Capturing and analyzing pattern diversity: an example using the melanistic spotted patterns of leopard geckos

Animal color patterns are widely studied in ecology, evolution, and through mathematical modeling. Patterns may vary among distinct body parts such as the head, trunk or tail. As large amounts of photographic data is becoming more easily available, there is a growing need for general quantitative methods for capturing and analyzing the full complexity and details of pattern variation. Detailed information on variation in color pattern elements is necessary to understand how patterns are produced and established during development, and which evolutionary forces may constrain such a variation. Here, we develop an approach to capture and analyze variation in melanistic color pattern elements in leopard geckos. We use this data to study the variation among different body parts of leopard geckos and to draw inferences about their development. We compare patterns using 14 different indices such as the ratio of melanistic versus total area, the ellipticity of spots, and the size of spots and use these to define a composite distance between two patterns. Pattern presence/absence among the different body parts indicates a clear pathway of pattern establishment from the head to the back legs. Together with weak within-individual correlation between leg patterns and main body patterns, this suggests that pattern establishment in the head and tail may be independent from the rest of the body. We found that patterns vary greatest in size and density of the spots among body parts and individuals, but little in their average shapes. We also found a correlation between the melanistic patterns of the two front legs, as well as the two back legs, and also between the head, tail and trunk, especially for the density and size of the spots, but not their shape or inter-spot distance. Our data collection and analysis approach can be applied to other organisms to study variation in color patterns between body parts and to address questions on pattern formation and establishment in animals.


Procedure for image acquisition
Geckos were anaesthetized using an open drop technique with cotton balls embedded with isoflurane. Anaesthetization of the geckos ensured that the animals were asleep while we took pictures for all the four datasets for each animal. Pictures were taken for the entire body of the animal -including the head and the tail if the entire animal fit into the frame -and for the two sets of legs (front and back) separately. Legs were held stretched out by two of us (NM and YC) and placed at the same angle from the body across picture sets. Legs were always positioned on top of the horizontal lines printed on the colored paper, while the body of the gecko was placed on the vertical line. Depending on the entire length of the animal, pictures of the tail were obtained together or separately from the picture of the main body. ID cards with information about the picture set were placed in the frame of all photos for use in identification during processing of images. A ruler was also placed in the frame of all the photos as a size reference. Photos were taken using a photography stand with the camera positioned directly at 12 cm above the gecko. Lighting was provided by the overhead lights in the room and by two additional lamps that were attached to the stand and directed towards the geckos. We used a Canon EOS Rebel T6i camera with focal length of 18mm exposure time of 1/80, aperture of f/8, ISO-400, and autofocus on. Settings were consistent across all photos and picture sets. To obtain pictures for different body parts (e.g., body vs legs), the animal was moved and then replaced every time for each picture set. We first obtained all the four sets for an animal before obtaining pictures for a new individual.

Description of patterns on different body part
Each of the four body parts (limb, trunk, head, and tail) presented different challenges so that the criteria for identifying melanistic spots were different for each type of body part. For example, the legs of the gecko were rounded and topographically complex with sharp angles due to the bones of the leg (see Figure S1, Panel G), whereas the head, trunk and tail were relatively flat. Thus the legs were unevenly illuminated and the algorithm for identifying the pattern of the legs included steps to brighten shadowed regions and darken bright regions. Shadows and glare were identified as contiguous areas of pixels that were darker or brighter than the mean pixel intensity at a spatial scale larger than the spatial scale of the spots. This method was effective since the melanistic spots on the legs were consistently smaller than the spatial scale of the leg contouring for all the gecko morphs. Spots identified by the algorithm almost always coincided with melanistic spots identified via eye inspection; only in a very few cases did the algorithm identify what looked like shadows as spots or vice versa (a comparison of photos and algorithmidentified spots for all patterns is added in the Supplementary Material). For the purposes of objectivity, consistency and reproducibility, the results of the algorithm were always used, even when there was a discrepancy between the algorithm and the pattern identified by eye. The trunk presented the challenge that dark crescent-shaped shadows of tubercules were difficult to distinguish from melanistic spots (see Figure S1, panel D). We found that these shadows were darker than melanistic spots in the blue channel and a satisfactory fraction of these shadows could be distinguished from the melanistic pattern by imposing additional blue channel criteria. Melanistic patterns varied considerably in their relative darkness, from gecko to gecko, body part to body part and even within a regional pattern (see Figure S1, Panels B, D, E and H). A qualifying melanistic pattern was identified by selecting pixels for each image that were sufficiently dark relative to the average pixel brightness of that image. The relative amount of darkness that was required to meet the threshold (see threshold section in "Methods") depended on the body part and the threshold rule was applied after the image was adjusted for shadows and glares. A lower threshold was chosen for the head because the fraction of melanistic pattern area on the head was typically much higher (for example, see Figure S1, Panel A) so that the average pixel intensity was closer to the intensity of the melanistic spots. This lower threshold was effective for the head patterning, whereas it would not work effectively for the identifying the patterning of the other body parts, since the head patterning had a relatively high contrast between melanistic and non-melanistic regions. Figure S1: Representative images of patterns found on the four types of body parts; head (A,B) trunk (C, D), tail (E, F) and limb (G, H, I). Head and trunk patterns commonly had very high contrast between spotted and non-spotted pixels (A, C) though not always (B, D) and welldefined spots were often connected by a 'thread' with especially light pigmentation (red arrows). The head and trunk patterns often contained spots with a mixture of low and high eccentricity, with spots of both a small, rounded, compact type and a more elongated stripe-like type (A, C). While some patterns could be found with relatively even pigmentation throughout the spots, some patterns had spots with variable pigmentation within and/or between spots (B, D, E, H). Shadows and other types of light effects coincided with the pigmentation pattern due to the rounded contour of the leg (G) and the high protrusion of the tubercules on the trunk (D). Other pattern "defects" included occlusion from wrinkles of the skin that, for example, would appear on some of the replicates but not others (I)

Identification of interior and edge spots:
Due to the fact that pictures were taken from above the geckos and due to the rounded shape of the gecko body, especially for the limbs, spots at the edges of the region were occasionally partially cut off from view. We call such spots whose boundary intersects with the edge of the body region "edge spots". Spots whose boundary lies entirely within the region of the body in the image are called "interior spots" (see Figure A2 in the Appendix for an example). Some of our measures should be computed with the full contour of the spot, but are not affected if edge spots are removed, while other measures do not require the full contour of the spot, but are affected if edge spots are removed. For example, the fractional melanistic area of spots depends only on the total melanistic area, rather than the spot size and shapes. Other measures such as the average size and eccentricity of spots require the entire contour of a spot but do not depend on the total number of spots. For this reason, some measures were computed for the entire spot pattern, including both interior and edge spots, and some measures were computed only for the interior spots that were not cut off by the edge of the region (see Table 3).

Differences between "lemon frost" and "normal" morphs
Our data set contained 20 geckos of the "normal" and five of the "lemon frost" morphotypes. Although this sample size is relatively small for the "lemon frost" and as such, the analysis of this section should be regarded as preliminary, a clear qualitative difference in the patterns of "normal" and "lemon frost" morphotypes can clearly be detected by eye, so that it is maybe not so surprising that we still obtain several statistically significant differences even with these small samples. The difference between the "lemon frost" morph and the "normal" is illustrated in Figure 6 (right panel), where we show the first two principal components of the head, trunk and tail patterns of the two morphs. For both the head and the trunk patterns, but not the tail patterns, some clustering is observable with lemon frost geckos tending to have smaller values for both principal components. This corresponds to smaller spots arranged in patterns with a shorter wavelength. In fact, when we investigated the differences between the "normal" and the "lemon frost" morphs for the head and trunk patterns (TableS1) we found that these differences are statistically significant for the first two principal components of the head patterns and the second principal component for the trunk. For individual indices, the differences are statistically significant in a few cases, most notably those that are associated with the shape of the spots (EE and EL). In both cases, the "lemon frost" spots are closer to circular than the "normal" spots. In contrast to this, the differences in tail patterns were not statistically significant in any of the indices or principal components; the smallest p-value of all indices was 0.24. We also investigated whether one of the morphs showed more variation in traits than the other (Table S1). With some exceptions, the sample standard deviations of the "lemon frost" morph are smaller than those of the "normal" morph, meaning that the "lemon frost" morph tends to have smaller variability than the "normal" one. This is particularly true for the head and trunk patterns, but there is again relatively little difference between the variances of the tail. Most of these differences are not statistically significant due to the small sample size with only 5 "lemon frost" individuals, but for the trunk, both MD and PL are significantly smaller (along with 5 other indices or principal components), and for head, MD is significantly smaller as well (with 4 other variables also). Both MD and PL are measures of the characteristic wavelength of the pattern, i.e. the typical distance between spots. Thus our results indicate that the "lemon frost" morph has less variability in this measure than the "normal" one for the head and trunk.

Differences between sexes
Finally, we also investigated the differences between female and male patterning, restricting the analysis to the head patterns of individuals with "normal" morphs. We concentrated on the head patterns because these had the smallest measurement error. None of our indices showed any significant differences (Table S2), consistent with the hypothesis that there is no significant difference between female and male melanistic patterns in captive bred individuals.