MATERIALS AND METHODS

Study species

The experimental procedure was carried out simultaneously for two ant species: Lasius niger (L., 1758) (Fig. 1) commonly termed as the black garden ant, widespread in Europe and northern America (Klotz, 2008); and Lasius neoniger Emery, 1893 known as the turfgrass ant, distributed in northern and Midwestern North America (E. O. Wilson, 1955). Both species are highly abundant in urban/suburban lawns and gardens and have similar diets. Natural diets are hemipteran honeydew and living and dead insects (Klotz, 2008; E. O. Wilson, 1955).
L. niger (Fig. 2) queens were collected in Münster (NRW, Germany) during a single nuptial flight on July 5, 2017. L. neonigerqueens were collected in Richmond (IN, U.S.A) during two nuptial flights on June 30 and July 12, 2017. Queens were housed in 18x150 mm test tubes half-filled with water and plugged with cotton. Most females initiated egg-laying after the first week. No food was provided until the emergence of the first workers’ cohort. Once workers were present, colonies were housed in 500 ml plastic boxes containing cotton-stoppered test tubes with water. Box walls were covered with fluon (polytetrafluoroethylene) to prevent escape. Colonies were housed in incubators at 25°C for L. niger and to 30°C for L. neoniger . Colonies were overwintered at 10°C (L. niger ) and 15°C (L. neoniger ) for two and one months, respectively, between January and February 2018. An acclimatization phase of 15 days prior and post overwintering period was allowed for gradual temperature transition. The higher temperature used in L. neoniger was due to concurrent experiments carried out with multiple, ecologically different, species.

General Data Analysis

All analysis and modelling were conducted for each species separately as the aim of this research was not to compare the species but to study the effects of the treatments simultaneously for both of them.
All data analysis was done in R version 3.6.2 (R Core Team, 2019). Type-II analysis of variance tables based on 1-d.f. chi square tests were used to assess the significance of terms in all models (e.g. lm, glm, lme) using the ’Anova’ function in the “car” library (Fox & Weisberg, 2018); these tests are marginal, i.e. each term is tested in presence of all other terms in the model and tests are not order-dependent. Data were summarized using libraries plyr (Hadley Wickham, 2010), dplyr (H Wickham, 2018) and Rmisc (Hope, 2013). Figures were made after the means and confidence intervals calculated from the raw data using ggplot2 (Hadley Wickham, 2016) with RColorBrewer (Neuwirth & Brewer, 2014).

Colony Growth

A pilot experiment was conducted in 2015/16 with L. neoniger . Twenty-four Colonies were monitored over six months; colonies were distributed among three treatments varying in the protein to carbohydrate content of their diet. Colony growth (the number of workers) increased near linearly with increasing parts of carbohydrates, where colonies fed a diet with a P:C ratio of 1:1 produced fewer workers than those fed a diet of 1:2, and those fed a diet of 1:3 produced the greatest number of workers. The experiment also documented that a weekly provision of ~125 µl of artificial diet (see below) was sufficient to maintain colony growth and survival. These pilot data informed our choice of the quantity of diet to provision and the P:C range.
Colonies in the present study were randomly assigned to treatments after the first census (approximately one month of growth). The treatments were feeding frequency and diet composition (protein to carbohydrate ratios, abbreviated as P:C henceforth) (Fig. 3). Feeding frequency had three levels: a 125 µl aliquot every other week (0.5), every week (1), or twice per week (2). The P:C had four levels representing a linear change in P:C from approximately 1:2 up to 1:16 with increments of rough doubling (see Fig. 3 for details), hereafter referred to as low carbohydrate (L), medium-low (ML), medium-high (MH), and high carbohydrate (H). The starting number of colonies was 96 for L. neoniger (8 colonies per treatment combination), and 86 for L. niger (7-8 colonies per combination).
Portions were 125 µl (127.33 ± 0.38 mg, mean ± SD) of isocaloric chemically defined diets prepared using casein, whey protein, egg powder and sucrose in agar solution (Dussutour & Simpson, 2008). Insect vitamins (Vandersant, MP Biomedicals, Germany) and a preservative (Methylparaben, Dephyte, Germany) were also added equally to the diets. For details on the diet constitution and recipes see Table S1. Food leftovers were removed after three days and new batches of food were prepared every three weeks. The development of the colonies was monitored over 10 months by counting larvae and workers on a monthly basis. Every colony census was done in a non-invasive way by visually inspecting the test tubes under a stereomicroscope (SZX7, Olympus, USA) at 40X magnification. Colonies were considered ‘dead’ only if the queen died.
Data were analysed to answer three related questions: 1) Did colony survival vary depending on treatments? 2) Did the dynamics of growth vary with treatments? and 3) Did overall growth vary with treatments?
G-tests (function ’G.test’) were used to assess differences in colony survival as related to treatments using the library RVAideMemoire (Hervé, 2014). The ’G.multicomp’ function was used for post-hoc comparisons - p-values were corrected for false discovery rate (FDR).
A linear mixed-effects model for each species, function ’lme’ from library nlme (Pinheiro, Bates, DebRoy, & Sarkar, 2014) was used to assess growth dynamics. Growth was approximated as the number of workers in each monthly census, excluding colonies that died. Feeding frequency, diet P:C and census number (the latter as a third-order polynomial, identified after initial visual inspection of the data) were the fixed effects (factorial), and colony ID was expressed as a random factor (96 colonies for L. neoniger and 86 for L. niger ). Additionally, census number was added as a first-order autocorrelation structure (corAR1) grouped by colonies.
To assess the effects of treatments on overall growth we used a two-way analysis of variance with the frequency of feeding and diet P:C as fixed effects and the difference in numbers of workers between the last and second census (i.e. the first census that had experienced the treatments) as the response variable.

Worker Phenotype

Several different worker traits were measured: head width, dry mass, lean mass and lipids content. These characteristics, individually and in aggregate, are typically responsive to nutrition and change with colony-level development and health (Smith & Tschinkel, 2009; Tschinkel, 1993). Head width of young workers (recognized by their lighter colouration, Fig. 2) was used as a proxy of body size (Schwander, Rosset, & Chapuisat, 2005). Head width images were taken at 56X magnification using a stereomicroscope (SZX7, Olympus, USA) with an attached camera (Retiga 2000R, Q-Imagine, Canada) and measured at the maximal distance across the eyes using iSolutions-Lite software (iMT Technology, USA). The number of ants measured per colony varied due to availability (8 ± 2.6, mean ± SD). Lipid extractions were done using gravimetry after repeatedly washing the samples with ether; we used the protocol of Tschinkel and Smith (2009). Samples were first dried for > 48h at 60°C. Due to the low mass of Lasiusworkers, we extracted a sample of ants, en masse , for each colony. The number of ants per colony was 8.8 ± 1.8 (mean ± SD). The lipid extraction gave three measures, dry mass, lean mass, and the proportion of dry mass that was lipids (i.e., the portion extracted). Both dry mass and lean mass were normalized as the mass of the sample divided by the number of ants in the sample. Linear models (’lm’ function) were used to analyse dry mass, lean mass and lipid content; head width was analysed using generalized linear models (’glm’ function) with Gamma distribution.

Stable Isotopes Pulse Experiment

In order to examine the flow of nutrients into a colony, we used a stable isotope pulse. Diets enriched with the heavy stable isotopes15N and 13C (when abbreviating carbon with C we use the superscript for atomic mass in order to differentiate from C for carbohydrates) were fed once and the flow of the isotopes to different castes was assayed over time. After the final census of L. neoniger , a subset of eight colonies in the ML treatment were randomly assigned to either of two treatment groups, those receiving the heavy isotope enriched diet (Table S1) or those receiving an unenriched (but otherwise identical) control diet. Five of the eight colonies were provided with the enriched diet, and the remaining three were fed the unenriched diet in order to control for background isotopic levels. The ML diet used in the previous experiment was modified so that isotopically labelled materials could be easily integrated. The basic methods of Shik et al. (2018) were followed. A small quantity of 15N labelled or unlabelled ammonium nitrate (labelled: Ammonium nitrate-15N, EB0056; unlabelled: Ammonium Nitrate SHBJ1050, Sigma-Aldrich, USA) and a small quantity of13C labelled or unlabelled glucose (labelled: D-Glucose-1-13C, MBBC0227V; unlabelled: D-(+)-Glucose, SLBJ0583V, Sigma-Aldrich, USA) were added so that the P:C ratio of the diet was not changed. The supplemental ammonium nitrate amounted to 0.9% (by dry mass) of the protein sources in the diet, and the supplemental glucose was approximately 4% of the carbohydrate sources in the diet. Colonies were offered 125 µL of the diet for 24 h. After this period, the food was removed and a random sample of approximately five workers and five larvae was withdrawn, frozen for 24 h, and then stored in a drying oven at 60°C (following the methods of Smith & Tillberg (2009)). Colonies were then sampled again after 96 h following the same procedure. Dried samples were homogenized in sterile aluminium capsules and elemental analysis was carried out at the University of California at Davis Stable Isotope Facility using a PDZ Europa ANCA-GSL elemental analyser interfaced to a PDZ Europa 20-20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, UK). The atomic per cent enrichment was calculated as the atomic per cent of the heavy isotope (relative to the lighter isotope) divided by the same quantity for the same sampling time and caste in the unlabelled colonies. The individuals in colonies fed enriched diets, overall, had 9% more 13C and 71% more 15N compared to individuals in colonies fed unlabelled diets. Each element was analysed separately using the atomic per cent excess as the response variable. Time (24 vs. 96 h) and developmental caste (worker vs. larva) were the categorical predictors. Due to heterogeneity of variances, these data were analysed using a Kruskal-Wallis test followed by a Dunn test (’dunnTest’ function in the FSA library (Ogle, 2018)) for pairwise differences using the Benjamini-Hochberg method.

Diet preference under field and lab conditions

Field trials were conducted between May and June 2018 in grassy lawns for both naturally occurring colonies of L. niger at the Botanical Garden of Münster (51°57’55.9”N 7°36’22.9”E), and for L. neoniger on the campus of Earlham College in Richmond, Indiana (39°49’15.2”N 84°54’47.0”W). A 7.5 cm microscope slide with 125 µl portions of all four P:C ratios (as described above) in random order, was placed near an active nest entrance. Baited nest entrances were at least 1.5 m apart and each was used only once. After placement, the number of ants at each food portion was counted every five minutes during 40 minutes. If other ant species were seen at any food portion at any location, then that time point and all subsequent time points at that location were not included in the analysis. We assumed that all food portions had an equal probability of discovery.
After food preference for the four P:C levels had been established, we performed an additional diet choice test with the species L. neoniger only. The diet was modified to ascertain whether P:C preference was driven by P content and/or C content. Levels of P were varied (holding C constant), and then separately, levels of C were varied (holding P constant). We used the midpoint for these diets to be the most preferred ’ML’ diet (see Fig. 3) and then halved and doubled either P or C. We then presented three sets of diets to field and lab colonies (as above) - all paired diets had either constant C or constant P (i.e. they were either on the vertical or the horizontal black line in Fig. 3).
Diet preference data collected from field colonies were square-root transformed, in order to homogenize variances, and analysed using a mixed-effect model to take into account the non-independence of repeated counts at each colony. The model included the ”bait” type (i.e., the diet P:C) as a fixed factor, and colony ID as a random factor (18 colonies for L. neoniger and 23 for L. niger ). A first-order autocorrelation structure (corAR1) was specified as “Time | colony ID / bait type”. As above, the ’lme’ function from the nlme library (Pinheiro et al., 2014) was used to analyse data.
Food preference trials were also conducted on two laboratory-rearedL. neoniger colonies (c.a. two years old with over one hundred workers). Food portions were placed in foraging arenas attached to colony containers and then foraging was filmed using iPhones and the Lapse-It application for time-lapse. Pictures were taken every 5 minutes for at least 3.5 hours and all ants at each bait were counted for every time period. Laboratory data were not statistically analysed due to low sample size. Both colonies showed the same preference to the four diet ratios, but only one colony of the pair responded with high recruitment to the variable P and C diets (Fig. S1).