Context: An increasing number of ecological studies have used chemical diversity as a functionally relevant, scalable measure of phytochemical mixtures, demanding more rigorous attention to how chemical diversity is estimated. Most studies have focused on the composition of phytochemical mixtures and have largely ignored structural concerns, which may have greater importance for ecological function. Here we explore the development of structural complexity and compositional diversity resulting from different biotic and abiotic interactions in Piper kelleyi Tepe (Piperaceae). We also describe how variation in structural complexity and compositional diversity differ between two congeners, P. kelleyi and Piper reticulatum. To better interpret these results, we have developed a hypothesis-driven framework for interpreting these dimensions of chemical diversity in phytochemical mixtures.

Approach: We used the tropical shrub, P. kelleyi, as a model system to examine interactions between ecological factors and dimensions of phytochemical diversity. We also compared compositional diversity and metabolic complexity in P. kelleyi and P. reticulatum using liquid chromatography and ¹H NMR respectively to examine tradeoffs between compositional diversity and structural complexity. A framework is provided to generate meaningful estimates of the structural complexity of phytochemical mixtures as measured by ¹H NMR.

Results and Conclusions: Piper is an abundant plant genus that supports diverse insect communities throughout the tropics. Subtle changes in understory forest light were associated with increases in herbivory that directly increased compositional diversity and indirectly decreased structural complexity in P. kelleyi. This was attributed to the production of oxidation products resulting from herbivory-driven decomposition of structurally complex defense compounds. This type of complex result would remain undetected using standard chemical ecology approaches and accounts for the detailed molecular changes that are likely to affect species interactions.

Synthesis: Our quantitative framework provides a method for considering tradeoffs between structural complexity and compositional diversity and the interpretation of analytical approaches for each. This methodology will provide new theoretical insights and a more sophisticated model for examining the ecology and evolution of chemically mediated interactions.

Piper reticulatum (Piperaceae) foliar samples were collected from individual plants > 10 m apart along both sides of the Sendero Surá (55-255 m), Sendero Arriera-Zampopa (110-804 m) and Sendero Tres Rios (155-2450 m) trails in La Selva Biological Station (LS, 10°25’ N 84°00' W), Costa Rica. In a prior experiment (Glassmire et al., 2019), we collected P. kelleyi from experimental plots of clones in Yanayacu Biological Station (YBS, 00°36′ S 77°53′ W) in Ecuador grown at high and low canopy height along an elevational gradient (2000-2400 m) of the Eastern Andes mountains. Canopy cover and direct light transmittance of P. kelleyi plants were measured using a Canon EOS Rebel-T4 camera with a hemispherical fisheye lens and processed using Gap Light Analyzer software (GLA software methods, Frazer et al., 1999). Herbivory was estimated from actual and estimated (pre-damage) leaf area using Image J before arcsine square root transformation (Glassmire et al., 2019). All Piper samples were finely ground using a tissue lyser (TissueLyser II, Quiagen; Hilden, Germany) and 100 mg portions from each individual plant were weighed before adding 10 mL of methanol (Fisher, Optima). Methanolic suspensions were sonicated for 10 min and extracted overnight with wrist-action shaking before filtering over a cotton plug and concentrating to dryness in vacuo. 

Nuclear magnetic resonance spectroscopy

Dried Piper extracts were re-dissolved in 1.00 mL d₁-methanol for deuterium exchange in order to minimize water signal (as HOD) in ¹H NMR spectra. An aliquot of this solution was diluted 1:10 using protonated methanol for LC-MS analysis before the remainder was dried in vacuo. A second deuterium exchange was performed before dissolving in d₄-methanol containing 0.05% tetramethylsilane (TMS) as an internal standard and acquiring ¹H NMR (32 scans) on a Varian MR-400 NMR. Raw FIDs were processed using MestreNova (Mestrelab Research, S.L. Santiago de Compostela, Spain), including auto-phasing, ablative baseline correction and global spectral deconvolution (GSD) peak picking. Spectra were exported to csv files and normalized to total peak area after removing solvent and TMS regions (0-0.5, 3.3-3.32, 4.79-4.92, and 7.25-7.28 ppm) before binning at 0.04 ppm. Bins with an area below 10^-4 were set to zero before calculating diversity measures.

Liquid chromatography-UV/VIS-mass spectrometry

Piper extracts were characterized via liquid chromatography-mass spectrometry (LC-MS), using an Agilent (Santa Clara, CA) 1200 analytical HPLC equipped with a binary pump, autosampler, column compartment and diode array UV/Vis detector, coupled to an Agilent 6230 Time-of-Flight mass spectrometer via an electrospray ionization source (ESI-TOF; gas temperature: 325 °C, flow: 10 L/m; nebulizer pressure: 35 psig; VCap: 3500 V; fragmentor: 165 V; skimmer: 65 V; octopole: 750 V). Extracts (0.20 μL) for P. kelleyi were co-injected with anofinic acid internal standard (1.00 uL, 0.2 M) and eluted at 0.500 mL/min through a Kinetex EVO C18 column (Phenomenex, 2.1 x 100 mm, 2.6 μ, 100 Å) at 40 °C. The linear binary gradient was comprised of buffers A (water containing 0.1 % formic acid) and B (acetonitrile containing 0.1 % formic acid) changing over 20 minutes accordingly: 0-1 min 20% B, ramp to 50% B at 6 min, ramp to 100% B at 12 min, 12-16 min hold at 100% B, 16-17 min ramp to 20% B, 17-20 min hold at 20% B. Raw LC-MS data were converted to mzML format using ProteoWizard (Kessner et al., 2008) before processing using the XCMS package in the R statistical programming software (Team, 2014). Chromatographic features were retention time corrected and aligned using density grouping. LC-MS features were sum-aggregated based on CAMERA classification (Kuhl et al., 2012) as to not overstate compositional richness. LC-UV chromatograms were integrated (l = 254) using Agilent MassHunter. This wavelength was chosen because it is representative of the relative concentrations of the NMR-apparent molecules in P. kelleyi, which allows for valid inferences about the composition of molecules in crude ¹H NMR spectra.

Compositional diversity vs. metabolic complexity

Compositional diversity (D_C) or metabolic complexity (D_M) can be calculated as Richness, Shannon or Simpson diversity, or beyond into higher q values using analogous methods for calculating Hill numbers for community data (Marion et al., 2015). Metabolic complexity arises in chemical mixtures, such as a crude plant extracts, from the aggregation of molecules, each with their own structural complexity, in proportion to their composition. Figure 1 outlines how the ¹H NMR spectra of individual molecules combine to yield crude mixture spectra within our framework for understanding the properties of chemical mixtures and their constituent molecules that yield metabolic complexity. If one were to analyze a hypothetical crude extract, separation methods such as liquid chromatography (LC) and gas chromatography (GC) provide a compositional profile of this mixture wherein each peak represents a different molecule (Fig. 1, compositional diversity). If each of these compounds is isolated to obtain a ¹H NMR spectrum, each spectrum reflects the complexity of that molecule (Fig. 1, structural complexity). The structural complexity index of a molecule is calculated in the same fashion as one would for compositional diversity, but instead of each peak representing a whole molecule, it represents a structural feature of that molecule; all peaks in a spectrum represent the spectral fingerprint of an entire molecule.

Recombining these individual ¹H NMR spectra according to their relative abundance yields a ¹H NMR spectrum that reflects the crude mixture before separation. Similar structural features of molecules will lead to overlapping ¹H NMR signals that increase abundance for those signals, lowering signal evenness. Dissimilar structural features will have non-overlapping signals, leading to higher peak richness (Fig. 1, structural dissimilarity). One can therefore view the diversity of crude ¹H NMR as a gross measure of metabolic complexity, incorporating the composition, complexity and dissimilarity of phytochemical mixtures.

Effective structural complexity (D_Seff)

Once we have obtained compositional diversity (D_C, from LC) and metabolic complexity (D_M, from ¹H NMR) as either Richness, Shannon or Inverse Simpson diversity, we can calculate effective structural complexity (D_Seff) as follows:

Having removed the compositional contribution to metabolic complexity, effective structural complexity represents the remaining structural contribution. In terms of richness, this could be thought of as the average number of ¹H NMR peaks per compound or as an abundance-weighted average for higher order Hill numbers. While seemingly simple, effective structural complexity represents both the structural complexity and dissimilarity of molecules in a mixture.

Metabolic complexity (D_M) of a mixture can be decomposed into beta dissimilarity (β_D) from mean structural complexity (D_S) of all constituent molecules in a mixture, weighted to their concentration, as described for mean alpha diversity using community data by (Jost, 2007):

Highly dissimilar mixtures will lead to high metabolic complexity, even when the constituents of those mixtures have low structural complexity. Structural dissimilarity (D_D) can be further decomposed from β_D and compositional diversity:

which yields a number between 0 and 1 that can be interpreted as the fraction of signal overlap between constituents of a phytochemical mixture per compound. Mixtures with high structural dissimilarity (near one) have very little signal overlap, while mixtures having low structural dissimilarity have high signal overlap per compound. While this decomposition provides theoretical insight into partitioning chemical diversity, it is impractical for experimental data where structures and spectra of individual phytochemicals within a mixture are unknown. However, it is possible to estimate compositional diversity (D_C) of phytochemical mixtures using hyphenated analytical methods (GC-, LC-), which can then be used to decompose structural complexity from metabolic complexity as in equation 1. If we combine equations 1-3, they simplify to give effective structural complexity as:

Effective structural complexity results from both structural parameters of chemical mixtures: structural dissimilarity and mean structural complexity. High D_Seff can result from highly dissimilar mixtures or highly complex individual compounds.

These data include the measures of phytochemical diversity generated in this study and the ecological data with which they are associated. ReadMe files will be uploaded with keys describing column names in tables.