Intraspecific morphological variation in the shieldtail snake Rhinophis philippinus (Serpentes: Uropeltidae), with particular reference to tail-shield and cranial 3D geometric morphometrics

2.1 Specimens and external morphology

Morphometric data were collected for 35 specimens of R. philippinus, comprising 16 males, 18 females, and one of unknown sex (Table 1). All specimens are spirit-preserved and stored in the permanent collections of the Natural History Museum, London, UK (BNMH) and the Muséum National d'Histoire Naturelle, Paris, France (MNHN). Tail length was used to infer sex of all specimens given that it was found to be strongly dimorphic (see below). Sex was determined for a few specimens by examination of the urogenital system via small incisions—as in the many other species of Rhinophis with dimorphic tail length (Cyriac et al., 2020; Gower, 2020), longer and shorter tails were always found in males and females, respectively. Eight linear measurements were taken using dial calipers to the nearest 0.1mm: tail length, head length (snout tip to posterior of fourth supralabial), head width (at posterior ends of fourth supralabials), midbody width, tail-shield middorsal length (shortest distance between posterior tip and anterodorsal midline limit of shield), tail-shield midventral length (shortest distance between posterior tip and anteroventral midline limit of shield), tail-shield height at base (shortest distance between anterodorsal and anteroventral midline limits of shield), tail-shield width at base (maximum transverse width of base of shield). Total length was measured by a ruler to the nearest 1 mm, and snout-vent length was determined by subtracting tail length from body length. Counts were taken of the total ventral scales (following Gower and Ablett (2006)), mean of left and right subcaudal scales, unpaired or fused subcaudal scales, and scales around the perimeter of the shield base.

2.2 Imaging cranial and tail-shield osteology

The heads and tails of all specimens were scanned using a Nikon Metrology X-Tek HMX ST 225 micro-CT scanner (BNMH specimens) and a v|tome|x L 240–180" model from GE Sensing & Inspection Technologies phoenix x|ray scanner (MNHN specimen). Reconstructed scan data were digitally segmented using Avizo Lite v.9.3 (FEI). These meshes were imported into Geomagic Wrap (3D Systems) and osteologically distinct, extraneous components such as vertebrae, quadrates, and lower mandibles were digitally removed. The outer surface of the bony tail-shields has many pits and holes that were filled manually using Geomagic in order to facilitate shape quantification procedures. Meshes were decimated to less than one million faces to reduce computational demand when landmarking, but still maintain enough detail to pick up any fine intraspecific variation. Two of the crania (BMNH 1964.1676 and BMNH 98.5.3.11) were mirrored using the “mirror” function because their left sides were damaged. Subsequent analyses were carried out on the left side of all crania. Two of the crania (BMNH 1964.1674 and BMNH 1968.5170) were too broken to be landmarked, and so were removed from the cranial dataset. All meshes used in this study are availble on Phenome10K (http://www.phenome10k.org/).

2.3 Geometric morphometrics

The complexity of the morphology of the cranium and irregular, monolithic nature of the tail-shield would be greatly underrepresented by linear measurements alone. Geometric morphometrics uses landmarks to pinpoint homologous points throughout these complex shapes to describe the geometry of the object. Landmarks were generated on the surface of the reconstructed meshes using Checkpoint (Stratovan).

2.3.1 Crania

Sixty-six Type I and Type II landmarks (Bookstein, 1992) were defined (Table S1) and placed on the left side of each cranial mesh (Figure 2). These landmarks served as homologous points, between which curve semilandmarks could be placed to delineate bones and regions of the crania. Here, we used a combined landmark and sliding semilandmark procedure to fully capture the complexities of the whole overall cranial shape. Fifty-six curves were placed on each cranial mesh (Table S2, Figure 2), and these were resampled such that each specimen had 818 curve semilandmark points in total. Curve semilandmarks were then slid using the “slider3d” function from the R package Morpho (Schlager, 2017) to minimize bending energy. This generated a dataset of homologous points that could be compared across the specimens.

2.3.2 Bony tail-shields

The external tail-shield of Rhinophis spp. is underlain by a closely matching osteological structure fused to the last few caudal vertebrae and referred to here as the bony tail-shield. Type II landmarks were placed at the anterodorsomedial and posteroventromedial ends of the bony shield, and on nine homologous points of the vertebrae fused to it, totaling 11 landmarks (Table S3, Figure 3b). These landmarks did not capture the overall shape of the bony shield, and so they were combined with sliding curve and surface semilandmarks. Two curves were used to delineate the bony shield, anchored by the two landmarks at the anterior and posterior ends (Figure 3). These were resampled so that each curve had 45 semilandmarks, a total of 90 curve semilandmarks per specimen. In order to capture the shape of the outer surface of the bony shield, a template was created to place surface semilandmarks onto all the specimens. One specimen (BMNH 1946.1.16.99), deemed to be among the specimens closest to the average bony-shield shape, was selected and duplicated with the landmark and curve semilandmark scheme detailed above. A specimen template was used instead of a generic hemispherical structure because variation is likely to be fine for an intraspecific dataset. Surface semilandmarks were placed on the template in a grid formation, with 16 rows spanning the dorsal surface of the bony shield. The template was then projected onto the other 34 bony shields in a semi-automatic patching procedure using the Morpho package in RStudio (Figure 3). The projected semilandmarks were checked using the “checkLM” function in Morpho to ensure an even distribution of points on all specimens fully captured the shape.

The inner surface of several bony-shield meshes was removed to prevent the surface semilandmarks falling underneath the dorsal surface. An additional curve was temporarily placed along the midline of the dorsal surface on eight specimens in order to redistribute the surface semilandmarks more evenly. A new template with this extra curve was used to project the surface semilandmarks onto those eight specimens. The surface semilandmarks were each expanded along their normals to differing extents, ensuring an even distribution that covered the entire shield outer surface. A total of 127 surface semilandmarks were projected onto the outer surface of each bony shield. Those surface semilandmarks, as well as the curve semilandmarks, were then slid, minimizing bending energy, using the “slider3d” function in Morpho. This procedure allows surface and curve semilandmarks to be transformed into geometrically homologous landmarks in order to compare shapes (Parr et al., 2012). The homologous landmarks and sliding semilandmarks were then compared across specimens using traditional morphometric methods.

2.4 Statistical analyses

Outlier plots were initially produced to ensure landmarks were placed in the correct order across all specimens. These plots also indicated that the cranium of one specimen (MNHN 6994) is substantially larger than the other specimens, and one of the smallest specimens (BMNH 1964.1701: Figure S1) has a notably less well ossified and more irregularly shaped bony tail-shield. These specimens were removed from further analyses, excluding the regression and linear dimensions analyses.

3.1 External morphology

Raw data are reported in Table 1. Results of the analyses performed on the external morphological measurements and counts show that the following measurements are not sexually dimorphic: snout-vent length (SVL; ANOVA: F = 0.174, p = 0.679; though note that the seven largest specimens are all female); head length (ANOVA: F = 0.036, p = 0.851), head width (ANOVA: F = 1.206, p = 0.280); midbody width (ANOVA: F = 0.208, p = 0.651); tail-shield middorsal length (ANOVA: F = 0.421, p = 0.521); tail-shield midventral length (ANOVA: F = 0.976, p = 0.331); tail-shield height at base (ANOVA: F = 0.424, p = 0.520); tail-shield width at base (ANOVA: F = 0.017, p = 0.896); total fused subcaudal scales (Kruskal–Wallis: H = 3.786, p = 0.052); total shield perimeter scales (Kruskal–Wallis: H = 0.220, p = 0.639).

In contrast, tail length is significantly dimorphic (ANOVA: F = 27.81, p < 0.0001), being longer in males than females. This dimorphism is still apparent when tail length is considered relative to size (SVL in this case: ANCOVA: F = 91.91, p < 0.0001). Numbers of ventral and mean of left and right numbers of subcaudal scales are also significantly sexually dimorphic (Kruskal–Wallis: H = 24.818, p < 0.0001 and H = 25.621, p < 0.0001, respectively). On average, females have more ventral scales (males: range = 157–166, average = 162.1; females: range = 169–179, average = 173.7) and fewer mean subcaudal scales (males: range = 5–6, average = 5.5; females: range = 3–4, average = 3.5).

3.2 Cranial and bony tail-shield allometry

Regressions of log₁₀-transformed centroid size against shape show that allometry has a significant effect on shape for both crania (R² = 0.238, p = 0.0001) and bony tail-shields (R² = 0.151, p = 0.0001).

The scatterplot of the regression of cranial log₁₀-transformed centroid size against shape (Figure S2) shows a strong linear relationship with little scatter. Larger crania tend to have a relatively lower, less dorsally arched cranial cavity, a longer occipital condyle stalk, and a slightly posteriorly flared exoccipital arch (Figure S2). The lack of separate clustering of the two sexes provides evidence that the allometric effect on cranial shape is not sexually dimorphic. MNHN 6994 has a much larger cranial centroid size and more extreme shape than the other sampled specimens, though it falls along the same linear relationship (Figure S2). Based on Procrustes distance from the mean, this specimen is an outlier, and it was removed from subsequent analyses.

The corresponding scatterplot of the regression of tail-shield log₁₀-transformed centroid size against shape also presents a linear relationship (Figure S3), although weaker at larger centroid sizes. At greater centroid sizes, tail-shields are more acutely conical. There is no indication of sexual dimorphism in the allometric effect on tail-shield shape, as shown by the lack of separate clustering of males and females. BMNH 1964.1701 is an outlier in the regression plot, which was expected given its much less well ossified and notably irregularly shaped tail-shield (Figure S1). This specimen was removed from subsequent analyses.

3.3 Cranial and bony tail-shield shape variation

Twenty-nine PC scores were required to explain 99% of cranial shape variation (Table S4) and differences in adjusted cranial shape at the minimum and maximum distributions along each axis are subtle (Figure 4a, Figures S4–S7). PC1 accounts for only 13.03% of the overall variation. Crania with highest positive values on this axis have a less spherical occipital condyle, a wider posterior process of the pterygoid, a taller septomaxilla and a more pronounced sloping of the premaxilla anteroventrally from the nasal process (Figure 4a and Figure S4). PC2, accounting for 11.96% of variation, is associated with differences in the dorsal longitudinal arch of the braincase, whereby crania at higher positive values along this axis have a lower dorsal arching in lateral view (Figure 4a and Figure S5); they also have a longer palatine and a relatively wider premaxilla. Specimens located at the negative extreme of PC2 are represented by three males (Figure 4a) with a clearly higher dorsal arch to the braincase. PC3, accounting for 9.22% of variation, is associated with differences in the position of the pterygoid, with crania at the far positive end of PC3 having relatively lower posterior ends of the pterygoids seen in lateral view (Figures S6 and S7).

No clustering of the sexes is displayed in the cranial morphospace of PC1 versus PC2 (Figure 4a). Thus, there is no evidence for sexual dimorphism in adjusted cranial shape in R. philippinus. Consequently, no inferences can be made as to the sex of the single specimen (BMNH 1964.1705) for which sex was not known. This was the only specimen with a somewhat intermediate relative tail length, and unambiguously male or female urogenital features could not be located.

Thirty PC scores were required to explain 99% of the variation in bony tail-shield morphology (Table S5). Identifying the aspects of shape variation explained by each PC axis is simpler than for cranial variation. PC1, accounting for 12.17% of variation, describes primarily the shape of the posterior projection of the shield in dorsal view, with shields at the far positive end of this axis being more broadly rounded with a blunter posterior edge, along with a less complex free edge and less extended transverse processes on the vertebrae (Figure 4B and Figure S8). PC2, associated with 10.36% of variation, relates to differences in the projection of the bony-shield's dome, with shields at the far positive end of this axis being more acutely conical (Figure 4b and Figure S9). PC3, accounting for 8.35% of variation, describes differences in the width and length of the shield, with those at the far positive end of this axis having a wider and shorter shield (Figures S10 and S11).

All specimens are distributed evenly throughout a tail-shield morphospace plot of PC1 versus PC2 (Figure 4b). There is no clustering of the sexes, such that there is no evidence of sexual dimorphism in the adjusted tail-shield shape in R. philippinus.

3.4 Covariation between cranial and bony-shield shape

Covariation between cranial and bony tail-shield shape was significant when using adjusted shape data (r-PLS = 0.85, p = 0.03) and shape data (r-PLS = 0.85, p = 0.001). Most of the covariation in the shape dataset was explained by the first axis (62.84%), with the majority of the covariation in the adjusted shape dataset spread over the first three axes (22.47%, 16.21%, and 12.05%).

The scatterplots of the results of each 2B-PLS analysis show a positive linear relationship between cranial adjusted shape and bony-shield adjusted shape. For the adjusted shape data analysis, crania at the positive end of the PLS axis are slightly more pointed with a less dorsally arched skull roof in lateral view. These cranial shapes are associated with more acutely conical bony shields (Figure 5a). No sexual dimorphism is apparent in this covariation scatterplot.

The shape data analysis finds a similar pattern of covariation between the cranial and bony-shield shape, though with significantly more pronounced differences in shape at the ends of each axis (Figure 5b). Here, crania with high positive PLS scores have less dorsally arched roofs in lateral view, relatively larger maxillae, an elongated snout, a posteriorly flared exoccipital arch, and a larger and more bulbous occipital condyle in lateral view. The minimum and maximum PLS scores for bony shields are associated with flatter and with more acutely conical shapes, respectively.

3.5 Sexual dimorphism tests for crania and bony tail-shields

Male and female specimens have very similar levels of adjusted shape variation (crania Pv = 0.00072 and 0.00063, respectively; tail-shield Pv = 0.01327 and 0.01342, respectively), with no significant difference in disparity (crania p = 0.182, tail-shield p = 0.925). Therefore, there is no evidence of sexual dimorphism displayed in the shape disparity of these structures in R. philippinus.

MANOVA found no adjusted shape differences depending on sex for both crania and bony tail-shields (F = 1.25, p = 0.171 and F = 1.379, p = 0.078, respectively). This is consistent with the lack of separation of male and female specimens in the morphospaces (Figure 4). The results of the analyses of variance performed on both the log₁₀-transformed centroid size of crania and bony shields are also non-significant (ANOVA: F = 0.003, p = 0.962 and F = 0.265, p = 0.618, respectively).

3.6 Correlation between centroid size and external morphology

SVL and cranial centroid size are moderately positively correlated (rho_S = 0.490, p = 0.004), while SVL and bony-shield centroid size are less positively correlated (rho_S = 0.341, p = 0.045). There is a low positive correlation also between tail length and cranial centroid size (rho_S = 0.192, p = 0.284), and between tail length and bony-shield centroid size (rho_S = 0.238, p = 0.168). Head length and cranial centroid size are moderately positively correlated (rho_S = 0.524, p = 0.002).

4.1 Sexual dimorphism

Our study found no evidence of sexual dimorphism in the size or shape of the head, cranium, external tail shield, bony tail-shield, or SVL in the available sample of R. philippinus. This finding might be considered surprising given that sexual dimorphism is prevalent in snakes, in general, particularly with regards to size (Fitch, 1981; King, 1989; Shine, 1994). Similar head and cranial shape and size in male and female R. philippinus may indicate a trade-off between ecological and sexual selection, should the latter exist in this species. Some studies have shown that the selective pressures generated by head-first burrowing and a fossorial lifestyle might have a larger impact on head and cranial morphology than does any sexual selection. Studies establishing the presence or absence of sexual dimorphism in other species of Uropeltidae could provide further tests of this possible trade-off. Among other head-first burrowing, limbless vertebrates, sexual dimorphism has been found in head or cranial shape in only some amphisbaenians (Hipsley et al., 2016) and in some caecilian amphibians. The causes and consequences are poorly understood in these typically little-studied animals. However, male Schistometopum thomense tend to have larger heads that are costly in terms of burrowing speed (Teodecki et al., 1998), and head dimorphism in Boulengerula boulengeri is associated with dimorphism in diet (Jones et al., 2006). Lack of sexual dimorphism in R. philippinus cranial shape suggests a homogeneity in burrowing ability and diet between males and females, something that is open to testing in the field and/or laboratory. It is important to note that sexual dimorphism is complex and can be generated from several different drivers that are not mutually exclusive, including sexual selection (Darwin, 1871; Olsson et al., 2002), male-male combat (Shine, 1978), or ecological niche divergence between the sexes (Slatkin, 1984; Vincent et al., 2004). Interpretations of the potential drivers of sexual dimorphism solely based on shape and size variation thus are tentative as acquisition of in vivo data are needed to test these hypotheses.

The lack of sexual dimorphism in tail-shield shape suggests that this structure in uropeltids, or at least R. philippinus, performs a common function in both sexes. Therefore, from this, it can be postulated that tail-shields serve no role in sexual selection in R. philippinus, and instead are more likely to function in locomotion and/or predator avoidance. Uropeltid snakes are known to be preyed on by boars and birds (Rajendran, 1985), and it has been argued that the cephalic mimicry of the tail-shield directs potential attacks from these predators away from the head and toward this more resilient structure (Cyriac & Kodandaramaiah, 2019; Gans, 1986a; Melvinselvan & Nibedita, 2016).

Although there is no dimorphism in external shield shape, there is pronounced sexual dimorphism in tail length, with males having longer tails than females. This can be explained by the uneverted hemipenes of male squamates being stored in the base of the tail (Shine et al., 1999). Male-male “tail wrestling” combat (Shine et al., 1999) seems unlikely in snakes with very short tails, as in Rhinophis, and has not been reported in uropeltids, but longer tails perhaps also help functionally in attaining cloacal apposition during initial stages of copulation. The latter could be tested with data on male reproductive success and/or body condition in relation to tail length (Sivan et al., 2020). As pointed out by Jins et al. (2018), the higher number of ventral scales in females can be explained by selective pressure for increased space for fetal gestation, all uropeltids being viviparous as far as is known (Rajendran, 1985).

4.2 Shape variation in crania and bony tail-shields

Variation in both cranial and bony-shield morphology in R. philippinus is significantly correlated with size, with allometry accounting for 23.8% and 15.1% of the shape variation, respectively. This allometric effect in the skull was characterized by larger crania being longer and narrower, particularly toward the posterior of the skull. Pointed heads are common in limbless fossorial vertebrates, often being more effective for head-first burrowing (Gans, 1994; Haas et al., 2006; Lopez et al., 1997; Rodrigues et al., 2016; Roscito & Rodrigues, 2010; Sherratt et al., 2014; Shimer, 1903). Pronounced allometry, with less acutely pointed crania and wider braincases in smaller specimens, is possibly explained by constraints in minimum functional size of sensory structures held in the back of the skull, such as the inner ear (Olori, 2010). Alternatively, larger crania may be relatively narrower to account for the greater effort required to move a larger body through the substrate.

MNHN 6994 is by far the largest specimen in the sample (total length 259 mm versus 211 mm for one other specimen and all others <200 mm), and allometric analyses identified this individual as also having a significantly larger cranial centroid size (66.3 mm versus mean of 47.9 mm in other specimens). There is no indication that the skull of MNHN 6994 (the holotype of the species) represents a species different from that of the other specimens because it is qualitatively similar in its osteology and external morphology, and does not depart from the same allometric regression line (Figure S2). However, the collection localities of all but two specimens in our sample are unknown, and we have little to no understanding of uropeltid demographics, ontogeny, and geographical morphological variation. Thus, it remains possible that our study lacks sufficient sampling of total lengths and or populations to detect other patterns of variation among specimens currently referable to R. philippinus.

Variation in cranial shape of R. philippinus is spread broadly across many axes, in contrast with findings from a wider squamate study in which almost two-thirds of cranial shape variation is explained by the first three PC axes (Watanabe et al., 2019). This is likely explained by the intraspecific sampling of this study and by the generally conserved cranial osteology within R. philippinus, including a lack of sexual dimorphism. The R. philippinus cranial variation across the first few axes encompass differences in the occipital condyle shape, the size and position of the pterygoid (though we note this could be an artifact of how the specimens are preserved) and the longitudinal arching of the braincase roof. Little variation is present in the shape of the snout, which may be highly conserved due to its function in head-first burrowing in these snakes. Cranial shape variation can be dependent upon the nature of the substrate in which a limbless fossorial organism burrows (Greenville & Dickman, 2009; Jackson et al., 2008). Most uropeltids, including R. philippinus (DJG, pers. obs.), tend to be found in moist, clay- and/or humus-rich soils (Rajendran, 1985), but no studies have been conducted to determine microhabitat tolerances or preferences in this family of snakes.

Larger (and presumably older) bony tail-shields are more acutely conical, but without more research, it is unclear how patterns of ossification may influence this shape and any functional consequences possibly associated with this. Variation in bony-shield shape was, as with the cranium, similarly spread over many axes, which is also likely to be explained by the intraspecific dataset and conserved shape of this structure, including lack of sexual dimorphism. Pyron et al. (2016) reported ontogenetic variation in the external keratinous shield of Rhinophis melanogaster, with smaller specimens having a relatively much smaller shield, but no quantitative data or analyses were presented. Broader taxon sampling will allow a better understanding of the patterns and causes of shield size and shape variation within and among species.

Although the overall variation in both the crania and the tail-shields is slight and subtle in R. philippinus, there is a significant covariation between the two structures, with a longer, narrower cranium being associated with a more conical tail-shield. However, given that this relationship is more exaggerated in the shape data that were not corrected for allometry, a large proportion of this covariation is likely due to covariation in size.

The use of geometric morphometrics provides a highly detailed approach to quantifying shape variation, especially with the use of sliding semilandmarks (Goswami et al., 2019). Although the number of landmarks and semilandmarks used on the tail-shield was sufficient to investigate the whole shape, it would have been useful to repeat the cranial analyses with the addition of surface semilandmarks to investigate whether that increased coverage better captured morphological variation.

It is interesting to note that centroid size, issued from the cranial and bony-shield shape analyses, is not a good proxy of the overall size of the individual R. philippinus, either for SVL or tail length. Furthermore, due to the decoupling of the head from the body in elongate animals (Sherratt et al., 2019), it might be expected that cranial centroid size would be a good proxy for head length; however, we found little correlation between these measurements. Thus, future analyses could substitute centroid size with tail length as a covariate in sex differentiation analyses, and SVL or total length for the impact of size on shape.

Our study provides new insights into intraspecific variation of this very poorly known genus and family of snakes and has prompted several questions worthy of future research. The main limitations of this study are the lack of ecological and behavioral data for R. philippinus, the lack of knowledge about geographic provenance of all but two of the sampled specimens, the lack of male specimens of total length >200 mm, and the paucity of information regarding the natural history of this species. Future studies of R. philippinus variation might aim to remedy these, as well as expand analyses of variation to other aspects of external morphology and osteology, including lower jaws. In addition, similar studies are required for other uropeltid taxa to determine whether the patterns found here extend to the whole family. Rhinophis species have some of the most extreme head and tail-shield morphologies within Uropeltidae and morphological variation within R. philippinus might not be typical. Some species of other uropeltid genera are more strikingly colored than R. philippinus and might be less dedicated burrowers, spending more time on the surface (Cyriac & Kodandaramaiah, 2020). Thus, similar studies of morphological variation for those species would provide a test of the hypothesis that morphological variation in the head and cranium of R. philippinus can be explained by strong stabilizing selection to maintain an optimum morphology for head-first burrowing. Finally, further studies of uropeltids in the field, in captivity and the laboratory are required to help identify causes of the patterns of morphological variation that are discovered.