Danio quagga and D. kyathit pigment pattern phenotypes overlap those of wild-type and mutant D. rerio

To better understand the cellular bases of pigment pattern differences between D. quagga and D. kyathit we first examined complements of pigment cells in young adult fish. We found that Danio kyathit and D. quagga had the same classes of pigment cells as wild-type D. rerio and these cell types were arranged similarly relative to one another in all three species: dark pattern elements contained melanophores and bluish iridophores and light pattern elements contained xanthophores and gold iridophores (Figs 1 and S1).

Because a proper study of pattern evolution requires a system that can address quantitative differences, we sought metrics that would describe patterns in Danio and be robust across species, stages, and lighting conditions. We focused on melanophores because each species has dark pattern elements that include these cells, and because melanophores are more easily documented than xanthophores or iridophores. Moreover, melanophore locations are correlated, or anti-correlated, with the distributions of other pigment cells owing to interactions among pigment cell classes essential for pattern formation [6,7,18,24–26,31,50], suggesting that documentation of just melanophore distributions could define a morphospace useful for describing pattern variation across species in which these interactions and relative cellular arrangements might be conserved. We therefore binarized patterns (melanophore element or not) and defined metrics to represent pattern phenotypes. These included estimates of pattern variation along dorsoventral (DV) and anteroposterior (AP) body axes, and the ratio between these metrics, most easily represented on a log scale, log₂(DV:AP variation). To estimate variation along the DV body axis, we calculated the average grey value for each row (of individually black or white) pixels, representing a single anteroposterior transect across the body. From these average grey values across many rows (transects) in each individual, we then calculated standard deviations, and used these as our estimate of DV variation. We performed the reciprocal calculations to obtain for each individual an estimate of AP variation (i.e., standard deviation of averaged pixel values across DV column transects). Patterns or pattern elements that are nearly uniform along an axis will approach a value of 0, whereas increasing heterogeneity along an axis will yield increasingly large estimates of variation (Figs 3A and S2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. A morphospace describing pattern variation.

(A) Schematic illustrating where idealized patterns fall in a morphospace defined by DV (dorsoventral) and AP (anteroposterior) pattern variation of melanized elements. Diagonals indicate different log₂ values for the ratio of DV variation: AP variation. (B) Mapping of pattern phenotypes in this morphospace revealed overlap of striped D. rerio, D. quagga, and D. nigrofasciatus, and their difference in morphospace from spotted D. kyathit or D. tinwini. Additional species occupied distinct regions of morphospace as well. Images of other Danio species have been published previously [4,16,19]. Points denote individuals and shapes denote observed phenotypic range. (C) Several single locus mutants of D. rerio lie between D. quagga and D. kyathit, or in the vicinity of D. kyathit. These phenotypes arise from null or hypomorphic alleles [22,27, 37,38,42,51,62–64,76,85,86,98–101], though aqp3a results from an activating mutation [53], and some mutants have yet to be characterized molecularly (cezanne, chagall, duchamp, ocelot) [8,84]. Points for each mutant denote individuals, or averages when images of multiple individuals were available.

https://doi.org/10.1371/journal.pgen.1009364.g003

We found overlap in pattern morphospace between three striped species, D. rerio, D. quagga, and D. nigrofasciatus, and two spotted species, D. kyathit and D. tinwini (Fig 3B). A species with three relatively broad and diffuse stripes, D. kerri, fell in between, whereas a species with particularly small and symmetrical light spots on a dark background, D. margaritatus, was separated from the other spotted danios. Species with vertical bars—D. aesculapii, D. choprae and D. erythromicron—occupied a distinct region of morphospace with higher AP variation. Finally, a species with a mostly diffuse, nearly uniform pattern, D. albolineatus, mapped close to D. kyathit. This latter grouping, in which two visually disparate patterns nevertheless occupied similar positions in morphospace, suggests that additional metrics ultimately will be needed to describe the broader range of phenotypes across the genus.

This pattern morphospace also allowed us to consider single locus pigment pattern mutants of D. rerio, representing a range of patterns, some of which have spots rather than stripes. Many of these mutants were situated either between D. rerio or D. quagga and D. kyathit, or, in the case of some spotted mutants, in the vicinity of D. kyathit (Fig 3C). These findings suggest the hypothesis that patterns of D. quagga or D. kyathit are separated by just a single mutational step, and identify a priori candidate genes that might contribute to the difference between these species.

Differentiation timing and morphogenesis differ during ontogeny of striped D. quagga and spotted D. kyathit

To understand the ontogeny of pattern differences between D. quagga and D. kyathit, we followed individual cell behaviors across daily image series in individual fish from larval through late juvenile stages. In comparison to D. quagga, melanophore numbers in D. kyathit increased more rapidly, and these cells reached higher total numbers (Fig 4A and 4B). This situation implies that spots in D. kyathit are not simply a consequence of fish having insufficient numbers of melanophores to fill a striped pattern; this was surprising in comparison with spotted mutants of D. rerio, in which melanophores are often fewer (e.g., ltk, gja5b, igsf11) [24,39,51] or of similar abundance (tjp1a, aqp3a) [52,53] to that of wild-type. In contrast to D. quagga, melanophores of D. kyathit were also more densely packed, achieving lower nearest-neighbor distances and higher densities within melanized pattern elements overall (Figs 4C and S3A). As pattern formation proceeded, new melanophores appeared as lightly melanized cells unpaired with other cells (Fig 4A), consistent with de novo differentiation from unmelanized precursors as occurs in D. rerio, rather than division of existing melanophores [23,54]. Melanin contents of melanophores, as inferred from two-dimensional areas of contracted melanin granules [54], did not differ between species during pattern ontogeny but were ultimately greater in adult D. kyathit than D. quagga (S3B Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Pigment pattern development in D. quagga and D. kyathit.

(A) Representative individuals imaged repeatedly during adult pattern formation. Images are aligned to show corresponding regions and rescaled to control for overall growth (frames and regions are selected from S1 Movie). Standard length (SL in mm) serves as a proxy for developmental stage as the relationship between development rate and days post fertilization (dpf) depends on rearing conditions [3]. By 7.0 mm SL, adult melanophores had started to develop in both species. Newly differentiating melanophores are marked by dashed squares, and the same cells are marked by dashed circles of the same color in subsequent images. Adult melanophores ultimately coalesced into dark pattern elements with movements by individual cells and rearrangements among cells particularly apparent in D. kyathit. As in D. rerio, melanophores were occasionally lost in both species (e.g., dashed triangle in D. kyathit) [25,99]. Densely arranged iridophores of the primary interstripe were evident by 7.0 mm SL in both species. Subsequently developing iridophores of stripes or spots (yellow arrowheads) appeared earlier in D. kyathit (7.0 mm SL) than D. quagga (9.2 mm SL), as did iridophores of secondary interstripes or “interspots” further ventrally (yellow arrowheads; 10.6 and 9.2 mm SL, respectively). To facilitate cell counting, fish were treated prior to imaging with epinephrine to contract melanin granules towards cell centers. Images shown are representative of 4 individuals of each species imaged throughout the stages of early adult pigment pattern development. (B–E) Pattern metrics during development of fish imaged repeatedly (splines over points), and in young adults and backcross progeny (most 22–25 mm SL, 4–5 months post-fertilization; see main text and S3 Fig). Individual points for melanophore nearest neighbor distances (C) represent median values calculated for all melanophores examined within each individual fish at time points with ten or more melanophores. *, P = 0.029 in B (F_1,21 = 5.52); ***, P<0.0002 in C (F_1,21 = 19.83), P<0.0001 in D (F_1,21 = 421.04), P<0.0001 in E (F_1,21 = 217.11). Scale bars, 100 μm.

https://doi.org/10.1371/journal.pgen.1009364.g004

Melanophore patterns emerged gradually. In both D. quagga and D. kyathit, a primary interstripe of iridophores formed near the horizontal myoseptum, followed by dorsal and ventral primary melanophore stripes (Fig 4A). In D. quagga, secondary melanophore stripes then appeared, but in D. kyathit, primary stripes subdivided into spots and additional spots emerged dorsally and ventrally, generating a much larger number of melanized pattern elements overall (Fig 4D and S1 Movie). Ontogenetic changes in pattern were also evident in DV and AP variation metrics (S3C and S3D Fig) and in the ratio of these metrics (Fig 4E). Although total areas of the flank covered by melanized pattern elements did not differ during larval stages of adult pattern formation, D. kyathit adults had less total area covered by melanized pattern elements than D. quagga, despite the former having more melanophores (S3E Fig). The higher density of melanophores achieved by D. kyathit was manifested at the level of individual cell behaviors, as development of spots in this species was accompanied by more pronounced movements of melanophores than were observed in D. quagga (S4 Fig; S1 Movie).

Locations of melanophores in D. kyathit and D. quagga presumably depend on interactions with iridophores and xanthophores as in D. rerio: interstripe iridophores promote the localization of melanophores in stripes and stimulate the development of interstripe xanthophores, which themselves help to organize melanophores; additional iridophores differentiate within stripes but are not known to be essential for stripe formation [6,24–26]. In both D. quagga and D. kyathit, onset of adult pattern development was marked by differentiation of iridophores in a nascent primary interstripe near the horizontal myoseptum (Fig 4A), as in D. rerio. Subsequent iridophore development differed between species, however, as new iridophores outside of the primary interstripe—contributing to dark pattern elements (stripes or spots) and additional light pattern elements (interstripes or “interspots”)—appeared earlier in D. kyathit than D. quagga (S6 Fig). Densely arranged interspot iridophores of D. kyathit developed over a broader area of the flank than interstripe iridophores of D. quagga, and as these cells differentiated, light interspot elements appeared in regions initially harboring only melanophores (S1 Movie). The inability to reliably distinguish individual iridophores in brightfield images precluded quantifying their numbers.

Adult xanthophores first became apparent at similar stages in the two species [S6A Fig; standard length (SL) of first appearance, F_1,6 = 2.36, P = 0.1754]. During ontogeny, interstripe elements of D. quagga and interspot elements of D. kyathit became populated with xanthophores, though these cells were somewhat fewer and less densely arranged in D. quagga. By adult stages, however, xanthophores in D. quagga were somewhat more numerous than in D. kyathit (S3F Fig). Visible xanthophore pigment was similar between species during larval through juvenile stages but slightly greater in D. kyathit adults than D. quagga adults (S3G Fig).

Finally, because dynamics of body growth relative to pigment cell development could affect pattern [55,56], we examined two measures of body dimensions: standard length (SL) and height of the body at the anterior margin of the anal fin (HAA) [3]. During pattern ontogeny, D. kyathit were ~6% larger in both dimensions than D. quagga but adult sizes were not significantly different from one another (S3H and S3I Fig). Though faster early growth could, in principle, cause a pattern to be broken into smaller elements (e.g., stripes becoming spots), the ~20% greater complement of melanophores in D. kyathit compared to D. quagga, and differences in melanophore morphogenesis as well as iridophore and xanthophore development, make it unlikely that differences in body size or growth trajectory play major roles in determining pattern differences between these species.

Together, these findings reveal several distinct cellular phenotypes associated with spotted and striped patterns (Table 1) consistent with species differences depending on variation at one or a few highly pleiotropic loci, or perhaps a larger number of loci each having more specific effects.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Qualitative summary of phenotypes.

Interpretations derived from numerical and statistical analyses in Figs 4, S3, S4, S6 and S1 Movie.

https://doi.org/10.1371/journal.pgen.1009364.t001

Genetics of pigment pattern variation in D. quagga and D. kyathit

Morphometric analyses showed that D. quagga and D. kyathit have phenotypes similar in morphospace to wild-type zebrafish and some single locus mutants of zebrafish, respectively (Fig 3C), suggesting the hypothesis of a simple genetic basis for the pattern difference between species. The phylogenetic proximity of D. quagga and D. kyathit [16] raised the possibility that interspecific mapping crosses could be used to test this hypothesis. We therefore generated F1 hybrids, finding them to be robust and fertile, and to have pattern phenotypes more similar to D. quagga than D. kyathit (Fig 5A). We then backcrossed F1 hybrids to D. kyathit to assess genetic correlations among phenotypes and to map genomic regions associated with pattern variation for establishing a lower bound on the number of contributing loci.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Hybrid and backcross progeny phenotypes of D. quagga and D. kyathit.

(A) Pattern variation morphospace showing locations of parental D. quagga and D. kyathit, F1 hybrids, and backcross progeny (BC_a). Images of D. quagga and D. kyathit patterns are details of the same individuals shown in Fig 2 and the F1 hybrid shows a typical pattern in this cross. (B) Principal components (PC) of phenotypic variation in BC_a progeny with percent variance explained noted in parentheses. Melanized element count and perimeter refer to entire spots or stripes. Melanized element coverage refers to total area covered by melanized elements relative to total imaged flank area. Likewise, melanophore and xanthophore densities were calculated relative to total area. Melanophore and xanthophore pigment refer to areas of contracted pigment granules, rather than entire cell sizes. For sex, arrow points towards values typical of males. NN, nearest neighbor.

https://doi.org/10.1371/journal.pgen.1009364.g005

Using a family of backcross progeny (BC_a, n = 158; Fig 5A), we examined correlations among metrics described above and additional metrics, which revealed four principal components (PCs) that explained ~73% of total phenotypic variance (Fig 5B). PC1 was most closely associated with overall pattern, as individuals having high scores in PC1 resembled D. quagga in overt phenotype and higher log₂(DV:AP variation) and DV variation, whereas individuals with low scores in PC1 resembled D. kyathit with higher AP pattern variation and melanized pattern element count. PC2 was associated with melanophore and xanthophore spacing and overall density, PC3 with melanophore element coverage and inferred melanin content, and PC4 with a deeper body (ratio of HAA to SL) in females and a narrower body in males.

To learn whether one or several loci contribute to pattern variation in this cross, we used quantitative trait locus (QTL) mapping. Parental fish were not inbred, so this approach will identify chromosomal regions associated with variation between D. quagga and D. kyathit, and also variation segregating within the D. kyathit stock to which F1 hybrids were backcrossed (Fig 6A). We genotyped siblings by RAD-tag sequencing and isolated 2,458 phased, polymorphic markers flanking SbfI restriction sites across the genome [57,58]. We then mapped these sites to the D. rerio genome using parameters to account for sequence divergence between closely related species [59].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Chromosomal regions associated with pattern and size metrics in backcross individuals.

(A) Crossing schemes for mapping variation in two independent families, BC_a and BC_b (see main text). (B–C) Log odds scores of example QTL (BC_a) for melanophore spacing (nearest neighbor distance, log₂(DV:AP variation), and number of melanophore elements across 25 chromosomes. Variation segegrating between species in blue and within D. kyathit in red. Dashed lines indicate q = 0.05 false discovery thresholds and asterisks indicate peak LOD scores. (E) Summary of regions associated with pattern variation and size in the same family referenced in B–C (BC_a), and defined by F_ST values for phenotypically extreme individuals from a second family (BC_b; see main text). Regions highlighted have peaks (q<0.05, LOD≥4) marked by vertical bars and widths defined by positions 1.5 LOD lower than peaks. Percents of phenotypic variance explained by QTL (%var) are indicated at right. Small, vertical hash marks on chromosomes denote BC_a maternal D. kyathit marker loci and positions of several genes having roles in pigment pattern formation in D. rerio are indicated as well.

https://doi.org/10.1371/journal.pgen.1009364.g006

These analyses revealed several genomic regions that contributed to phenotypic variation, allowing us to reject the hypothesis that a single major effect locus is responsible for the species-level transition between stripes and spots. Three QTL were linked to variants present in the hybrid parent, representing alleles that differ between the D. quagga and D. kyathit backgrounds: melanophore nearest-neighbor distance mapped to a broad region of Chr15 (peak LOD = 6.8; q<0.05; Fig 6B and 6E); log₂(DV:AP variation) mapped to a broad region of Chr18 (peak LOD = 4.7 with marginal significance at q<0.05; Fig 6C and 6E); and PC3—associated with melanized pattern element coverage and melanin content—mapped to Chr23 (peak LOD = 5.0; q<0.05; Fig 6D and 6E). Four additional QTL were linked to variants segregating within the D. kyathit background (Figs 6E and S6). These were associated with variation in melanophore density (Chr8), melanized area per cell (Chr8), and xanthophore density (Ch11). An additional QTL (Chr20) was associated with log₂(DV:AP variation), DV variation, number and perimeter of melanized pattern elements, and PC1 (Figs 6C and 6E and S6).

Several of these chromosomal regions harbor known pigment pattern genes. For example, the very broad region associated with log₂(DV:AP variation) between D. quagga and D. kyathit on Chr18 includes slc24a5 and mc1r, though functions for these genes in melanophore differentiation are not obviously relevant to pattern element shape [60,61]. A region associated with several metrics on Chr20 includes kita, required for melanophore development, whereas the region associated with PC3 on Chr23 includes edn3b and alkal2a, required for iridophore development and thus indirectly for melanophore pattern [17,20,22,39,62–64]. Spontaneous alleles of kita and edn3b or their homologues have been found affecting pigmentation in natural or domesticated populations of other species [65–70], though these analyses tend to argue against a major effect role for kita between D. kyathit and D. quagga, as variation identified at this locus segregated within the D. kyathit background rather than between D. kyathit and D. quagga.

Among phenotypes with significant peaks, log₂(DV:AP variation) was associated with the genotype inherited from the hybrid sire on Chr18 and the D. kyathit dam on Chr20. To determine if these regions represented variants with additive effects or an epistatic interaction, we compared individuals with each of the four possible genotype combinations (Fig 7A, upper plot). Three genotype combinations had phenotypes midway between the two parental species whereas one combination was associated with striped patterns more similar to D. quagga, indicating a strong epistatic interaction between these QTL. This epistasis was manifested in morphospace as a shift upward toward the region occupied by D. quagga and other striped species (Fig 7B). Epistasis between Chr18 and Chr20 extended to other phenotypes as well. For example, after accounting for the minimum three stripes, individuals with both stripe-associated alleles averaged about half as many additional melanized pattern elements as siblings with other genotypes (~2.8 vs. ~5.5, respectively; Fig 7A, lower plot).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Epistatic relationships between QTL influence pattern phenotype.

(A) Genotypes at Chr18:5246981 and Chr20:23025453 linked to variants affecting log₂(DV:AP variation) and melanophore element count in BC_a progeny. Genotypes are listed as maternally-inherited D. kyathit allele followed by paternally inherited allele with phenotype-associated alleles underlined. Bars indicate medians ± interquartile range. Overall analyses of variance for allelic combinations in BC_a progeny: upper, F_3,154 = 7.26, P<0.0001; lower, F_3,154 = 10.07, P<0.0001). Means not significantly different from one another (P>0.05) in post hoc Tukey-Kramer comparisons are indicated by shared letters above data points. (B) Three combinations of Chr18 and Chr20 alleles overlapped in morphospace. When striped alleles on Chr18 and Chr20 were together, a phenotype significantly more similar to that of D. quagga developed. Ellipses indicate 95% confidence intervals of means. (C) Genotype at Chr7:27412951 impacts log₂(DV:AP variation) and melanophore element count in BC_b progeny selected for striped and spotted phenotypes. Analyses of variance: left, F_1,34 = 9.30, P = 0.0044; lower, F_1,34 = 8.91, P = 0.0052. (D) Morphospace position was impacted by variants linked to maternally inherited D. kyathit alleles at Chr7:27412951. Most siblings with the A allele had patterns similar to D. quagga whereas progeny with the G allele were more similar to D. kyathit. (compare with representative parental species values in A). Statistical analyses in A and C used DV:AP variation and ln(melanophore elements), which stabilized among-group differences in residual variance evident in original values.

https://doi.org/10.1371/journal.pgen.1009364.g007

Finally, to further assess roles for inter-species QTL on Chr15 and Chr18, and to potentially identify other QTL, we generated independent F1 hybrids and a second backcross family (BC_b) to D. kyathit (Fig 6A). To account for any parent-of-origin effects, we used a hybrid female in BC_b as opposed to the hybrid male used in BC_a. We reasoned that major-effect loci should be identifiable by genotyping backcross progeny that most resembled the parental phenotypes. Extreme sampling approaches can use fewer individuals while offering similar power as larger scale experiments that incorporate many uninformative intermediate phenotypes [71]. We therefore generated 457 full siblings and selected individuals representing opposite ends of the pattern spectrum (n = 18 striped and n = 18 spotted, 3.5% at each extreme) allowing calculation of fixation indices, F_ST, across the genome to assess relative allele frequencies between and within phenotypic classes (Fig 7C and 7D).

Remarkably, analyses of BC_b individuals failed to recover any of the genomic regions identified in the first cross and instead revealed broad segments of elevated F_ST on Chr7, Chr9, and Chr19 (Figs 6E and S6). Inspection of haplotypes revealed that alleles driving the extreme F_ST values were transmitted both by the hybrid dam (Chr7) and the D. kyathit sire (Chr9 and Chr19), indicating that both interspecific and intraspecific variants affected pattern in this second backcross, as in BC_a. QTL-linked variants on each of these three chromosomes affected both log₂(DV:AP variation) and melanophore element count (Figs 7C and 7D and S8). All three stripe-associated alleles co-occurred in 8 of 18 striped individuals, whereas three spot-associated QTL co-occurred in 10 of 16 spotted individuals suggesting that our extreme phenotype sampling approach selected for combinations of multiple variants impacting pattern.

Though genes previously associated with pigment pattern were not found within the region of high F_ST on Chr9, near the peak F_ST values on Chr7 and Chr19 were located tight junction protein 1a (tjp1a) and gap junction protein alpha 4 (gja4), respectively. The products of these genes interact physically and D. rerio mutants of each have spotted phenotypes [52,72] (Fig 3C), making them viable candidates for contributing to the naturally occurring and complex genetics of pattern variation across D. quagga and D. kyathit.

Ethics statement

All animal research was conducted according to federal, state and institutional guidelines and in accordance with protocols approved by the Institutional Animal Care and Use Committee of University of Washington (4094–01) and the Animal Care and Use Committee of University of Virginia (4170). Anesthesia and euthanasia used MS-222.

Fish stocks and rearing conditions

D. quagga and D. kyathit were obtained from the pet trade and reared subsequently in the same conditions and recirculating water system used to maintain D. rerio (~28°C; 14L:10D). Though DNA sequences of D. quagga type specimens have not been described, fish used in this study were morphologically indistinguishable from published photographs of D. quagga [48] (S1A Fig). Fish were crossed by pair-wise natural matings to produce larvae of each species for analysis, or F1 progeny, which were then backcrossed to other D. kyathit to produce crosses for genetic mapping (BC_a and BC_b). Adult fish were anesthetized prior to imaging and fin clipping for subsequent DNA extraction. Live D. rerio used for imaging (Fig 1) or morphospace analysis (Fig 3C) were wild-type strain NHGRI-1 [83], cezanne^utr17e1 [84], chagall^vp36rc1 [84], edn3b^vp30rc1 [17], duchamp^utr19e1 [49], igsf11^utr15e1 [51], kita^b5 [62], leo^t1 [76] ltk^j9s1 [85], mitfa^w2 [38], ocelot^vp37rc1, tspan36^wpr21e1 [27]. Phenotypes of additional D. rerio mutants were analyzed from published images: alkal2a^ya340, alkal2a^ya342 [64], aqp3a^tVE1 [53], asip1 [42], gdf6a^s327 [37], gja4 [29], kitlga^tc244b [22], srm^t26743 [86], tjp1a^twl1 [52].

Phenotyping of adults and longitudinal imaging during development

Anesthetized fish were imaged using a Zeiss AxioObserver inverted microscope (BC_a), a Zeiss AxioZoom stereo zoom microscope, or a Nikon D810 digital single lens reflex camera with 105 mm Nikkor macro lens as appropriate for the size of the fish (BC_b). Images used for BC_a and daily image series were recorded after first treating fish with 10 mM epinephrine to contract pigment granules towards cell centers. Morphometrics (standard length, height at pelvic fin, etc.) were measured in ImageJ. Images were then isometrically scaled and aligned in Adobe Photoshop 2018 to ensure that subsequent analyses used the same region across fish and developmental stages. The region of interest was centered slightly anterior to the vent and extended dorsoventrally to encompass the three melanophore stripes in a mature adult zebrafish. Regions of interest were exported as RGB images and melanophore location and size were measured in ImageJ [87] across eight body segments. For repeated imaging of pigment pattern development, the region of interest was truncated anteroposteriorly (relative to adult images) to allow for cell tracking and quantification at earlier developmental stages.

To segment melanophores from RGB images, we identified objects of the appropriate color and size after removing lighting artifacts and background. To reduce background variability from lighting artifacts such as reflection and iridescence from iridophores, we removed bright outliers within a 25-pixel radius. We then isolated the red channel to better distinguish black melanophores from xanthophores and the underlying skin and muscle (which has a light pink color). We then applied automatic local thresholding using the Sauvola method (originally designed to detect numerous black text characters on a light page) with radius set to 25 and the k parameter set to 0.34. Segmented objects with area between 100 and 1500 pixels were retained as melanophores. Yellow/orange xanthophores with high red values and low blue values were similarly segmented after subtracting the red channel from the blue channel, using the Sauvola method with the same parameters, and selecting objects between 10 and 200 pixels. Improperly segmented cells including uncontracted melanophores were manually corrected in ImageJ and replaced with circles of diameter 15 pixels. Melanophore nearest neighbor distances were calculated using the spatstat package in R [88].

For measurements of melanophore movement, segmented melanophores from eight daily image series (n = 4 D. quagga; n = 4 D. kyathit) spanning 18 to 45 dpf were used. Melanophores were converted to points of uniform size to allow image registration optimization in ImageJ using the Linear Stack Alignment with SIFT plugin with a rigid transformation, inlier ratio of 0.20, and default parameters for all other values. Aligned melanophores were tracked using the TrackMate v5.2.0 plugin [89] in ImageJ allowing for up to 100 px displacement over 24 hours. Cells tracked across seven or more days with no more than one missing day were included in analyses.

Pigment patterns were quantified using binary images with each pixel segmented according to whether it was part of a melanophore pattern element. Melanophore elements of epinephrine-treated fish were defined as all pixels within 20 pixels of the center of a melanophore identified using the previously described segmentation parameters. This distance approximates the average distance between nearest neighbor melanophores measured previously. To obtain binarized pigment patterns for fish not treated with epinephrine, regions of interest were thresholded manually to contain similar amounts of segmented pixels as epinephrine-treated images. A number of images, mostly drawn from previous publications describing species or mutants, required manual segmentation due to image compression for publication or uneven lighting. Pattern morphospace location was defined by horizontal and vertical variation, which were calculated as the standard deviation of the 8-bit gray value profiles along the respective axis of each binary image. When possible, images of multiple individuals were used to account for phenotypic variability present within species or between individuals mutant for the same locus.

First appearance of adult xanthophores was determined by inspection of color brightfield images of daily image series for the presence of orange carotenoids pigmentation, evident in fish that had been treated with epinephrine. Appearances of iridophores outside of primary stripes were recorded when sparsely arranged iridophores were first evident in brightfield micrographs within the prospective ventral primary stripe and when first evident where near the ventral edge of the flank where prospective ventral secondary interstripe or interspot elements first form.

For display, images were color-balanced and levels adjusted in Adobe Photoshop 2020. In Fig 1, backgrounds of whole fish were partially desaturated for color to better represent the natural appearance of the pigment pattern.

Genomic DNA extraction and genotyping

DNA was extracted from adult fin clips using a DNeasy 96 Blood & Tissue Kit (Qiagen) or a MagAttract HMW DNA kit (Qiagen). DNA concentrations were measured with a Qubit 2.0 fluorometer and standardized prior to genotyping. Fish for each mapping cross were genotyped using RADseq. The BC_a cross used single digest RADseq with high fidelity SbfI (NEB) as previously utilized with D. kyathit [16]. Libraries were sequenced on two high output lanes of an Illumina NextSeq using paired end 75 bp reads. The BC_b cross employed double digest RADseq with high fidelity PstI (NEB) and EcoRI (NEB) as described previously [90] and was sequenced on one lane of an Illumina HiSeq with paired end 150 bp reads. Reads were aligned to the zebrafish genome (GRCz10) using the Burrows-Wheeler Aligner and 19-base kmers (bwa mem -k 19). Reads with multiple best mapping locations and reads mapping to annotated repetitive elements were excluded from further analyses. Mapping to the zebrafish genome allowed the use of existing annotations for inferring genes likely to be present within intervals defined by QTL. This approach is reasonable in the present context as the karyotype of 25 pairs of metacentric or submetacentric chromosomes is largely conserved across diploid species within Cyprinidae, which includes genus Danio, and comparison of zebrafish to common carp Cyprinus carpio revealed largely conserved intrachromosomal structure between even these divergent taxa [91–93]. Nevertheless, testing of individual candidates identified in this manner warrant further validation of candidate gene linkage with QTL as well as finer scale analysis of synteny relationships across these species. Genotypes were called using the ref_map.pl script from the Stacks pipeline with default parameters [94,95]. Genomic regions in the BC_b cross with loci having significantly different genotype frequencies were identified from F_ST values calculated by the Stacks pipeline from genotype calls.

QTL mapping and statistical analyses

For the BC_a cross, principal components of the mappable phenotypes were calculated in R using prcomp() from the ‘stats’ package. Genotypes of progeny were phased based on parental genotypes using Joinmap (Kyazma). Loci with any missing genotypes were excluded from analyses. QTL mapping was then performed using the ‘qtl2’ package in R [96]. Phased genotypes were mapped separately based on parent of origin in the BC_a cross. QTL were identified by Haley-Knott regression using the scan1(), scan1perm(), and find_peaks() functions in the ‘qtl2’ package. QTL windows were defined using a 1.5 LOD drop from the most-associated marker.

Parametric and non-parametric analyses of other quantitative data were performed in JMP 14 (SAS Institute, Cary NC).