FOX and ETS family transcription factors regulate the pigment cell lineage in planarians

The diverse colors of life derive from organic pigments that selectively absorb particular wavelengths of light. These molecules are often produced by specialized cells that acquire unique structural and functional properties during embryonic development. Most vertebrate pigment cells, for example, arise from the neural crest (NC), a multipotent population of cells that delaminate from the neuroepithelium and migrate throughout the embryo to adopt a wide variety of fates (Pavan and Raible, 2012). A subset of NC cells undergo dorsolateral migration away from the neural tube, colonize the overlying ectoderm, and differentiate to form melanocytes in birds and mammals or chromatophores in amphibians, reptiles and fish (Mills and Patterson, 2009). Melanocytes produce the dark, tyrosine-derived pigment melanin, and then transport it via dendrites into neighboring keratinocytes, where it protects DNA from ultraviolet radiation (Cichorek et al., 2013). Melanocytes also impart color to feathers and hair. Chromatophores synthesize a variety of pigments, including melanin (melanophores), carotenoids and/or pteridines (erythrophores and xanthophores) and purines (iridophores), but generally retain them intracellularly rather than distributing them to other cell types (Eom et al., 2015; Mills and Patterson, 2009). Many chromatophores also exhibit a dendritic morphology, facilitating regulated aggregation and dispersal of pigment granules to effect rapid color changes for purposes such as camouflage (Aspengren et al., 2009).

Some animals retain the capacity to generate new pigment cells beyond embryogenesis. In zebrafish, for instance, the alternating horizontal stripes of the skin are formed from black melanophores and yellow-orange xanthophores, with iridescent iridophores conferring a shiny appearance (Lister, 2002). Xanthophores are derived directly from embryonic NC cells, while the melanophore and iridophore lineages include larval or adult stem cells associated with the peripheral nervous system (Parichy and Spiewak, 2015). Larval melanophores can be reformed from otherwise quiescent stem cells following ablation (Yang and Johnson, 2006), and new melanophores also arise from undifferentiated precursors during fin regeneration in adults (Rawls and Johnson, 2000). Similarly, adult stem cells at the base of hair follicles drive melanocyte production during the hair regeneration cycle in mammals (Nishimura et al., 2002).

Despite the fact that melanocytes are by far the most widely studied pigment cells, numerous aspects of their biology remain unknown. For example, they have been found in the inner ear, heart and brain, in addition to the epidermis (Goldgeier et al., 1984; Mjaatvedt et al., 2005; Roberts and Linthicum, 2015), and their lineage was recently shown to encompass ventrally migrating NC cells better known for generating Schwann cells (Adameyko et al., 2009). Even the widely cited photoprotective function of melanin in the skin might be an oversimplification, given several recent studies indicating that it can act as a potent photosensitizer (Chiarelli-Neto et al., 2014; Noonan et al., 2012; Premi et al., 2015; Takeuchi et al., 2004). Although recent progress has been made in organisms such as zebrafish (Parichy and Spiewak, 2015), non-melanocyte pigment cells and the molecular mechanisms that regulate their development are even less well understood. We have much to learn, therefore, about the biology of organic pigments and the cells that produce them.

Here we conducted a lineage analysis of pigment cells in adult freshwater planarians (lophotrochozoan flatworms) (Newmark and Sanchez Alvarado, 2002; Reddien and Sanchez Alvarado, 2004). Like many other animals, planarians exhibit highly variable pigmentation between species (Hyman, 1951). Biochemical and genetic evidence indicate that the body pigment in Schmidtea mediterranea is composed of ommochromes and cyclic tetrapyrroles called porphyrins (Hase et al., 2006; Stubenhaus et al., 2016). Enzymes in the ommochrome and porphyrin biosynthesis pathways are expressed in a candidate pigment cell population that is highly dendritic in morphology, and RNAi knockdown of some of these genes led to white animals (Stubenhaus et al., 2016; Wang et al., 2016). Porphyrins react with oxygen to form free radicals when exposed to light. Thus, prolonged light exposure ablates S. mediterranea pigment cells, resulting in an otherwise healthy, white animal (Stubenhaus et al., 2016).

In the current study, we take a systematic approach to defining the dynamics and regulation of the S. mediterranea pigment cell lineage by performing whole-animal mRNA sequencing (RNAseq) at multiple time points during light-induced depigmentation and subsequent repigmentation. This analysis revealed ten pigment cell markers that can be divided into two general categories: ʻdendritic' markers exhibit a unique expression pattern revealing the highly arborized morphology of the pigment cells; whereas the more numerous ʻpunctate' markers exhibit more focused RNA localization that is likely to reflect confinement to the cell body. Both categories of markers are expressed in the same subepidermal space and exhibit some degree of overlap at steady state, suggesting that they are co-expressed in the same cell type. When animals were challenged to make pigment cells de novo during regeneration or repigmentation of depigmented animals, dendritic markers appeared first, suggesting that they are involved in pigment biosynthesis pathways activated early during pigment cell differentiation. Finally, using single-cell RNAseq (scRNAseq) datasets, we identified three novel regulators of pigment cells: foxF-1, ets-1, and an FGFR-like molecule fgfrL-1. Our work provides the first molecular insight into pigment cell lineage specification in planarians.

In order to discover transcripts expressed in planarian pigment cells, we used RNAseq to compare gene expression in fully pigmented control animals, fully depigmented animals obtained by prolonged light exposure (Stubenhaus et al., 2016), and animals beginning to repigment after being returned to a dark environment (Fig. 1A; see Materials and Methods). In total, 436 transcripts were found to decrease by ≥2-fold with depigmentation (Table S1), 75 of which were cloned and screened by whole-mount in situ hybridization (WISH) (Fig. S1) for patterns consistent with pigment cell-specific or enriched expression (Stubenhaus et al., 2016; Wang et al., 2016). From this screen, two classes of pigment cell markers were identified.

The first class revealed the dendritic morphology of pigment cells, and included the known pigment biosynthesis enzymes ALAS, PBGD-1 and KMO-1 (Stubenhaus et al., 2016). Light-induced loss of these markers was confirmed by WISH (Fig. 1B,B′). Two of the remaining three dendritic class genes had strong homology to the enzymes phosphoenolpyruvate carboxykinase (Smed-pepck) and β-lactamase (Smed-lactb). The second class of markers exhibited a more punctate WISH pattern that was more evenly distributed across the animal than for the dendritic class (Fig. 1C,C′). Cells expressing punctate markers clearly outnumbered cells expressing dendritic markers, but they were also ablated by light exposure, even more rapidly than dendritic cells (Fig. 1C, Fig. S2). Two of the four punctate markers had orthology to a 60 kDa lysophospholipase [Smed-lysoLP-1 (Sugimoto et al., 1998)] and a threonine dehydratase II (Smed-thrD-1). The mammalian ortholog of Smed-lysoLP-1 is mainly expressed in liver and kidney tissue (endoderm), and plays crucial roles in the hydrolysis and transacylation of multiple phosphatidylcholine derivatives (Sugimoto et al., 1998). The remaining transcripts did not exhibit detectable homology and were named from their transcript numbers (Fig. 1B,C). In total, we identified ten markers potentially defining two different subpopulations of light-sensitive pigment cells.

We confirmed previous observations that KMO-1 and PBGD-1 are expressed in the same cells (Stubenhaus et al., 2016), and also determined that all dendritic markers were coincident by double fluorescent in situ hybridization (dFISH) (Fig. 2A). Similarly, we observed near complete overlap between different punctate class markers (Fig. 2A). Notably, FISH revealed that the punctate marker SmedASXL_005875 was expressed in cells with a dendritic morphology not evident by colorimetric WISH (Fig. 2A). Interestingly, 37±7% of cells expressing punctate markers also expressed dendritic markers, whereas 92±5% of cells expressing dendritic markers co-expressed punctate markers. We concluded that dendritic and punctate markers reflect two distinct cell types or states (see the Discussion), yet their overlapping expression also suggested that they might define cells in the same lineage.

We next analyzed the spatial location of dendritic and punctate markers in transverse sections using known markers for stem cells (piwi-1), epidermal progenitors (prog-1 and AGAT-1), and muscle (collagen) (Eisenhoffer et al., 2008; Reddien et al., 2005; Witchley et al., 2013; Zhu et al., 2015). Both dendritic and punctate markers were subepidermal, and interdigitated with muscle and epidermal progenitor cells (Fig. 2B), as predicted by previous electron microscopy studies of pigment cells in a related freshwater planarian species (Palladini et al., 1979). Interestingly, we observed a small but significant overlap between punctate and dendritic markers and early (prog-1⁺) and late (AGAT-1⁺) epidermal progenitors (between ∼2% and 14%; Fig. 2B), suggesting that the lineal origin of the pigment cells might be epithelial stem cells (also known as zeta-neoblasts) (van Wolfswinkel et al., 2014). However, when we ablated zeta-neoblasts using zfp-1(RNAi), we did not observe effects on either class of pigment cell markers (Fig. S3; see Discussion).

Following head or tail amputation, regenerating tissue (the blastema) is initially unpigmented for 10-20 days before eventually developing full pigmentation. This allowed us to monitor the birth and/or migration of new pigment cells. To assay the dynamics of pigment cell markers in the blastema, trunk fragments regenerating new heads and tails were labeled for dendritic and punctate markers for 21 days post-amputation (Fig. 3A). Dendritic markers were detected in significantly higher numbers than punctate markers for the first ∼8 days of regeneration, and in lower numbers thereafter (Fig. 3A,B).

The emergence of dendritic and punctate markers in blastemas could be due to migration of pre-existing pigment cells. Alternatively, we reasoned that the predominance of dendritic markers at early stages of regeneration might reflect de novo formation of new pigment cells via a lineage in which cells expressing dendritic markers give rise to cells that express punctate markers. We tested these possibilities by birth-dating cells with the thymidine analog BrdU. Trunk fragments were injected with BrdU at 1 day post-amputation, and subsequently assayed for the appearance of dendritic and punctate markers as well as BrdU in head and tail blastemas (Fig. 3C). At day 4 of regeneration (3 days after the BrdU pulse), new pigment cells (BrdU⁺) were evident, and all of these expressed dendritic, but not punctate, markers (Fig. 3C,D). By day 7 of regeneration, BrdU was evident in cells expressing either punctate or dendritic markers, as well as cells co-expressing both markers (Fig. 3C,D). A similar trend continued through day 14, at which time the majority of the BrdU signal was in cells expressing punctate markers (Fig. 3D). These data show that new pigment cells are formed during regeneration, and are consistent with a lineage in which the newly born cells progress from expressing dendritic to punctate markers (see Discussion).

Similar to regeneration, animals that are fully depigmented after light exposure take ∼20-30 days to fully repigment (Fig. 4A) (Stubenhaus et al., 2016). We monitored the generation of new pigment cells in this context by dFISH with dendritic and punctate markers (Fig. 4A). As early as day 3 of repigmentation, increases in both classes of markers were evident (relative to residual pigment cells at day 0; Fig. S3B). In these new cells, we observed a very high degree of co-expression, peaking at >80% at day 5 of repigmentation (Fig. 4A). Thus, the earliest-born pigment cells express dendritic markers, as was observed during regeneration. In order to determine how pigment class markers correlate with the actual production of visual pigment, we performed single-worm dFISH combined with ʻlive' imaging immediately following fixation (Fig. 4B). Visual pigment cells in live images were circled and directly mapped to the same worm following dFISH for dendritic and punctate markers. We observed that until day 7 of repigmentation, all cells that were pigmented express dendritic markers. It was not until after day 11 that we observed any significant numbers of pigment⁺ punctate⁺ dendritic⁻ cells. These data suggested that, as pigment cells mature and produce high levels of pigment, they downregulate dendritic markers of the pigment synthesis pathway.

To further understand the timing of birth and differentiation of pigment cells during repigmentation, we injected animals at the end of depigmentation (day 0 of repigmentation) and monitored the chase of the label into each class of pigment cell marker. Similar to pigment cell regeneration, dendritic marker⁺ cells were labeled with BrdU first by 5 days, with a subsequent chase into dendritic⁺ punctate⁺ cells by 7 days and punctate⁺-only cells by 14 days (Fig. 4C). However, considerable overlap with both classes of markers was seen throughout, suggesting a continuum of differentiation.

Both regeneration and repigmentation assays demonstrated that dendritic⁺ cells precede punctate⁺ cells and might constitute a progenitor for mature (punctate) pigment cells. Such a ‘progenitor state’ would be consistent with other cell lineages in planarians, such as the gut, brain, muscle, protonephridia, eyes, and epidermis, all of which have a postmitotic progenitor-like cell state en route to final differentiated cell types (Currie et al., 2016b; Eisenhoffer et al., 2008; Forsthoefel et al., 2012; Lapan and Reddien, 2012; Scimone et al., 2014, 2011; Witchley et al., 2013; Zhu et al., 2015). For these other lineages (particularly the epithelium), stem cell ablation using gamma-irradiation (Dubois and Wolff, 1947; Hayashi et al., 2006) leads to progenitor loss several days later (Eisenhoffer et al., 2008; Zhu et al., 2015). Thus, we reasoned that cells expressing dendritic markers might be irradiation sensitive if they define a progenitor for mature pigment cells. To test this hypothesis, we lethally irradiated fully pigmented animals and assayed changes in dendritic and punctate markers for 14 days (near the terminal point of survival for these animals). Surprisingly, we detected no changes in the number of cells expressing dendritic or punctate markers, demonstrating that there is little turnover of pigment cells during tissue homeostasis (Fig. S4). To determine whether expression of dendritic markers constitutes a progenitor state only when animals are challenged to make new pigment cells, we lethally irradiated animals 8 days after ablating pigment cells by light exposure. In these repigmenting animals, the number of cells expressing punctate markers remained constant, whereas the number of cells expressing dendritic markers significantly decreased (Fig. S5). These data suggested that dendritic markers might define an immature progenitor within the pigment cell lineage, at least when animals are forced to generate new pigment cells.

An alternative to dendritic⁺ cells being progenitors is that cells can dynamically switch between cell states (see Discussion). To test whether dendritic markers can be dynamically controlled, we performed RNAi to PBGD-1, which results in white animals without the loss of pigment cells (Stubenhaus et al., 2016). As animals became whiter, we observed significant increases in the numbers of cells expressing the dendritic marker KMO-1, as well as in the number of dendritic⁺ punctate⁺ cells (Fig. 4D). Combined with a concomitant decrease in punctate⁺-only cells, this suggested that pigment class markers can be dynamically controlled.

Using available single-cell RNAseq (scRNAseq) datasets (Molinaro and Pearson, 2016; Wurtzel et al., 2015), we identified 35 cells with detectable expression of PBGD-1. RNAi knockdown of this gene produces white animals (Stubenhaus et al., 2016). When the PBGD-1⁺ cells were hierarchically clustered, a cluster of three cells stood out as being likely pigment cells in that they expressed all dendritic and punctate markers (Fig. 5A; see Discussion). Importantly, the three identified pigment cells also co-expressed the known pigment cell regulator and forkhead transcription factor albino (Wang et al., 2016) and an FGF receptor (FGFR)-like molecule, Smed-fgfrL-1, also required for normal pigmentation (Dr Francesc Cebria, personal communication; transcript number SmedASXL_006982). These three cells were previously categorized as ʻgut' (Wurtzel et al., 2015), and together with the observation that lysoLP-1 and thrD-1 are liver-enriched genes in mammals (Leoncini et al., 1998; Sugimoto et al., 1998), this could suggest that pigment cells may have an endodermal origin. In addition, all three pigment cells highly express the putative endodermal-specific ʻgamma-neoblast' marker hnf4, which indicated that planarian pigment cells might originate from stem cells with endodermal characteristics (Martín-Durán and Romero, 2011; van Wolfswinkel et al., 2014). We tested this idea by RNAi knockdown of hnf4, and although the experiment was complicated by the health defects that occur when RNAi animals are also exposed to light, we did observe some changes in the intensity of dendritic marker staining (Fig. S6). Thus, it is interesting to speculate that planarian pigment cells have endodermal, rather than ectodermal, origins (see Discussion).

We next analyzed the co-expression of transcription factors that were highly correlated with PBGD-1 in the scRNAseq data (Pearson correlation ≥0.3) (Fig. 5B). In total, we found 17 transcription factors that met this criterion (Table S2). We performed an RNAi screen of the transcription factors co-expressed in the pigment cells, and found two that were required for normal pigmentation pattern or intensity of pigmentation, similar to Smed-fgfrL-1 and albino: a FoxF homolog (Smed-foxF-1) and an ETS-1 family homolog (Smed-ets-1) (Fig. 5C-G).

We assayed RNAi knockdown of the above genes for alterations of both dendritic and punctate markers in uninjured animals using pooled dFISH probes. Following knockdown of either albino or foxF-1, both punctate and dendritic markers were virtually ablated (Fig. 5F,G). fgfrL-1(RNAi) animals also showed reduction, but not complete ablation, of these markers. By contrast, ets-1 knockdown predominantly affected cells expressing punctate markers (Fig. 5F,G). It should be noted that RNAi of each factor was eventually lethal to the animals, with the exception of albino; however, we did not observe overt defects in other known cell types outside of the pigment lineage (Fig. S7).

These RNAi experiments were performed on steady-state animals, where we have shown the relationships between dendritic and punctate markers to be somewhat different than when animals are challenged to remake pigment cells de novo (Fig. 4). Therefore, we next tested the role of each factor in the regeneration of new pigment cells. Both foxF-1(RNAi) and albino(RNAi) animals showed total ablation of both dendritic and punctate markers during regeneration (Fig. 6A). ets-1(RNAi) animals, on the other hand, showed reduced numbers of cells expressing both marker classes compared with controls (Fig. 6A,B). We conclude that ets-1 functions to maintain the punctate cell state in homeostatic animals, but is required for pigment cell production during regeneration. fgfrL-1(RNAi) animals showed more modest, yet measurable, defects in the regeneration of punctate markers only (Fig. 6A,B).

In summary, we have identified multiple transcriptional regulators of pigment cells, as well as a potential signaling pathway involving an FGFR-like molecule that is likely to be involved in pigment cell specification and differentiation.

Our characterization of gene expression profiles in S. mediterranea pigment cells provides new insight into the developmental origins of body pigmentation in planarians. Specifically, we have identified a series of molecular markers for the pigment cell lineage and shown that production and maintenance of pigment cells during regeneration and tissue homeostasis is controlled by an FGFR-like molecule and the FOXF-1 and ETS-1 transcription factors.

We identified and characterized two classes of pigment cell markers exhibiting spatial and temporal differences in expression (Figs 1-4). These data are consistent with two different models for the pigment cell lineage: (1) all pigment cells are equivalent and capable of expressing dendritic markers, punctate markers, or both in response to developmental cues (Fig. 7A); (2) cells expressing dendritic markers are precursors for terminally differentiated pigment cells, which express only punctate markers (Fig. 7B). We favor the former model for several reasons. First, previous electron microscopy studies reported that mature pigment cells have a dendritic morphology in planarians (Palladini et al., 1979), as in many other animal species. Second, although there are clear differences in the number of cells expressing dendritic versus punctate markers during physiological cell turnover in adult animals (at the resolution of WISH), these markers show significant overlap (Fig. 2). Furthermore, FISH revealed that at least one punctate marker (SmedASXL_005875) was expressed in cells exhibiting a dendritic morphology (Fig. 2A), suggesting that the punctate expression pattern might reflect confinement of mRNAs to the cell body, rather than a difference in cell morphology. Third, scRNAseq revealed co-expression of all dendritic and punctate markers in all three pigment cells analyzed (Fig. 5A). Finally, RNAi knockdown of the dendritic markers KMO-1 or PBGD-1 uniformly alters or eliminates normal bodily pigmentation (Stubenhaus et al., 2016), rather than leading to a salt-and-pepper appearance as might be expected if only a fraction of pigment cells was affected.

We found that cells expressing dendritic markers are the first to appear during both the pigmentation of newly regenerated tissue (Fig. 3) and the repigmentation of depigmented tissue (Fig. 4A-C). Although these observations are formally consistent with a dendritic precursor model (Fig. 7B), we instead propose that they reflect upregulation of pigment biosynthesis pathways necessary for ommochrome and porphyrin production in new pigment cells. Indeed, three of the six dendritic markers we identified (but none of the punctate markers) are known pigment biosynthesis enzymes [ALAS, PBGD-1 and KMO-1 (Stubenhaus et al., 2016)]. Once new pigment granules have been synthesized, expression of these enzymes might be downregulated in concert with upregulation of punctate class genes. The functional significance of this switch awaits further characterization of the punctate markers, but it is interesting to note that it might be reversible, as we observed a dramatic upregulation of dendritic class markers when pigment biosynthesis was inhibited via RNAi knockdown of PBGD-1 (Fig. 4D).

dFISH with dendritic or punctate markers together with markers for progenitors of the epithelial lineage revealed a low but significant overlap in expression (particularly for prog-1⁺ ʻearly progenitors'; Fig. 2B). Because the epithelial lineage appears to have a dedicated stem cell class, i.e. the zeta-neoblast (van Wolfswinkel et al., 2014), we tested whether zfp-1(RNAi) (zeta-neoblast-ablated) animals could make new pigment cells. They could (Fig. S3), suggesting that the pigment cell lineage is not downstream of zeta-neoblasts [although these experiments were complicated by the poor health of zfp-1(RNAi) animals, which was compounded by depigmentation/amputation].

Unlike RNAi knockdown of zfp-1, RNAi knockdown of hnf4 did affect the staining intensity of dendritic cells during repigmentation and regeneration (Fig. S6). Moreover, lysoLP-1, thrD-1 and hnf4 were strongly expressed in all three pigment cells analyzed by scRNAseq (Fig. 5A). This raises the interesting possibility that the pigment cell lineage might overlap with the lineage of endodermal tissues such as the gut. It will be interesting to test this hypothesis in the future, when endodermal neoblasts are better defined.

By analyzing irradiated animals, we determined that pigment cells do not exhibit significant turnover during the course of the 2 weeks that animals can be assayed (Fig. S4). This implies that, once an animal is fully pigmented, stem cells no longer drive more than a very low rate of new pigment cell production. In turn, this might suggest a feedback loop in which mature pigment cells repress pigment cell specification (Fig. 7). The existence of such a regulatory mechanism might explain the variable RNAi phenotypes observed for fgfrL-1 and ets-1 during tissue homeostasis and regeneration (Figs 5, 6).

Our data also provide insight into the molecular mechanisms regulating the pigment cell lineage. The FGFR-like molecule encoded by fgfrL-1 is required for normal pigment cell specification (Figs 5, 6), similar to the function of a homologous pathway in the invertebrate chordate Ciona (Curran et al., 2009; Racioppi et al., 2014; Squarzoni et al., 2011; Thomas and Erickson, 2009). The identity and source of the ligand for this receptor are presently unknown, but it is interesting to speculate that it might be secreted by muscle, both because the pigment cells are interdigitated with the muscle layer (Fig. 2) (Palladini et al., 1979) and because many other patterning signals are secreted by muscle cells in planarians (Lander and Petersen, 2016; Scimone et al., 2016; Witchley et al., 2013).

RNAi knockdown of foxF-1 led to complete ablation of both dendritic and punctate class markers during tissue homeostasis and regeneration (Figs 5, 6), suggesting that this transcription factor is a key regulator of pigment cell production. The related family member Foxd3 controls melanophore specification in zebrafish and chick by regulating expression of the microphthalmia-associated transcription factor (Mitf) (Curran et al., 2009; Racioppi et al., 2014; Squarzoni et al., 2011; Thomas and Erickson, 2009). There are multiple Mitf homologs in S. mediterranea EST databases, raising the possibility that this regulatory mechanism is conserved in planarians. RNAi knockdown of the transcription factors albino and ets-1 also affected pigment cell specification, with albino showing a strong RNAi phenotype like foxF-1 (Figs 5, 6) (Wang et al., 2016), and ets-1(RNAi) predominantly affecting the expression of punctate markers, like fgfrL-1 knockdown during regeneration (Figs 5, 6). Identifying the regulatory hierarchy of these genes represents a key goal for future studies.

Pigment cells function in processes as diverse as mating, mimicry, camouflage, and adaptation to UV, all of which are subject to strong selective forces (sexual and natural). In planarians, the body pigment cells may be involved in the sensing of light [the ʻnon-ocular light response' (Birkholz and Beane, 2017)], as well as camouflage from predators, although neither has been rigorously tested. In our studies, animals that survived the depigmentation process appeared perfectly normal, suggesting that there is no essential function for pigment cells in S. mediterranea (Stubenhaus et al., 2016).

As observed in many other animal lineages, cave-dwelling planarians have lost body pigmentation, further supporting the notion that pigment cells are dispensable for core biological functions. It remains unknown whether this reflects loss of pigment cells or pigment biosynthesis enzymes (Hyman, 1951). Thus, it will be interesting in future studies to characterize the evolutionary mechanisms leading to the loss of pigment cells or biosynthetic pathways in various cave planarian species by conducting a careful examination of the markers and regulators that we have reported here.

Asexual Schmidtea mediterranea strain CIW4 were reared as previously described (Sánchez Alvarado, 2002). For irradiation experiments, planarians were exposed to 60 Gray (Gy) of γ-irradiation from a ¹³⁷Cs source (Pearson and Sánchez Alvarado, 2010). For depigmentation experiments, planarians were exposed to light at intermittent intervals as previously described (Stubenhaus et al., 2016). Animals were exposed to white light from LED light strips for 15 min, followed by 225 min of recovery in the dark. This cycle was repeated six times per day for 8-11 days, or until animals had lost all visible body pigment. A fan was used to prevent temperature elevation during light exposure, and animals were maintained in 50 µg/ml gentamycin and rinsed daily throughout the process. Size-matched control animals were kept in the dark in a 20°C incubator. Following depigmentation, animals were kept in the dark in an incubator for up to 33 days to allow for repigmentation. Planarians were prepared for bright-field imaging by incubation in 2% HCl for 15 s, then rinsing with PBS. Animals were immediately imaged using a Leica M165 fluorescence dissecting microscope.

RNAi experiments were performed using previously described expression constructs and E. coli HT115 (Newmark et al., 2003). Bacteria were grown to an OD₆₀₀ of 0.8 and induced with 1 mM IPTG for 2 h. Bacteria were pelleted, mixed with liver paste at a ratio of 333 µl liver to 100 ml original culture volume, and frozen in aliquots. This dilution was used in all experiments except for the zfp-1 knockdown repigmentation experiment, in which 500 µl liver was mixed with 100 ml original culture volume. The negative control, ʻcontrol(RNAi)', was the gfp sequence as previously described (Cowles et al., 2013). RNAi food was fed to 1-week-starved experimental worms twice weekly for a total of two to eight feedings. The total number of feedings, as well as the number of days after the last feeding, are denoted in each figure; for example, 6fd3 indicates six feedings and 3 days since the final feeding. Amputations were performed 3 days after the final feeding, with only the trunk fragment used in experiments. For the combination RNAi and repigmentation experiments, animals were fed twice during depigmentation: zfp-1(RNAi) animals were fed on day 0 and day 3 of depigmentation, whereas hnf-4(RNAi) animals were fed on day 5 and day 8. All animals were size-matched between experimental and control worms.

Whole-mount ISH (WISH), double fluorescent ISH (dFISH), and immunostainings were performed as previously described (Currie et al., 2016a; Lauter et al., 2011; Pearson et al., 2009). Colorimetric WISH stains were imaged on a Leica M165 fluorescence dissecting microscope. dFISH and BrdU stains were imaged on a Leica DMIRE2 inverted fluorescence microscope with a Hamamatsu back-thinned EM-CCD camera and spinning disc confocal scan head. BrdU (Sigma, B5002-5G; 25 mg/ml) was dissolved at 25 mg/ml in 50% ethanol and injected into the gut of each planarian using a Nanoject II microinjector (Drummond) and stained as previously described (Zhu et al., 2015). Planarians for BrdU experiments were maintained in a high salt concentration [2.5 g/l of Instant Ocean (∼43 mM)] beginning 1 week prior to injection. For regeneration experiments, trunk fragments were injected with BrdU 1 day after amputation of heads and tails. Cell counts and colocalizations were quantified using the Cell Counter function in ImageJ (http://rsb.info.nih.gov/ij/). Positive cells were visually distinguished manually. Significance was determined by a two-tailed, unequal variance, pairwise Student's t-test. Images were post-processed in a similar manner using Adobe Photoshop. Brightfield images of live worms in Fig. 4B were processed by changing the gamma to enhance visualization of pigment.

RNA deep sequencing (RNAseq) was performed on wild-type (WT) animals as well as fully depigmented animals (8 days of depigmentation by light exposure) and partially repigmented animals (1, 2 and 8 days of repigmentation in the dark on starved animals). Experiments were performed in biological triplicate, sequenced to a depth of ∼40 million reads per sample, and multiplexed on an Illumina HiSeq2500 with 50 bp, single-end reads. Seventy-five genes of interest were identified based on high 8 day WT versus depigmented fold change (Table S1). Raw scRNAseq data (including stem cells, neurons, gut, epithelial, muscle and parapharyngeal cells) were obtained from NCBI Sequence Read Archive (PRJNA276084) and Gene Expression Omnibus (GSE79866) (Molinaro and Pearson, 2016; Wurtzel et al., 2015). Reads were aligned to the SmedASXL transcriptome assembly under NCBI BioProject PRJNA215411 using bowtie2 (Langmead and Salzberg, 2012) with 15 bp 3′ trimming. Violin plots were produced using modified source code from Macosko et al. (2015) and heatmaps were produced using the modified heatmap.3 source code from Molinaro and Pearson (2016). Heatmaps in Fig. 1A and Fig. 5B have expression levels row-normalized to WT values for ease of visualization. The heatmap in Fig. 5A has log₂ normalized expression counts. Pigment lineage transcription factors were identified based on expression correlation with PBGD-1 (Pearson correlation ≥0.3; Table S2) in the published scRNAseq data (Molinaro and Pearson, 2016; Wurtzel et al., 2015).

We thank Dr Francesc Cebria for sharing fgfrL-1 information (transcript sequence and phenotype); and Dr Jochen Rink for sharing anti-PIWI-1 antibody.