The nutritional landscape of host plants for a specialist insect herbivore

Abstract Nutrition has far‐reaching effects on both the ecology and evolution of species. A substantial body of work has examined the role of host plant quality on insect herbivores, with a particular focus on specialist–generalist dynamics, the interaction of growth and other physiological attributes on fitness and tritrophic effects. Measures of plant quality usually involve one or two axes of nutritional space: typically secondary metabolites or elemental proxies (N and C) of protein and carbohydrates, respectively. Here, we describe the nutrient space of seven host plants of the specialist insect herbivore, Manduca sexta, using an approach that measures physiologically relevant sources of nutrition, soluble protein and digestible carbohydrates. We show that plant species differ markedly in their nutrient content, offering developing insect herbivores a range of available nutrient spaces that also depend on the age of the leaves being consumed. The majority of host‐plant species produce diets that are suboptimal to the herbivore, likely resulting in varying levels of compensatory feeding for M. sexta to reach target levels of protein to ensure successful growth and development. Low‐quality diets can also impact immune function leading to complex patterns of optimization of plant resources that maximizes both growth and the ability to defend from parasitoids and pathogens. This study is the first to quantify the nutrient space of a suite of host plants used by an insect herbivore using physiologically relevant measures of nutrition.

and fitness (Chen, 2008;Cotter, Simpson, Raubenheimer, & Wilson, 2010;Simpson & Raubenheimer, 2001; and that these effects can cascade upward to higher trophic levels (Sznajder & Harvey, 2003), in many systems we still lack a basic understanding of the nutritional landscape offered to insect herbivores over different spatial and temporal scales.
Quantifying nutritional quality of different diet sources can be challenging, as diets are multifaceted. Organisms must balance main nutritive components that are required to generate energy (carbohydrates, protein, and lipids) with each other, but also with micronutrients and the presence of secondary metabolites that may be detrimental to performance. The Geometric Framework (GF) is a conceptual framework that defines a given diet in multidimensional space, where each axis represents the level of an individual component, which can include macro or micronutrients, and even secondary metabolites (Simpson & Raubenheimer, 2001). This framework has been widely used to examine a variety of different topics in nutritional ecology, from target intake rates (Raubenheimer, Simpson, & Mayntz, 2009), behavior (Raubenheimer & Simpson, 1993), and optimal foraging theory (Simpson, Sibly, Lee, Behmer, & Raubenheimer, 2004). Here, we use the GF to explore nutrient space across host-plant species and plant tissues on two important axes (Behmer, 2009;Deans, Behmer, Fiene, & Sword, 2016): plant soluble protein and digestible carbohydrates.
Elemental analysis of plants has long been the standard for quantifying protein and carbohydrate components with nitrogen acting as a proxy for protein and carbon acting as a proxy for carbohydrates (Berner, Blanckenhorn, & Körner, 2005;Deans et al., 2016;Telang, Booton, Chapman, & Wheeler, 2001). Though this approach has generated useful research outcomes (Joern et al., 2012), researchers have recently made the case for using more direct measurements of soluble protein and digestible carbohydrates (Deans et al., 2016). This argument stems from the fact that some percentage of carbon and nitrogen atoms exist in non-nutritive compounds in plant tissue (i.e., cellulose, lignin, and allelochemicals), making linkages from elements to nutritive biochemicals difficult. Though current methods (Deans et al., 2018) do not account for species-specific differences in gut biochemistry, they likely make better proxies for nutrient content than traditional elemental analysis (Deans et al., 2018). Here, we take an insect-centric approach and use methods outlined by recent work (Deans et al., 2016) to examine soluble protein and digestible carbohydrates in a suite of host plants used by a specialist insect herbivore: Manduca sexta (Sphingidae).
Manduca sexta has a wide distribution across tropical and temperate regions of the Nearctic and is classified as a specialist herbivorefeeding on a variety of host plants within Solanaceae (Yamamoto & Fraenkel, 1960), but has also adopted a nonsolanaceous host (Proboscidea spp., Martyniaceae) (Mechaber & Hildebrand, 2000) (  (Diamond & Kingsolver, 2011), fitness , and predator avoidance (Mira & Bernays, 2002), but to our knowledge, the nutritional landscape of the range of host plants available has yet to be explored, despite strong evidence for interactions between nutritional components and performance, fecundity, and immune function , 2011Wilson et al., 2019 for a given species, and only one of a few to use these nutrient assays to show the insect-centric nutrient space using the geometric framework (Deans et al., 2016;Li, Volenec, Joern, & Cunningham, 1996;Ojeda-Avila, Woods, & Raguso, 2003;Sánchez, Rivero, Ruiz, & Romero, 2004;Stieger & Feller, 1994). By examining patterns of variation in nutrition across a wide range of host plants, we gain deeper insight into the ecological and evolutionary dynamics that shape fundamental relationships between plants, herbivorous insects, and third trophic levels.

| Lyophilization and material processing
Our protocol for sample preparation and analysis is adapted from work by Deans et al. (2018). After reaching −80°C, samples were taken out of the freezer and placed in two Eppendorf tube trays. The trays of samples were stacked into the main chamber of a lyophilizer (Virtis Freezemobile 6) where they were processed for a minimum of 48 hr, at a pressure between 100 and 150 Millitorr and a temperature of −40°C. After lyophilization, samples were stored at room temperature. The shelf-stable samples were then preground using a ceramic mortar and pestle and then placed into a Mills grinder that was set to run on the highest level of grinding for 30 s.

| Soluble protein
We measured 20 mg of lyophilized leaf tissue in duplicate into 1.5 ml Eppendorf tubes and added 500 μl of 0.1 M NaOH. We then sonicated tubes for 30 min (Branson 2510, Branson Ultrasonics) and placed tubes in a 90°C water bath (VWR 5L Avantor) for 15 min.
Tubes were then centrifuged for 10 min at 13,000 rpm (7,558 g, VWR Galaxy 16D, Avantor), and the supernatant transferred to another 1.5 ml Eppendorf tube. We pipetted 300 µl of 0.1 M NaOH to the Eppendorf tube with the pellet and centrifuged for another 10 min at 13,000 rpm (7,558 g). Then, we transferred the supernatant to the other Eppendorf tube and added 11 µl of 5.8 M HCl.
We vortexed tubes for a few seconds each, so HCl would distribute evenly throughout the sample and measured the pH of three randomly chosen samples (out of a batch of 4-8, depending on the run) to ensure that we had achieved a neutral pH. We pipetted 90 µl of 100% TCA solution into samples and placed them on ice for 30 min, to facilitate protein precipitation. We then centrifuged samples again (at 13,000 rpm for 10 min), removed the supernatant, and washed pellets with 100 µl of −20°C acetone. Acetone was removed and samples were allowed to air-dry in a 15°C refrigerator for approximately 10-20 min. Samples were visually inspected to ensure that excess acetone had evaporated, after which protein pellets were resuspended with 1 ml of 0.1 M NaOH. This process was facilitated by vortexing, placing samples back in the 90°C hot water bath, and by adding 20-40 µl of one normal NaOH when needed. Samples were then diluted by mixing 50 µl of the resuspended solution to 950 µl of deionized water. We then added 60 µl of each new diluted solution to a 96-well plate in triplicate (to generate technical replicates of each sample-these were subsequently averaged to generate the raw values for further analyses) and added 100 µl of deionized water to each sample wells. Between 4 and 8 individual leaf samples (in triplicate) were run per batch. We also added 160 µl of IgG standard solutions (0-12.16 µg of protein across six samples) in triplicate to the plate. A total of 40 µl of BioRad protein assay dye reagent was added to each well, and the entire plate was incubated at 23°C for 5 min. Absorbance was measured at 595 nm in a spectrophotometer (Thermoscientific Multiskan GO, Thermo Fisher Scientific), and we used 2nd-order polynomial regressions for soluble protein standard curves because these models provided better fits than linear models.

| Digestible carbohydrates
We measured 20 mg of lyophilized leaf tissue and added them to glass vials with 1 ml of 0.1 M H 2 SO 4 in duplicate. Vials were placed in a water bath at 100°C (a beaker with deionized water on a hot plate) for an hour. After samples cooled, we transferred them from the glass vials to 1.5 ml Eppendorf tubes and centrifuged them for 10 min at 13,000 rpm. We transferred 15 µl of supernatant from each sample to glass test tubes with 385 µl of deionized water each; we also prepared six glass test tubes containing 0-0.1875 µg/µl glucose (our standard curve). We added 400 µl of 5% phenol to test tubes and pipetted 2 ml of concentrated H 2 SO 4 (95%-98%) onto the surface of the solution. Test tubes were allowed to sit for 10 min, after which we vortexed them on a low setting and allowed them to sit for another 30 min. Finally, 1 ml of each sample and standard was transferred to cuvettes and run in duplicate on the spectrophotometer at 490 nm. We used linear regression models for digestible carbohydrate standard curves.

| Data analysis
All analyses were performed in R (www.r-proje ct.org; 3.5.0 'Joy in Playing

| Soluble protein and digestible carbohydrate content
Host-plant species differed significantly in both the percentage of soluble protein (F = 3.3943, NumDF = 6, DenDF = 41.666, p = .008; Figure 1a; Table 2) and digestible carbohydrate (F = 11.841, NumDF = 6, DenDF = 66.350, p < .0001; Table 3) after controlling for the time period between when a plant was first transplanted (for protein -we didn't control for this in carbohydrates because of issues with singular fits in model generation) and when we sampled leaves (Table 2). Plants species differed more in digestible carbohydrate content than in soluble protein content, and also had higher variance ( Figure 1). N. attenuata had the lowest amount of soluble carbohydrates of all species with an average of 10.3 ± 6% while D. wrightii had the highest with an average of 43.5 ± 15% (Figure 1a).
Additionally, N. tabacum had low levels of soluble protein with an average of 6.9 ± 5% while C. annuum had the highest at 15.6 ± 9.7%. On average, young leaves had higher p:c ratios than old leaves (mostly due to higher protein levels; Figure 1a

| D ISCUSS I ON
The importance of interacting nutritional components of diet has long been recognized as a fundamental factor that shapes the ecol-  appropriately), and the importance of phenology and among-and within-plant nutritional heterogeneity in structuring herbivore-plant interactions (Bernays & Chapman, 1994). However, one of our main results is striking: None of the common host plants we measured provide p:c ratios for M. sexta that are optimal for growth.
We measured p:c ratios of individual leaves on multiple individuals from a given species-each leaf can be visualized in nutrient space as a rail-a line with a slope that is the p:c ratio (Simpson & Raubenheimer, 1995). M. sexta larvae move among different leaves on a single plant throughout development (Casey, 1976;Potter, Bronstein, & Davidowitz, 2012) and have leaves of all ages to feed on. In Figure 4, we plot the nutrient rails of both young and old leaves for all species, which illustrates a few key points about the system. First, we show in previous work that M. sexta caterpillars from a long-standing colony in our laboratory need p:c ratios of approximately 1:1 for optimal growth . When combined with results from the current study in Figure 4, a few key points emerge. Importantly, almost all plant species we measured demonstrated p:c ratios that were less than optimal (low in protein).
Some plant species (N. attenuata and C. annuum) provide diet sources that would allow larvae to reach target intakes without substantial compensatory feeding, but other common host plants (N. tabacum and D. wrightii) typically have leaves that are relatively low in soluble protein, meaning that larvae would have to eat more leaf material to reach target values (Figure 4). These patterns suggest a mechanism that explains high reported leaf damage in some species (e.g., N. tabacum and D. wrightii;Casey, 1976;Potter et al., 2012;Wilson & Woods, 2015): Larvae are forced to eat extra leaf material to acquire enough protein, even on younger leaves with higher protein content (Figure 4), though this pattern may be complicated by the inclusion of secondary metabolites. However, previous work has shown that M. sexta is relatively tolerant to defensive compounds produced by plants in Solanacae (Wilson, Tseng, Potter, Davidowitz, & Hildebrand, 2018;Wink & Theile, 2002). Second, there are substantial differences in the nutritional environments available to larvae developing on different host species (highlighting the potential impact of oviposition site selection), but also the range of nutrient space available to larvae moving between young and old leaves.
Work in other plant species has demonstrated large shifts in nutrient composition over longer developmental times (Deans et al., 2016), and the impact of lag time between when we sampled leaves and when plants were first transplanted echoes this pattern (Tables 2   and 3). Combined, these results suggest that while there are marked differences among old and young leaves and different host plants, the nutritional landscape is complex for both female M. sexta moths deciding when and where to lay eggs (Potter et al., 2012) and for developing larvae trying to maximize growth and other components of fitness. Larger colored points represent the centroid for each group, with their size being scaled to the summed standard deviation for each axis (a larger centroid point indicates more variance in nutrient space for a given species). The diagonal dotted line represents a 1:1 ratio between carbohydrate and protein content Though growth and body size are often good proxies for fitness (Arnold, 1983), other physiological attributes can have consequences for survival and reproduction. Previous work has shown a strong link between different aspects of the insect immune system and nutrition (Cotter et al., 2010;Wilson et al., 2019 (Diamond, Hawkins, Frederik Nijhout, & Kingsolver, 2010). These results are contrary to the expectations from our data (Figure 4), where P. parviflora appears to be a higher quality host plant than N. tabacum (closer to a 1:1 p:c ratio). This apparent contradiction draws attention to the complex nature of these interactions and highlights the idea that nutrition interacts with other plant traits (such as secondary metabolites) to affect herbivore fitness (Chen, 2008). Though P. parviflora may be nutritionally superior to N. tabacum, it also has physical defenses (sticky trichomes) that make it difficult to process, particularly for young larvae (Mira & Bernays, 2002), in addition to a suite of defensive secondary compounds that could affect M. sexta performance, immune function, and fitness.
Ultimately, a given host plant is a combination of components including macro-and micro-nutrients and secondary metabolites, so while there are clear differences among host plants in two axes (soluble protein and digestible carbohydrates), other axes that were not measured in this study could have significant impacts on herbivore performance and fitness.
Large-scale spatial and temporal variation in host plant quality and M. sexta adaptation may also drive patterns of host use dynamics. For example, different genetic lines of M. sexta varied in their performance on N. tabacum and P. louisianica, with wild lines being better adapted for high performance on the novel (and presumably lower quality) P. louisianica . It is still unclear whether M. sexta is locally adapted to different host species across its wide range, and though M. sexta is an incredibly well-studied species (e.g., Kanost, Jiang, & Yu, 2004), surprisingly little is known about its population structure and connectivity.
Spatial and temporal variation in host plant quality is also likely important in structuring interactions between M. sexta and its hosts.
For example, localized variation in microclimate or soil chemistry can impact plant quality, both in terms of nutrient composition (Mankin & Fynn, 1996)  The slopes of Blue lines represent the 1:1 ratio of protein to carbohydrates. Solid lines represent nutritional rails for old leaves whereas dashed lines represent nutritional rails for young leaves. Gray lines are raw values whereas black lines represent the median value for each group. The black dot represents the average target intake (1.587 g of carbohydrates and 1.894 g of protein) of a M. sexta larvae over their development (as measured by previously- Wilson et al., 2019). Rails to the left of the 1:1 P:C rail indicate that the larvae are protein constrained on that host-plant species and rails to the right of the 1:1 P:C rail are carbohydrate constrained one species in particular (N. attenuata) had a nutrient composition that is of higher quality compared with other species (a more optimal p:c), it also has lower total nutrient content compared with other species, meaning that M. sexta larvae (and other insect herbivores) would need to eat substantially more leaf material to reach intake targets. This study is one of only a few that has used the geometric framework to examine the nutrient space of plants from an insectcentric nutritional perspective (soluble protein and digestible carbohydrates), and the first to our knowledge to describe the nutrient space of a suite of host plants used by a single insect herbivore. By better understanding the physiologically relevant nutritional landscape provided to insect herbivores by host plants, we gain insight into the ecological and evolutionary dynamics that shape the interactions between these two key groups in terrestrial ecology.

ACK N OWLED G M ENTS
This study was supported in part by National Science Foundation grant IOS-1053318 to G.D. and by the Center for Insect Science

National Institutes of Health Postdoctoral Excellence in Research
and Teaching (PERT) grant K12GM000708. We would like to thank Shuzo Oita with his methodological support, as well as the laboratory of Dr. Patricia Stock for their use of equipment.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
JKW and GD conceived experiments and the project. JKW, LR, and JD carried out experiments and analyses. JKW led the writing of the manuscript, with support from LR, JD, and GD. All authors contributed critically to the drafts and gave final approval for publication.

DATA AVA I L A B I L I T Y
All data and code for analyses (R scripts) are archived and available on Zenodo (https ://doi.org/10.5281/zenodo.3420883).