Study system.
We studied two species of Asota (A. eusemioides and A. heliconia) across three species of Ficus (F. septica, F. pachyrrhachis and F. hispidioides). Because of the complexity of the taxonomy of Asota (Holloway 2022), we became aware that A. heliconia was in fact a duplex of two cryptic species that are partially overlapping across the wide range of what was originally considered one species (Holloway 1988, Supplementary Information). However, these two species show segregation with altitude in New Guinea. This enabled us to establish that we were dealing with only the lowland member of the pair in Madang. A more detailed account is presented in Supporting Information. Our field collections also showed that caterpillars from each of the two Asota species were largely restricted to only one or two Ficus species. We found A. eusemioides only on F. hispidioides and F. pachyrrhachis, while almost all A. heliconia caterpillars collected for our experiment were from F. septica. Only two A. heliconia caterpillars were found on another Ficus species (F. hispidioides); they were excluded from our study.
Field experiment.
We conducted a two-part experiment in Ohu, Madang Province, Papua New Guinea (− 5.140° 145.410° 200 m) from February to April 2018. Because conditions and rearing approach were consistent across the entire study period, we combined experimental data from the first and second parts for analysis so that four individual trees from each Ficus species were involved (n = 12). While a larger host sample size would have been preferable, replication across insects on the same host plant individual is required in our study. Metabolic analysis of resource and herbivores can only establish compound fate if all resource compounds have equal possibility to be found further down the trophic chain. And as we were mainly interested in how insects vary in their responses to a common resource, maximising variation at the resource level is not desirable. Moreover, designing complex experiments in remote tropical rainforests are often shaped by logistical limitations. Thus, given our research focus and the resources at hand, we decided to maximise replication at the level of the organism being studied (i.e., the insect) and control for repeated measures at all steps.
Larval development in Asota. The first part of our experiment was aimed at rearing caterpillars on each host to adulthood; this allowed accurate identification of species and developmental stage, and confirmation that body size increased predictably across instars (Dyar 1890). We selected two individual trees from each host species (n = 6) and attempted to rear five individual first instar caterpillars (collected from the appropriate host species in the surrounding forest) to adulthood (n = 30). Individual caterpillars were placed on a single branch of the Ficus species from which they were sampled and enclosed in a fine breathable mesh for rearing (Fig. S2). Larvae were reared on young, fully expanded and healthy leaves until their last instar, and frass was regularly removed (at least daily) to keep conditions as clean as possible. Caterpillars were followed from 13/02/2018 until 01/03/18, by which time 21 out of 30 larvae had entered pupation. Eleven larvae were reared to adulthood and stored at − 20°C before being freeze dried. Caterpillar body length was measured to the nearest 0.1 cm on a daily basis. We also collected 10 leaves from a separate branch of the same tree before placing the caterpillars (n = 48; leaves from one tree were excluded and one tree was sampled for eight leaves). These leaves were placed in an ice box and transferred to a − 20°C freezer before being freeze dried at the Binatang Research Centre.
Chemical variation across sample types. For the second part of our experiment, we selected an additional two individual trees from each host species, for a total of 12 trees. For each tree, we aimed to place 15 individual caterpillars of the appropriate Asota species on individual branches covered by mesh bags. Due to the local availability of first instar caterpillars, however, we ended up with 29 caterpillars on F. septica, 32 on F. hispidioides and 31 on F. pachyrrhachis, for a total of 92 caterpillars. As with the first part, caterpillars were allowed to feed only on young and healthy leaves.
Once caterpillars had moulted into the last instar, ca. one half (n = 44) were freeze dried and collected as larval samples while the other half (n = 48) were reared to adults. The former were starved for six hours prior to freeze drying to ensure that there was no contamination from plant metabolites. Caterpillars in the rearing treatment were allowed to feed for 24 hours after their final moult before being placed into pots with fresh leaves. Their frass was collected every 45 minutes for four and a half hours. The caterpillars were then returned to their host plants to pupate naturally, and adult moths were freeze dried after emergence. A total of 17 caterpillars from the rearing treatment died in the last instar and were removed; 15 of the 17 deaths were due to parasitoid attack. We also removed four caterpillars and one adult (and their frass samples) due to possible degradation to give a total of 70 insect samples.
Our aim was to collect six frass samples per caterpillar, for a total of 30 frass samples per tree. Because some samples were pooled in the chemical analysis due to low mass, we ended up with 170 frass samples instead of 180. In these cases, samples from the same individual—but not across individuals—were pooled. Frass was frozen immediately upon collection, as were the leaves upon which caterpillars had fed, and were later freeze dried. As the leaves on which the caterpillars fed were in generally poor condition and potentially influenced by caterpillar feeding itself, we again collected and freeze dried 10 leaves from a separate branch of the same tree to serve as baseline comparison.
After accounting for excluded specimens, a total of 362 samples were used for subsequent analyses. The number of samples for each tissue type and adult body part are reported in Table 1. All materials were sent to the University of Turku for chemical analyses.
Table 1
A table of tissue types collected in the study. For adults we analysed three different sections of the body. Note that 17 caterpillars and their associated frass sampled were removed for statistical analysis (please see Chemical variation across sample types).
Sample Type
|
n
|
Adult
|
39
|
Caterpillar
|
31
|
Leaf
|
106
|
Tree
|
12
|
Frass
|
108
|
Chemical analyses.
We first grouped our samples based on tissue type (i.e., leaves, frass, caterpillars and adults). Adult samples were then further dissected based on body parts into three subgroups: i) body (which, for this study, was composed of the head, thorax and abdomen), ii) wings, and iii) legs and antennae (Table 1). All samples were ground into fine powder by a ball mill. We macerated 10 mg with 1800 µl methanol overnight in cold room, and then extracted via sonication in a water bath for 30 min, centrifuged at 14,000 rpm for 10 minutes. Supernatants were decanted to new Eppendorf tubes and the methanol was evaporated in an Eppendorf concentrator. Samples were then dissolved in 1000 µl of 5 mM aq. HCl, filtered through 0.20 µm PTFE filters and pipetted into an UHPLC wellplate prior to the UHPLC–MS/MS analyses.
UHPLC–MS/MS analyses were conducted on an Acquity UPLC system coupled with a DAD detector (Waters Corporation, Milfor, MA, USA) and a hybrid quadrupole-Orbitrap mass detector (Q Exactive™, Thermo Fisher Scientific GMbH, Bremen, Germany) via HESI source (H-ESI II, Thermo Fisher Scientific GmbH, Bremen, Germany). The column was Acquity UPLC BEH Phenyl
(30 × 2.1 mm i.d.; 1.7 µm; Waters Corporation, Wexford, Ireland). The mobile phases consisted of acetonitrile (A) and 0.1% formic acid (B). The eluent profile was as follows: 0–0.1 min, 3% A in B (isocratic); 0.1–3.0 min, 3–45% A in B (linear gradient); 3.0–3.1 min; 45–90% A in B (linear gradient), 3.1–4.0 min 90% A in B (isocratic); 4.0–4.1 min; 90 − 3% A in B (linear gradient), 4.1–4.2 min; 3% A in B (isocratic). The eluent flow rate was 0.65 ml/min and injected volume 5 µl. The mass spectrometer was operated in a positive ionization mode with mass range of m/z 120–1000 and with lock mass. The following parameters were used for the positive ionization: spray voltage set at 3.0 kV, N2 sheath gas flow rate at 60 arbitrary units, N2 aux gas flow rate at 20 arbitrary units, capillary temperature 380°C and S-lens RF level at 60. Orbitrap resolution for full scan was 70,000 (full MS) with an automatic gain of 3×106. Data dependent MS/MS-spectra was obtained with a resolution of 17,500.
Post analysis data handling was done using Thermo Xcalibur Qual Browser software (Version 3.0.63, Thermo Fisher Scientific Inc., Waltham, MA, USA), Compound Discoverer 3.1 (Thermo Fisher Scientific Inc., Waltham, MA, USA) and MZmine version 2.53 (Katajamaa et al. 2006; Pluskal et al. 2010) to achieve quantitation of all possible ions (as the extracted ion area / mg dry weight of tissue). Compound Discoverer utilized the following parameters: untargeted metabolomics workflow template with mass tolerance of 5 ppm, intensity threshold of 30%, S/N threshold of 3, minimum peak intensity of 1×106, and maximum element count of C×100, H×200, Cl×4, N×10, Na×4, O×100, P×3 and S×5. For peak detection the following parameters were used: filter peaks true, maximum peak width of 0.5 min, remove singlets true, minimum # scans per peak 5 and minimum # isotopes 1.
Tentative identification of the detected molecules was based on the use of the GNPS analysis environment and its feature based molecular networking (FBMN) (Horai et al. 2010; Pence and Williams 2010; Wang et al. 2016; Nothias et al. 2020). All UHPLC-MS/MS datafiles were converted to mzXML format using ProteoWizard MSConvertGUI (version 3.0.19316). LC-MS feature detection and alignment for GNPS was done using the following MZmine methods and parameters:
-
Mass detection: MS1 noise level of 1×105 and MS2 noise level of 0
-
ADAP chromatogram builder: minimum group size in # of scan of 5, group intensity threshold of 3.0×105, minimum highest intensity threshold of 1.0×105 and m/z tolerance of 0.01 Da
-
Chromatogram deconvolution: local minimum search algorithm with chromatographic threshold of 3.0×105, search minimum in RT range of 0.03 min, minimum relative height of 4.0%, minimum absolute height of 1.0×105, minimum top peak ratio of 1.4 and peak duration range of 0.01–0.50 min. Chromatogram deconvolution also used MS2 scan pairing at m/z range of 0.01 Da and RT range of 0.05 min.
-
Isotopic peak grouper: m/z tolerance of 0.01 Da, RT tolerance of 0.05 min and maximum charge of 3
-
Join aligner: m/z tolerance of 0.01 Da, weight for m/z at 75, retention time tolerance of 0.05 min and weight of RT at 25
-
Gap filling: peak finder (multithreaded) method with intensity tolerance of 25.0%, m/z tolerance of 0.001 Da and retention time tolerance of 0.01 min
-
Feature list row filter: minimum peaks in an isotope patter selected with value of 2 and keep only peaks with MS2 scan (GNPS) selected
GNPS parameters used for the FBMN are as follows: parent mass tolerance of 0.01 Da, ion tolerance of 0.01 Da, minimum pairs cos of 0.7, minimum matched peaks of 6, network TopK of 10, minimum cluster size of 2 and maximum connected component size of 100. The FBMN was additionally visualized with Cytoscape (version 3.8.1).
Alkaloids found to be indicative of sample type (see Statistical analyses for details on alkaloid selection) were further analysed by MS/MS to elucidate their structures on the basis of accurate masses of the molecules and fragments, and their corresponding double bond equivalents. Potential alkaloid-like compounds were statistically analyzed and 17 focal compounds were selected for closer manual examination of their molecular ions (based on CCA scores on the first two axes, please see below for more detail), MS/MS -spectra and corresponding double bond equivalent (DBE) which led to structural characterization of an isoquinoline alkaloid (A5), three pyridoindoles (A9, A16 and A28), four seco-phenantroindolizidines (A17, A22, A24, A35), three seco-phenantroindolizidines (A25, A36, A43), two seco-phenantroindolizidine-N-oxides (A27 and A32), one phenantroindolizidine (A30) and three alkaloid-like compounds (A37, A38 and A41) that we could not classify further.
A5, characterized as dihydro-dimethoxy-dihydroisoquinolinium, showed a molecular ion at m/z 224.09176 and a matching molecular formula C11H13NO4. UV absorption maxima were observed at 204, 231 and 340 nm. Highly conjugated A30 was characterized as dimethoxy-dihydro-dibenzo-pyrroloisoquinolinol at m/z 346.14421, molecular formula C22H25NO3 and UV absorption maxima at 259, 281 and 340 nm.
A9, A16 and A28 showed molecular ions at m/z 232.10796, 215.08168 and 231.07640, respectively, and molecular formulas C12H13N3O2, C12H10N2O2 and C12H10N2O3. A9 was characterized as amino-tetrahydro-pyridoindole-carboxylic acid with UV absorption maxima observed at 225, 256 and 362 nm. A16 was characterized as dihydro-pyridoindole-carboxylic acid with UV absorption maxima observed at 250, 283 and 361 nm. A28 was characterized as hydroxy-dihydro-pyridoindole-carboxylic acid.
A17 and A22 shared the molecular formula C23H27NO5 with molecular ions at m/z 398.19639 and 398.19640, respectively. These alkaloids were characterized as isomers of dihydroxy-trimethoxy-seco-phenantroindolizidine. MS/MS-spectra obtained for A17 and A22 had major fragments at m/z 70 matching dihydropyrrole and at m/z 218 matching dimethoxyphenol-dihydropyridine. UV absorption maxima obtained for 17 at 241, 266 and 343 nm and for 22 at 235 and 269 nm. A24 was characterized as dihydroxy-seco-phenantroindolizidine with molecular ion at m/z 308.16440 and with molecular formula C20H21NO2. For A24, MS/MS-spectra showed major fragments at m/z 188, 186, 70, 107 and 212, bearing similarities with other seco-phenantroindolizidines. A35 had molecular ion at m/z 350.21112 and was characterized as dimethoxy-methyl-seco-phenantroindolizidine with molecular formula C23H27NO2.
A27 and A32 had both major MS/MS fragment at m/z 86, characterized as N-hydroxydehydropyrrol, indicating presence of N-oxide moiety in the structure. A27 was characterized as hydroxy-dimethoxyphenyl-methoxyphenyl-hexahydro-indolizine-4-oxide with a molecular formula C23H27NO5, molecular ion at m/z 398.19605 and UV absorption maxima at 245, 261 and 343 nm. A32 was characterized as dimethoxyphenyl-methyloxoniophenyl-hexahydro-indolizine-4-oxide with matching molecular ion at m/z 382.20112 and molecular formula C23H27NO4.
A25 was characterized as hydroxy-trimethoxy-seco-phenantroindolizidine at m/z 382.20120 with molecular formula C23H28NO4 and matching MS/MS-spectra had major fragments at m/z 70 and 218 (Lee et al. 2011). For A36 molecular ion was obtained at m/z 352.19038 and molecular formula C22H25NO3 with matching characterization of hydroxy-dimethoxy-seco-phenantroindolizidine (Stærk et al. 2002). MS/MS fragmentation revealed several fragments at m/z 235, 121, 84, 266, 135, 125, 86, 334, 202 and 159. A43 was characterized as trimethoxy-seco-phenantroindolizidine with molecular ion at m/z 366.20663 and molecular formula C23H27NO3 (Stærk et al. 2002). MS/MS fragmentation revealed major fragments at m/z 70 and 202.
Statistical analyses.
Larval development in Asota. We used linear mixed models, implemented using the R package ‘nlme’ (Pinheiro et al. 2022), to test how caterpillar body size increased across instars. Body length was the response variable, instar and caterpillar species were the fixed categorical explanatory variables, and individual caterpillar was used as the random explanatory variable to account for multiple measures of the same individual dates as it grew. We also tested the size differences among instars of A. eusemioides across its two hosts (F. hispidioides and F. pachyrrhachis) using similar mixed models, but with host species instead of caterpillar species as the explanatory variable.
Chemical variation across sample types. Based on shared conditions for collection and rearing, we used the combined data from the first and second parts of the experiment in subsequent analyses. For the purposes of initial data exploration, we first ran a partial Principal Components Analysis (pPCA) to visualise the total variability in chemical composition between different tissue types and body parts. The effect of the individual was removed by conditioning the ‘community’ matrix on a vector coding for the individual sampled in cases of paired measures. Alkaloid concentration in area per g dry weight (DW) was log transformed. Mean values across individual host or insect were used when multiple samples of the same type were taken from the same individual.
Our experimental design necessitated the collection of frass from insects subsequently reared to adults, causing us to have some paired data points. To explore and control for this non-independence we ran four analyses. We first ran a standard CCA (i) in which the response variable was the same ‘community’ matrix dataset used in the pPCA, and the explanatory variables were tissue type and host Ficus species. Because A. eusemioides is found on both F. hispidioides and F. pachyrrhachis while A. heliconia is restricted to F. septica, the variable ‘insect species’ is collinear with ‘host species’. Host species provides greater resolution and information content and is preferred for data exploration. An additional CCA (ii) included an extra explanatory variable, a vector called ‘individual’, was used to group any paired observations (e.g., frass and the insect from which it was collected). Next, we performed pCCA (iii) using the same set of variables from ii), with the effect of the individual removed by conditioning the ‘community’ matrix on a vector coding for the individual sampled. Finally, to further control for the possible influence of the individual on compound selection, we ran a standard CCA (iv; with the same formulation as above) on a reduced data set in which alternating adult or frass data points were removed when both were recorded from the same individual. For all CCAs model simplification proceeded through stepwise permutation tests (999 permutations) in both directions, and adjusted R2 was used as the stopping criterion. The significance of explanatory variables were summarised as an ANOVA table. All multivariate analyses were conducted in the R package ‘vegan’ (Oksanen et al. 2020) which implements the CCA following Legendre and Legendre (2012).
Additionally, sampling multiple body parts from the same adult allowed us to compare uptake of chemicals in the same individual. To control for the effect of individual, we visualised the adult body part dataset by using pCCA and adding a vector that grouped all samples taken from the same individual (as used in pPCA). Model simplification and significance followed that above as used for tissue type.
Differences in compound occurrence across tree species and sample groupings were initially tested using linear mixed models implemented in ‘lme4’ (Bates et al. 2015). These models were tested for dispersion, outliers, distribution of residuals and zero-inflation using the R package ‘DHARMa’ (Hartig 2020). Because our dataset comprised many zeros, we included a single zero-inflation parameter applying to all observations using the R package ‘glmmTMB’ (Brooks et al. 2017). A Gaussian distribution of errors was appropriate for all models. The response variable, alkaloid concentration (peak area/g DW), was log transformed, and the explanatory variables used for the models depended on which dataset was being used. For the tissue type dataset, we used host tree species and sample type (with adults split into each species) as explanatory variables. For the body part dataset, we used sample type (with adults split into each species) as the explanatory variable. We only split adult moths by species, assigning caterpillars and frass to species would create too many multiple comparisons for meaningful interpretation. We performed Tukey’s HSD tests for linear combinations of each explanatory variable using the R package ‘multcomp’ (Hothorn et al. 2008). As our central question necessitated the use of the same individual across or within developmental stages, we included ‘individual’ as a random effect to all our mixed models.