Coral snakes predict the evolution of mimicry across New World snakes

Batesian mimicry, in which harmless species (mimics) deter predators by deceitfully imitating the warning signals of noxious species (models), generates striking cases of phenotypic convergence that are classic examples of evolution by natural selection. However, mimicry of venomous coral snakes has remained controversial because of unresolved conflict between the predictions of mimicry theory and empirical patterns in the distribution and abundance of snakes. Here we integrate distributional, phenotypic and phylogenetic data across all New World snake species to demonstrate that shifts to mimetic coloration in nonvenomous snakes are highly correlated with coral snakes in both space and time, providing overwhelming support for Batesian mimicry. We also find that bidirectional transitions between mimetic and cryptic coloration are unexpectedly frequent over both long- and short-time scales, challenging traditional views of mimicry as a stable evolutionary ‘end point' and suggesting that insect and snake mimicry may have different evolutionary dynamics.


Mimicry (with corals = 0)
Neighborhood size (km) Standardised residuals of mimetic snakes for a single randomisation compared to the simulated means for each grid cell confirm that the null distributions perform well (nearly all grid cell values between ±2 s.d. and minimal spatial pattern to outlier cells). Yellow and warmer colours denote significant positive deviation in all panels. b, Standardised residuals of observed coral snakes to simulated means confirm no complex interaction effect between coral snake and overall species richness. c, Standardised residuals of observed mimicry data under a conservative georeferencing regime are qualitatively similar to results with non-georeferenced records placed randomly within collection state (Fig. 1d). d, Observed species richness distribution of mimetic species with colour polymorphism (n = 46 species). e, Standardised residuals of polymorphic mimics show a more complicated geographic pattern in which polymorphism is high both in areas with many coral snake species (western Amazon, Central America) as well as in allopatry to coral snakes (western United States).  (Fig. 2a). For all regression plots, black points are observed data and grey dots are expected values from simulation. b, Weak but nonsignificant positive correlation between the number of polymorphic mimics and number of coral snakes, excluding cells with no coral snakes. c, Weak but nonsignificant positive correlation between polymorphism and coral snakes when including cells with no coral snakes. d, More conservative placement of non-georeferenced occurrences shows the same result as our original georeferencing pipeline (Fig. 2a). e-h, Repeated richness analyses at a 2° grid cell resolution. i, Abundance plot including grid cells with no coral snakes show increased ratios of 0 but do not change the overall conclusion that mimetic snakes are often more abundant than coral snakes. j-k, Repeated abundance analyses at a 2° grid cell resolution. l, Using only georeferenced occurrence records show the same pattern as the full record set (Fig. 2b). Ratio of mimics to models (log 2 ) More models More mimics Rank order abundances Ratio of mimics to models (log 2 ) More models More mimics Rank order abundances Ratio of mimics to models (log 2 ) More models More mimics Rank order abundances Ratio of mimics to models (log 2 )

More mimics
Rank order abundances Coral snake species Mimetic snake species

Supplementary Figure 4. Red-black banded (RBB) warning coloration in Elapidae.
Character reconstructions for the snake family Elapidae using three different methods show multiple independent origins of RBB coloration. Marked nodes are those reconstructed as redblack banded.

Philodryas georgeboulengeri
Philodryas Million years before present

Supplementary Figure 5. Red-black banded (RBB) mimetic coloration in Dipsadines.
Character reconstructions for the subfamily Dipsadinae (family Colubridae) using three different methods show multiple independent origins of RBB coloration. Marked nodes are those reconstructed as red-black banded. Tree fragment on the right follows from the top of the tree fragment on the left.

Supplementary Figure 6. Red-black banded (RBB) mimetic coloration in Colubrines.
Character reconstructions for the subfamily Colubrinae (family Colubridae) using three different methods show multiple independent origins of RBB coloration. Marked nodes are those reconstructed as red-black banded. Tree fragment on the right follows from the top of the tree fragment on the left.

Sibynophis collaris
Grayia tholloni   It should be noted that although our spatial dataset is only ~83% complete due to lack of recorded observations in museums, we do not expect these missing species to fundamentally alter our conclusions. The species missing from the richness dataset are generally endemics with very small ranges (as widespread species are most likely to be represented in collections) and the species missing from the abundance dataset are generally quite rare. Thus, the numerical impact of missing data in our analyses is expected to be minimal, as both species richness and abundance grid cell totals would only be negligibly increased by the inclusion of sparsely distributed data given the large overall size of both datasets.
We note that our main temporal tests do not explicitly include full ancestral range reconstructions to validate "ensured historical sympatry" between coral snakes and their mimics.
Rather, our temporal correlation simply tests for the earliest opportunity in which these clades could have been in sympatry in the New World. Because the fossil record for New World snakes is patchy and sparse (although coral snake vertebrae are known from the Miocene of Nebraska, USA, so coral snakes were historically distributed further north than their current distribution), we originally assumed nothing about the historical distribution of coral snakes other than the timing of their earliest possible arrival to the New World from the molecular split with Asian coral snakes. Although the absolute values of all divergences times may be skewed, that bias should not systematically affect the relative position of these values to each other, which is the metric of interest in our assessment of the timing of RBB origin in coral snakes versus New World colubrids. Although we then reconstructed latitudinal midpoints of geographic ranges as continuous variables across the tree to generate a profile of the geographic distribution of ! 10 coloration changes, we note that these are not "full range" constructions -they are are only latitudinal midpoints. Thus, we urge that this analysis should be interpreted conservatively.
Because there is considerable geographic conservatism to clade distributions (e.g., there is a large Dipsadine clade that does not expand out of the tropics), we think that this analysis is informative about the relative rate of gain and loss in the tropics and near the margins of the present-day coral snake distribution. Beyond this coarse generalisation, though, we encourage caution in making anything more significant out of this finding.
Additionally, although many taxa are missing from the tree due to lack of DNA sequence information, the overall rarity of the RBB coloration suggests that the future addition of missing taxa will mainly serve to confirm or increase -rather than decrease -the number of transitions recovered here. Transitions could be pushed further back in time by the fortuitous, balanced placement of missing taxa specifically spanning transition points, but numbers of reconstructed origins are far more likely to increase rather than decrease without any changes to the timing of the origin of mimicry within clades (as we found by adding taxa to the Sonorini clade).
We also note that there are two New World mimetic species each on single, long branches (Colubridae: Neckband snakes, Scaphiodontophis annulatus; and Aniliidae: South American pipe snakes, Anilius scytale) for which we have no information to reliably date the RBB origin. The branch lengths for these species are ~40 and ~80my respectively, and our stochastic character mapping simulations produce essentially uniform distributions of the timing of the RBB origin across each branch. Thus, we do not consider either taxon to be an exception to our finding that RBB coloration in non-venomous snakes arose after the earliest possible arrival of coral snakes ! 11 in the New World, but rather that due to their unique topology, we have no information about the timing in these two lineages to reconstruct these origins with any confidence.