Importance of local vs. geographic variation in salt marsh plant quality for arthropod herbivore communities

Authors


Correspondence author. E-mail: laurie.marczak@cfc.umt.edu

Abstract

Summary

  1. An important recent advance in food web ecology has been the application of theory regarding spatial gradients to studies of the factors that affect animal population dynamics. Building on extensive studies of the Spartina alterniflora food web at the local scale, we hypothesized that geographic variation in S. alterniflora quality is an important bottom-up control on food web structure and that geographic variation in S. alterniflora quality would interact with the presence of predators and top omnivores to mediate herbivore densities.
  2. We employed a four-factor fully crossed experiment in which we (i) collected plants from high- and low-latitude locations and grew them in a common garden and varied (ii) plant fertilization status (mimicking the plant quality differences due to marsh elevation), (iii) mesopredator density and (iv) omnivore density.
  3. Our results suggest that the single most important factor mediating insect herbivore densities is local variation in plant quality – induced in our experiment by fertilization and demonstrated repeatedly as a consequence of marsh elevation.
  4. Top-down effects were generally weak and in those cases where predators did exert a significant suppressing effect on herbivores, that impact was itself mediated by host-plant characteristics.
  5. Finally, despite observed variation in plant quality with latitude, and the separately measurable effects of this variation on herbivores, geographic-scale variation in plant quality was overwhelmed by local conditions in our experiments.
  6. Synthesis. We suggest that a first-order understanding of variation across large latitudinal ranges in the Spartina alterniflora arthropod food web must begin with local variation in plant quality, which provides strong bottom-up forcing to herbivore populations. A second-order understanding of the arthropod food web should consider the role of predation in controlling herbivores feeding on low-quality plants. Finally, while latitudinal variation in plant quality probably explains some variation in herbivore densities, it is probably more of a response to herbivore pressure than a driver of the herbivore dynamics. Although extrapolating from local to geographic scales presents multiple challenges, it is an essential task in order for us to develop an understanding that is general rather than site-specific.

Introduction

An important recent advance in food web ecology has been the application of theory regarding spatial gradients to studies of the factors that affect animal population dynamics (McGeoch & Price 2005; Post 2005). Historically, ecologists debated the importance of top-down (Hairston, Smith & Slobodkin 1960) and bottom-up (Ehrlich & Birch 1967) factors in regulating herbivore populations, but most ecologists now agree that these factors interact, sometimes in complex ways, to influence population dynamics (Hunter & Price 1992). With this new, nuanced perspective, ecologists are now asking how variation in abiotic factors and community composition across landscapes affects the relative importance of top-down and bottom-up forces (Walker & Jones 2001; Denno, Lewis & Gratton 2005). Most of this work has focused on local and regional scales, with few studies at large geographic (= continental) scales that span large gradients in climate, oceanographic drivers or soil type (but see Blanchette et al. 2008; Marczak et al. 2011; McCall & Pennings 2012). As a result, we currently have a poor understanding of the extent to which factors mediating variation in community structure at local scales also matter at geographic scales. Thus, an important task for ecologists is to expand our studies of local landscapes to the geographic scale in order to better understand geographic variation in community structure.

Salt marshes on the Atlantic Coast of the United States are ideal systems for comparing local and geographic variations in community structure. Atlantic Coast salt marshes consist of relatively simple communities of plants and animals that are broadly similar in composition across a large range of latitude and climate from Central Florida through to Maine (Pennings, Siska & Bertness 2001). In particular, lower elevations of marshes throughout this geographic range are dominated by a single plant species, salt marsh cordgrass, Spartina alterniflora (Bertness 2007), with its associated herbivores (Denno et al. 1987; Pennings et al. 2009). This apparent simplicity, however, conceals considerable variation at both local and geographic spatial scales.

At the local scale, S. alterniflora plants close to creekbanks are larger, richer in nitrogen, lower in phenolics and more palatable to herbivores than the plants from the high marsh (Ornes & Kaplan 1989; Goranson, Ho & Pennings 2004; Denno, Lewis & Gratton 2005). Within local marshes, planthoppers migrate seasonally to creekbanks, departing as winter conditions make this habitat unsuitable (Denno 1983). Mesopredators (spiders) also migrate from high-marsh to low-marsh habitats seasonally (Finke & Denno 2006). As a consequence of this variation in plant quality and predator distribution, top-down and bottom-up forces on herbivores vary both seasonally and across the marsh landscape (Denno, Lewis & Gratton 2005).

At the geographic scale, plants from high latitudes are again richer in nitrogen, lower in phenolics and more palatable to herbivores than conspecific plants from the same habitats at low latitudes (Pennings, Siska & Bertness 2001; Siska et al. 2002; Salgado & Pennings 2005). Planthopper herbivores show little variation in density with latitude, but herbivorous snails are more abundant at low latitudes (Pennings et al. 2009). Spiders are present at all latitudes, but top omnivores (katydids) are most abundant at low latitudes (Pennings et al. 2009). The consequences of geographic variation in plant quality and predator distribution for herbivore populations are poorly understood (but see Marczak et al. 2011; McCall & Pennings 2012).

An additional complication in understanding the geographic variation of arthropod community structure is that the trophic impacts of omnivores are ambiguous. Omnivores may sustain their population levels when prey are scarce by eating plants (Dayton 1984; Eubanks & Denno 2000), resulting in increased predation pressure on herbivores when herbivores are present. Alternately, omnivores may become satiated faster by eating at multiple trophic levels, reducing their per capita consumption of any particular prey (Eubanks & Denno 2000). Omnivory may thus either enhance or attenuate top-down effects (Denno et al. 2002; Ho & Pennings 2008), and variation in the abundance of omnivores or the strength of their interactions may be important factors in determining the structure of food webs across large latitudinal gradients.

In sum, we have a reasonable understanding of how top-down and bottom-up forces interact to explain variation in herbivore densities over small elevational gradients (creekbank vs. mid-marsh) within single marshes. In contrast, although we know that plant quality and predator/omnivore densities vary geographically, we have little understanding of how these factors interact to mediate geographic variation in herbivore densities. Previous work suggests that enhanced plant nutrient status may allow herbivores to escape from predator control either through nutritional benefits leading to rapid population growth (Denno et al. 2002) or because predator search times are increased on plants with higher biomass (Olmstead et al. 1997). Here, we compare the importance of local vs. geographic controls on arthropod communities, using Spartina alterniflora and associated arthropods as a model system. To explore how local and geographic variation in plant quality and food web composition might affect Spartina herbivore dynamics, we conducted a mesocosm study that varied (i) plant provenance (plant quality derived from the latitudinal gradient), (ii) fertilization status (mimicking plant quality differences due to marsh elevation), and the presence of (iii) mesopredators (the spider Pardosa littoralis) and (iv) omnivores (the katydid Orchelimum fidicinium). To further understand the results of this mesocosm experiment and extrapolate to more complicated food webs in the field, we conducted predation trials in the laboratory to determine the rates at which different species from the Spartina food web feed on each other (a subset of the most abundant taxa from field collections: four spiders, a beetle and Orchelimum adults as predators and three additional prey species). Building on extensive studies of the S. alterniflora food web at the local scale that have shown a strong bottom-up effect of local variation in plant quality (Denno et al. 2002, 2003; Denno, Lewis & Gratton 2005), we hypothesized that geographic variation in S. alterniflora quality was also an important bottom-up control on food web structure and that local and geographic variation in S. alterniflora quality would interact with the presence of predators and top omnivores to strongly mediate herbivore densities.

Materials and methods

The Spartina food web

On the Atlantic Coast of North America, salt marshes dominated by Spartina alterniflora (henceforth Spartina) range from peninsular Florida to Canada (Denno et al. 1996; Pennings, Siska & Bertness 2001). Spartina plants from high latitudes (RI to ME, 41–43°) are more nutritious (% N) and tender than conspecific plants from low latitudes (GA to FL, 30–31°) (Salgado & Pennings 2005; McCall & Pennings 2012). Moreover, the concentration of phenolics in Spartina is lower in plants from high (2.45 ± 0.13% dry mass) vs. low (3.16 ± 0.10%) latitudes (Siska et al. 2002). Preference tests in the laboratory using a variety of herbivores demonstrated that both polar extracts (polar extracts containing allelochemicals in artificial diet) and live leaf tissue from high-latitude Spartina are far more palatable (more diet or leaf area consumed) than their low-latitude counterparts (Pennings, Siska & Bertness 2001; Siska et al. 2002; Salgado & Pennings 2005). In sum, high-latitude plants in these marshes are known to be more nutritious, softer, less defended and more palatable than low-latitude plants.

Spartina also varies in quality within individual salt marshes. Spartina grows across an elevational range of c. 0.5–2.0 m (McKee & Patrick 1988) within a single marsh. Tall-form Spartina plants (about 1.5 m tall) occur along estuarine creeks while the short form (about 0.5 m) occurs in high-marsh habitats (Valiela, Teal & Deuser 1978). Tall-form plants close to creekbanks are larger, richer in nitrogen (tall-form shoot tissue, 1.0–1.69%N; short-form shoot tissue, 0.8–1.53%N; data summarized in Ornes & Kaplan 1989), lower in phenolics and more palatable to herbivores than plants from the high marsh (Ornes & Kaplan 1989; Goranson, Ho & Pennings 2004; Denno, Lewis & Gratton 2005). Nitrogen is more available to plants at creekbanks than in the high marsh (Mendelssohn & Morris 2000), and fertilization experiments have demonstrated that the nutritional status of tall Spartina (TS) and short Spartina (SS) generally accounts for the majority of the phenotypic differences between the height forms (Valiela, Teal & Deuser 1978). Consequently, fertilization is a common experimental practice used to mimic variation between tall- and short-form plants in greenhouse experiments (Denno et al. 2002).

By far, the most common herbivores of Spartina are the delphacid planthoppers Prokelisia marginata and P. dolus (Denno et al. 2002). Prokelisia marginata and P. dolus (hereafter, Prokelisia) are multivoltine, with a summer generation time of 5–6 weeks (Stiling & Rossi 1997). They are phloem sap-feeders that can reach densities exceeding several thousand adults per m2 with nymphal densities greater than 10 000 per m2 (Denno 1983). A number of invertebrate predators are common in Spartina stands, including spiders and coccinellid beetles. The hunting spiders Hogna modesta (hereafter Hogna) and Pardosa littoralis (hereafter, Pardosa) are particularly common and are known to suppress planthopper populations (Denno et al. 2004) except when planthoppers experience ‘outbreaks’ on high-nitrogen host plants (Denno et al. 2004; Huberty & Denno 2006). In our mesocosm experiment, we used Pardosa as a well-studied representative of the Spartina predator community that is abundant across the entire latitudinal range of our study; other predators were included in predation trials to help us generalize the experimental results across the Spartina community. The most common true omnivore in the Spartina zone of south-eastern salt marshes is the tettigoniid katydid Orchelimum fidicinium (hereafter, Orchelimum) (Pennings et al. 2009). Although historically regarded as an herbivore (Teal 1962), Orchelimum, like most tettigoniids, is omnivorous and readily eats small arthropods (Jiménez et al. 2012). Orchelimum is univoltine, with adult densities reaching 9.6 individuals per m2 (Stiling, Brodbeck & Strong 1991). At high latitudes, Orchelimum is replaced by the smaller tettigoniid Conocephalus spartinae (hereafter, Conocephalus) (Wason & Pennings 2008), which is also omnivorous (Bertness, Wise & Ellison 1987; Bertness & Shumway 1992; Goeriz Pearson et al. 2011).

Mesocosm experiment

The mesocosm study addressed putative local and geographic drivers of herbivore density by varying four factors in a fully crossed design: (i) plant provenance (constitutive plant quality derived from the latitudinal gradient), (ii) fertilization (causing plant quality differences to mimic those related to marsh elevation), and the presence of a dominant species of both (iii) mesopredator (the spider Pardosa littoralis) and (iv) omnivore (the katydid Orchelimum fidicinium). We collected Spartina from five high-latitude sites and five low-latitude sites (Table 1) and established mesocosms with two levels each of fertilizer level (fertilized and unfertilized), mesopredator density (0 or 3) and omnivore density (0 or 1). These treatments were crossed in a full factorial design for a total of 80 mesocosms with five replicates per treatment where each replicate contained field-collected plants from a single site of the appropriate latitude (site was thus nested within latitude). Each mesocosm was stocked with 25 herbivores (Prokelisia). Initial Prokelisia, spider and katydid densities were selected to represent average densities that occur naturally on the marsh (Denno et al. 1996; Ho & Pennings 2008).

Table 1. Sources of Spartina alterniflora plants used in mesocosms: site names and locations
Site nameStateDecimal latitudeLatitude category
Nelson IslandMA42.44High
Great NeckMA42.42High
100 AcresRI41.46High
Rumstick CoveRI41.43High
Cottrell MarshCT41.20High
BaruchSC33.22Low
Ace BasinSC32.33Low
EuloniaGA31.54Low
Airport/Dean CreekGA31.23Low
AmeliaFL30.40Low

Plants (short-form plants from the marsh platform) were collected from field locations between 18 and 22 May 2009 and established in mesocosm containers (each mesocosm consisting of four Spartina stems from a single collection site in a single 30-cm diameter pot; soil was a 1 : 1 mixture of sand and potting soil) in an outdoor greenhouse with a roof to block rain but no walls, thereby keeping plants at close to ambient temperature and humidity, at the University of Georgia Marine Institute on Sapelo Island GA (31°27′ N; 81°16′ W). Beginning on June 2 of 2009, mesocosms assigned to the fertilizer treatment received 9.5 g of fertilizer (Ultra Vigoro Plant Food, 12-5-7, Madison, Wisconsin, USA) every week for 4 weeks prior to the beginning of the experiment and every second week after the experiment had begun. On 19 June 2009, after plants had acclimated to greenhouse conditions and responded to initial fertilizer treatments, we took preliminary measurements of all plants (number of green, yellow [naturally senescing] and damaged leaves, mean percent damage to leaves, chlorophyll content measured with an OPTI-Sciences CCM-200 chlorophyll metre) and placed 5 g of (dry weight) dead Spartina stems and leaves at the base of each plant to provide habitat structure for spiders. Each mesocosm was fitted with a mesh cage consisting of lightweight fabric supported by bamboo stakes. The mesh cages reduced incident light by c. 18%. Although this design placed plants from high and low latitudes in a common garden, differences in the palatability of high- vs. low-latitude Spartina plants persisted for more than a year and five clonal generations in a previous common garden experiment, we saw no evidence for local feeding preferences among herbivores (Pennings, Siska & Bertness 2001), and differences observed in the common garden were sufficient to explain differences in palatability observed in freshly collected field plants (Salgado & Pennings 2005). Thus, we expect that the ‘latitudinal signal’ of plant quality was fully maintained in this common garden experiment.

On 21 June 2009, we field-collected spiders (Pardosa littoralis) and katydids (Orchelimum sp.) in Georgia and held them in the laboratory. On 22–23 June 2009, we collected Prokelisia from the field in Georgia and placed 25 individuals in each mesocosm. We allowed the Prokelisia to disperse within each mesocosm and added spiders (Pardosa) the following day. Katydids (Orchelimum) were introduced 4 h after the addition of spiders. Once mesocosms were fully stocked according to the treatment (26 June 2009), we positioned each mesocosm haphazardly across the greenhouse. A limited number of variables (number of Prokelisia, Orchelimum and Pardosa) were measured 2 weeks into the experiment to assess the potential for outbreak dynamics; variables relating to plant quality could not be determined because of the risk of escape by arthropods. Mesocosms were broken down and resampled after 6 weeks (12 August 2009) once it became apparent that herbivores were reproducing rapidly in some treatments and consuming entire plants.

We used mixed-model nested anovas to assess the effect of treatments for individual response variables where site was nested within latitude (random) and latitude, fertilizer, spider density and katydid density were fully crossed fixed factors. Relative growth rate of plants was calculated as the natural log +1 of the final number of green leaves minus the natural log +1 of the initial number of green leaves, divided by the duration (days) of the experiment. Katydid relative growth rates were calculated analogously as the natural log of final mass minus the natural log of initial mass, divided by the duration (days) of time in the experiment. Growth rates could not be similarly calculated for spiders because we were unable to mark individuals; a group estimate of change in biomass was hindered by low survival rates (many zeros). Accordingly, we used survival as our variable of interest for spiders. We estimated initial plant biomass allometrically based on the height of all shoots (cm) in each plant mesocosm at the start of the experiment; final plant biomass was determined by clipping, drying and weighing all the above-ground live biomass (g). Leaves from each of these plants were lyophilized and analysed for total nitrogen content at the University of Georgia Chemical Analysis Laboratory, Athens, Georgia, USA. We used log (χ + 1) and square root transformations where necessary to improve the normality and heterogeneity of variances. Where data were unbalanced, we employed Satterthwaite's approximation (as recommended by Quinn & Keough 2002) that results in fractional denominator degrees of freedom.

A potential weakness of the anova approach is that it does not account for the fact that some variables (e.g. herbivore numbers, plant traits) are both responding to and simultaneously driving other variables. Thus, we also analysed results of the mesocosm experiment using structural equation modelling (SEM) that allows a variable to be simultaneously influenced by other variables and cause variation in a dependent variable (Grace 2006). We were particularly interested in using SEM to test the hypotheses that there would be an interaction between fertilization and top-down effects, that any effect of fertilization on Prokelisia would be mediated through increased plant biomass and N content and that fertilization would decrease damage to plants by increasing plant biomass (a dilution effect). Building an SEM model consists of several consecutive steps. It starts with a priori identification of the causal relationships between the interplaying variables, followed by the estimation of the path parameters performed by screening the matrix of covariances over the hypothetical model. Finally, model fit is determined by comparing the predicted matrix of covariances with that from the original data. Parameters of the model were estimated using amos 7.0 (Amos Development Corporation, Mount Pleasant, South Carolina, USA) with the maximum likelihood method, and the model fit was tested by the likelihood chi-square value.

Predation trials

To broaden our understanding of the trophic interactions and to determine the rates at which different species within the Spartina food web feed on each other, we conducted cannibalistic trials and predation trials in the laboratory including animals not used in our experimental food web but which were numerically dominant in the field at the time of the experiment. Trials for cannibalism included Orchelimum adults that were enclosed with 5th instar Orchelimum juveniles (last instar before maturation). The Orchelimum adults were 24% larger (tibia or body length) than the juveniles. Predation trials included six predators (four spiders, a beetle and Orchelimum adults), as well as the mirid bug Trigonotylus sp. and the lygaeid bug Ischnodemus badius as potential prey. Trials were conducted in June and July of 2009 and 2010. Animals were collected from the field by hand, sweep net or vacuum sampler. Replicates of the different treatments were run as individuals of different species became available, and the number of replicates varied among species combinations due to the availability of animals (= 27 for Orchelimum adults on Prokelisia and = 3–15 for all other trials; refer to Fig. 7 for details). Upon collecting, animals were acclimated for 24 h to laboratory conditions and to standardize levels of hunger for field caught animals. Individual trials were run for up to 24 h in 850-mL glass jars at a constant room temperature of 25 °C and photoperiod (14 : 8 day : night). Each jar was stocked with a 15 cm long Spartina leaf that served as a substrate for the animals. In each case, consumers were allowed to feed ad libitum and were used only once. Since the containers we used were of small volume, we assume that the predator–prey encounter rates were so high that predation rates were not limited by the number of prey offered, but only by the ability and motivation of predators to subdue and consume prey. This motivation, however, may have been occasionally modified by nonpredatory mortality of prey that occurred after 8 h of the trial (all trials with the soft-bodied mirid Trigonotylus and some with Prokelisia). Some trials were stopped at 8 h if predators were depleting prey or if prey appeared stressed in the jars. Consumption rates for longer assays were pro-rated by duration and all data reported as number consumed in 8 h. Results from 2009 and 2010 were broadly consistent across years and across Prokelisia size categories, but distinct among different predator–prey combinations; we therefore pooled data among years and across different sizes of Prokelisia prey to increase sample sizes. We conducted one-way anovas for different predator–prey combinations within each of three prey groupings (Prokelisia, Orchelimum, mixed trials) with predator–prey combination treated as a fixed factor.

Results

Initial plant quality

At the start of the experiment, fertilized plants had 8.8% greater total biomass (F1,56 = 45.19, < 0.0001), 22.5% more green leaves (F1,56 = 47.22, < 0.0001) and 58% higher levels of chlorophyll (F1,56 = 24.24, < 0.0001) than unfertilized plants. In contrast, there were no initial differences by latitude of plant origin in plant biomass (F1,56 = 0.10, P = 0.76), chlorophyll content (F1,56 = 0.35, = 0.57), number of green leaves (F1,56 = 0.27, = 0.27) or plant damage (mean percent damage, F1,56 = 1.0, = 0.32).

Plant responses

Fertilization increased the percent foliar nitrogen in Spartina leaves at the end of the experiment by 50–100% (F1,45.2 = 130.81, < 0.0001) over unfertilized plants. The resulting differences in N content (control 1%; fertilized 1.8%) are within the range of those observed between short- and tall-form Spartina alterniflora in South Atlantic salt marshes (refer to data from Ornes & Kaplan 1989 summarized under Materials and methods in this report). Fertilized plants had a lower percentage of damage to individual leaves (Fig. 1a) than unfertilized plants; the proportion of green leaves that were damaged followed the same patterns and statistical significance (Table 2). Fertilized plants also exhibited greater overall plant biomass (Fig. 1b) than unfertilized plants. Low-latitude plants had greater biomass (Fig. 1b and Table 2) than high-latitude plants. At the end of the experiment, plants from high-latitude sources were 22.4% higher in foliar nitrogen than those from the low-latitude sources. The presence of katydids in mesocosms reduced final plant biomass (Fig. 1b and Table 2). We recorded the lowest plant growth rates (negative RGR) in unfertilized plants from high latitudes in mesocosms containing katydids; in contrast, the growth of plants from low latitudes was enhanced by fertilizer application, but not consistently affected by omnivore presence (Fig. 2 and Table 2). There were no important main effects of spider presence on any plant variables (Table 2). High-latitude plants lacking either consumer had depressed nitrogen contents relative to plants with spiders (latitude × mesopredator × omnivore interaction, Table 2). This effect was particularly striking for fertilized plants, but only at high latitudes (latitude × fertilized × mesopredator × omnivore, Table 2). We suspect that this effect was due to herbivore feeding depressing plant nitrogen content (Denno et al. 2000), especially on the relatively smaller high-latitude plants, but did not investigate this further.

Figure 1.

Mesocosm experiment. Main effects for models without significant interactions. (a) Mean area (as a proportion) of individual leaves showing Orchelimum damage (F1,22.6 = 27.19, < 0.0001). (b) Total above-ground plant biomass (fertilized: F1,56 = 142.8, < 0.0001; latitude: F1,8 = 8.43, = 0.02). All data are back-transformed lsmeans and 95% confidence intervals.

Figure 2.

Mesocosm experiment. Interactive effects of latitude, omnivore and fertilizer for relative growth rate (RGR) of green leaves (lat × fert × om, F1,56 = 6.74, = 0.012).

Table 2. Mesocosm experiment. Results from nested mixed-model ANOVAS for the following variables: plant biomass, percent foliar nitrogen, relative growth rate (RGR) of green leaves, the proportion of leaves that were green, the proportion of leaves that were damaged and the percentage of damage to individual leaves. Since measures of plant damage were scored as the proportion or percent of omnivore damage, estimations could only be made for mesocosm combinations which included omnivores – absent estimations are indicated by blank cells. We used log (χ + 1) and square root transformations where necessary to improve normality and homogeneity of variances Where data were unbalanced, we used Satterthwaite's approximation, which results in fractional denominator degrees of freedom. Den d.f. = denominator degrees of freedom (numerator degrees of freedom for all effects = 1). All models included a random site effect (nested within latitude). Effects that are statistically significant at <0.05 are highlighted in bold – no post hoc corrections for multiple tests have been employed
Source of variationPlant biomassPercent foliar NRGR green leavesGreen leaves (proportion)Damaged leaves (proportion)Percent damage to leaves
Den d.f. F P Den d.f. F P Den d.f. F P Den d.f. F P Den d.f. F P Den d.f. F P
Latitude88.43 0.02 9.06.81 0.028 823.28 0.0013 7.41.250.307.72.240.187.28.22 0.023
Fertilizer56142.83 < 0.0001 45.2130.32 < 0.0001 5616.97 < 0.0001 53.615.76 0.0002 22.933.88 < 0.0001 22.627.19 < 0.0001
Mesopredator560.080.7945.24.84 0.03 564.36 0.041 53.60.230.6322.91.350.2622.61.920.18
Omnivore566.77 0.012 45.51.570.22561.800.1953.60.260.61      
Lat × fert560.490.4945.20.010.98561.380.2553.62.040.1622.92.750.1122.60.980.33
Lat × meso561.230.2745.21.820.18567.34 0.0089 53.61.250.2722.90.140.7122.60.090.77
Lat × om560.060.8145.55.24 0.027 560.540.4753.60.200.65      
Fert × meso560.610.4448.37.09 0.011 56<0.010.9553.60.880.3522.90.180.6822.60.900.35
Fert × om560.010.9348.20.010.91563.150.0853.63.250.08      
Meso × om562.170.1546.33.710.06564.73 0.034 53.61.330.25      
Lat × fert × meso561.150.2948.31.100.30561.510.2253.60.020.9022.90.050.8422.60.110.74
Lat × fert × om561.120.3048.22.420.13566.74 0.012 53.60.180.68      
Lat × meso × om561.250.2746.39.45 0.0035 562.480.1253.6<0.010.97      
Fert × meso × om561.980.1745.51.920.17561.640.2153.60.050.83      
Lat × fert × meso × om560.430.5245.56.77 0.013 560.190.6753.61.150.29      

Herbivore responses

After 2 weeks, Prokelisia populations were elevated on high-latitude plants (P = 0.003) and tended to be suppressed (= 0.097) by spiders (Fig. 3a). At this time, Prokelisia populations were suppressed in the presence of spiders only on unfertilized plants; on fertilized plants, Prokelisia attained similar populations regardless of spider predation pressure (Fig. 3b). These basic patterns continued to the end of the experiment (Table 3), but levels of statistical significance varied. At the end of the experiment, Prokelisia populations did not differ between high- and low-latitude plants, were enhanced several-fold by fertilizer and were reduced by omnivores but not by spiders (Fig. 4c). The impacts of both omnivores and spiders depended on whether or not plants were fertilized: both consumers tended to suppress Prokelisia only on unfertilized plants Fig. 4a,b), although this pattern was only significant for omnivores. Over all treatment combinations, Prokelisia population density was positively related to leaf nitrogen content (Fig. 5a).

Figure 3.

Mesocosm experiment. Abundance of Prokelisia after 2 weeks in experimental mesocosms. (a) Main effects of latitude (F1,9.2 = 15.24) and mesopredator presence (F1,53.23 = 2.84). Open bars represent low latitude and mesopredator absence, respectively, while closed bars represent high latitude and mesopredator presence, respectively. (b) Significant interaction of fertilizer and mesopredator presence (fertilized × mesopredator, F1,53.2 = 5.2, = 0.027). Data are back-transformed lsmeans and 95% confidence intervals.

Figure 4.

Mesocosm experiment. Abundance of Prokelisia in mesocosms at the end of the experiment. (a) Interaction of fertilizer and mesopredator presence. (b) Interaction of fertilizer and omnivore presence. (c) Main effects. Data are back-transformed lsmeans and 95% confidence intervals.

Figure 5.

Mesocosm experiment. (a) Relationship between percent foliar N and Prokelisia density (R2 = 0.16, < 0.0001,). (b) Relationship between percent foliar N and RGR of Orchelimum (R2 = 0.27, P = 0.022).

Table 3. Mesocosm experiment. Results from nested mixed-model ANOVAS for the following variables: total abundance of Prokelisia nymphs, Prokelisia adults, all Prokelisia combined, the relative growth rate (RGR) of grasshoppers and proportion of spiders surviving. Estimations of grasshopper RGR and spider survival could only be made for mesocosm combinations which contained these animals – absent estimations are indicated by blanks cells. We used log (χ + 1) and square root transformations where necessary to improve normality and homogeneity of variances. Where data were unbalanced, we used Satterthwaite's approximation, which results in fractional denominator degrees of freedom. Den d.f. = denominator degrees of freedom (numerator degrees of freedom for all effects = 1). All models included a random site effect (nested within latitude). Effects that are statistically significant at <0.05 are highlighted in bold – no post hoc correction for multiple tests has been employed
Source of variationProkelisia nymphsProkelisia adultsAll Prokelisia (log)RGR grasshoppersSpider survival
Den d.f. F P Den d.f. F P Den d.f. F P Den d.f. F P Den d.f. F P
Latitude80.370.5682.340.1681.730.222.60.190.707.70.120.73
Fertilizer5618.26 < 0.0001 5625.14 < 0.0001 5625.69 < 0.0001 17.118.33 0.0005 41.72.200.15
Mesopredator560.000.96560.250.62560.450.50420.30.710.409   
Omnivore565.66 0.0208 563.400.071565.47 0.023    41.20.000.99
Lat × fert562.110.15560.340.56561.750.1917.10.010.9841.71.150.29
Lat × meso565.34 0.025 561.060.31562.070.1620.30.010.94   
Lat × om562.450.12561.450.23561.190.28  .41.20.000.97
Fert × meso563.090.084563.430.069563.490.06716.41.000.33   
Fert × om564.01 0.050 564.69 0.035 565.43 0.024    41.70.25 
Meso × om560.340.56560.730.40561.050.31     0.62
Lat × fert × meso562.290.14560.570.45561.170.2816.40.630.44   
Lat × fert × om560.030.86562.130.15561.120.30   41.71.810.19
Lat × meso × om560.030.86560.240.63560.320.57      
Fert × meso × om560.110.74560.080.78560.110.74      
Lat × fert × meso × om560.050.82561.440.24560.550.46      

Predator and omnivore responses

Spider survival to the end of the experiment was unrelated to any treatment variables (Table 3). The growth rate of katydids was positive for fertilized plants but negative across all unfertilized plants (Table 3). Over all treatment combinations, the growth rate of katydids was positively related to leaf nitrogen content with a transition from negative to positive growth at leaf nitrogen contents of around 1.5% (Fig. 5b). Neither source latitude of plants nor the presence of spiders affected katydid growth rates (Table 3).

SEM analysis

Overall, the SEM analysis supported our initial prediction and general finding from anova that fertilization (or local nutrient status) was a strong mediating factor on top-down effects in the Spartina mesocosms. As we initially predicted, SEM analysis showed that the effect of fertilization on Prokelisia was in large part mediated through an increase in plant biomass (Fig. 6a,b); larger plants simply represented greater food availability for herbivores. We do not believe that this rules out a direct effect for nitrogen content as shown in Fig. 5, because the SEM considered high- and low-latitude plants separately and so had less power and a reduced range of N content within each analysis, but it emphasizes that plant nutrition affects both size and N content and that both can affect herbivore populations. We also predicted that fertilized plants would experience lower levels of damage essentially via a dilution effect attributable to an overall increase in plant biomass. In our experiment, fertilized plants did experience less omnivore damage – a result confirmed by both anova (Fig 1b) and SEM analyses. The SEM also supports the hypothesis that this was due to omnivore effects being diluted among the greatly increased biomass of plants, particularly at low latitudes (Fig. 6b). At the same time, an increase in Prokelisia densities reduced omnivore leaf damage in both high- and low-latitude plants, probably by providing an alternative food for Orchelimum.

Figure 6.

Mesocosm experiment. SEM model of the experimental Spartina food web imitating (a) high latitudes and (b) low latitudes. The model is consistent with the data (= 0.70, chi-square d.f. = 0.78). Path coefficients describe standardized values showing relative effects of variables upon each other. Arrow width is proportional to the strength of the path coefficient; one headed arrows represent causal relationships; nonsignificant relationships are marked with a dotted line.

Predation trials

All five predators tested (three spiders, a beetle and Orchelimum adults) ate Prokelisia, but the spider Hogna ate 3–4 times more Prokelisia than the other predator species (Tukey–Kramer HSD < 0.05, Fig. 7a). Both Hogna and Pardosa spiders ate fifth-instar Orchelimum, but did so at very low rates, and Marpissa and a salticid spider ate no Orchelimum (Fig. 7b). Adult Orchelimum did not feed on fifth-instar conspecifics (Fig. 7b). Other arthropod herbivores (Ischnodemus, Trigonotylus) common in the Spartina community were consumed at moderate rates in predation trials by at least one potential consumer, and Marpissa spiders were vulnerable to intraguild predation from the larger Hogna spiders (Fig. 7c).

Figure 7.

Predation trials. Predator success over 8 h with (a) Prokelisia or (b) fifth-instar Orchelimum as the prey item and (c) in addition predator–prey pairings using other intermediate-sized arthropods common in the Spartina community as prey items. Numerals indicate sample size in each trial.

Discussion

It is now generally agreed that top-down and bottom-up forces interact to affect populations of herbivores (Gruner 2004; Stiling & Moon 2005; Bertness et al. 2007; Sala, Bertness & Silliman 2008). In our mesocosm experiment, bottom-up sources of variation in plant quality determined food web structure. This effect, however, was strong at the local scale but weak at the latitudinal scale. Top-down effects on consumers were driven by the omnivorous katydid in our study rather than the strictly carnivorous spider. In those cases where predators did exert a significant suppressing effect on herbivores, that impact was itself mediated by host-plant characteristics. Thus, in this case, we agree with the paradigm that ‘plants set the stage on which herbivorous insects and their enemies interact’ (Denno, McClure & Ott 1995; Denno et al. 2002).

Local, not geographic, sources of bottom-up control

Our mesocosm experiment indicated that the Spartina food web is strongly structured by local variation in plant quality – a result supported by recent field experiment that noted increases in multiple functional groups in response to in-situ fertilization in both high- and low-latitude Spartina marshes (McCall & Pennings 2012). In our more-controlled greenhouse experiment, fertilized plants (mimicking variation in plant quality and form across the local elevation gradient) were consistently larger and supported more herbivores, but showed less damage (due to dilution of herbivore damage and increased numbers of alternate prey for omnivores) than unfertilized plants. This result is consistent with a number of studies that have documented that herbivores prefer and perform better on higher-quality plants from the creekbank vs. the mid-marsh and that this result also tracks the variation in plant quality and height noted with field-collected plants (Denno et al. 2002; Goranson, Ho & Pennings 2004; Wimp et al. 2010).

In contrast, latitudinal variation in plant quality produced only weak or transient effects on herbivore populations. While Prokelisia populations were initially elevated on high-latitude plants, this effect was no longer apparent by the end of the experiment. Because latitudinal variation in plant quality is maintained for an extended period of time when plants are grown in a common garden (Salgado & Pennings 2005), we do not believe that the latitudinal plant quality signal was artificially weak in our experiment. While the effects of latitude were not as dramatic as that of fertilizer addition, low-latitude plants were consistently larger, but lower in nitrogen, at the end of the experiment. Although high-latitude plants are higher in nitrogen, lower in phenolics and more palatable than low-latitude plants (Pennings, Siska & Bertness 2001; Siska et al. 2002; Salgado & Pennings 2005), and although these differences are sufficient to cause geographic variation in herbivore body size (Ho, Pennings & Carefoot 2010), they were nevertheless overwhelmed in our experiments by fertilizer-induced nutritional status and food web composition. These findings are similar to those in our previous work with the high-marsh shrub Iva frutescens (Marczak et al. 2011), where we found that latitudinal variation in plant quality had much smaller effects on herbivore populations than latitudinal variation in top consumers and competition.

We anticipated interactions between top-down effects and the nutrient status of Spartina in our mesocosms, and we subsequently observed that local (fertilized) differences in the nutritional status of Spartina modified the effects of both katydids and spiders on herbivore populations. Predators were able to suppress herbivores only on unfertilized plants, possibly because herbivores were concentrated on these smaller plants and had lower reproductive rates. At the same time, damage from omnivore populations was greatest on plants that did not receive fertilizer, again probably because omnivores were concentrated on smaller, unfertilized plants and had fewer insect prey as an alternative to feeding on plants. Control of consumers by local variation in plant quality is consistent with some previous work on the Spartina food web. For example, Denno et al. (2002) found that enhancing the nutrition of host plants did not strengthen top-down effects on Prokelisia despite field-based evidence that predator densities were also elevated on higher-quality plants. Instead, predators more effectively suppressed Prokelisia populations on poor-quality host plants (Denno et al. 2002). In contrast, when Prokelisia are feeding on high-quality plants (in the case of the present experiment, fertilized plants), they escape from effective predator control. Denno (2002) has previously argued that this result is the key to understanding the seasonal migration of Prokelisia populations from the high to the low marsh in high-latitude locations where high-quality plants are only available seasonally in the low marsh due to severe winter freezes.

Similarly, Bertness et al. (2007) demonstrated that mixed insect communities had little effect on Spartina productivity in relatively low-nutrient marshes in Rhode Island (US), while increasing levels of eutrophication triggered outbreaks or aggregations that led to herbivore control of marsh productivity in eutrophic marshes. Nitrogen-rich Spartina are characteristic of low-marsh habitats (Denno 1983; Ornes & Kaplan 1989) and herbivores are thought to be generally N-limited (White 1983; Huberty & Denno 2006), as we also found (Fig. 5a,b). Field studies have demonstrated that the high nitrogen content of low elevation Spartina plants encourages mass colonization, enhances both survival and fecundity, and promotes rapid population expansion of Prokelisia planthoppers (Olmstead et al. 1997), and that nitrogen enhancement results in a greater abundance of herbivores, predators and parasitoids across a large range of latitude (McCall & Pennings 2012). Our mesocosm experiment also demonstrated these outbreak conditions on fertilized plants. In contrast, latitudinal variation in plant quality did not affect Prokelisia populations, omnivore growth rates or the effect of predators on herbivore populations.

Weaker top-down effects

Since the early studies by Teal (1962), salt marsh ecosystems have been considered systems under strong bottom-up control, with primary productivity patterns being largely driven by physical conditions and nutrient availability. More recent work has championed top-down regulation of productivity in salt marshes by herbivores (snails: Silliman & Bertness 2002; insects: Finke & Denno 2004; geese: Kuijper & Bakker 2005; crabs: Altieri et al. 2012). Control of herbivore populations meanwhile appears to be regulated by a combination of bottom-up and top-down factors. In our mesocosm experiment, while both katydids and spiders successfully suppressed Prokelisia populations, these consumer effects were consistently modified by bottom-up conditions and consumer effects did not cascade to benefit plants.

The effects of plant architecture on predator capture efficiencies have been noted in salt marsh and other systems (Landis, Wratten & Gurr 2000). It is possible that in our experiment, predator suppression of herbivores on low-nitrogen plants was due to the differences in physical structure and size of fertilized and control plants that altered the foraging efficiency of invertebrate predators. In particular, predator foraging success is likely to have been higher on smaller, unfertilized plants. Alternately, suppression of Prokelisia populations on unfertilized plants by katydids and spiders (Fig. 4a,b) might have occurred because Prokelisia populations were reproducing poorly and individuals were less able to escape predators due to feeding on a nutritionally poor food source.

Weak or absent trophic cascades

In our mesocosm study, Orchelimum reduced Prokelisia abundance; however, this did not cascade to indirectly benefit plants, probably because Orchelimum also negatively affected plants by directly consuming them. Control of herbivore abundance by spiders was generally weak (statistically discernible at the experiment mid-point but not evident at its conclusion), and this effect was not strong enough to cascade to positively affect plant growth or other characteristics.

While Orchelimum are capable of consuming substantial quantities of Prokelisia planthoppers (Fig. 7), they appear to have relatively inflexible diets that require them to also consume plant material (Jiménez et al. 2012), and they may continue to eat high-quality plant tissue even when invertebrate prey are also available. Several studies have suggested (Denno & Fagan 2003; Matsumura et al. 2004) that even a small addition of protein in the diet of omnivores can result in large improvements in nutrition and growth. Jiménez et al. (2012) found that Orchelimum grew better on a mixed diet including both prey and plant material than on either single diet alone. We found that Orchelimum exhibited greater relative growth rates on plants with greater foliar nitrogen (Fig. 5b). This could have been because Orchelimum performed better when eating plants with higher nitrogen content, or because Orchelimum were able to eat more prey on fertilized plants because Prokelisia were more abundant. In either case, both results are consistent with the idea that Orchelimum is severely limited by N availability on a diet of low-N Spartina. At the same time, Orchelimum grew better on a mixed diet than on a diet of pure animal prey (Jiménez et al. 2012), indicating that while plants and herbivores may be somewhat substitutable foods (Van-Rijn & Sabelis 2005) for other marsh omnivores (Armases crabs, Ho & Pennings 2008), they are complementary for Orchelimum. Whether omnivory acts to strengthen or weaken trophic cascades (Polis & Strong 1996; Eubanks & Styrsky 2005) may thus depend on how readily an omnivore can switch food sources. This may be particularly true in the case of true omnivores (those that consume both plant and animal tissue) where a preference for plant over animal tissue will serve to cancel cascading effects of consuming herbivorous prey.

Extrapolating to the more diverse community in the field

Our mesocosm experiments necessarily utilized a limited number of species. Although these were selected because they were among the more common taxa in the field, this still raises the question of whether the results can safely be extrapolated to the more diverse field community. The predation trials suggest that they can, albeit with an important caveat. First, all the moderately sized arthropod herbivores that we tested (Prokelisia, Ischnodemus and Trigonotylus) were readily eaten by common predators. Second, with the exception of a salticid spider that we did not identify to species, all the predators that we tested (three other spiders, a beetle and Orchelimum) readily consumed at least one of the herbivore species tested. Thus, although the precise predation rates in any particular setting would depend on the densities of the different taxa and the differences in vulnerability of particular herbivores to particular predators, the predator–prey interactions in this community appear to be relatively generalized. As a result, the various less-common herbivore species are likely to be affected by predators and top omnivores in much the same way that Prokelisia was affected.

The major caveat with extending our results to the more diverse field community is that the multiplicity of predatory taxa in the field (e.g. at least six common spiders and another six or more rare species, Wimp et al. 2010; authors personal observations) greatly increases the potential for intraguild predation between predators. Depending on variation in predator density, and on the extent to which various predators interfere with vs. facilitate each other, our mesocosm results may imprecisely estimate predation rates in the field (Schmitz, Beckerman & O'Brien 1997; Finke & Denno 2004). No doubt the details of the trophic ecology of each of the predators differs somewhat. Nevertheless, because almost all the consumers tested (with the exception of one spider) readily ate the common arthropod herbivores, it is most likely that the relatively diverse predator assemblage found in the field has a negative effect on Prokelisia populations that is broadly consistent with the mesocosm results based on the single common predator Pardosa. In sum, although there is much to be learned about how herbivore and predator diversity affects the functioning of the Spartina arthropod community, we are confident that our mesocosm experiments, despite being stocked with only a few species, provide a robust first-order approximation of how the more diverse field community functions.

Fully understanding geographic variation in the Spartina food web will need to consider the full suite of herbivores including snails (Silliman & Zieman 2001), crabs (Altieri et al. 2012) and vertebrates (Buchsbaum, Valiela & Swain 1984). A complete consideration of these herbivores, and their interactions with insect herbivores, awaits further study; however, we speculate that the principles that we have outlined here for insect herbivores also apply to these other herbivores. For example, herbivorous Sesarma crabs are most abundant near high-quality, tall-form Spartina plants (Teal 1958), and per capita snail effects on Spartina are greatest on tall-form plants (Silliman & Bertness 2002). Snails experience the strongest top-down control from predators on high-quality creekside instead of low-quality platform plants, but this is simply because their predators are of marine rather than terrestrial origin (Silliman & Bertness 2002). Snail herbivory is most important at low latitudes due to a turnover in the most common taxa across latitude (Pennings & Silliman 2005), paralleling the many salt marsh insect herbivores that are more abundant at low latitudes (Pennings et al. 2009). Thus, for snails and crabs, the primary factors controlling local and geographic variation in herbivore abundance may be the same factors that affect insect herbivores: local variation in plant quality and geographic variation in species composition. This suggests that we are close to a conceptual unification of the factors mediating the distribution and abundance of salt marsh herbivores at both local and geographic scales.

Comparing local and geographic patterns

Taken as a whole, our results suggest that the single most important factor mediating Prokelisia densities is local variation in plant quality between short- and tall-form plants. Predation pressure can be important, but only on low-quality plants, because herbivores escape from predator control on high-quality plants. Similarly, although plants demonstrably vary in quality across latitude, and this has measurable effects on herbivore performance (Ho, Pennings & Carefoot 2010), the effects of latitudinal variation on herbivore abundance tend to be overwhelmed by other factors.

Our experiments did not address factors extrinsic to the food web such as climate, but the short growing season and colder winters at high latitudes are the most likely reasons why high latitudes are characterized by a reduced number of generations in multivoltine species such as Prokelisia (Friedenberg et al. 2008), a reduced body size of some univoltine species such as Orchelimum (Wason & Pennings 2008; Ho, Pennings & Carefoot 2010) and a reduced density of some other members of the food web (Pennings et al. 2009; McCall & Pennings 2012).

In sum, we argue that a first-order understanding of variation in the Spartina arthropod food web at geographic scales consists of two variables: geographic variation in climate, which mediates the density and body size of different members of the food web in ways that are not yet well studied; and local variation in plant quality, which provides strong bottom-up forcing to herbivore populations. A second-order understanding of the arthropod food web needs to also consider predation, which is most effective at controlling herbivores feeding on low-quality plants (Denno, Lewis & Gratton 2005). Finally, latitudinal variation in plant quality probably explains some variation in herbivore densities and body size, but is probably more of a response to herbivore pressure (Pennings et al. 2009) than a driver of it.

Acknowledgements

We thank the National Science Foundation (DEB-0296160, DEB-0638813, OCE06-20959) for funding and two anonymous reviewers for their constructive commentary. This work is a contribution of the Georgia Coastal Ecosystem Long-Term Ecological Research programme and contribution number 1030 from the University of Georgia Marine Institute.

Ancillary