- Top of page
- Materials and Methods
- Supporting Information
Chemistry plays a dominant role in species interactions, especially between plants and herbivores. Measurement of chemical concentrations important to species interactions provides insight into the mechanisms driving plant allocation of resources in response to herbivory. The underlying factors that drive plant chemical responses to herbivory provide a framework within which we can test ecological and evolutionary theory (Dyer, 1995; Thompson, 1999; Agrawal, 2001; Guisan & Thuiller, 2005; Mooney et al., 2010; Agrawal et al., 2012a).
In particular, a key area of interest for understanding the ecology and evolution of plant–herbivore interactions is inducible plant defenses. Inducible plant defenses are thought to have evolved as a mechanism to minimize the cost of production of defenses, given the unpredictability of herbivore attack (Herm & Mattson, 1992; Karban & Baldwin, 1997). The ability of plants to rapidly mount defensive responses to herbivore damage is important in reconciling variations in plant metabolic tradeoffs, reproductive success, community composition and higher level trophic interactions (Herm & Mattson, 1992; Malcolm, 1995; Malcolm & Zalucki, 1996; Karban & Baldwin, 1997; Thaler et al., 2001; Kessler & Baldwin, 2002; Mooney et al., 2010; Agrawal et al., 2012a).
In general, studies examining the inducibility of plant defenses rely on the sampling of combinations of damaged and undamaged plants over time, or repeatedly sampling the same plant tissue to determine changes in plant chemistry. Both of these methods require destructive sampling of plant tissue for chemical analysis. The former approach relies on the assumption that all plants respond similarly, and necessitates a large and sometimes impractical sample size to appropriately replicate the time points. The latter method relies on the assumption that the damage from tissue collection itself does not elicit further chemical responses. This method also becomes impractical when examining a rapidly induced chemical response, because it requires multiple collections from the same tissue source over a short time period.
Reflectance spectroscopy provides an alternative approach to the investigation of rapid phytochemical changes because the method is noninvasive, relatively inexpensive, fast and can be applied to living tissue (reviewed in Foley et al., 1998). The use of reflectance spectroscopy to characterize foliar chemistry is based on the influence of illumination on the harmonic vibrations between atoms or among groups of atoms, specifically C–H, N–H and O–H bonds, the main constituents of organic material, at specific wavelengths in the visible, near-infrared (NIR) and shortwave infrared (SWIR). Spectral measurements of organic material collected under uniform and stable illumination provide the basis for the estimation of the chemical composition of a sample (Shenk et al., 1992). The calibration is accomplished by pairing spectra with reliable chemical measurements, from which chemical data are modeled as a function of spectra using multivariate methods. The calibration model is validated by comparing relationships between observed and predicted values to determine the robustness of the model. This calibration model can then be used to predict phytochemical concentrations of the remaining unknown samples on the basis of their spectral signature alone. An often underappreciated benefit of spectroscopy is the ability to collect information on multiple foliar constituents simultaneously from a single spectral measurement.
Common milkweed (Asclepias syriaca) is an ideal system to test the ability of spectroscopy to track rapid changes in phytochemistry. Milkweed has been studied extensively to understand the relationships between phytochemical variations and the ecology and evolution of plant–insect interactions (Malcolm, 1991, 1995; Mooney et al., 2010; Agrawal et al., 2012a), with the prime phytochemical focus being cardenolides. Cardenolides are phloem-mobile, steroidal toxins that disrupt cellular ion transport (Malcolm, 1991; Agrawal et al., 2012a), and damage to common milkweed is known to rapidly increase foliar cardenolide concentrations (Malcolm & Zalucki, 1996; Agrawal et al., 2012a). Cardenolides are potent chemical defenses in milkweed plants and are often implicated in the reduction in performance of herbivores specializing on milkweed (Malcolm, 1995; Malcolm & Zalucki, 1996; Zalucki et al., 2001; Agrawal, 2005; Agrawal et al., 2012b). Although most prevalent in the family Apocynaceae, cardenolides are also found in a number of other plant families (Kreis & Müller-Uri, 2010; Agrawal et al., 2012a).
Although cardenolides are an important component of milkweed–insect interactions, multiple plant variables (e.g. nitrogen, C : N, leaf toughness, trichomes and latex) are also known to influence herbivore performance in this system (Agrawal & Fishbein, 2006). For example, in addition to cardenolides, milkweed is also characterized by the exudation of latex in response to damage that negatively affects the feeding of chewing insects by gumming their mouthparts (Dussourd & Eisner, 1987). Moreover, latex can contain cardenolide concentrations more than two orders of magnitude greater than those in leaves (Zalucki et al., 2001). It has been suggested that the increase in foliar cardenolide concentrations in response to damage is caused by the rapid influx of latex, as opposed to localized foliar upregulation of cardenolide biosynthesis or latex-independent transport into the leaf (Zalucki et al., 2001; Agrawal et al., 2012a). However, current methodological approaches make the elucidation of the mechanism responsible for a rapid increase in cardenolides difficult. The ability to nondestructively determine real-time, in vivo chemical changes in response to environmental perturbation can greatly enhance our understanding of the relationships among phytochemical variation and species interactions, and the role they play in ecological and evolutionary theory.
Here, we describe a novel approach to quantify rapid phytochemical changes in response to damage. Utilizing recent advances in spectroscopy, we characterize the induction profile of cardenolides in A. syriaca in response to damage, noninvasively track these changes through single plants and examine the influence of latex on foliar cardenolide concentrations.
- Top of page
- Materials and Methods
- Supporting Information
Predictive models produced using PLSR accurately characterized fresh leaf cardenolide concentrations in A. syriaca, with average R2, RMSE and bias values of 0.851, 0.221 μg mg−1 and 0.007, respectively (Fig. 1a–c). The distribution ranges for R2, RMSE and bias reported also suggest that the model predicting cardenolides is relatively stable, as all models produced a high R2 (range, 0.54–0.95), RMSE values (range, 0.11–0.36 μg mg−1) between 5% and 15% of the mean and minimal bias (range, −0.16 to 0.18). The range of cardenolides reported in the plants used in this study, determined by standard chemical analysis, ranged from 1.0 to 3.1 μg mg−1, and was closely matched by the range of cardenolides predicted via spectroscopy: 0.9–3.3 μg mg−1 (Fig. 1d). Calibration coefficients for raw spectra are reported in Supporting Information Table S1. Foliar water status, determined by NDWI, was not affected by damage treatment in plants used to build the calibration (t = −0.21, P = 0.82) or in plants with notched petioles (t = 0.67, P = 0.41). Petiole notching, however, visibly reduced latex flow to damaged areas on leaves (J. Couture, pers. obs.).
Figure 1. Error distribution of (a) R2, (b) root-mean-square error (RMSE) and (c) bias for validation data generated via cross-validation using 1000 random subsamples of the data for calibration (70%) and validation (30%). (d) Observed vs predicted values of cardenolide (CG) concentrations of Asclepias syriaca.
Download figure to PowerPoint
Substantial variation existed in reflectance among damaged and undamaged leaves and across collection periods (Fig. 2a). Overall, the evaluation of the strength of the PLSR loadings by wavelength, determined using VIP selection, indicated the highest contribution to the model of cardenolide concentrations in the general areas of major peaks from purified digitoxin, especially in the visible and NIR regions (Fig. 2b). The general pattern of variation in reflectance of damaged, relative to undamaged, leaves was similar for plants used either to build the model or repeatedly sampled (Fig. 2c,d). Relative variation in reflectance was most pronounced within 24 h and generally largest in the visible and SWIR regions (Fig. 2c,d). Relationships between foliar traits that might co-vary with cardenolides and cardenolide concentrations themselves were nominal (Table 1).
Table 1. Pearson correlations among foliar levels of cardenolides, nitrogen, leaf mass per area, phytochemical reflective index, normalized differential vegetation index and normalized differential water index
|Cardenolides||1.00|| || || || || |
|Nitrogen||−0.313||1.000|| || || || |
|LMA||−0.324||−0.558||1.000|| || || |
|PRI||0.180|| 0.493 ||−0.502||1.000|| || |
|NDWI|| 0.306 ||−0.168||−0.029||0.001||−0.202||1.000|
Figure 2. (a) Average, ± SD, and minimum and maximum Asclepias syriaca leaf reflectance. (b) Variables important to the projection (VIP) plot showing the relative strength of the contribution of partial least-squares regression (PLSR) loadings by wavelength to the model predicting cardenolide concentrations (red line) overlaid on the first derivative of the reflectance spectra of purified digitoxin (black line). (c) Relative variation in reflectance of leaves at 0.25, 1, 72 and 124 h post-damage for plants used to build the predictive model relating leaf reflectance and cardenolide concentrations. (d) Relative variation in the reflectance of leaves at 0.25, 1, 72 and 124 h post-damage from repeatedly sampled plants.
Download figure to PowerPoint
Foliar cardenolide concentrations varied significantly among collection times and treatments (Table 2). Cardenolide concentrations increased over the first 24 h in response to damage and then decreased to levels similar to constitutive, undamaged plants (Fig. 3). Estimates of cardenolide concentrations derived from spectra for the seven unharvested, but damaged, A. syriaca plants matched the induction profile of the harvested plants, in which concentrations were determined by standard chemical analysis (open circles in Fig. 3). Importantly, this shows that estimates of cardenolide concentration based exclusively on reflectance spectroscopy of intact live leaves exhibit the expected trends based on the analytical chemistry of harvested samples.
Table 2. Two-way ANOVA examining the effects of damage (undamaged vs damaged) and time on foliar cardenolide concentrations of Asclepias syriaca of plants used to build the calibration model
|Treatment||df|| F || P |
|Damage||1, 78||77.7||< 0.001|
|Time||5, 78||28.6||< 0.001|
|Damage × time||5, 78||6.5||< 0.001|
Figure 3. Cardenolide (CG) induction profile of Asclepias syriaca collected at 0.25, 1, 24, 72 and 124 h post-damage. Damaged (closed circles) and undamaged (closed triangles) values were generated from the predictive model. Repeat measured (open circles) values were generated by repeatedly collecting reflectance from the subset of seven plants damaged at the same time as those used to build the predictive model, but not themselves included in model building. Constitutive (dotted line) is the average cardenolide concentration, determined using the predictive model, of the 10 plants harvested before damage. Values are means ± SD.
Download figure to PowerPoint
Comparison of intact and modified leaves, whose petioles were notched to reduce latex flow, revealed that the enhanced expression of foliar cardenolides in response to damage was mostly eliminated in leaves in which latex flow was reduced (Table 3, Fig. 4). This result suggests a strong relationship between cardenolide concentrations and latex, and further suggests that latex exudation is a probable mechanism for the increase in leaf-level cardenolides in response to tissue damage.
Table 3. Repeated-measures ANOVA of the effect of petiole modification and time on foliar cardenolide concentrations of Asclepias syriaca
|Treatment||df|| F || P |
|Petiole modification||1, 59||64.8||< 0.001|
|Time||4, 59||41.1||< 0.001|
|Petiole modification × time||4, 59||19.7||< 0.001|
Figure 4. Cardenolide (CG) induction profile of Asclepias syriaca collected at 0.25, 1, 24 and 72 h post-damage in leaf pairs in which one leaf had its petiole modified to reduce latex flow (closed circles) and the other petiole remained intact (open circles). Constitutive (dotted line) is the average cardenolide concentration of leaves before damage. Values are means ± SD.
Download figure to PowerPoint
- Top of page
- Materials and Methods
- Supporting Information
We successfully characterized phytochemical variation in milkweed in response to damage using reflectance spectroscopy. Our approach follows on from numerous recent advances in chemical spectroscopy in which multivariate modeling techniques, such as PLSR, are increasingly being utilized to estimate foliar biochemical, nutritional and morphological traits (Gillon et al., 1999; Petisco et al., 2006; Asner & Martin, 2008; Kleinebecker et al., 2009; Asner et al., 2011; Serbin et al., 2012). Most studies predicting foliar chemistry, however, use dried and ground leaf material. Although spectroscopy on dried leaf material saves time and the cost of processing and chemically analyzing samples, it is not possible to repeatedly sample the same leaf. As a consequence, reflectance spectroscopy on green leaves offers great promise for the characterization of plant responses to perturbation.
Our findings are similar to those of Ebbers et al. (2002), who were able to accurately characterize the concentrations of specific terpenoids and phenolics from fresh leaves of Eucalyptus. In our study, water absorption features (1450 and 1950 nm) partially obscured some spectral features of a purified cardenolide, digitoxin. Nevertheless, we were able to predict cardenolide concentrations of A. syriaca, with some of the wavelengths of importance occurring in areas identified for the purified digitoxin standard. Model coefficient loadings strongly contributing to cardenolide predictions, determined using VIP analysis, appeared in general areas of major features of purified digitoxin, the analytical standard, in particular in the visible and SWIR regions, and on the shoulders of the major water absorption features (1450 and 1950 nm, Fig. 2b). Specifically, key spectral features associated with cardenolide concentrations appeared at wavelengths of c. 670, 819, 1159, 1389, 1494 and 1890 nm.
The alignment of spectral features of purified digitoxin with PLSR coefficient loadings was not exact, but this result was not surprising as digitoxin, although a cardenolide, is not known to be present in A. syriaca. The value in this analysis is comparative: digitoxin is a commonly used cardenolide standard with strong similarities to the cardenolides in A. syriaca. Cardenolides found in the genus Asclepias are different from digitoxin (e.g. stereochemistry) and vary among themselves (e.g. differences in the number of hydroxyl groups on the steroid nucleus) in chemical characteristics (Malcolm, 1991; Agrawal et al., 2012a). These variations are probably responsible for differences in digitoxin spectra and VIP values in Fig. 2(b). Ultimately, the detection of cardenolides can be complicated by water absorption features, multiple forms of the compound of interest and other leaf morphological characteristics (e.g. leaf wax cuticular structure or trichomes). The ability to predict the phytochemical composition of fresh leaves represents an important step in the accurate estimation of real-time changes in plant metabolism in response to biological perturbation and environmental variation (Serbin et al., 2012).
In response to damage, leaves of A. syriaca can rapidly increase their cardenolide concentrations, and we observed that cardenolide levels peaked after 24 h in response to mechanical damage. Different types of damage elicit different cardenolide induction profiles in A. syrica depending on damage type (i.e. herbivory vs simulated herbivory), herbivore feeding guild (i.e. chewing vs sucking) and herbivore identity within feeding guilds (Agrawal et al., 2012a). Foliar tissue removal generally elicits an influx of latex, and has been suggested to be the reason for the rapid elevation of cardenolide levels in response to damage, as milkweed latex can contain cardenolide concentrations orders of magnitude larger than those of foliar tissue (Zalucki et al., 2001). Moreover, localized cardenolide biosynthesis is less likely to occur as quickly in response to damage as does increased latex flow to a damaged area (Agrawal et al., 2012a). By notching leaf petioles and reducing latex flow into the leaf, we observed that the induction of foliar cardenolide concentrations was mostly eliminated. Such an analysis was facilitated by the ability to measure cardenolides nondestructively (i.e. using reflectance spectroscopy), whilst simultaneously measuring and manipulating multiple factors. Although a more detailed understanding of the genetic and enzymatic upregulation of cardenolide biosynthesis is needed to reveal the underlying cause of the increased cardenolide expression in milkweed leaves following damage, our findings are consistent with the conclusion that the rapid induction of foliar cardenolides in response to damage is probably driven by latex influx (Zalucki et al., 2001; Agrawal et al., 2012a).
The correlations of several plant traits and commonly used vegetation indices derived from reflectance spectroscopy (i.e. PRI, NDVI and NDWI) with cardenolide concentrations were relatively low, which indicates that our method appears to be sensitive to cardenolide concentrations rather than other widely measured characteristics that are also correlated with plant performance. The plant trait exhibiting the strongest relationship with cardenolide concentration was LMA (−0.324), potentially suggesting a trade-off between the allocation of leaf carbon to growth or metabolite synthesis. Overall, LMA varied among all collection periods for damaged plants (F = 5.7, P = < 0.001), but was not statistically significantly different in the first three collection periods post-damage (data not shown). Moreover, the inclusion of leaf mass and LMA as covariates in the ANOVAs of cardenolide concentrations did not alter the significant effects of time or treatment on total cardenolide amounts (data not shown). The minimal relationship of cardenolides with foliar nitrogen and LMA, and other commonly used reflectance-derived vegetation indices (Table 1), demonstrates that our ability to detect the rapid increase in foliar cardenolide production over a short (< 24 h) time period is not a byproduct of the relationships among other plant variables.
A key component of this study is the ability of spectroscopy to provide chemical and morphological information regarding multiple plant constituents simultaneously with a single spectral measurement. We were able to generate foliar nitrogen and LMA with the same spectrum as used to determine cardenolides. Although the calibrations used to determine foliar nitrogen and LMA were generated from models built using Populus spp. leaves (Serbin et al., 2012), the measurements reported (1.2–5.2% leaf nitrogen and 48.7–94.8 LMA) fall within the values reported in the literature for A. syriaca (Nagel & Griffin, 2004; Zehnder & Hunter, 2007). The ability to simultaneously determine changes in multiple plant constituents on fresh, rather than harvested, leaves undoubtedly provides a more complete understanding of the physiological changes in plants in response to herbivory.
By repeatedly sampling reflectance from the same leaf, we successfully tracked changes in phytochemical concentrations, thereby reducing the need for progressive harvesting of foliage over the course of an induction study. Although not identical, the reflectance patterns of leaves used to build the model and leaves repeatedly sampled were similar, suggesting that changes in leaf optical properties varied in a predictable manner. To our knowledge, this is the first study demonstrating the ability of spectroscopic measurements to follow a rapid chemical induction in vivo in response to foliar damage. Few studies have explored the ability of leaf optical properties to nondestructively characterize real-time changes in plant chemical properties (Bilger et al., 1989; Gamon & Surfus, 1999). We have built on these previous studies by demonstrating the utility of leaf optical characteristics, in our case reflectance spectroscopy, to describe variations in plant responses to foliar damage.
The rapid induction of plant defenses in response to herbivory has received considerable attention as a mechanism by which plants can avoid the costs of producing defenses whilst maintaining their resistance benefits (Karban & Baldwin, 1997). As such, the inducibility of defenses has been used to help elucidate theories in plant ecology and evolution (Herm & Mattson, 1992; Rasmann & Agrawal, 2009). Our results demonstrate a unique application of existing technologies in spectroscopy to characterize phytochemical variation in response to damage. Our findings highlight the potential of these techniques for the study of plant–environment interactions, specifically because of the rapid and repeatable determination of multiple plant traits simultaneously.
Our study demonstrates the capacity to estimate cardenolide concentrations in milkweed using reflectance spectroscopy, but we recognize possible limitations in our measurement approach for the calibration estimates of cardenolide concentrations. Our chemical assay relies on the measurement of the absorbance of a chromophore produced from the reaction between the butenolide portion of cardenolides and TNDP. Although this method is highly correlated with cardenolide concentrations produced via more sensitive chemical assays, such as HPLC (Rasmann et al., 2009), TNDP can also react with foliar components other than cardenolides, including ketones, pregnane glycosides and plant pigments (Malcolm et al., 1989). In addition, our analytical approach does not discern among individual cardenolides; thus, we are unsure of the sensitivity of our model to the prediction of specific cardenolides. Ultimately, the comparison of models using estimates produced via multiple methodologies will demonstrate whether more sensitive chemical assays used for calibration data are needed to reduce residual error in predictive models and to demonstrate the ability of our approach to predict individual cardenolides. However, our results are within the range of total cardenolide concentrations reported in the literature, and replicate the expected cardenolide induction profile through time as a consequence of simulated herbivory seen by Malcolm & Zalucki (1996).
Our analysis was based on plants from one population, and so the range of cardenolide prediction of our model does not completely capture the full range of variability in cardenolides across all genotypes of A. syriaca in northern North America (Bingham & Agrawal, 2010; Vanette & Hunter, 2011). As such, our model probably breaks down at cardenolide concentrations outside the range sampled in our study; however, the cross-validation results, combined with the nominal relationships of cardenolides with foliar nutrient and morphological traits, suggest that our model is reliable within the range of cardenolide concentrations measured here, and is robust against variation in plant nutritive status and physical characteristics. The inclusion of a wider concentration range of cardenolides from a variety of genotypes, and species, of milkweed, together with more sensitive chemical information, would probably increase model precision at lower concentrations. Regardless of these known limitations, we have demonstrated the effectiveness and ability of repeatedly measured spectroscopic data to accurately characterize phytochemical variation in fresh milkweed leaves.
Our work demonstrates that ecologically relevant secondary metabolites can be quantified and tracked successfully in fresh foliage using reflectance spectroscopy in the visible, NIR and SWIR wavelengths. In addition, we have shown that variation in these compounds in response to environmental perturbations can be detected in vivo through the resampling of the same leaf. Although based solidly in sound analytical chemistry, the expansion of research relating hyperspectral data and multivariate modeling to a broad array of secondary metabolites and physical traits can provide novel insights into the ecology and evolution of plant–herbivore interactions by rapidly and simultaneously measuring multiple plant characteristics.