Study Site and Study Organism. The study was conducted at La Selva Biological Station, province of Heredia, Costa Rica, between February and March 2021. La Selva, managed by the Organization for Tropical Studies (OTS), encompasses a 1,536-hectare protected lowland area comprising a mix of primary and secondary forests and abandoned plantation areas (McDade et al. 1994). In the study site, short-tailed bats are one of the most abundant groups of fruit bats (Carollia, Phyllostomidae). Carollia bats are the primary seed dispersers of pepper plants (Piper, Piperaceae), a diverse genus of flowering plants (Fleming 2004; Maynard et al. 2019; Santana et al. 2021).
Secondary Metabolite Selection. Based on previous phytochemical studies in Piper spp. (Salehi et al. 2019) and commercial availability, we selected the alkaloid piperine (285.34 g/mol, Sigma-Aldrich), the polyphenolic tannic acid (1701.20 g/mol, Sigma-Aldrich), the phenolic eugenol (164.20 g/mol, Sigma-Aldrich), and the terpene phytol (296.53 g/mol, Cayman) to test how different classes of secondary metabolites affect the foraging behavior and digestive physiology of bats. We conducted the experiments described below using concentrations of 0.1%, 2%, and 3% of the dry weight of an artificial diet for each metabolite. Similar ranges of concentrations have been reported for these and other secondary metabolites in Piper spp. plants (Salehi et al. 2019).
Bat Capture and Maintenance. We captured bats using mist nets placed in clearings and secondary forest sites. Upon capture, we released juvenile bats and pregnant females, while adult male and non-reproductive female bats were transferred to individual flight cages (2x1x1 m) within the forest. We utilized 30 bats for the study, housing them in three groups of ten individuals during three rounds of trials (see below). All bats acclimated to the flight cages the first two nights, where they were fed on a maintenance diet of water, agar powder (Eco-Taste), mashed ripe banana, soy protein isolate powder (Bulk Supplements), NaCl, CaHPO4 (Eisen-Golden Laboratories), vegetable oil, and wheat germ (Bob's Red Mill), using the ingredient proportion suggested in Denslow et al., (1987). After finishing each nightly trial, bats were fed 35 g of unsupplemented maintenance diet and water ad libitum. We cleaned the bottom of the cages daily with a diluted bleach solution (1/10). After finishing all the trials, we released the bats at the capture site.
Objective 1. The Effects of Secondary Metabolites on the Foraging Behavior of Captive Bats. To assess the effect of metabolite identity and concentration on C. perspicillata preference, we performed non-choice trials to measure the amount of food consumed by each bat within a 30-minute timeframe. Each bat received the four metabolites and two controls (unsupplemented maintenance diet) in a randomized sequence, with either one treatment or control given per night. These trials were repeated with the three groups of bats, and each group received the metabolites in a different concentration in the artificial diet: group 1 received 0.1% of the metabolites, group 2 received 2%, and group 3 received 3% (N = 10 bats per treatment/compound concentration). We offered approximately five grams, equivalent to about 0.8 grams of dry weight of the experimental diet in a plastic Petri dish at around 7:00 p.m., when bats are actively foraging in their natural habitat. After 30 minutes, we recorded the weight of the Petri dishes.
Objective 2. The Effects of Secondary Metabolites Consumption on the Bat Fecal Metabolome. After recording the amount of food consumed for objective 1, Petri dishes were returned to the flight cages, allowing the bats to consume the remaining diet. Approximately three hours later, we collected the fecal samples resulting from the initial five grams of food offered. To separate feces from urine, we positioned a fine plastic mesh above the cage floor, lined with a plastic sheet, ensuring the fecal samples remained on the mesh while the urine passed through it. We used a clean spatula to collect the fecal samples and stored them in plastic microcentrifuge tubes. Fecal samples were not collected from bats that did not entirely consume the diet. We stored fecal samples at − 80°C for later analysis in the laboratory, except during transport on dry ice from Costa Rica to Virginia, USA.
Fecal Metabolome Analysis. Frozen fecal samples were dried using a Speedvac at 65°C, 100 mTorr for four hours. As bat fecal samples are potentially contaminated with Histoplasma capsulatum, we decontaminated them by adding 1 mL of isopropanol to approximately 10–20 mg of fecal samples. The isopropanol was evaporated using a Speedvac at a temperature of 65°C, vacuum 100 mTorr for around three hours. Then, we added 10 uL of 1 µg/µL ribitol as the internal standard and 500 µl of 80% methanol. We vortexed the samples for 5 seconds, sonicated them for 15 minutes, and then shook them for two hours on an orbital shaker at 140 rpm at room temperature. We centrifuged the samples at 13000 g for 15 minutes. We collected 400 µl of the supernatant from each sample into a glass micro insert and evaporated the solvent at 65°C, vacuum 100 mTorr for one hour. For the derivatization reaction, we added 40 µL of 20 mg/mL methoxyamine hydrochloride in pyridine and incubated for 90 minutes at 60°C. Then, we added 40 µL MSTFA (N-methyl-N-(trimethylsilyl)trifluoroacetamide) + 1%TMCS (chlorotrimethylsilane) reagent and incubated for 90 minutes at 60°C. Finally, we injected the samples into an Agilent 7820 gas chromatograph paired with a 5977 mass spectrometer equipped with an HP5-MS column (Agilent Technologies, Santa Clara, CA, USA). We converted all Agilent files (.D) to AIA format (.CDF) using ChemStation and processed chromatography data, including peak peaking and alignment, using the R package metaMS (Wehrens et al., 2014). We saved every metabolite spectra as .msp files and matched them using the NIST MS search 2.0. In each chromatogram, we removed the peaks detected in the blanks and normalized the instrument response by dividing the peak area of each peak by the peak area of the internal standard and the dry weight of each sample.
Statistical Analysis. All the analyses were performed using R version 4.2.1 (R Core Team 2021). To investigate the effects of the four selected secondary metabolites (piperine, tannic acid, eugenol, and phytol) on the foraging behavior of captive bats, we first determined a consumption ratio. This ratio was calculated by dividing the total amount consumed (g) per treatment per individual bat by the average food consumption (g) recorded in the two unsupplemented controls offered to the same bat. The resulting ratio provides insight into the relative preference of each bat for particular secondary metabolites compared to the controls. Ratios greater than one indicate a preference for the treatment over the control, ratios less than one indicate a preference for the control, and ratios of one suggest no preference, implying no deterrent effect of the treatment. Bats not participating in the control trials were excluded from subsequent analyses. We assessed the distribution of the ratios using the 'shapiro.test()' function. Given the non-normal distribution of the ratios, we conducted a non-parametric one-sample Wilcoxon test using the function ‘wilcox.test()’, with the mu parameter set to 1, to assess whether the ratios of consumption at a given concentration (0.1, 2, and 3% dw) were significantly different from 1.
To examine how secondary metabolite consumption affects the composition of the bat fecal metabolome (objective 2), we used a set of non-metric multidimensional scaling (NMDS) analyses with the Bray-Curtis dissimilarity index. One model was conducted for each concentration and set of trials. To assess the statistical support for differences in composition across compounds, we first examined the homogeneity of group dispersions (PERMDISP2) using the 'betadisper()' function, followed by a permutational multivariate analysis of variance (PERMANOVA) using the 'adonis2()' function with 999 permutations. To identify the individual excreted metabolites that distinguish the fecal metabolome of bats that ingested different secondary metabolites at various concentrations, we used random forest classification models followed by variable selection using the Boruta algorithm, using the ‘randomForest()’ and ‘Boruta()’ functions respectively. Random forest and Boruta analyses suggest a set of candidate molecules that potentially differ in the fecal metabolome of bats that consumed different secondary metabolites at a given concentration. Then, we tested the direction, i.e., increase or decrease of each excreted metabolite among the treatments, and the statistical support for differences among treatments using a set of generalized linear mixed models (GLMMs) for each excreted metabolite. The GLMMs included each excreted metabolite identified in the Boruta analysis as the response variable, ingested secondary metabolite identity (treatment) as the predictor variable, and bat identity and trial date as random effects in the models coded as random intercepts, i.e., (1|Bat) + (1|Date). Finally, for each excreted metabolite suggested in random forest and Boruta analyses, we obtained the chemical taxonomy classification using the package ‘ClassyFireR’ (Djoumbou Feunang et al. 2016).
Multivariate analyses, including NMDS, PERMANOVA, and multivariate homogeneity of variances (PERMDISP2) were conducted using the ‘vegan’ package (Oksanen et al. 2020). Random forest classification models and variable selection via the Boruta algorithm were performed using the ‘randomForest’ (Liaw et al. 2015) and ‘Boruta’ (Kursa and Rudnicki 2017) packages. The set of GLMMs was constructed using the ‘glmmTMB’ package (Brooks et al. 2017).