Hosts can reduce the negative fitness effects of parasite infection by avoiding contact with parasites or by resisting infection after contact. Because of their shared outcome, avoidance and resistance have been hypothesized to trade off with one another. Assuming these defenses carry fitness costs, hosts are expected to have high levels of one defense or the other, but not both. Alternatively, avoidance and resistance may covary positively, if, for example, they complement one another or are genetically or mechanistically linked. Testing these hypotheses requires measuring avoidance and resistance independently, which is challenging because they are functionally linked. In this study, we separated avoidance and resistance of the host C. elegans against the bacterial parasite S. marcescens and tested for a correlation between them. We phenotyped a panel of 12 genetically divergent hosts using two distinct bacterial strains and multiple experimental contexts. We found no evidence of a correlation between avoidance and resistance. This result suggests that avoidance and resistance can covary independently. Moreover, we found strong genetic specificity not only for resistance, but also for a measure of avoidance, motivating further research to examine the coevolutionary dynamics of avoidance.

Overview

We examined the relationship between avoidance and resistance to two distinct strains of the opportunistic bacterial parasite S. marcescens across 12 genetically divergent strains of C. elegans. We chose parasite strains that we expected to differ in their virulence and the avoidance they elicit. We assayed resistance independently of avoidance by measuring survival after directly dosing hosts in liquid, which prevents hosts avoiding the parasite. We used two assays to measure avoidance: a lawn-leaving assay (Pradel et al., 2007) and a choice assay (Schulenburg & Müller, 2004). These are two common designs used for assessing behavioral responses to bacteria, and we expected they would capture different aspects of behavioral variation. We used data from these assays to test whether resistance and avoidance covaried across C. elegans strains. Each assay was conducted in a blocked design, with a different subset of host strains tested in each of four to five experimental blocks that were conducted on different days. Each block contained all bacterial treatments (see Table S1 for block assignments and Supplementary Text – Section S1 for details on statistical handling of block effects).

Strains

Hosts

The 12 genetically divergent strains of C. elegans were acquired from the Caenorhabditis Natural Diversity Resource (CaeNDR, https://caendr.org/). The strains were as follows: CB4856, CX11314, DL238, ED3017, EG4725, JT11398, JU258, JU775, LKC34, MY16, MY23, and N2. We assigned host strains a number label from 1 to 12 to improve interpretation across the figures (see Table S1). CaeNDR recommends using this set of strains as a starting point to determine whether genetic variation exists in ecologically relevant traits across the species (Cook et al., 2017). In a previous study, these strains showed heritable variation in avoidance behavior (lawn-leaving, see below) of the S. marcescens strain Db10 (Amoroso et al., 2024).

Upon receipt of the strains, we followed the CaeNDR protocols for cleaning the strains (see https://caendr.org/data/protocols), and we cryopreserved them at −80°C. Before experiments, we thawed the strains and maintained them on ample food (Escherichia coli strain OP50) for 14–17 days at 20°C prior to an assay to allow populations to recover from freezing.

Parasites

The two S. marcescens strains used in this experiment were Sm2170 and Db10. Sm2170 is a red-pigmented strain that is known to be highly virulent, killing the greatest proportion of hosts among a set of five S. marcescens strains in a previous study (Schulenburg & Ewbank, 2004). Prior studies found that this parasite strain is sufficiently virulent to select for avoidance and resistance in the host (Morran et al., 2011; Penley et al., 2017, 2018; Penley & Morran, 2018). The second strain, Db10, is unpigmented and elicits a high degree of avoidance from C. elegans (Amoroso et al., 2024; Pradel et al., 2007). The virulence of Db10 and Sm2170 has not been directly compared, but Db10 appears to be less virulent (Pradel et al., 2007). Stocks of the bacterial strains were maintained at -80ºC in glycerol and thawed for proliferation prior to each experiment.

Experimental Design

Quantifying Resistance

To isolate resistance (post-contact defenses) from avoidance (pre-contact), we measured infection outcomes after controlling for contact, i.e., by directly dosing the host with a known concentration of bacterial cells. We accomplished this in a liquid environment, where hosts indiscriminately ingested a suspension of the parasite. Then we monitored survival over the subsequent days (Schulenburg & Ewbank, 2004; Schulenburg & Müller, 2004). We used two doses of each of the two parasite strains to maximize the potential variation in resistance that we could capture across the 12 host strains (Fig. 1A).

For each C. elegans strain, L4 larval stage nematodes were individually sorted into wells of a 96-well plate using a Union Biometrica Large Particle Flow Cytometer (WormSorter) that initially contained 50 µl of sterile S-medium. Each well was inspected microscopically to confirm the presence of one live nematode. We excluded wells with more than one nematode or a dead nematode. Then the hosts were dosed with an additional 50 µl of one of five treatments: a high or a low dose of Sm2170, a high or a low dose of Db10, or a control dose of OP50. The high and low doses were prepared by pelleting 24-hour LB cultures of the two parasite strains (initial bacterial concentrations approximately 3.99 +/- 0.43 x 10⁸ CFU/ml for Sm2170; 4.29 +/- 1.2  x 10⁸ CFU/ml for Db10) and resuspending them in one-eighth (high dose) or one-quarter (low dose) volume of S-medium. When added to the 50 µl of S-medium already in the wells, the high dose was concentrated 4x from the initial culture’s cell counts, and the low dose was concentrated 2x. For the OP50 control, a 24-hour culture in LB was pelleted and resuspended in an equal volume of S-medium. Because it included no nutrients, S-medium did not promote additional growth of the bacteria in the wells over the course of the experiment. Host strains were assigned to wells column-wise in a balanced design across four to five plates, and bacterial strain doses were administered row-wise (see Figure S1 for details). For each combination of host strain, parasite strain, and treatment, the WormSorter filled on average 28.8 +/- 7.45 wells with a single L4-stage host. Each plate was covered with Breathe-Easy film and incubated at 20ºC, shaking at 160 rpm to promote gas exchange in the liquid.

After 48 hours, we inspected each well with an inverted microscope and scored individual hosts as alive if there was any sign of movement (Schulenburg & Müller, 2004). We also recorded whether larvae were present in the well, indicating that the host had successfully reproduced. Individuals that were missing at 48 hours were excluded. We obtained survival data for a mean of 27.6 +/- 7.5 individuals per combination of host strain, parasite strain, and treatment, ranging from a minimum of 10 to a maximum of 46 individuals per combination.

Quantifying Avoidance with a Lawn-leaving Assay

We first measured avoidance using a lawn-leaving assay (Fig. 1B), which is modeled after past studies of avoidance of strain Db10 (Amoroso et al., 2024; Pradel et al., 2007). We prepared the lawns by pipetting 30µL of 6-hour LB cultures of Db10 or Sm2170 onto the center of 60-mm Petri dishes of nematode growth medium (NGM-Lite, US Biological, Catalog No: N1005) with 0.7% agarose added to discourage C. elegans’ burrowing. Using an inoculating loop, lawns were spread to cover a 25-mm-diameter circle that had been drawn on the bottom of the Petri dish with a template to standardize the lawn’s size, shape, and location. The plates were incubated overnight at 28ºC.

Synchronized populations of nematodes were raised on OP50 for 48 hours at 20ºC until they reached the L4 larval stage. They were washed from the plates and into a tube with M9. They were washed three times with 2 ml of M9 to remove OP50, and then approximately 20 individuals were transferred in ~3 µl of M9 to the center of the bacterial lawn using a pipette. The precise number of hosts added was immediately confirmed by microscopy.

After hosts were added, the plates were maintained at 20ºC. After 18 and 24 hours, we counted the number of individuals located on and off the lawn of bacteria. Nematodes touching any part of the lawn were counted as on the lawn. The baseline expectation for non-parasitic bacteria is that nearly all individuals would be located on the lawn of bacteria at this timepoint, as demonstrated by the control treatment in a previous study (Amoroso et al., 2024). This experimental protocol follows that of the previously published study (Amoroso et al., 2024), except that the transfer of nematodes was done in liquid, which is more efficient than picking individual nematodes.

Quantifying Avoidance with a Choice Assay

We also measured behavioral responses to S. marcescens in a binary choice assay (Fig. 1C), which is a common experimental design for studying C. elegans’ bacterial preference (Glater et al., 2014; White et al., 2019; Zhang et al., 2005). Here, we grew cultures of Db10 and Sm2170 in LB for 6 hours. We also grew a culture of OP50 overnight. Onto each 100-mm Petri dish containing nematode growth medium with 0.7% agarose, we added two 10-µl spots of bacterial culture: one spot of OP50, and the other spot of either Db10 or Sm2170. The bacterial spots were added on opposite sides of the dish, each approximately 1 cm from the inner wall of the dish. The dishes were incubated overnight at 28ºC to grow into small lawns.

C. elegans populations were prepared as for the lawn-leaving assay, and approximately 50 L4-stage nematodes were added to the center of the agar, equidistant from the two bacterial spots. An initial count of nematodes was recorded within the first 10 minutes of plating. The number of nematodes located on each lawn and on neither lawn was counted at 3 hours, 21 hours, and 27 hours after plating.

Data Analysis

Analyses were run in R (v. 4.4.1) (R Core Team, 2023) using RStudio (v 2024.04.2+764) (Posit Team, 2023). Data were processed and graphics produced with the package tidyverse (Wickham et al., 2019).

We first used the liquid resistance data to examine the effects of each parasite strain on survival compared to the OP50 control. Survival was coded as binary (1/0). We fit a generalized linear mixed model (‘glmer’ function in the package lme4 (Bates et al., 2015)) with a binomial family that included bacterial strain and C. elegans strain as fixed effects, and plate as a random effect. To examine variation in resistance to the parasites across host strains, we fit a second binomial GLMM to the data only for the parasite treatments, with host strain, parasite strain, and dose as fixed effects. We controlled for plate with a random effect.

We then used the avoidance data to examine variation across host and parasite strains. For the lawn-leaving data, we fit a binomial generalized linear mixed model to the counts of hosts on and off the parasite lawns, using host strain, parasite strain, and count time as predictors, and including a random effect for plate. For the choice assay, we calculated a ‘choice index’ for each plate by subtracting the number of hosts on the S. marcescens lawn from the number on the OP50 lawn and dividing by the total, including hosts on neither lawn. A negative choice index indicates attraction to the parasite, and a positive choice index indicates avoidance. We fit a linear mixed model that included host strain, parasite strain, count time, and block as fixed effects, and plate as a random effect.

Model fits were assessed with the ‘check_model’ function and its extensions in the package performance (Lüdecke et al., 2021) and tested for overdispersion where appropriate. We estimated marginal means (‘emmeans’ function from the package emmeans (Lenth, 2023)) and where appropriate, converted them from log-odds to percentages for ease of interpretation (‘plogis’ function). Models were compared using AIC scores (‘AIC’ and ‘compare_performance’ functions).

We took two approaches to examine correlations between avoidance and resistance. The first was to run pairwise correlations (‘cor’ function) between all the avoidance and resistance trait variables to determine whether any individual measures of avoidance and resistance were correlated. We performed a Bonferroni adjustment to correct for multiple comparisons. The second was to conduct a principal components analysis (PCA) on the avoidance traits and a separate PCA on the resistance traits to capture a single axis of variation within each defense category (‘princomp’ function). We then fit a linear model to the major axis for avoidance and the major axis for resistance. For strain-level estimates of each defense trait, we took the mean trait value across replicates for each host strain for each treatment, and logit-transformed the variables that were measured as proportions.