Protein-Trap Insertional Mutagenesis Uncovers New Genes Involved in Zebrafish Skin Development, Including a Neuregulin 2a-Based ErbB Signaling Pathway Required during Median Fin Fold Morphogenesis

Skin disorders are widespread, but available treatments are limited. A more comprehensive understanding of skin development mechanisms will drive identification of new treatment targets and modalities. Here we report the Zebrafish Integument Project (ZIP), an expression-driven platform for identifying new skin genes and phenotypes in the vertebrate model Danio rerio (zebrafish). In vivo selection for skin-specific expression of gene-break transposon (GBT) mutant lines identified eleven new, revertible GBT alleles of genes involved in skin development. Eight genes—fras1, grip1, hmcn1, msxc, col4a4, ahnak, capn12, and nrg2a—had been described in an integumentary context to varying degrees, while arhgef25b, fkbp10b, and megf6a emerged as novel skin genes. Embryos homozygous for a GBT insertion within neuregulin 2a (nrg2a) revealed a novel requirement for a Neuregulin 2a (Nrg2a) – ErbB2/3 – AKT signaling pathway governing the apicobasal organization of a subset of epidermal cells during median fin fold (MFF) morphogenesis. In nrg2a mutant larvae, the basal keratinocytes within the apical MFF, known as ridge cells, displayed reduced pAKT levels as well as reduced apical domains and exaggerated basolateral domains. Those defects compromised proper ridge cell elongation into a flattened epithelial morphology, resulting in thickened MFF edges. Pharmacological inhibition verified that Nrg2a signals through the ErbB receptor tyrosine kinase network. Moreover, knockdown of the epithelial polarity regulator and tumor suppressor lgl2 ameliorated the nrg2a mutant phenotype. Identifying Lgl2 as an antagonist of Nrg2a – ErbB signaling revealed a significantly earlier role for Lgl2 during epidermal morphogenesis than has been described to date. Furthermore, our findings demonstrated that successive, coordinated ridge cell shape changes drive apical MFF development, making MFF ridge cells a valuable model for investigating how the coordinated regulation of cell polarity and cell shape changes serves as a crucial mechanism of epithelial morphogenesis.


Introduction
Skin conditions generate between 12% to 43% of medical visits [1,2]. In the United States, skin disorders are estimated to affect one third of the population at any time, with an estimated total annual cost of nearly US$100 billion [3]. Patients with skin disease frequently suffer physical discomfort and pain, and often experience diminished quality of life and psychological distress [4][5][6]. Medically, skin conditions are challenging to treat: skin is an exposed, physically vulnerable external barrier whose continuous turnover can impede long-lasting healing. Because there is a limited variety of clinical treatment methods, many of which are non-specific immune modulators such as steroids [7], new therapeutic targets for skin conditions could have important health and economic benefits [8]. Strategies for identifying novel integument genes and expanding our understanding of incompletely characterized integument loci offer avenues for subsequent interventional approaches.
The zebrafish (Danio rerio) is an excellent vertebrate model for understanding development and disease. Zebrafish not only share significant genomic similarity with humans [9,10], they also generate numerous transparent, externally fertilized embryos ideal for in vivo imaging and for phenotype-based gene discovery ("forward genetics") [11,12]. In addition to traditional chemical mutagenesis [13,14], forward genetic screening uses insertional mutagenesis methods, including retroviruses [15,16] and the more recently developed gene-breaking transposon (GBT) technology ( Fig 1A) [17]. GBT mutagenesis generates mRFP-tagged, Cre recombinaserevertible insertional alleles with 97% knockdown of endogenous transcript levels [17]. Zebrafish GBT insertional mutagenesis has already identified and characterized new genes, expression patterns, and phenotypes in the heart [18,19], vasculature [20], and muscle [21]. GBT insertional mutagenesis has also been used to dissect genetic links between brain and behavior [22]. However, it had not previously been applied to studying skin biology.
Fin fold development is an important event in larval epidermal development. Zebrafish larvae have an unpaired median fin fold (MFF) and a pair of pectoral fin folds (PFFs). The MFF is an ancient structure whose origins predate the evolution of paired pectoral fins [35,36]. MFF development begins at 18 hours post-fertilization (hpf) when midline basal keratinocytes undergo profound cell shape changes that drive median epidermal ridge (MER) formation along the ventral and dorsal caudal midlines. Midline basal keratinocytes adopt a wedgeshaped cross-sectional profile by reducing their basal surfaces and expanding their apical surfaces. Those shape changes, in conjunction with loss of contact with the underlying mesoderm, push up the midline basal keratinocytes (ridge cells) to create the MER [37,38]. Additional keratinocytes are then recruited to the proximal base of the MER. There they initiate MFF formation by elevating the MER perpendicularly to the larval mediolateral body axis. As the MFF extends further from the larval body, a sub-epidermal (dermal) space forms between the resulting apposed epidermal sheets. During MFF extension, the ridge cells of the initial MER remain at the tip of the MFF. As we show here, ridge cells' basal surfaces have re-expanded by 30 hpf during a second phase of cell shape changes. The single cells at the tip of the apical MFF (cleft cells) maintain an overall wedge shape while the growing basal surface invaginates, forming an intracellular cleft as a distal extension of the dermal space. In contrast, progressive apical and Gene-breaking activity occurs when an endogenous locus with a GBT insertion is transcribed. The vectorsupplied splice acceptor (SA) in the 5' protein trap cassette intercepts the endogenous splicing machinery and transcript, redirecting them to read directly into an AUG-free mRFP sequence (*mRFP). That event generates a fusion transcript by tagging the 5' portion of the endogenous transcript with mRFP. When translated, the mRFP fusion transcript will produce a potentially mutagenic truncated protein product. Simultaneously, the 3' exon trap cassette uses the vector-supplied splice donor (SD) to create a GFP fusion transcript with the remaining downstream endogenous transcript. GBT alleles are revertible because loxP sites (blue diamonds) flank the cassettes, allowing the mutagenic elements to be excised in the presence of Cre recombinase. (B-E') At 24 hours post-fertilization (24 hpf), GBT-generated mRFP fusion proteins from megf6a mn0325Gt (B), grip1 mn0078Gt (C), fras1 mn0156Gt (D), and hmcn1 mn0263Gt (E) localize along MFF edges. (B', E') Both megf6a mn0325Gt (B') and hmcn1 mn0263Gt (E') localize within a narrow region along the MFF edge (blue arrowheads). (C', D') grip1 mn0078Gt (C') and fras1 mn0156Gt (D') localization also follows the fin fold edge (blue arrowheads), though they are distributed somewhat more diffusely than are megf6a mn0325Gt (B') and hmcn1 mn0263Gt (E'). (F-H) Whole-mount in situ hybridization (WISH) in 24 hpf wild-type embryos reveals similar MFF expression patterns of endogenous grip1 (F), fras1 (G), and hmcn1 (H) genes. The mRFP fusion protein localization patterns observed in the respective GBT lines recapitulate endogenous gene expression (C, D, E). basal surface growth leads the two to three ridge cells on either side of the cleft cell to elongate, adopting a flattened, epithelial morphology with a rectangular cross-sectional profile.
Little is known about the genetic control of the processes that direct MER formation [38,39], but studies indicate that MER cells play crucial roles during MFF development and maintenance. As MFF development proceeds, MFF basal keratinocytes, including MER cells, secrete extracellular components into the dermal space. Those components then assemble into specialized extracellular structures, including cross-fibers [37,40] and collagenous actinotrichia [41,42], which provide the MFF with structural support against mechanical stresses. While basal keratinocytes along the entire proximo-distal length of the MFF secrete actinotrichia components [42], MER cleft and ridge cells are the predominant producers of basement membrane (BM) components such as Laminin [43,44] (see below) and BM-dermal anchorage proteins such as Fras1 [44,45] (see below). By 48 hpf, fin mesenchymal cells (FMCs), a type of dermal fibroblast, have finished invading the MFF to further support and organize the dermal space and its ECM [40,46].
Genetic evidence increasingly demonstrates that zebrafish MFF development models multiple processes affected in human skin diseases. Two prominent examples are fraser syndrome 1 (fras1) and fraser-related extracellular matrix 2a (frem2a), encoding extracellular matrix (ECM) components mutated in human Fraser Syndrome patients. During human and mouse development, correct BM-dermal junction formation requires FRAS1 and FREM2, especially in limb buds and craniofacial regions. FRAS1 and FREM2 mutations cause transient, locally restricted fetal skin blistering and later lead to characteristic malformations such as cryptopthalmos and syndactyly [47][48][49]. Likewise, mutations in the corresponding zebrafish homologs fras1 and frem2a lead to transient skin blistering in developing larval fins [44]. Additional examples are the zebrafish mutants in the BM component laminin α5 (lama5) [43] and its transmembrane receptor integrin α3 (itga3) [44], as well as in the intracellular integrin modulator kindlin-1 (fermt1) [50]. All of these mutants are characterized by impaired MFF epidermal integrity, and they model the corresponding epidermal defects characteristic of human epidermolysis bullosa (lama5, itga3) and Kindler Syndrome (fermt1).
Our study used GBT technology to identify eleven zebrafish integument loci expressed in either larval epidermis or FMCs. Three of those loci (fras1, hmcn1, grip1) are novel revertible alleles of known disease-related genes that will provide new, Cre-regulatable molecular tools for skin biology studies. We also identified neuregulin 2a (nrg2a) as a novel essential regulator of apical MFF development and MFF ridge cell polarity. nrg2a is a zebrafish homolog of the epidermal growth factor (EGF) superfamily member NEUREGULIN 2 (NRG2). Although an earlier morpholino (MO) knockdown study indicates that nrg2a participates in dorsal root ganglion development [51], nrg2a has not been investigated in conjunction with skin or MFF development. We show here that zebrafish nrg2a mn0237Gt/mn0237Gt insertional mutants display specific defects in MFF ridge cell organization, leading to aberrant apical MFF morphology reminiscent of ErbB2 (erbb2 -/-) mutants [52]. ErbB receptor tyrosine kinases (RTKs) mediate signaling by EGF superfamily ligands, such as the Neuregulins (NRGs), through several different signal transduction pathways, including the mitogen-activated protein kinase (MAPK) ERK, and/or phosphatidylinositol-3-kinase (PI3K) and AKT [53]. Our pharmacological inhibition, erbb expression, and activated ERK (pERK) and AKT (pAKT) immunofluorescence studies were consistent with this novel Nrg2a-dependent MFF pathway operating through ErbB2/3 and the PI3K -AKT signal transduction cascade. Furthermore, the nrg2a phenotype led us to discover an early role for the epithelial cell polarity regulator and tumor suppressor lethal giant larvae 2 (lgl2) in MFF development. Previous mutant studies established that lgl2 promotes epidermal integrity by attenuating ErbB signaling during late larval development (96-120 hpf), but did not find an earlier developmental role for lgl2 despite having documented its expression from 24 hpf onwards [52,54]. We show here that MO-directed knockdown of lgl2 in nrg2adeficient larvae rescues the MFF defects caused by loss of functional Nrg2a two days before lgl2's tumor suppressor role. Therefore, we have identified a previously undocumented negative impact of lgl2 on ErbB signaling that regulates epithelial polarity and keratinocyte biology during epidermal morphogenesis in the apical fin fold.

Results
Gene-breaking Transposon (GBT) alleles of eleven genes with previously known and unknown expression in epidermal or fin mesenchymal cells Locating expression within a tissue of interest is a key step toward identifying new genes involved in tissue-specific biology. Using the online GBT database and image catalog zfishbook.org (http://www.zfishbook.org) [55] as an initial selection tool, we screened 350 GBT insertional alleles [17] for mRFP expression within the skin of the MFF (Fig 1B-1E; Fig 2B, 2D, 2F, 2H, 2J, 2L, and 2N) and/or covering the larval body (Fig 2A, 2C, 2E, 2G, 2I and 2K) [37,38,56]. That expression prioritization screen identified eleven gene-break alleles (GBT lines) with mRFP localization in the integument (Fig 1B-1E; Fig 2). We refer to these as the initial Zebrafish Integument Project (ZIP) GBT lines. The ZIP lines' RFP localization patterns [17] fell into three broad categories. The first was expression in MFF edges (Fig 1B-1E); the second was expression throughout some or all of the epidermis (Fig 2A-2N); and the third was expression in FMCs (Fig 2I-2N; Fig 3). Because we identified these eleven ZIP lines among the initial cohort of GBT insertional alleles, the GBT-disrupted genes in several ZIP lines had been identified during our initial GBT mutagenesis study [17]. We completed sequence-based identification of ZIP genes not identified in the previous study (Table 1).
Of the eleven genes identified (Table 1; Fig 1B-1D'; Fig 2), three have established roles in skin development and epidermal-dermal junction formation: fras1, grip1, and hmcn1 [44,62]. As described above, FRAS1 is a large basement membrane (BM)-associated ECM protein connecting BM with underlying dermal connective tissue. The cytoplasmic protein GRIP1 is required for properly localizing FRAS1 to the basal side of epithelial cells [63]. In human and mouse, recessive mutations in FRAS1 [47,48] and GRIP1 [63][64][65] are characterized by malformations resulting from epidermal/epithelial blistering during fetal development. Zebrafish fras1 mutants are characterized by corresponding blistering in the developing fin folds, though zebrafish grip1 mutants have no overt morphological phenotype [44]. Hmcn1/Fibulin 6 is a highly conserved member of the Fibulin ECM protein family, with epithelial cell-anchoring roles in C. elegans [66]. Zebrafish genetic studies have shown Hmcn1 is essential for epithelialdermal anchorage in developing fin folds [44].
To varying degrees, five other ZIP alleles have reported zebrafish or mammalian integument expression: col4a4, ahnak, capn12, nrg2a, and msxc. Col4a4 encodes one of the six subunits of We identified the corresponding GBT-tagged locus for each ZIP line. Consistent with the random nature of RP2 GBT insertion, no two insertions were on the same linkage group (LG). All eleven ZIP loci have mammalian homologs. Ten have human homologs, and one (msxc mn0245Gt ) has a murine homolog but no reported human homolog. Mutations in two of those ten human homologs, FRAS1 and GRIP1, are known to be involved in Fraser Syndrome, a human disease whose phenotype includes blistering. Three ZIP alleles are novel integument genes: arhgef25b mn0175Gt , fkbp10b mn0316Gt , and megf6a mn0325Gt . Although several ZIP alleles ( §) were reported in an earlier publication, this study initially identified their expression patterns and includes previously unpublished data for each. § Sequence identified in Clark et al., 2011 [17]. ‡ Novel integument genes.
type IV collagen, a major BM structural component [8,62,67]. Originally isolated from bovine muzzle epidermis as Desmoyokin, AHNAK [68] shuttles between the cytoplasm and the plasma membrane in keratinocytes and other epithelial cells [69]. Ahnak mn0196Gt -mRFP shows similar membrane localization (Fig 2G and 2H). CALPAIN12, a member of the Calpain family of calcium-dependent, non-lysosomal cysteine proteases, is expressed in murine skin [70]. Calpain activity has also been implicated in wound healing [71]. Transcriptional profiling data indicate that NEUREGULIN 2 (NRG2) is expressed in the epidermal basal layer [72]. msxc encodes the transcription factor MsxC and is expressed in zebrafish FMCs [39,61]. Finally, ZIP lines arhgef25b, fkbp10b, and megf6a did not have documented integument expression. arhgef25b mn0175Gt , fkbp10b mn0316Gt , and megf6a mn0325Gt therefore represent novel skin genes-and are novel revertible alleles suitable for future genetic studies. fras1 and hmcn1 GBT alleles phenocopy known ENU mutants and are revertible To determine which ZIP loci were required for early zebrafish skin or fin fold development, we bred each expressing insertion to homozygosity through a standard inbreeding scheme. For each line, we screened offspring of intercrossed heterozygous parents during the first five days of development (120 hpf) for abnormal skin or fin fold morphology present only in mRFPpositive (GBT-expressing) larvae and absent from their wild-type siblings [17] (an "mRFPlinked" phenotype). Of the eleven lines screened, three displayed recessive, mRFP-linked phenotypes: fras1 mn0156Gt (Fig 4B), hmcn1 mn0263Gt (Fig 4C), and nrg2a mn0237Gt (Fig 5B). Comparisons to published phenotypes of ENU-derived alleles indicated that fras1 mn0156Gt and hmcn1 mn0263Gt represented novel alleles of known skin blistering mutants pinfin (pif) and nagel (nel) [44,56], respectively (Table 1, Fig 4A-4C).
Because the previously reported zebrafish nrg2a sequence contained a putative non-coding first exon but did not contain an N-terminal signal sequence [51], we investigated whether additional nrg2a N-terminal exons were present. Through 5'RACE and zebrafish genome database searches, we identified two alternative N-terminal exons, exons 1B and 1C. While the putative exon 1A contains only non-coding sequence and is used for the previously reported N-terminally truncated Nrg2a isoforms [51], the newly annotated exons 1B and 1C give rise to alterative N-termini of 145 (1B) or 37 (1C) amino acid residues, respectively ( Fig 6A; S2 Fig). Exon 1B has homology to N-termini of NRG2 homologs in other species, whereas exon 1C shows no homology to any other species. SignalP4 software analysis [77] (http://www.cbs.dtu. dk/services/SignalP/) indicated that only the exon 1B-encoded isoform contains an N-terminal signal sequence. In addition, NucPred software analysis [78] (http://www.sbc.su.se/~maccallr/ nucpred/) predicted that the exon 1C-encoded isoform would be capable of nuclear localization (S2 Fig). The gene-breaking cassette that generates the mn0237Gt allele is located in the intron between exon 1C and exon 2 ( Fig 6A). According to RT-PCR analyses, endogenous 1B and 1C transcripts were present in both wild-type (nrg2a +/+ ) and heterozygous (nrg2a +/ mn0237Gt ) siblings, but absent in homozygous mutant (nrg2a mn0237Gt/mn0237Gt ) siblings ( Fig 6B   and 6C). Conversely, 1B-mRFP and 1C-mRFP fusion transcripts were present in mutant and heterozygous siblings, but absent from wild-type siblings (Fig 6B and 6C). We note that we were unable to amplify the sequence corresponding to non-coding exon 1A [51]. Together, these transcript data indicate that mn0237Gt is a null or near-null loss-of-function allele (see also Discussion).
Both nrg2a-mRFP fusion transcripts ( Fig 5G) and Nrg2a mn0237Gt -mRFP fusion protein ( Fig  5E and 5H) were epidermally localized in 24 hpf nrg2a +/mn0237Gt heterozygotes. WISH analyses using a probe for mRFP revealed that the fusion transcript was prominently expressed along the MFF edges (Fig 5E and 5F)-that is, within the apical MFF-and was weakly expressed throughout the entire epidermis ( Fig 5G). WISH analyses of endogenous nrg2a transcripts in wild-type embryos showed the same pattern (24 hpf; Fig 5H and 5I), thus demonstrating that the mn0237Gt insertional allele recapitulates endogenous nrg2a expression.
Co-labeled MFF transverse sections further revealed that the Nrg2a mn0237Gt -mRFP fusion protein was present in the basal layer of keratinocytes, together with the nuclearly localized basal keratinocyte marker ΔNp63 [32]. In contrast, the EVL, characterized by expression of the krt4:EGFP transgene [79], lacked Nrg2a mn0237Gt -mRFP fusion protein (Fig 5J-5N). Higher magnification analyses confirmed that the punctate distribution of Nrg2a mn0237Gt -mRFP fusion protein (Fig 2B; Fig 5E) reflected nuclear localization of the fusion protein (Fig 6C-6F). In addition, Nrg2a mn0237Gt -mRFP fusion protein could be detected in the cytoplasm of basal keratinocytes (Fig 5J-5N; Fig 6C-6F). Both of those findings regarding Nrg2a mn0237Gt -mRFP subcellular localization were initially surprising because Nrg2a is a growth factor (but see Discussion). However, wild-type embryos injected with synthetic mRNA encoding the exon 1C version of the Nrg2a-mRFP fusion protein showed similar cytoplasmic localization. Exogenous exon 1C-mRFP fusion protein was present in both keratinocyte nuclei and cytoplasm (Fig 6K-6N), mimicking the distribution of the transgene-encoded fusion protein in nrg2a +/mn0237Gt heterozygous embryos (Fig 6C-6F). Exogenous exon 1B-mRFP fusion protein also showed some cytoplasmic localization (Fig 6G-6J). These results were consistent with the presence of one or more putative nuclear localization sequences (NLS) in the predicted amino acid sequence encoded by nrg2a exon 1C (S2B Fig), and suggest that the complex subcellular localization of the Nrg2a-mRFP fusion protein is caused by differential distributions of the different N-terminal isoforms.

nrg2a mutants display altered epithelial organization of MFF ridge cells
To further characterize the MFF defect in nrg2a mn0237Gt/mn0237Gt mutants, we compared transverse sections of median epidermal ridges (MER) from nrg2a mn0237Gt/mn0237Gt mutant larvae with those of wild-type siblings. As described above, the MER consists of a central cleft cell, characterized by a cleft-like invagination of the dermal space, and ridge cells to either side of the cleft cell. To visualize the shapes of individual cleft and ridge cells, we crossed nrg2a mn0237Gt into a Tg(Ola.Actb:Hsa.hras-egfp) vu119 (hras-egfp) ubiquitously expressed membrane-bound EGFP background (Fig 7A-7C). Furthermore, to specifically label cleft cells, we crossed nrg2a mn0237Gt into an ET37 background, in which EGFP is specifically expressed in FMCs and cleft cells (Fig 7D and 7E). We also performed transmission electron microscopy (TEM; Fig 8;  S4 Fig) to examine the cleft and ridge cells in finer detail. Because the gross morphological changes characterizing nrg2a mn0237Gt/mn0237Gt mutant MFFs were visible by the second day of development (Fig 5B), we conducted our analyses between 30 and 52 hpf.
During normal MFF morphogenesis, MER cells undergo characteristic consecutive cell shape changes, accompanied by changes in the organization of the developing dermal space between the two apposed epidermal sheets. During the initial steps of fin fold elevation along the body midline, MER ridge cells acquire wedge-like shapes by expanding their lateral and/or apical domains at the expense of the basal domain [38]. But at 30 hpf, re-enlargement of ridge cell basal domains had partially reversed the earlier shape changes. Ridge cells acquired an intermediate shape, more cuboidal than wedge-like, while the dermal space was curved ( Fig  7A). MER morphogenesis continued through 36 hpf (Fig 8A, 8E, and 8F; S4A Fig). By 52 hpf, the dermal space had straightened and ridge cells had elongated, acquiring a flattened shape with large, equally-sized apical and basal domains, but small lateral domains (Fig 7B and 7D; Fig 8C).
Cleft cells in nrg2a mn0237Gt/mn0237Gt mutants were only mildly affected. They expressed the ET37 marker (Fig 7E) and retained a typical cleft shape (Fig 7E; S4B Fig) (Fig 7C and 7E; Fig 8B, 8D-8G). The bulges consisted either of two ridge cells sharing an extended lateral border (Fig 8G and 8H), or of a single ridge cell with an extended basal domain (S4C and S4D Fig). In both cases, the extent of apical domains was reduced. These findings suggest that Nrg2a regulates ridge cells' apicobasal organization during apical MFF morphogenesis, thereby promoting or maintaining apical domain identity at the expense of basolateral identity.
To determine whether Nrg2a might have a corresponding function in more proximal (non-MER) epidermal cells of the MFF, we compared 52 hpf and 96 hpf mutant and wild-type siblings via TEM. However, we did not observe consistent epidermal differences, malformations, or cell-cell junction defects in these regions of nrg2a mn0237Gt/mn0237Gt mutants compared to wild-type siblings (data not shown).

GBT mutagenesis further develops the zebrafish as a skin biology model
By selecting for skin-and MFF-localized mRFP, we identified four genes with little or no previously appreciated connection to skin biology: megf6a, nrg2a, arhgef25b, and fkbp10b (Fig 1B  and 1B'; Fig 2A, 2B, 2E, 2F, 2K and 2L). In addition, we isolated GBT alleles of several known skin-related genes ( Table 1; Fig 1C-1E'; Fig 2). In zebrafish embryos, apical MFF basal keratinocytes express fras1, grip1, and hmcn1 [44,45], and MFF mesenchymal cells (FMCs) express msxc [39]. ahnak and capn12 have not yet been investigated in zebrafish, but have been detected in mammalian stratified squamous epithelia [70,91]. col4a4 encodes one (A4) of the six possible constituents (A1-A6) of collagen IV, which forms large networks integral to BMs. Zebrafish col4a4 expression has thus far only been analyzed via RT-PCR of whole-body RNA extracts, without corresponding spatial resolution [8]. However, in humans, cutaneous collagen IV largely employs subunits other than COL4A4 [67], and COL4A4 loss-of-function elongated ridge cells (red arrowhead) and a straight dermal space (ds). (J) A sibling embryo treated with 20 μM PD168393 displays basal bulging of MFF ridge cells towards the center the fin fold (red arrowheads) and a corresponding serpentine-like folding of the dermal space (ds), resembling the defects of the nrg2a mutant (compare with Fig 8D). (K, L) Whole-mount in situ hybridization (WISH) for egfra (K) and erbb3a (L) in wild-type embryos at 34 hpf (lateral views of tail) reveals epidermally expressed erbb3a transcripts (L), whereas egfra transcipts are absent in the epidermis, but present in the somites (K). (M-P) Anti-pAKT immunofluorescence of wild-type (M, N) and nrg2a mutant (O, P) embryo at 34 hpf; transverse sections through MER region of MFF, counterstained with DAPI (N, P). The wild-type embryo (M, N) displays pAKT localization in distal epidermal MFF cells (ridge cells and cleft cells; arrowheads), whereas pAKT levels in more proximal epidermal MFF cells are much lower. In addition, pAKT is localized at the tight junctions of the outer EVL (arrows), consistent with previously described roles of pAKT to phosphorylate tight junction proteins ZO-1 and Occludin, and to tighten the junctions [122]. In the nrg2a mutant (O, P), pAKT signals in ridge and cleft cells are strongly reduced (arrowheads), while pAKT signals at EVL tight junctions are unaltered (arrows). doi:10.1371/journal.pone.0130688.g009 Protein-Trap Skin Loci and a Novel Neuregulin 2a-ErbB Pathway mutations, such as Alport Sydrome, largely affect epithelia other than the skin [92]. Thus, zebrafish col4a4 mn0275Gt cutaneous expression offers a new tool for investigating evolutionary changes in tissue-specific expression of these important BM components and the impact of those changes on the biology of specific organs.
With fras1, grip1, and hmcn1, we identified GBT alleles of loci with documented functions in zebrafish MFF development and connections to human skin disease. In humans and mice, recessive loss-of-function mutations in FRAS1 and GRIP1 cause the rare, clinically overlapping congenital disorders Fraser Syndrome and Ablepharon Macrostomia Syndrome [47,48,64,65]. Both are characterized by malformations resulting from epidermal or epithelial blistering during fetal development due to compromised anchorage of epidermal and other epithelial basement membranes to underlying connective tissues such as the dermis. hmcn1 has recently been proposed as an additional Fraser Syndrome gene, though that hypothesis awaits confirmation [44]. fras1 mn0156Gt/mn0156Gt and hmcn1 mn0263Gt/mn0263Gt homozygotes fully phenocopied the respective ENU-induced mutants [44] (Fig 4A-4C). Consistent with previous MO knockdown results, grip1 mn0078Gt/mn0078Gt homozygotes lacked an overt skin phenotype, likely due to partial functional redundancy with zebrafish grip2 [44].
In contrast with the ENU-induced alleles, however, the fras1 mn0156Gt and hmcn1 mn0263Gt GBT alleles are revertible in a Cre-dependent manner (Fig 4D and 4E). That reversion capability offers important experimental advantages. For instance, coupling transgenic methods for temporally and/or spatially restricted Cre expression with GBT reversion would allow investigators to parse more finely the spatiotemporal requirements for Fras1 and Hmcn1 in MFF development. Similarly, tissue-specific Cre expression could be used to rescue the lethal craniofacial defects of fras1 mutants [93], thereby permitting investigations into possible later defects in fras1 mutants, including possible later consequences of the (non-lethal) epidermal blistering. In mammals, transient epidermal blistering in Fras1 mutant embryos is thought to abrogate crucial epithelial-mesenchymal interactions, leading to characteristic later defects such as syndactyly (fused digits) or cryptophthalmos (fused eye lids) [47][48][49]. Thus far, it has not been possible to investigate possible later consequences of embryonic epidermal blistering in the zebrafish model due to the lethal craniofacial defects. But as outlined above, such studies could now be carried out with the new revertible GBT allele. The 11 ZIP lines described here (Table 1; Fig 1B-1E'; Fig 2) comprise 3% of the original 350 GBT lines [17]. Continued expression cataloging and gene identification will build a comprehensive, in vivo spatiotemporal anatomical expression atlas for integument development and diseases, revealing relationships among anatomy, gene expression, and protein localization. That unique resource would be a valuable supplement to standard anatomical atlases [94,95] as well as a comparative resource for mammalian skin biology studies.
The nrg2a mn0237Gt allele Of the 11 identified loci, 3 gave a morphologically visible larval skin phenotype when their respective gene-breaking alleles were bred to homozygosity (fras1, hmcn1, nrg2a). The question of whether any of the other identified loci are required during later stages of skin biology requires further investigation since our analyses only extended through 120 hpf. Because the fras1 and hmcn1 mutant phenotypes had already been characterized, we focused on the nrg2a mutants, which displayed specific defects during MFF morphogenesis.
As with the other Neuregulin family members [96], zebrafish nrg2a exists in multiple isoforms due to differential promoter usage and alternative splicing. Previously reported isoforms [51] would have lacked the N-terminal sequences described here because exon 1A is a noncoding exon. Such 1A isoforms would therefore lack the first 145 amino acid residues of the 1B isoform and the first 37 N-terminal amino acid residues of the 1C isoform that we describe here (S2 Fig). We successfully amplified 1B and 1C transcripts from 5' RACE and RT-PCR analyses, but were unable to successfully amplify the 1A isoform. Those results suggest that the 1B and 1C versions are the predominant nrg2a transcripts generated during the stages relevant to MFF development. It should also be noted that Honjo et al. used MOs targeting an internal splice site present in all isoforms, rather than the 1A-specific translational start region, leaving open the question of which isoform(s) are required for dorsal root ganglia development [51].
Most important for our present work, however, is the fact that all three nrg2a alternative first exons, 1A-C, are located upstream of the GBT insertion site that generates the nrg2a mn0237Gt allele. Consequently, nrg2a mn0237Gt/mn0237Gt mutants have truncated nrg2a transcripts which lack the sequences encoded by exon 2 onwards due to the GBT cassette's transcription termination sequence (Fig 1A). Accordingly, the 1B and 1C transcripts give rise to Nrg2a-mRFP fusion proteins in which the N-terminal 138 or 29 amino acids encoded by exons 1B or 1C, respectively, are directly fused to mRFP. 1A transcripts are not translated at all because the endogenous translational start codon is localized in exon 2. Therefore, nrg2a mn0237Gt is most likely a null or near-null loss of function allele.
It is also noteworthy that out of all three N-terminal isoforms, only the longer and phylogenetically conserved 1B isoforms contain an N-terminal signal sequence involved in co-translational protein translocation to the endoplasmatic reticulum and subsequent secretion [97]. However, all endogenous isoforms share an internal transmembrane domain [51] which can target the proteins to the cell membrane [98] from where their biologically active ectodomains can be released via proteolytic processing [96]. Nrg2a mn0237Gt -mRFP fusion proteins lack that internal transmembrane domain because they only contain exon 1-encoded Nrg2 sequences. The absence of that domain in Nrg2a mn0237Gt -mRFP fusion proteins may contribute to the cytoplasmic and nuclear localization of the 1C fusion protein, which also lacks an N-terminal signal sequence, but instead contains at least one nuclear localization sequence (S2 Fig). This suggests that the unexpected distribution of the fusion proteins might be a special feature of the targeted locus and a consequence of the lack of crucial internal domains in the encoded proteins, thus demonstrating an expected limitation (loss of protein trafficking signals due to truncation) of GBT technology to recapitulate the subcellular distribution of the endogeneous protein counterparts. On the other hand, we cannot rule out that the RFP localization observed in nrg2a mn0237Gt embryos does reflect the actual subcellular distribution of the endogenous Nrg2 proteins. Indeed, nuclear localization has also been reported for several other RTK ligands, such as different Fibroblast growth factors (FGFs) and the Neuregulin relative EGF, as well as their receptors [99][100][101][102][103]. Studies with isoform-specific Nrg2a antibodies or full-length Nrg2-RFP fusion constructs will be needed to give more definitive answers.

Nrg2a/ErbB3 signaling is an essential regulator of ridge cells' apicobasal organisation and MFF morphogenesis
In mice, loss of Nrg2 leads to growth retardation and reduced reproductive capacity [104]. We were unable to address whether the loss of Nrg2a in zebrafish has similar consequences because nrg2a mn0237Gt/mn0237Gt mutants die during larval stages, well before the onset of significant somatic growth and sex differentiation. The reason for this larval death is currently unknown. However, the defects during MFF morphogenesis described here are most likely not the primary cause, because other mutants can survive in the complete absense of median fins (M.H., unpublished observations).
Altered MFF morphology similar to that of nrg2a mn0237Gt/mn0237Gt mutants has only been reported for one other mutant to date-erbb2 -/-, which harbors a loss-of-function mutation in EGF/NRG co-receptor ErbB2. However, the MFF morphological defects of erbb2 -/mutants have not yet been analyzed in additional detail [52]. The initial steps of MFF formation and elevation involve basal detachment of midline keratinocytes and their transition from a cuboidal to a wedged morphology, which is marked by shrinkage of their basal domains [37,38]. The distal-most basal keratinocyte in the tip of the elevating fold becomes the cleft cell; the two to three basal keratinocytes adjacent to each cleft cell become ridge cells. As we have documented here, ridge cells then transition a second time, re-elongating their basal domains at the expense of their apical domains to reverse their wedged morphology and return to their initial rectangular/cuboidal morphology (Fig 6A). Finally, ridge cells then become progressively flatter by extending their basal and apical sides at the expense of their lateral domains (Fig 6F and 6H).
This second phase of cell shape changes is affected in nrg2a mutants: nrg2a mutant ridge cells display disproportionately expanded basolateral domains and bulge basally, deforming the dermal space into serpentine shapes (Fig 6C, 6E, 6G and 6I; S3C-S3F Fig). In addition, ectopic and enlarged actinotrichia in the distal-most regions of the MFF (Fig 5H) suggested increased basal activity such as secreting ECM components into the dermal space. Together, these findings suggest that coordinating apical and basal extension during ridge cells' second shape change requires active Nrg2a -ErbB signaling which plays a "pro-apical" role during that transition by promoting maintenance of the apical domain and/or antagonizing basolateral epithelial domains to counterbalance the "basalization" of ridge cells that occurs during their transition from a wedged back to a rectangular shape. These shape changes during the second phase of MFF morphogenesis mainly occur in the distal-most MFF cells (ridge and cleft cells), whereas more proximal MFF basal keratinocytes remain rectangular during their initial recruitment into the MFF [37]. Accordingly, later MFF morphogenetic steps require less-pronounced shape changes by those proximal basal keratinocytes, which could explain why they are not affected in nrg2a mutants. Consistent with the spatially restricted defects in the mutant, distal MFFs in wild-type embryos show stronger endogenous nrg2a expression (Fig 5H and  5I). Ridge cells' higher pAKT levels (Fig 9M-9P) also suggest that the apical MFF receives stronger Nrg2a signaling than do proximal MFF cells. In contrast, expression of the likely Nrg2a receptor gene erbb3a is fairly homogeneous throughout all basal epidermal cells ( Fig  9K). In conclusion, we have good evidence that the distal restriction of the defects in the mutants is due to spatially restricted signaling in wild type embryos. Nevertheless, it remains unclear why only the ridge cells, but not the cleft cells, are affected in nrg2a mutants.
Nrg2a -ErbB signaling and Lgl2 display antagonistic effects during the separate processes of epidermal morphogenesis and homeostasis Identifying the molecular and cellular mechanisms underlying the pro-apical and/or anti-basal effects of Nrg2a-ErbB signaling will require additional investigation. The expression of erbb3a, but not egfra/erbb1, at the appropriate locations (Fig 9), together with previously reported erbb2 epidermal expression in zebrafish embryos and the requirement of erbb2-/-for proper MFF morphogenesis [52], suggests that Nrg2a may signal via Erbb2/3 receptor heterodimers, which would be consistent with biochemical findings [53]. In addition, reduced pAKT levels, but normal pERK levels in nrg2a mutant MFF keratinocytes (Fig 9; S5 Fig) points to Nrg2a -ErbB signaling via the PI3K -AKT signal transduction pathway, rather than the MAPK/ERK pathway, as has also found for Neuregulin signaling in other instances [53,85,87,88].
Our finding that concurrent loss of Lgl2 activity significantly alleviated MFF defects in nrg2a mutants suggests that Nrg2a's pro-apical effects extend to antagonizing Lgl2 pro-basal activity. As a homolog of the Drosophila cell polarity gene lethal giant larvae (lgl), lgl2 is a cell polarity regulator required for maintaining the basolateral domain in epithelial cells [89,90]. Likewise, the epidermal defects of zebrafish lgl2 mutants can also be interpreted in terms of loss of basolateral identity. Basal keratinocytes in lgl2 mutants not only lack hemidesomomes, basal domain junctions involved in attachment to underlying BM [54], but also display an even more pronounced loss of basal characteristics by undergoing EMT. Such events point to a tumor-suppressor role for Lgl2 [52]. Overactive ErbB signaling and its PI3K -AKT signal transduction branch, on the other hand, have well-known oncogenic effects, promoting EMT and hyperproliferation [105][106][107]. Thus, the roles of the Nrg2a -ErbB3 -AKT axis and Lgl2 during MFF morphogenesis identified in this study suggest that in addition to the established antagonistic relationship between ErbB signaling and Lgl2 during later phases of epidermal homeostasis [52] and carcinogenesis, the Nrg2a -ErbB3 -AKT signaling axis and Lgl2 have as-yet unappreciated antagonistic, and more moderate anti-or pro-basal effects, respectively, on keratinocytes during earlier steps of epidermal morphogenesis. To our knowledge, our data also provide the first indication that AKT/pAKT is involved in regulating apicobasal organization and epithelial cell polarity during the development of any species.
Strikingly however, despite their early and late antagonisms, Lgl2 and Nrg2a per se are uniquely required during different stages of epidermal morphogenesis and homeostasis. Lgl2 is required during late stages: epidermal defects in lgl2 mutants only become apparent between 108 and 120 hpf [52], and MFF formation is unaffected. In contrast, Nrg2a is only required early: mutants develop MFF defects between 30 and 36 hpf (Fig 6), and we did not detect any later-stage body epidermis defects corresponding to those of lgl2 mutants (data not shown). Yet concomitant loss of Lgl2 rescues the early MFF defects of nrg2 mutants (Fig 10), while concomitant loss of ErbB2 signaling rescues the late defects of lgl2 mutants [52]. Genetically, these findings argue that Lgl2 is epistatic to Nrg2a -ErbB signaling during early MFF morphogenesis: in double-deficient zebrafish embryos, the Lgl2 phenotype (no MFF defects) was dominant over the nrg2a phenotype. This suggests that Nrg2a -ErbB acts at least partly by blocking Lgl2 activity, and that the MFF phenotype of Nrg2a-deficient embryos is at least partially caused by Lgl2 overactivity. That proposition fits with previous findings that EGF treatment suppresses LGL2 transcription in human cell culture systems [108], but confirming it will require analyzing Lgl2 protein levels in nrg2a mutant embryos [108]. ErbB signaling, on the other hand, appears to be epistatic to Lgl2 during later epidermal homeostasis and carcinogenesis: in ErbB2/Lgl2-double deficient zebrafish larvae, the erbb2 phenotype (no EMT, no keratinocyte hyperproliferation) was dominant over the lgl2 phenotype [52]. Those findings, together with increased levels of ErbB signaling mediator pERK in lgl2 mutants, had led to the conclusion that Lgl2 acts by blocking ErbB signaling, and that the MFF phenotype of Lgl2-deficient embryos is at least partially caused by ErbB overactivity. Together, our data and those of Reischauer et al. [52] suggest the existence of a mutual negative feedback mechanism in which Lgl2 blocks ErbB signaling and vice versa. However, these opposing interactions occur at different developmental stages, and may involve different ErbB signaling molecules. Thus, early MFF ridge cell morphogenesis involves Nrg2a, most likely acting through ErbB2/3 and the PI3K -AKT signal transduction pathway (Fig 9), whereas later carcinogenesis involves the ERK MAPK signal transduction pathway [52], most likely activated by EGF and ErbB1/2.

Neuregulins may also be involved in mammalian epithelial morphogenesis and carcinogenesis
Although studies of Neuregulins have primarily addressed their roles during neuronal or neural crest development [109,110], expression of the NRG ligand family has also been described in epithelial contexts, most notably in mammary gland epithelium. Neuregulins and their ErbB receptors are expressed in murine embryonic mammary gland epithelium [111]. Expression of Neuregulins 1, 2, 3, and 4 has also been reported in breast cancer cell lines [112] and ductal carcinomas [113]. There is also some evidence that NRG2 has an epidermal role. A transcriptional profiling study of differential gene expression during human epidermal differentiation found that NRG2 was expressed in basal keratinocytes, while suprabasal cells expressed the known NRG2 receptor ERBB3 and its likely co-receptor ERBB2 [72]. In light of these findings, our data regarding the role of Nrg2a during zebrafish MFF development may also provide new insights into regulation of epithelial polarity and morphogenesis during mammalian epithelial development and carcinogenesis. Overall, GBT technology provides a valuable leap forward in using the vertebrate zebrafish model to identify important molecular players and to gain new insights into the genetic control of skin biology and disease.

Larval and adult care and maintenance
Adult fish and embryo care was performed according to standard protocols approved by the on-site Institutional and Public Animal Care and Use Committees. Wild-type stocks were derived from the offspring of adult zebrafish obtained from Segrest Farms (Segrest Farms, Florida, USA). Non-GBT transgenic marker lines used were enhancer trap ET37 [46,60], ubiquitously expressed membrane-bound EGFP Tg(Ola.Actb:Hsa.hras-egfp) vu119 [75,114] and Tg (krt4:GFP) gz7 [79].
in vivo imaging of live larvae Live larvae were imaged for mRFP fluorescence on an Axio ImagerZ.1 ApoTome microscope (Zeiss) with AxioVision software (Zeiss). Z-stack images were flattened using the AxioVision Multi-Image Projection (MIP) tool. SCORE imaging techniques [115] were used to hold live larvae and optimized for high-quality fluorescent imaging as follows: live larvae were transferred into 2.5-2.6% methylcellulose and drawn into borosilicate glass capillaries (World Precision Instruments WPI #1B120-3). 80% glycerol was used as the imaging medium between capillary and cover slip. Brightfield images of live larvae were obtained either by mounting the specimen in 1.5% methylcellulose and imaging with a Zeiss Axiophot microscope using the Zeiss AxionCam HRc camera and Axiovision software, or by SCORE imaging on a Zeiss Axio-plan2 microscope using a Canon PowerShot A640 camera with Remote Capture Task and Canon Image Browser software.

Cre mRNA phenotype rescue
Cre-mediated excision of GBT mutagenic cassettes was performed as previously described [17]. fras1 and hmcn1 reversion experiments were scored at 48 hpf; nrg2a reversion experiments were scored at 96 hpf. Results were graphed with PRISM software (Graphpad); p-value and significance were calculated with t-tests.

Sectioning of larvae
Larvae for cryosections were fixed in either 4% PFA or EAF (40% ethanol, 5% acetic acid, 4% formaldehyde in PBS) for 3h at RT, washed several times with PBST, embedded in 15% sucrose/1% agarose in PBS, incubated in 30% sucrose/PBS overnight and mounted in tissue freezing medium (Leica). 10 or 20 μm sections of tails were obtained using a Leica CM1850 cryostat. Larval heads were used for genotyping.

EGFR/ErbB inhibitor treatment
A 10 mM stock of PD168393 (EMD Millipore Calbiochem #513033) was prepared with DMSO and diluted in embryo water to concentrations of 1 and 20 μM. 30 to 40 embryos were randomly selected for each treatment group and placed into Petri dishes (35-1007, BD Falcon). To examine the effect of inhibitor on the earliest stages of documented MFF phenotype through early pectoral fin fold development, PD168393 treatment and control groups were incubated from 24 hpf through 52 hpf. Larvae were gently transferred into fresh embryo water after incubation. Blinded treatment and control groups were and scored at 96 hpf with a Stemi 2000-C (Zeiss) dissecting microscope. Scoring was performed at 96 hpf to verify that inhibitorinduced morphological changes persisted beyond the end of the treatment window itself. Results were graphed with PRISM software (Graphpad); p-value and significance were calculated with Chi-square analysis. For TEM analysis samples, 20 μM-treated wild-type embryos were randomly selected immediately following incubation and removed into Trump's fixative, as were age-matched sibling controls.

Larval preparation for Transmission Electron Microscopy (TEM)
Embryos younger than 48 hpf were sorted by mRFP fluorescence. Larvae 48 hpf or older were sorted by MFF phenotype and fluorescence. Sorted or selected embryos were transferred into glass Petri dishes for all subsequent processing (Corning) and fixed in Trump's fixative (4% formaldehyde, 1% glutaraldehyde in 0.1M phosphate buffer (pH 7.2)) for a minimum of 24 hr. Following fixation, zebrafish were secondarily fixed with 1% osmium tetroxide and 2% uranyl acetate, dehydrated through an ethanol series and embedded into Embed 812/Araldite resin. After a 24 hr polymerization at 60°C, ultrathin sections (0.1 μm) were stained with 2% uranyl acetate and lead citrate. Micrographs were acquired using a Jeol J1400 transmission electron microscope (JEOL, Inc., Peabody, MA) at 80 kV equipped with a Gatan digital CCD camera (Gatan, Inc., Warrendale, PA) and Digital Micrograph software. Minor adjustments were made in ImageJ or Keynote.
Supporting Information S1 Fig. mRFP fusion protein distribution in GBT lines recapitulates the endogenous expression patterns of arhgef25b, fkbp10b, and msxc. Panels (A, A', C, C', E, E') show the same in vivo fluorescence images of mRFP localization in GBT lines arhgef25b mn0175Gt (A, A'), fkbp10b mn0316Gt (C, C') and msxc mn0245Gt (E, E') as in Fig 2E, 2F, 2K, 2L, 2M, and 2N, respectively. Panels underneath show images of wild-type embryos of comparable stages (indicated in hpf) and orientations following whole-mount in situ hybridization (WISH) with arhgef25b (B, B'), fkbp10b (D, D'), or msxc (F, F') RNA probes. mRFP localization in the epidermis covering the yolk in the arhgef25b mn0175Gt gene-break allele (A) corresponds to endogenous arh-gef25b gene expression (B). Both arhgef25b mn0175Gt mRFP localization and endogenous arhgef25b gene expression are also observed in the MFF (A', B'). Furthermore, just as the genebreak alleles fkbp10b mn0316Gt (C, C') and msxc mn0245Gt (E, E') show mRFP localization in fin mesenchymal cells (FMCs), fkbp10b and msxc genes show endogenous FMC expression (C-F'; also compare with MsxC mn0245Gt -mRFP fusion protein localization in Fig 3E-3H). mRFP localization to the pectoral fin buds observed in the fkbp10b mn0316Gt allele (C) parallels endogenous fkbp10b expression in the pectoral fin buds (D). fkbp10b also shows expression in the posterior region of the notochord (D'). The weaker WISH signal in anterior (earlier specified) regions of the notochord compared to the mRFP signal (C') points to transient expression of the endogenous gene in notochord cells, and higher stability of the GBT-generated mRFP fusion protein than the endogenous transcript. The endogenous msxc gene is also expressed in the maculae of the inner ear and in the pectoral fin buds (F), confirming that the msxc mn0245Gt GBT allele and its resulting mRFP fusion protein also recapitulate endogenous expression in tissues other than the skin (E). (TIF)