Stress responses upon starvation and exposure to bacteria in the ant Formica exsecta

Animals

For this study we used a population of the ant F. exsecta located in southwestern Finland in the municipality of Raseborg, Prästkulla (FI) (59°58′44.6″N 23°20′50.9″E). Our sampling procedures are non-destructive and do not cause harm to the field colonies. No federal laws apply to the collection and scientific study of insects in Finland, and the study-site at Prästkulla is not under protection. The specific population consists of nests with multiple reproductive queens (polygyne), hence the relatedness, and consequently the genetic similarity, among workers is low, yet genetic differences between individual nests is not expected to be high (Sundström, Seppä & Pamilo, 2005). In August 2013, we collected adult workers and nest material from inside nine nest mounds, and transferred the workers, and ca. 5dl of nest material immediately into nest boxes (30 × 20 × 15 cm), lined with Fluon^® (Whitford, Runcorn, UK) in order to prevent the ants from escaping. The ants were allowed to acclimatize to the laboratory conditions for 24 h at ambient room temperature (23 °C), and were fed daily ad libitum with a standardized diet based on agar, honey and egg (Bhatkar & Whitcomb, 1970).

Experimental design

To start the experiment we transferred 20 monomorphic worker ants from each nest into eight Fluon^® coated pots (Ø seven cm, h: five cm) with a plaster bottom (eight pots per nest). To maintain humidity, an open 1.5 ml tube filled with water, and a piece of cotton was placed in each pot, so that the relative humidity was maintained at ca. 75%. The ants were also allowed to acclimatize to the experimental pots for 24 h before the start of the experiment, and were fed with ∼100 μl of the standard diet. After 24 h of acclimatization dead ants were replaced from the main colony nest boxes to maintain a starting group size of 20 ants per treatment.

At the start of the experiment the eight pots per colony (72 pots, 20 ants each) were divided into four groups that corresponded to four Exposure treatments (18 pots per Exposure) including a control (factor denoted as Exposure). For these four Exposures we supplemented ∼100 μl standard diet either with 50 μl of clear LB-medium (Exposure denoted as control-exposure), or with 50 μl of concentrated bacteria growth culture (Exposures: S. marcescens, E. coli and P. entomophila), leading to a final volume of 150 μl of food, which was provided in a detached cap of a plastic test tube. Each day, a new cap was used and the old food was removed from the pots. In order to provide a standardized amount of live bacteria each day, the growth-cultures were renewed daily by inoculation with 30 μl of overnight culture in 10 ml LB-medium at 37 °C without shaking. We concentrated the bacterial culture by spinning the overnight culture tubes for 3 min at 8,000 rpm, removed all liquid LB-medium and subsequently suspended the remaining pellet in one ml clear LB-medium, which led to a final concentration of ca. 10⁸ cfu/ml for each Exposure.

The experiment featured two phases. For the first 7 days (first phase), the ants in both pots of each Exposure (18 pots per Exposure) were fed daily ∼150 μl diet according to the assigned Exposure. After 7 days, the second phase started (factor denoted as Treatment), during which the ants in one of the two pots of each Exposure (nine pots per Exposure and Treatment) were deprived of food for the next 7 days (denoted as starved), whereas those in the other pot continued to receive food on their assigned Exposure (denoted as continuously exposed). This design allowed us to assess the combined effects of oral bacterial exposure and starvation, as the effect of starvation could have been negated, if oral exposure had followed the starvation step.

Throughout the experiment mortality was recorded daily, and dead ants were removed from the pots. To assess the change in gene expression induced by the treatments, we randomly selected three ants per pot for qPCR analysis on day nine after the beginning of the experiment (i.e., 2 days after the onset of the second phase). Pilot experiments suggested that starvation-induced mortality set in around 2 days after removal of the food, as we observed a 30% drop in survival within the first 2 days, which then petered out during the following days. Thus, we expected effects on gene expression to be pronounced at this time point. Each colony constituted one biological replicate per Exposure/Treatment combination. For each pot, three ants were placed in 300 μl Isol-RNA Lysis Reagent (5 PRIME, Hamburg, DE), and cut into small pieces. The samples were then stored at −80 °C until further processing.

Gene expression analysis

To extract the RNA we homogenized the thawed samples in 600 μl Isol-RNA Lysis Reagent (5 Prime) with two stainless steel beads using a TissueLyser (Qiagen, Hilden, Germany). We then added 400 μl Isol-RNA Lysis Reagent and 150 μl chloroform (Sigma, St. Louis, MO, USA). After mixing we centrifuged the samples for 10 min at 13,000 rpm at 4 °C, and transferred the upper, transparent phase, containing the RNA, to a new, autoclaved 1.5 ml tube, and supplemented with 500 μl isopropanol (Sigma, min. 99%). After mixing, we let the suspended RNA precipitate over night at −20 °C, and then centrifuged the samples for 30 min at 13,000 rpm at 4 °C to sediment the RNA. After removal of the supernatant, we washed the pellet on ice with 500 μl 80% EtOH (Altia Oyj, Helsinki, Finland), and centrifuged for 10 min at 13,000 rpm at 4 °C. After drying the pellet we dissolved the RNA in autoclaved ddH₂O. We measured the concentration and quality of the RNA photospectrometrically with a NanoDrop (Peqlab, Fareham, UK), and eliminated possible gDNA contamination by DNase digest (DNase I, RNase-free; Thermo Scientific, Waltham, MA, USA) before cDNA synthesis with a blend of oligo(dT) and random primers (iScript cDNA Synthesis Kit; Bio-Rad, Hercules, CA, USA). For cDNA synthesis we used one μg RNA for each sample, and afterward diluted the resulting 20 μl cDNA in 80 μl autoclaved ddH₂O.

As target genes we chose eight genes representing immune defense mechanisms on a broad scale. We included two storage protein coding genes Vitellogenin 1 (Vg1) and Arylphorin (Aryl), both of which also have immunological functions (Amdam et al., 2004; Zhu et al., 2009), one Insulin Receptor (IR3), two antimicrobial peptides Lipopolysaccharide-binding protein (LPS-bp) and Hymenoptaecin (Hyme), the two cascade molecules Pro-Phenoloxidase (PPO) and Toll-receptor (Toll), as well as the enzyme Lysozyme C (LysC), which has a double role in digestion and immune defense (Cançado et al., 2007; Kajla et al., 2010). A list of the primer sequences is provided in the Table S1.

We designed qRT-PCR primers using the online Primer3 internet-based interface (http://www.ncbi.nlm.nih.gov/tools/primer-blast/) (Untergasser et al., 2012). Primers were designed by the rules of highest maximum efficiency and sensitivity (Table S1), to avoid formation of self- and hetero-dimers, hairpins, and self-complementarity. Gene-specific primers were designed on the basis of the sequences obtained from F. exsecta transcriptome (Johansson et al., 2013). Q-RT-PCR was performed on 384-well plates on a CFX384 Touch™ Real Time PCR Detection System (Bio-Rad, Hercules, CA, USA) using iQ™ SYBR^® Green Supermix (Bio-Rad, Hercules, CA, USA), with an initiation of 3 min at 95 °C, 40 cycles of 15 s at 95 °C for denaturation, followed by 45 s at 58 °C for annealing/extension, and a final step for 7 min at 95 °C. All the Q-RT-PCR assays were run using two technical replicates, which were assessed for consistency and outliers (one case), and subsequently averaged before normalization. Non-detects (no amplification signal within 40 qPCR cycles) were set to the maximal cycle number (i.e., C_t = 40), or removed in case the second technical replicate showed amplification.

Statistical analysis

We tested for the effect of ingestion of bacteria (Exposure), and subsequent starvation (Treatment) on the survival of the ants by using mixed effects Cox proportional-hazard regression with Exposure (control-exposure, E. coli, P. entomophila, S. marcescens) as fixed factor, Treatment (Continuously exposed, Starved) as time-dependent factor. The time-dependent factor captures the fact that the ants of both Treatments were essentially treated the same during the first phase of the experiment. Furthermore we included the interaction between the two factors. To account for differences among the colonies, we used colony as random intercept. As each colony was exposed to each Treatment and Exposure only once, there is only one unique combination of colony, Treatment and Exposure, which corresponds to the pot. As each pot reflected the colony, specifying pot as a random slope would allow for different reaction norms of the colonies to each treatment. Unfortunately, random slopes are not yet available in the survival packages for R. A random intercept term is not appropriate, as it would fix the same slope for all pots (i.e., colonies), and further reduce the power of the analysis, potentially leading to the acceptance of a false null hypothesis. We therefore decided to omit pot as a random factor. In case of a significant interaction between Treatment and Exposure, we examined pairwise mortality risks as post hoc analysis, correcting for multiple comparisons using false discovery rates (Benjamini & Hochberg, 1995). Due to the limited sample size we focused the statistical inference on a set of planned comparisons: (i) between starved and continuously exposed individuals within each Exposure and (ii) between each Exposure and the control exposure within each Treatment. We used a single model including all Exposures to allow the reader the comparison among all Exposures, just without the aid of statistical inference. We tested for overall effects of Treatment, Exposure and the interaction through likelihood ratio tests. We compared the full model against an intercept only model, and against a model with dropped interaction term.

Gene expression was analyzed on the level of normalized C_t values. Three ants were pooled from each pot, which led to a final sample size of nine samples per Treatment and Exposure. To test for differences in gene expression across Exposures and Treatments, we first performed Principal Component Analysis (PCA) on the inverted normalized C_t-values. We inverted the C_t-values because they are negatively correlated with the transcript level (i.e., higher C_t-values indicate lower gene expression levels). Normalization of the C_t-values was done with the NORMA-gene algorithm, which does not require reference genes (Heckmann et al., 2011). We assessed the data for suitability for a PCA: The Kaiser-Meyer-Olkin measures of the eight genes ranged between 0.61 and 0.84, and thus indicated a sampling adequacy acceptable for PCA (Field, Miles & Field, 2012). A Bartlett’s test of sphericity also indicated sufficiently strong correlations between the genes (Χ² = 321.4, p < 0.0001), and the determinant of the correlation matrix (det = 0.009) indicated that the correlations were not too strong for a PCA (i.e., det > 0.00001; Field, Miles & Field (2012)).

We used an unrotated PCA for component selection, and consulted the scree plot of the Eigen values for the relative weights of the components. Two components had an Eigenvalue >1.0 (Table S2), but the scree plot showed a high relative weight only for PC1, whereas the weight of PC2 was not distinctively higher compared to the subsequent components (Fig. S1). A preliminary analysis including PC1 and PC2 showed no significant effects of PC2 for any of the factors (Tables S3 and S4). We therefore decided to use only PC1 for the analysis. The PC scores were then used as dependent variables in a linear mixed effects model to test for an association of PC1 with Exposure and Treatment. Colony was added as a random intercept. In case of a significant interaction of Treatment and Exposure, we examined pairwise gene expression differences as post hoc analysis, correcting for multiple comparisons using false discovery rates. We used the same procedure for post hoc comparisons as for the survival data. To assess the contribution of each gene to the effects found for PC1, we here arbitrarily define the loadings as strong (loading 0.67–1.0), moderate (loading 0.33–0.66), or weak (0.0–0.32). Because the PCA indicated multicollinearity among the genes, we did not use a gene-by-gene analysis as a basis for our conclusions, but provide one as Supplementary Material (Fig. S2; Table S5).

We assessed the validity for a cox-proportional hazard model with a proportional hazard test (Grambsch & Therneau, 1994), which did not reveal any indication for non-proportional hazards (Chi-squared: 9.17, p = 0.24). As a method to test for the presence of proportional hazards is not available for mixed effects Cox-proportional hazard models, we specified a default cox-ph model, with Colony specified as frailty term (Therneau, Grambsch & Pankratz, 2003). The validity of the linear mixed effects models was assessed visually through inspection of the residuals and the leverage, which showed no evidence for misspecification of the model. For all statistical tests we used a threshold level of significance of α = 0.05. Statistical analysis was carried out in R version 3.0.2 (R Core Team, 2015), using the packages survival (Therneau, 2013), coxme (Therneau, 2018), psych (Revelle, 2016), lme4 (Bates et al., 2015), lmerTest (Kuznetsova, Brockhoff & Christensen, 2017), and multcomp (Hothorn, Bretz & Westfall, 2008).

Survival

Overall the Cox-proportional hazard regression revealed significant effects of the treatments on the survival of the ants (Likelihood ratio test: Chi-square = 476.89, df = 7, p < 0.0001). The interaction was close to significant (Likelihood ratio test: Chi-square = 7.41, df = 3, p = 0.06). Given the borderline effect, we decided to retain the interaction term in the model as the log likelihood was slightly higher when the interaction term was included (loglik_Full = −4032.0, loglik_Reduced = −4035.7).

Starvation reduced the survival of the ants in all exposure regimes (Fig. 1; Tables 1 and 2). Exposure to S. marcescens significantly reduced survival compared to the respective controls, both under continuous feeding and starvation (Fig. 1A, comparisons between black and red continuous lines, and dotted lines, respectively; Tables 1 and 2). No significant changes in mortality emerged when the ants were continuously fed and exposed to E. coli or P. entomophila (Figs. 1B and 1C, comparison between black and red continuous lines; Tables 1 and 2). However, when these ants were starved, they survived significantly longer if they had been exposed to E. coli or P. entomophila, than if they had been exposed to the control-exposure (Figs. 1B and 1C, comparison between black and red hatched lines; Tables 1 and 2).

Gene expression

From the PCA on gene expression we retained one component, which explained 51% of the variation (Table S2). The loadings of all genes were positive, and either moderate (Vg1: 0.60; PPO: 0.66; Hyme: 0.36; LPS-bp: 0.62), or strong (Aryl: 0.76; IR3: 0.88; LysC: 0.82; Toll: 0.89). This shows that all genes were regulated in the same direction and to a similar degree (apart from Hymenoptaecin), and that the PC1 scores therefore reflect the general level of gene expression. Overall the mixed effects model revealed significant effects of the treatments on gene expression (Likelihood ratio test: Chi-square = 177.2, df = 2, p < 0.0001), with a significant interaction term (Likelihood ratio test: Chi-square = 130.81, df = 3, p < 0.0001). Continuous exposure to S. marcescens led to a decrease in gene expression levels, compared to the control-exposure, whereas the opposite occurred when the ants were starved (Fig. 2, comparison between black symbols, and white symbols, respectively; Tables 3 and 4). By contrast, ants that were continuously exposed to E. coli or P. entomophila showed higher gene expression levels than ants of the control-exposure (Fig. 2, comparison between black symbols). When the ants had been exposed to E. coli, this also occurred after starvation, but not when they had been exposed to P. entomophila (Fig. 2, comparison between white symbols; Tables 3 and 4).

Ants that had received either the control-exposure or P. entomophila treatment showed lower gene expression levels after starvation, than during continuous exposure (Fig. 2, comparison between black and white symbols within regime). By contrast, the ants that had been exposed to S. marcescens, showed significantly higher gene expression levels in starved, than in continuously exposed individuals. Only ants that had been exposed to E. coli showed no difference in gene expression between starved, and continuously exposed individuals (Fig. 2; Tables 3 and 4).