An omnivore vigor hypothesis? Nutrient availability strengthens herbivore suppression by omnivores across 48 field sites
Data files
Jan 23, 2023 version files 31.74 KB
-
data_omnivore_vigor.xlsx
28.26 KB
-
README.md
3.49 KB
Abstract
- Nutrients regulate herbivore growth from the ‘bottom-up’ via improved plant vigor and food quality. Nitrogen also affects ‘top-down’ control of herbivores by moderating attraction of predators and the rates at which they consume herbivorous prey.
- Tri-trophic consequences of nitrogen availability are more challenging to predict among omnivorous natural enemies who feed on both plants and herbivores, limiting our ability to predict net outcomes of nutrient availability in food webs.
- In a two-year field survey of insects on zucchini host plants at 48 sites, I predicted that both herbivores and foliar-feeding omnivores would increase with nutrient availability, while predators would not.
- My results revealed positive relationships between omnivores and foliar nitrogen concentrations, while predators had neutral responses to foliar N. Surprisingly, herbivores declined with increasing foliar N across the field sites.
- Greenhouse experiments reinforced these patterns, as herbivore growth inversely correlated with soil N concentrations in communities that included foliar-feeding omnivores. Conversely, herbivore growth was uncorrelated with soil N on plants with generalist predators, nor on predator-free plants.
- These results suggest that omnivores mount strong and consistent responses to nitrogen in plant tissues in a variety of ecological contexts. In environments where omnivorous arthropods can thrive, their recruitment to nitrogen-rich plants may increase predation and thereby counterbalance and stabilize ‘bottom-up’ increases in herbivore performance supported by enhanced foliar nutrition.
Methods
Field survey
To examine how natural populations of herbivores, omnivores and generalist predators respond to plant quality along a nutrient gradient, I performed observational field surveys. Between 2020 and 2021, my students and I sampled arthropods, soils and foliar tissues at 56 farm fields that varied in fertility management practices and formed a gradient of nutrient availability across Georgia, South Carolina, and North Carolina, USA (Appendix Fig. S1, Appendix Tables 1,2). The number of sites surveyed in 2020 (18) were considerably fewer than those sampled in 2021 (38) due to restricted mobility during the Covid-19 pandemic.
At each site, we randomly selected 10 unique zucchini plants along a transect spanning each field and visually surveyed arthropods on six haphazardly selected leaves (two new leaves, two middle-aged leaves, and two old leaves), stems, and in all open flowers. Herbivores and natural enemies were counted and identified on-site with the finest resolution possible. Most common taxa (noted above in the study system) were identified to species or genus, while some taxa are more challenging to identify on-site, and were identified at the family level, or order (spiders). Fewer than 1% of the arthropods we observed were impossible to confidently assign to functional trophic groups (decomposers, herbivores, foliar-feeding omnivores, generalist predators), and these were excluded from analyses.
We then collected 4.5 L soil at five locations along the transect from the 0–15cm profile, homogenized it in a bucket, collected a subsample for chemical/physical analyses, and saved the remaining soil for the greenhouse experiment (described below). We collected foliar tissue from five plants per field from the 3rd leaf below the apical meristem. Foliar tissue samples were collected at 48 of the 56 sites visited; some farmers preferred that I not remove plant tissue. Soil and foliar tissue samples were sent to the Agricultural and Environmental Service Laboratory at the University of Georgia to determine percent total nitrogen concentrations (organic + inorganic N) with an Elementar Vario Max Total Combustion Analyzer (Langenselbold, Germany).
Greenhouse experiment
To determine how nitrogen availability influences suppression of herbivores by omnivores and generalist predators, I performed a greenhouse predation experiment. From a randomly selected subset of 25 farms, I planted zucchini seeds in three pots of soil collected per site (for a total of 75 pots) on 27 September 2021. The three pots from each site were placed in a single cage, and the 25 cages were randomized on the greenhouse bench. After four weeks of growth, I introduced 25 mixed-age cotton melon aphids (A. gossypii) to the 3rd leaf from the apical meristem and enclosed each plant in an individual cage. To reduce error from differential growth rates in mixed-age colonies, I returned after 24 and 48 h to subsidize or reduce colonies on plants so that all plants contained 25 aphids before data collection began. It is difficult to introduce aphid colonies at uniform stages of development, and some introduced colonies inevitably include more late-instar individuals than others, and will therefore reproduce more rapidly regardless of treatment. Including this brief acclimation period where individuals are culled from colonies with faster initial growth rates helps to even the age structure before assays begin, which reduces this random source of error (Blubaugh et al. 2018, Blubaugh et al. 2021). While colonies were establishing and acclimating, I monitored baseline colony growth across the plants for 5 days before introducing predators.
Once aphid colonies were established (after 5 days), I initiated three predation treatments on three separate benches, each containing 25 plants grown in the farm soils from the survey, completely randomized in cages 12 inches apart. In my first treatment, I introduced two omnivorous minute pirate bugs per cage (O. insidiosus, Evergreen Growers Supply, Portland, Oregon, USA); in the second treatment, one convergent ladybeetle (a generalist predator) was introduced (Hippodamia convergens Guérin-Méneville, Carolina Biological, USA). Natural enemies were introduced at natural densities similar to those most commonly observed in surveys (1.85 Orius/plant, and 1.25 ladybeetles/plant), and they were replaced if found dead in cages during daily inspections. In the last treatment, aphids grew on 22 plants (three of the 25 plants failed to germinate) in predator-free cages. I arranged each of the predation assays together in separate blocks because aphid growth rates can be affected by alarm pheromones emitted by adjacent colonies when exposed to natural enemies (Vandermoten et al. 2012). Therefore, to minimize error related to alarm pheromones and prioritize quantifying fertility-mediated effects for each predator, I spatially separated each of the predation assays (Appendix Fig. S4). I placed each predator treatment on the same bench because space constraints required that the 25 cages be immediately next to each other, potentially within range of detection of alarm pheromones coming from nearby predator treatments. I counted aphids on all plants and terminated the experiment 96 h after introducing the natural enemies (as in Desneux and O’Neill 2008), as some of the aphid colonies had grown large enough that host plants were beginning to die.
Statistical analyses
All analyses were performed in R version 3.6.0. To examine how nitrogen availability influences top-down and bottom-up forces shaping herbivore pressure in the field, I ran path analyses using the ‘piecewiseSEM’ package in R (Lefcheck, 2016). My hypothetical model included all possible links between foliar N and counts of omnivores, generalist predators, and herbivores. Each of the component generalized linear models (predicting counts of omnivores, generalist predators, and herbivores, respectively) were modeled with a Poisson distribution and had normally distributed residuals. Since relatively few repeat visits to farms were made between 2020 and 2021, and because all farmers rotated their crops to different locations between years, I considered each visit to a unique field to be an independent replicate in the model and therefore included no random effects. I used nitrogen concentrations in foliar tissue as my predictor variable for host plant quality in the field survey (rather than soil nitrogen) because it is a more reliable indicator of plant quality during the brief visit. Fertilizer application timing, planting dates, and mineralization rates varied considerably across the farms we visited, leading to wide variation in available soil nitrogen that did not correlate with nitrogen available in plant tissue to herbivores and omnivores at the time of the visit. I evaluated the overall fit of my path model with d-separation tests on Fisher’s C statistic and AIC (Shipley, 2013). I gradually eliminated non-significant predictors in the series of path models until I minimized AIC, indicating the best-fit series of interactive models.
In my greenhouse experiment, I performed simple linear regressions on aphid growth in each of my predator treatments with soil N concentration as the predictor variable. Unfortunately, technical problems with the analysis of the greenhouse plant tissue samples meant that I could not include foliar N in the model, as I did in the field surveys, and had to instead use less-reliable and more-widely varying estimates of nitrogen in the soil (Ransom et al. 2020). However, because planting dates and data collection were uniform across the samples in the controlled greenhouse experiment, soil N should reasonably predict nitrogen available to the herbivores we introduced. Percent soil N and dry plant biomass positively correlated (Appendix Fig. S2), indicating that plant growth (i.e. plant vigor) is indeed linked with soil nutrient availability.
My response variable was the relative proportionate growth rate of aphid colonies 96 h after predators were introduced (i.e. I divided the difference in aphid counts after 96 h by the initial number of aphids in each cage when predators were introduced; Leather and Dixon 1984). Greenhouse space limitations required spatial segregation of my predator treatments and precluded explicit tests of interactions between soil N and predator feeding guilds (compromising independence between predator treatments), so I analyzed each of the three predator treatments separately. Normality assumptions and homogeneity of variance assumptions were checked by examining histograms and residual plots. I also analyzed aphid counts for all plants across the experiments together during the 5-day baseline period of colony establishment, prior to predator introduction, with Poisson regression. Percent soil N was a predictor variable, and aphid counts five days after aphid introduction as the response variable, with farm identification nested in block as a random effect.
Usage notes
Microsoft Excel, program R