Sampling Localities. Eulaema nigrita males were collected from 14 February to 4 March 2020, in six remnants of Brazilian savanna, three from forest habitats (gallery and seasonal semideciduous forests) and three from woody savanna (Cerrado strict sense) located in the Triângulo Mineiro region, Minas Gerais State, Brazil (Fig. 1a). The forested habitats were located in the following areas: GF – Glória Experimental Farm (18° 57' 03'' S, 48° 12' 22 '' W), PER – Panga Ecological Reserve (19° 10ʹ 04” S, 48° 23ʹ 41” W) and IF – Irara Farm (19° 08' 39''S, 48° 08' 46 '' W). GF is a farm with 685-ha, mostly used for agricultural activities that present a seasonal semideciduous forest remnant of approximately 30-ha, continuous with gallery forests (Lopes et al. 2011). PER has 403.85 ha (Cardoso et al. 2009) composed of a vegetation mosaic including shrubby grassland, palm swamp, wet grassland, dense cerrado woodland, stricto sensu cerrado, evergreen and gallery forest (Gonçalves et al. 2021). The forest habitats, evergreen and gallery forest, represents 11% of the total area (Gonçalves et al. 2021). IF is a private property with a seasonal semideciduous forest reserve of 22.3 ha immersed in a gradient including “cerradão” and gallery forest. IF remnant shows few signs of anthropic intervention and is surrounded by areas used for livestock and soybean crop (Lopes et al. 2012).
The seasonal semideciduous forest remnants studied have composition and floristic richness dependent on successional stage (Lopes et al. 2012). These variations could have a direct influence on the availability of the odor resource, and consequently on the chemical profile of perfumes available to male orchid bees.
The woody savanna remnants sampled were: CCPIU – Caça e Pesca do Itororó Club (18° 59' 20.89” S, 48° 18' 12.57” W), EUC - Eucatex Lumber Company (19° 02' 21.3" S, 48° 05' 32.7" W) and BR497 – a public reserve located between the Uberlandia-Nova Ponte road (19° 02' 04.0" S, 48° 32' 04.6" W). The CCPIU is a private reserve with approximately 127 ha, predominating the woody savanna and palm swamps (= Vereda). EUC is also a private reserve area with a typical woody savanna environment, approximately 100 ha in size and surrounded by forested environment within the savanna landscape. The third fragment sampled is an area located on the BR497 road has a size of 316 ha, with a predominantly savanna environment inserted within a matrix dominated by large areas of monocultures and small forest remnants.
Sampling of bees. We collected 11 males at each study area (except PER with 10 males) using bait traps containing the synthetic compound 1,8-cineole. Five bait traps were exposed at each locality for at least 3h, between 07:00 and 12:00 at a distance of 5–10 m between them and 1.5 m above ground. Chemical baits were covered by metallic mesh to keep away bees from having direct access to the synthetic compound. We collected males with entomological nets and placed in individual vials to cool them in a cooler. Later, we freeze the bees in a laboratory freezer. Perfume were sampled on the same day by cutting off individual right hind legs and placed them in a chromatographic vial for direct headspace.
Chemical analyses of perfumes. We used the conventional headspace technique with a gastight syringe to collect volatile compounds. One leg of each individual was placed in headspace vials (Agilent, Vial, HS, crimp, RB, 20ml). Vials were equilibrated for 5 min at 40°C. Later, we collected 0.1 ml of the headspace fraction and immediately injected into the gas chromatograph. The chromatography/mass spectrometry (GC/MS) was done at the ESTES – Escola Técnica de Saúde in the University of Uberlândia, Brazil, using an HP-5MS column (Agilent, 30 m length, 0.25 mm ID, 0.25 µm film thickness) and a model 5577 E selective mass detector (© Agilent Technologies, Santa Clara, California, United States). We programmed the equipment according to the specifications used in Zimmermann et al (2009) for chemical analysis of volatile and semi-volatile compounds in orchid bees, with some modifications. Samples injection was splitless (5 min), carrier gas was helium; flow rate – 1 mL/min, the temperature program was from 40°C to 150°C at 3°C/min and later up to 250°C at 5°C/min. The MS detector was a quadrupole with voltage at 1.2 kV, temperature – 250 C, and the mass spectral data acquisition scan interval was 1.0 s and data were collected over a mass range of 40–400 u.
We processed GC–MS data (retention index - RI, mass spectra and associated retention time) using the MassHunter GC/MS Acquisition software vB.07.00 (Agilent Technologies). To identify the compounds, retention times of integrated chromatographic peaks and mass spectra were compared and referenced with mass spectra and retention times of the compounds recorded in Adams (2017) and NIST (2017) libraries. Chromatogram peaks were detected and integrated using the Agile 2 integrator. We performed the same step for all the samples. Only peaks with an area greater than or equal to 1% of the largest peak were included in subsequent analyses. We found Lipids and aliphatic hydrocarbons previously reported as compounds derived from labial-cephalic glands secretions in males of other two Eulaema species (Zimmermann et al. 2006) (ex. some carboxylic ester, fatty alcohol). These compounds were excluded from the analyses because they are not exogenous substances (Eltz et al. 2005; Eltz et al. 2007; Zimmermann et al. 2009; Eltz et al. 2015; Brand et al. 2020).
We performed a Spearman rank correlation to test the association between the number of compounds and the integrated ion currents of each peak across all individuals (amount of perfume).
Wing Wear Measurements. Left and right forewings of each individual were removed at the wing base and mounted on glass slides. The slides were photographed using a Nikon COOLPIX P520 digital camera and images were saved as high-resolution .jpg files. For each wing image, wing wear (ww) was measured calculating the wing wear percentage by measurements of the total area of each wing using the ImageJ® program (Rasband, WS, ImageJ, US National Institutes of Health, Bethesda, Maryland, USA, 1997–2014) and subsequently, the missing area due to wear was calculated. To calculate ww, we used the following equation: % = [(Rww + Lww) * 100] / (RW + LW), where Rww and Lww are the right and left wing wear, respectively; RW and LW represent the total area of each wing. Based on the wing wear percentage (Fig. 2), we classified individuals in categories as follows: category 0 for individuals with no ww; individuals with ww > 0% and up to 0.3% in category 1; ww > 0.3–0.6% in category 2; ww > 0.6–0.9% in category 3; ww > 0.9–1.9% in category 4; and individuals with ww greater than 1.9% in category 5. These categories used were based on the wing wear model proposed by Rebêlo and Garófalo (1991). The method of wing wear percentage was developed by Pacheco and cols. (unpublished data) in order to gain greater precision in the wing wear measurements.
Body Size Measurements. Individual head was removed and carefully mounted on microscope slides. We photographed each slide with individual's head towards the lens of the camera. By saved images, we measured the maximum head width using the ImageJ® program.
Bouquet Similarity Analysis. To visualize differences in the perfume chemical profile between sampled physiognomies, we performed a non-metric multidimensional scaling (NMDS) analysis. We calculated Bray-Curtis index based on the similarity/dissimilarity of relative contributions of each absolute peak area (integrated ion currents) (Legendre and Legendre 1998). The resulting similarity matrix was ordinated on a two-dimensional graph by MDS. The points in the MDS plot represent distances that exactly match the rank order of dissimilarities between samples in the underlying similarity matrix. Deviations from this match are expressed in terms of "stress", where values <0.15 indicate a good representation of the overall structure of the matrix (Clarke and Warwick 2001).
To test the hypothesis that the factors “physiognomy” and “locality” had no effect on the perfume composition similarities, we used two-way PERMANOVA permutation tests (Anderson 2017). Later, we identified volatile compounds responsible for creating the similarity/dissimilarity patterns observed through the SIMPER algorithm, which shows the percentage contribution of each compound to the overall similarity within population. The index weighs both the extent and consistency (in comparisons of individual pairs) by which each compound contributes to the overall similarity (Clarke and Warwick 2001).
Effect of Wing Wear and Body Size on Perfume Complexity and Amount. To analyze the number of compounds in relation to morphological measurements, we used a Generalized Linear Model (GLM) with poisson distribution (link = log), fitting the error distribution with quasi-poisson distribution when necessary. Wing wear percentage and head width were the explanatory variable. Since there was no effect of head width on the number of compounds (response variable), we excluded this factor from the final model. After testing the effect of wing wear percentage on perfume complexity, we used a GLM to assess differences between the averages of the wing wear categories through a contrast analysis.
We evaluated the significance of the variables as follows: first, through the ANOVA model comparison function, we compared complex models with the simpler ones obtained by combining the variables; if simplification did not cause significant changes, we accept the simplest models, maintaining the principle of parsimony (Crawley 2013). Then, we submitted all fitted models to a residual analysis to verify the adequacy of the modeling and the normality of the error distribution. For all the analyses, we considered P < 0.05 values as statistically significant. We executed the same method described above substituting the amount of perfume as response variable instead the number of compounds.
All the analyses were performed in the RStudio 1.3.1093 program (R Development Core Team 2020; © 2009–2020 RStudio PBC). For perfume similarities analyses we used the packages “vegan” (Oksanen et al. 2018), “permute” (Simpson et al. 2019), “lattice” (Sarkar et al. 2018), “mgcv” (Wood 2020), “nlme” (Pinheiro et al. 2018), and “labdsv” (Roberts 2019).