # 900–108), 1% PenStrep (Corning #30–002 CI), and 1% Glutamine (ThermoFisher # 25030164). A375 cells were obtained from ATCC (RRID:CV-CL_0132) and grown at 37 °C, 5% CO₂. Cell identity was confirmed by fingerprint every 2 years and tested for mycoplasma approximately every 2–4 weeks using Lonza’s second generation myco PLUS kit. ZMEL1 cells were grown at 28 °C, 5% CO₂ and tested for mycoplasma approximately every 2–4 weeks (Heilmann et al., 2015). Human recombinant TGFB1 (R&D 240-B-002, Minneapolis, MN, USA) was reconstituted at 20 µg/mL in sterile 4 mM HCl containing 1 mg/mL BSA according to manufacturer’s instructions. To activate the TGFb pathway, cells were serum starved for 2 hr, then treated with 10 ng/mL human recombinant TGFB1 or 4 mM HCl containing 1 mg/mL BSA vehicle control for an additional 2 hr.

RNA-seq was performed in triplicate using A375 or ZMEL1 melanoma cells. RNA was collected from adherent cells using the Qiagen RNeasy Plus Mini Kit (Qiagen #74134). RNA quality was confirmed using a Fragment Analyzer. One microgram of RNA was ribodepleted using NEBNext rRNA Depletion Kit (NEB #E6310). Ribodepleted RNA was fragmented, reverse transcribed, and library prepped (NEB #E7530, NEB #E7335). Samples were sequenced on an Illumina Hi-Seq 4000 sequencer. Quality control of RNA-Seq datasets was performed by FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) (RRID:SCR_014583) and Cutadapt (RRID:SCR_011841) to remove adaptor sequences and low quality regions (Martin, 2011). The high-quality reads were aligned to Ensembl build version GRCh38 of human genome or zebrafish Ensembl GRCz11 annotation (RRID:SCR_002334) using Tophat 2.0.11 (RRID:SCR_013035) without novel splicing form calls (Trapnell et al., 2009). Transcript abundance and differential expression were calculated with Cufflinks 2.2.1 (RRID:SCR_014597) (Trapnell et al., 2010). FPKM values were used to normalize and quantify each transcript; the resulting list of differentially expressed genes were filtered by log fold-change and q-value. Pathway analysis was conducted using Gene Set Enrichment Analysis (GSEA) version 4.1.0 Hallmark gene sets (Mootha et al., 2003; Subramanian et al., 2005). Deseq2 (RRID:SCR_015687) was used to create differential expression heatmaps and volcano plots. Read counts less than 10 were excluded.

A375 cells were fixed directly in 15 cm plates with 11% formaldehyde and collected by scraping. Approximately 100,000,000 cells were used per condition. Cells were lysed using lysis buffers with protease inhibitors (Roche #05056489001, Basel, Switzerland) and sonicated such that fragmented chromatin was 200–300 bp long. Optimal chromatin length was confirmed by gel electrophoresis. Prior to antibody addition, 50 µL chromatin was collected for input sample. The remaining sonicated chromatin was incubated overnight at 4 °C with 10 µg antibody attached to Dynabeads (Invitrogen #10004D). Samples (including inputs) were washed with wash buffers and eluted for 6 hr at 70 °C, treated with RNaseA (Sigma #R4642) and Proteinase K (Life Technologies #AM2546, Carlsbad, CA, USA), and purified using Zymo ChIP DNA Concentrator kit (Genesee Scientific #11–379 C, San Diego, CA, USA). Libraries were end repaired (VWR #ER81050, Radnor, PA, USA), polyA tailed (Invitrogen #18252–015, NEB #M0212L), adaptor ligated (NEB #E7335), size selected using Ampure XP beads (Beckman Coulter #A63881, Brea, CA, USA, Life Technologies #12027), and PCR amplified (NEB #M0531). Libraries were run on an Illumina Hi-Seq 4000 sequencer. The following antibodies were used: H3K27ac (Abcam #4729; RRID:AB_2118291), SMAD2/3 (Abcam #202445), MITF (Sigma #HPA003259; RRID:AB_1079381), ATF3 (Abcam #207434; RRID:AB_2734728), JUNB (CST #3753, Danvers, MA, USA; RRID:AB_2130002). Using HOMER analysis (RRID:SCR_010881) we confirmed that JUNB and ATF3 binding motifs were present under their respective ChIP peaks (ATF3, p=1e^–4983) (JUNB, p=1e^–15697). All ChIP-Seq datasets were aligned to Ensembl build version GRCh38 of the human genome using Bowtie2 (version 2.2.1; RRID:SCR_016368) with the following parameters: --end-to-end, -N0, -L2086. MACS2 version 2.1.0 (RRID:SCR_013291) peak finding algorithm was used to identify regions of ChIP-Seq peaks, with a q-value threshold of enrichment of 0.05 for all datasets (Langmead and Salzberg, 2012; Zhang et al., 2008). Uropa (Universal Robust Peak Annotator) is utilized to annotate ChIP-seq peaks to neighboring genes according to Ensembl gene annotation (Kondili et al., 2017). The parameters are defined as proximal promoter: 500 bp upstream – 50 bp downstream of TSS; distal promoter: 2 k bp upstream – 500 bp downstream of TSS; enhancer: 100 k bp from TSS. The genome-wide transcription factor SMAD2/3, JUNB, ATF3 occupancy profile figures were generated by deeptools2 according to two computation modes (Ramírez et al., 2016). In the reference-point mode, a set of genomic positions (e.g. the center of ChIP peak) are used as anchor point, 2 kb upstream and downstream of these position are plotted in the profile figure. HOMER analysis was performed to confirm transcription factor binding under peaks (Heinz et al., 2010). The hg19 genome was used with a random set of background peaks for motif enrichment.

Firefly luciferase reporter constructs (pGL4.24) were created by cloning the full and mutated TGFb enhancers upstream of the minimal promoter using BglII and XhoI sites (see Figure 2—figure supplement 2 for sequences). A375 cells were plated in opaque-walled 96-well plates (Thermo Fisher #136101) and approximately 5000 cells were co-transfected with 100 ng firefly and 10 ng Renilla luciferase plasmids using Lipofectamine 3000 (Invitrogen #L3000008). After 48 hr cells were serum starved for 2 hr and treated with 10 ng/mL TGFB1 or 4 mM HCl containing 1 mg/mL BSA vehicle control for an additional 2 hr. Firefly and Renilla luciferase were then measured using the Dual-Glo Luciferase Assay (Promega #E2920, Madison, WI, USA) according to the manufacturer’s instructions. Luminescence was read on a Synergy Neo plate reader and the ratio of firefly to Renilla luminescence was calculated. Empty firefly luciferase vector was used as a negative control and Renilla luciferase was used as control for transfection efficiency. Experiments were performed in biological triplicate with three technical replicates each.

Two TIE:EGFP;mpeg:mCherry;Tg(mitfa:BRAF^V600E);p53^-/-;mitfa:BFP tumors were dissociated for 30 min with occasional chopping using 0.075 mg/mL Liberase (Sigma #5401119001) in DMEM/F12 (Gibco # 11580546), and the reaction was inactivated by adding a solution of 15% FBS in DMEM/F12. Dissociated samples were filtered through a 40 µm filter and resuspended in FACS buffer (PBS/5% BSA). Casper zebrafish skin was used to exclude autofluorescent cells, and mpeg:mCherry⁺ zebrafish skin was used to set the mCherry gate. SYTOX red dead cell stain was used as a live/dead marker (Invitrogen #S34859). Cells were sorted into FACS buffer in two separate tubes: mpeg:mCherry⁺ TIE:EGFP⁺ and mpeg:mCherry⁺ TIE:EGFP^-. RNA was extracted from cells sorted by FACS using the Direct-zol RNA MicroPrep Kit (Zymo Research #2060), and reverse transcribed to cDNA using the SuperScript VILO cDNA Synthesis Kit (Invitrogen #11754–050). qPCR was performed using SYBR Green qPCR Mix (ThermoFisher # 4309155), and samples were run using the QuantStudio 6 Flex system (ThermoFisher). Mitfa primer sequences (FP: 5’CTACGACAGCCCAAACAAGG, RP: 5 ’GCCATTGTCATGTTCGTCCA). Sox10 primer sequences (FP: 5 ’ACGCTACAGGTCAGAGT CAC, RP: ATGTTGGCCATCACGTCATG). Data was analyzed using the delta delta Ct method. Ct values were normalized to those of housekeeping gene, b-actin. B-actin primer sequences (FP: CGAGCAGGAGATGGGAACC, RP: CAACGGAAACGCTCATTGC).

To calculate significance of electroporation and flow analysis assays, a two-tailed unpaired t-test with Welch’s correction was performed using GraphPad Prism version 9.0.2 (RRID:SCR_002798) for Mac (GraphPad Software, San Diego, California USA, https://www.graphpad.com). To calculate significance of luciferase assays, a 2-way multiple comparison ANOVA test was performed using GraphPad Prism version 9.0.2 for Mac. For the qPCR experiment in Figure 3—figure supplement 2, 2-tailed unpaired t-tests was performed using GraphPad Prism for Mac.