Like many top consumers, parasites can regulate feeding of their prey via trait-mediated means. If parasites modify the feeding behavior of ecologically important grazers, they may have cascading effects on the structure and functioning of whole plant communities. The extent to which parasites can influence plant communities in this way is largely dependent on the strength of their behavioral alteration, their prevalence in host grazers, and the density of those hosts. Recent experiments and comparative surveys in southeastern USA salt marshes revealed that common larval trematode parasites suppress the per capita grazing impacts of the marsh periwinkle (Littoraria irrorata), generating a trophic cascade that protects foundational marsh plants from drought-associated overgrazing. Here, we conducted a field manipulation wherein we modified grazer host density while holding infection prevalence constant at an ecologically relevant level (20%) to determine whether the indirect, facilitative effects of parasites on marsh plants varied with the density of grazers. We found that parasites had significant positive impacts on marsh net primary productivity at moderate densities of snails (≥50 snails/ 0.5 m²), but that the positive effects of parasites were negligible at lower densities. Our results confirm the findings of previous studies that parasites can protect marsh plants from overgrazing at sufficiently high prevalence but show that their ability to do so depends on host density.

To determine how trematode infection influences marsh productivity at different levels of grazer host density, we conducted a field manipulation where we modified grazer host density while keeping infection prevalence constant. In June 2016 we established 0.5 m² caged plots in a structurally homogenous swath of smooth cordgrass (Spartina alterniflora) marsh within the Hoop Pole Creek Clean Water Reserve in Atlantic Beach, North Carolina, USA (34°42’25.12” N, 76°45’1.14” W). The site was characterized by a relatively uniform elevation, very low snail densities (<1 adult snail per m²), and minimal visible signs of snail grazing.

Snails used in the experiment were collected from a marsh die-off area where snails were abundant and infection prevalence was known to be high (Morton & Silliman 2020). Collected adult snails (shell length > 15mm) were transported back to the lab where their infection status was determined using a previously described cercariae shedding method that produces no false-positives (Morton 2018, Morton & Silliman 2020). We marked the shell of each infected snail with a red dot using a non-toxic, water-resistant paint pen while uninfected snails were marked with a blue dot (Henry & Jarne 2007). Snails were kept in separate aquaria and provisioned with damp cordgrass wrack for ~3 weeks until they were deployed in the field.

Roofless cages (0.7 × 0.7 × 1 m) were constructed from untreated wooden posts and galvanized hardware mesh. A strip of copper tape was applied to the inner base of each cage, just above the sediment, to discourage snail escapes. Caged plots were spaced at least 1-m apart to assure independence of replicates—a design confirmed from past studies (Silliman & Zieman 2001, Silliman & Bertness 2002, Morton & Silliman 2020). Cages were buried 10 cm into the substrate to prevent snail and mud crab migration in and out of cages, and to inhibit belowground connections between plants inside and outside of the cages. Each plot was assigned to one of 8 snail density treatments (20, 40, 50, 60, 70, 80, 90, and 100 snails/per 0.5 m²) and one of two parasite addition treatments (0 and 20% infection prevalence). This resulted in 16 total treatments (n = 4 replicates per treatment). The snail densities used spanned the full range of adult snail densities observed within marsh die-off areas at this site. The 20% infection prevalence used in the experiment reflected the average naturally occurring summertime infection prevalence value for snails within local die-off areas (Morton & Silliman 2020). Uncaged plots marked at the corners with colored PVC flags (n = 4) and partial cages with one open side (n = 4) served as cage controls.

Before the beginning of the experiment, we removed any mud crabs and snails from cages. We took measurements of several marsh characteristics in all plots at the beginning of the experiment. We counted all S. alterniflora stems and measured the heights of ten randomly selected stems in each plot. Random selection of stems was accomplished by tossing a plastic dowel into plots and measuring the first 10 stems touching the dowel. We constructed a height-to-biomass regression by collecting 30 cordgrass stems of varying sizes from the marsh directly adjacent to our experimental site. The stems were washed, their length measured, and were dried at 70˚C until they reached a constant weight. We used the resulting height-to-biomass regression to estimate standing cordgrass biomass in each plot. We also counted all juvenile (< 0.5 cm in diameter) and adult fiddler crab burrows at both the beginning and end of the experiment, because these organisms are known to influence cordgrass growth by oxygenating the sediment through burrowing (Bertness 1985, Daleo et al. 2007, Angelini & Silliman 2012, Gittman & Keller 2013, Raposa et al. 2018).

Infected and uninfected snails within each plot were counted twice weekly and replaced as necessary to maintain the assigned snail density and infection prevalence treatments for the 3-month duration of the experiment. During monitoring, any predatory mud crabs found within plots were removed and their burrows plugged with marsh sediment to discourage successful re-occupation.

We took final metrics of marsh vegetation characteristics in September 2016, 12 weeks after the beginning of the experiment. All snails were removed and dissected to confirm infection status. Because snail grazing had dramatically reduced stem densities in many plots, we were able to measure all stems within each plot to generate final plot biomass estimates by calculating and summing individual biomass estimates. Belowground biomass cores (15-cm diameter x 25-cm height) were also taken from the center of plots. Cores were washed and root material separated from rhizomes. Roots and rhizomes were dried to a constant weight at 70˚C and weighed.

The possible influence of caging artifacts on marsh invertebrates and cordgrass growth was assessed using paired t-tests. Specifically, we compared the number of total fiddler crab burrows, final stem density, initial aboveground biomass, final aboveground biomass, change in aboveground biomass, and number of flowering stems between cage controls and open controls.

Because fiddler crabs can positively influence cordgrass growth (Bertness 1985, Gittman & Keller 2013), we examined the relationship between burrow density and cordgrass stem density using a linear model. We validated model fit by examining model residuals, their distribution relative to fitted values, and normal Q-Q plots (car package, Chambers and Hastie 1992).

In accordance with our hypotheses, models including the interaction between parasite infection (uninfected versus infected) and snail density (20, 40, 50, 60, 70, 80, 90, or 100 snails per 0.5 m²) were fit for each response metric (belowground biomass, aboveground biomass, shoot density, and number of flowering stems). Additionally, all models initially included fiddler crab burrow density as a covariate. However, this covariate was ultimately not significant for any model and was therefore dropped and each model re-run to include only the interaction term. We analyzed the interactive effects of infection status and snail density on belowground biomass using a two-way analysis of variance (ANOVA), with data log-transformed prior to analysis to meet model assumptions. Change in aboveground biomass was analyzed with a linear model, with initial aboveground biomass values for each plot subtracted from corresponding final values. Model assumptions were verified through assessments of homogeneity of variance (Levene’s test, P > 0.05) and examination of fitted residuals and normal Q-Q plots. The number of cordgrass inflorescences was modeled with a negative binomial generalized linear model (glmmTMB, Brooks et al. 2017). Model appropriateness and fit were confirmed through examination of simulated scaled residuals (DHARMa, Hartig 2022).

For each response metric, the significance of the interaction model was examined in an Analysis of Deviance Table using Wald chi-square tests (car package, Chambers & Hastie 1992). For all models, Tukey’s post-hoc comparisons were used to assess pairwise differences for any significant treatment or interactive effects in the models (emmeans package, Lenth et al. 2021). All analyses were performed in the R statistical computing environment (v. 4.1.3; R Core Team 2018).