Artificial Butterfly Models
We used artificial butterflies to assay predation of H. doriswarning colorations in three distinct geographic locations with known divergence in H. do ris wing colorations (Panama versus French Guiana red hindwing pattern). Artificial butterfly experiments in natural populations have proven to be efficient means to record predator attacks for several Heliconius species and warning colorations (Arias et al., 2016; Chouteau et al., 2016; Finkbeiner et al., 2018; Seymoure et al., 2018). At each of the three locations in Panama and French Guiana, we used artificial butterflies of threeH. doris warning colorations and the cryptic Pierella hyceta , which we used as a palatable control following the method in Chouteau et al., (2016). This model allowed us to obtain additional data on the intensity of selection at each locality, however it also provided a comparative insight on the selective advantages of crypsis versus aposematism.
Standardized photographs of the ventral and dorsal wings of each butterfly were used and printed on two-sided matte photographic paper. (Epson C135041569 paper and L110 Printer). In order to produce a high volume of standardized models, a silicon mold (Mold Star, Smooth-on) was fabricated using clay bodies that were shaped to resembleHeliconius bodies. The paper wings were inserted into each mold along with a thin 20 cm metal wire before pipetting a mixture of high melting point wax with a black dye and then left to solidify. The different colors on the printed wings were calibrated in Photoshop (Adobe Inc.) and then contrasted with the colors on actual H. doris wings by measuring the reflectance spectra of red, black, yellow and blue using a spectrophotometer (HR2000+ES, Ocean Optics) and a deuterium/halogen light source (DH-2000; Ocean Optics) connected to a 3.175-mm diameter sensor (QR600-7-UV125BX; Ocean Optics) inserted in a miniature black chamber. Reflectance spectra were taken at 90º for all colors except for the blue structural coloration which was taken at 45º incidence relative to a 99% reflectance standard (300-700 nm; Spectralon) and to a dark current. Spectra were recorded with SpectraSuite 1.0 software (Ocean Optics). color spectra from real and printed wings were then compared using the method described by Osorio (1998) in Avicol v.6 software (Gomez, 2006). We contrasted blue, black, red and yellow, under two main avian vision systems: blue tit (Parus caeruleus ) for UV vision, with cone proportion and sensitivity as described by Hart et al., (2000), and wedge-tailed shearwater (Puffinus pacificus ) as described by Hart (2004) for violet (V) vision. Photoreceptor activity was computed from the Weber fraction (Osorio, 1998), and set to 0.05 for all artificial models. Small gap light conditions, as defined by Endler (1993) from French Guiana were included in all calculations (Thery et al., 2008). Chromatic (Delta S) and achromatic differences (Delta Q) for all colors were found to be under the noticeable threshold for avian vision in UVS and VS (<1.00 Just Noticeable Difference units, as in Llaurens et al., (2014), thereby confirming the accuracy in color of our printed wings to real wings (See Table 1).
Using the attached thin metal wire, models were placed on leaves, trunks or twigs in visible, well-lit areas at 10m intervals along a 4km transect in each site. The placement of each model was carried out so as to mimic the natural perching behavior of Heliconius butterflies and provide a visible target for potential avian predators. The distinct model morphs were placed along the transect in a regular order. From 376 to 416 models were placed per site and left for 72 hours, after which, models were collected. Damage was clearly visible in the malleable wax bodies and paper wings of several models. Damages were catalogued as either (i) “invertebrate attack” when bearing the visible fine marks of arthropod mandibles, often on the wax bodies, (ii) “Avian Attack” when bearing the characteristic U or V shape marks on the wax or (iii) “Unknown Predator” when a severe attack was evident but a specific mark was not found, such as when wings were torn or wax bodies broken in pieces. Models that bore attack marks characteristic of invertebrates were not included in the data analysis (n= 97 out of 2,271), as there is currently no literature regarding invertebrates carrying the cognitive capacity necessary to make the associations between unpalatability and warning color patterns central to Müllerian mimicry. Furthermore, missing models were also excluded from the analyses as we are unable to determine if they were displaced by falling forest debris, human action or attacked by natural predators.