Wettability and morphology of proboscises interweave with hawkmoth evolutionary history

ABSTRACT Hovering hawkmoths expend significant energy while feeding, which should select for greater feeding efficiency. Although increased feeding efficiency has been implicitly assumed, it has never been assessed. We hypothesized that hawkmoths have proboscises specialized for gathering nectar passively. Using contact angle and capillary pressure to evaluate capillary action of the proboscis, we conducted a comparative analysis of wetting and absorption properties for 13 species of hawkmoths. We showed that all 13 species have a hydrophilic proboscis. In contradistinction, the proboscises of all other tested lepidopteran species have a wetting dichotomy with only the distal ∼10% hydrophilic. Longer proboscises are more wettable, suggesting that species of hawkmoths with long proboscises are more efficient at acquiring nectar by the proboscis surface than are species with shorter proboscises. All hawkmoth species also show strong capillary pressure, which, together with the feeding behaviors we observed, ensures that nectar will be delivered to the food canal efficiently. The patterns we found suggest that different subfamilies of hawkmoths use different feeding strategies. Our comparative approach reveals that hawkmoths are unique among Lepidoptera and highlights the importance of considering the physical characteristics of the proboscis to understand the evolution and diversification of hawkmoths.


Contact angle measurements
Our first method of measurement stems from a previous custom-designed LabVIEW program used to measure contact angle of fibers (Sun et al., 2022;Zhan et al., 2022).For this code, we needed to specify the region of interest (green rectangle in Fig. S1a) to cover the menisci on either side of the proboscis at the water-air interface.The menisci are detected as seen in Fig. S1b.
Simultaneously, the traced contours are plotted in Cartesian coordinates for the contact angle measurement, as shown in Fig. S1c.Then, the red "start" cursor (  ,   ) is selected where the meniscus meets the surface of the proboscis at the water-proboscis interface, and the blue "end" cursor (  ,   ) is used to specify the range of data for the parabola fitting.Depending on the magnification of the images, data points within 20 to 30 pixels in width (20 ≤ |  −   | ≤ 30) give the best parabola fitting,  =  2 +  + . Figure 2d shows the data as red squares and the black curve corresponding to the best parabola fitting.Thus, the contact angle can be given by  =  − tan −1 | ′ (  )|,  ′ (  ) = 2  +  is the slope where the meniscus meets the surface of the proboscis.
During our analyses of hawkmoth contact angles, we noticed thatbecause of the hydrophilicity of the hawkmoth proboscis -the videos were showing two contact angles: one on the lateral sides, and another on the legular bands of the proboscis.This made the measurements from the LabVIEW code less reliable.Because of that, we opted to also measure the contact angles using the angle tool in Fiji-ImageJ (Schindeling et al., 2012;Schneider et al., 2012).To ensure the measurements from the angle tool were repeatable, we measured both sides of the proboscis and average them.The measures made on both sides were highly consistent (Fig. S2) which ensured that the method was reliable.
Further, to ensure our measures were robust, we compared the measurements taken from the LabVIEW code and the ImageJ measurements on a few select species.Given that the LabVIEW measures were slightly higher than the ones made with ImageJ, we performed a conservative correction to add the possible error measurement to our data.We divided the LabVIEW measure by the ImageJ measure to have a ratio of difference between these two methods, which showed that, on average, measures LabVIEW were 1.12 times larger than ImageJ measures (Fig. S3).We then multiplied the standard deviation of the final dataset by 1.12.Although that procedure increased the error in our data, it also ensured the reliability of our data.It is important to highlight that our results would not change regardless of the method used.All proboscises would still be entirely hydrophylic.Thus, the correction was solely to add a source of error to the data, keeping the data reliable.

Fig. S2.
Comparison of the measurements between the right and left side of the proboscis of the hawkmoths probed in our experiment.As can be seen, both sides had consistent measures.

Fig. S3
. Comparison between the methods to measure contact angle.LabVIEW showed consistently higher angles for the advancing contact angle.To account for this source of error, we multiplied the ratio between the averages of both measurements by the standard deviation in the main analysis.

Measuring the radius of curvature of the galea
The ImageJ macro begins by having the user select nine points on the curvature of interest on the image of interest.Each three points is considered one set, where the last point of the first set is shared as the first point in the subsequent set.When all the points have been placed, there will be four sets of data.
Within a set, the code first calculates the distance between points by finding the side of a triangle.Then, by using the distances between the three points in a set, we can calculate the perimeter of the triangle.Using both the perimeter and the distances, we calculate the area of the circumcircle of the triangle using the following equation: where p is perimeter and d is the distance of a given side of the triangle.Next, we calculate the circumradius using the following equation: To transform that measure from pixels to millimeters (or any other scale), we divide the measure by the ratio between pixels and millimeters obtained from the 'Set scale' function in ImageJ.To ensure the measurement is realistic, we found the centroids in x and y of the triangle and used the radius of the circumcircle to draw the circle.In a set, the dots are connected to make the triangle visible to the user.
The process is then iterated to other sets of points.The process of placing the points of interest can change the metric being calculated.For instance, if the points are placed along various parts of a structure, it will measure how the radius of curvature changes along the curve of the structure of interest.If the points are placed in the same area multiple times, the same area will be measured four times, which can then be averaged.We placed the points of interest multiple times in the same structure and averaged the measurements of radius of curvature.Depending on the position of the cross-section of the proboscis, some sets did not provide realistic measures of the curvature.Thus, we only used the measurement if we could reliably average two sets of data.
The code and a user manual will be uploaded as supplementary materials alongside the paper.
Table S2.Average advancing contact angle with standard deviation along the proboscis of 13 species of hawkmoths.The normalized distance from the head denotes the region of the proboscis, with smaller values being closer to the head, while larger values are closer to the tip of the proboscis.These values were used to generate Figure 3 in the text.Table S1.Average values (± standard deviation) of all the measurements taken from 13 species of hawkmoths (Lepidoptera: Sphingidae).Slope of change denotes the slopes of a relationship between average advancing contact angle and the section of the proboscis in which the angle was measured.Measurements of radii of curvature and tapering angle were made from CT-scans, thus representing the measurements taken from one individual per species.

Calculation of proboscis dipped volume in Xanthopan morgani
To calculate how much the nectar pooled in the orchid Angraecum sororium would rise when the proboscis is dipped, we used data published in Wasserthal (1997), Arditti et al. (2012), and an image of A. sororium taken from the Herbarium Jany Renz.From Wasserthal (1997), we extracted the volume of nectar in the flower and the range of proboscis lengths.From Arditti et al.
(2012), we used the image of X. morgani with the proboscis uncoiled to measure

Table S3 .
Akaike Information Criterion values of the evolutionary models fitted to the phylogenetic linear regression models.Models with ∆AIC ≥ 2 were considered poorly fit.When competing models had ∆AIC ≤ 2, we chose the model with the fewest parameters.