Food dissemination in ants: robustness of the trophallactic network against resource quality

Social insects, and ants in particular, live in large and complex societies. Information sharing and the presence of feedback loops between workers (Schafer et al., 2006) allow for a collective exploitation of the best quality food source (e.g. Beckers et al., 1993) while coping with the colony needs (Portha et al., 2002; Sorensen et al., 1985). The individual and collective feeding behaviour in ants is influenced by the food source characteristics such as the sucrose concentration (Beckers et al., 1992; Cassill, 2003; Josens et al., 1998; Price et al., 2016; Reid et al., 2012) and the nature of the food (Markin, 1970; Sorenşen et al., 1981; Sorensen and Vinson, 1981), as well as by the level of starvation of the colony (Josens and Roces, 2000). However, the way in which these parameters impact food dissemination inside the colony and, in particular, the individual patterns of trophallaxis exchange, their dynamics and the structure of the associated networks, have been largely overlooked (Cassill, 2003; Greenwald et al., 2015, 2018; Sendova-Franks et al., 2010). Yet, different parameters, such as starvation, affect these dynamics and networks, facilitating food recruitment, which speeds up the dynamics of food accumulation and dissemination at the intranidal level (Buffin et al., 2012; Mailleux et al., 2010; Sendova-Franks et al., 2010). At low or intermediate starvation level, the crop is still partially laden (Cassill and Tschinkel, 1999b) and the propensity of ants to recruit and to exchange food increases with food concentration (Cassill, 2003). Moreover, this propensity also increases with starvation level (Cassill, 2003; Mailleux et al., 2011) and leads to a large total number of trophallactic events and to a fast food dissemination through the nest (Buffin et al., 2012; Sendova-Franks et al., 2010). This increase in speed is partially due to reorganization of the intranidal process of food spreading, such as a higher proportion of ants that are both donors and receivers when highly starved compared with satiated. In addition, fluid intake rate and crop filling increase with starvation (Josens and Roces, 2000) and food of a lower quality is enough to stimulate the foragers to collect food (Mayor et al., 1987; Mc Cabe et al., 2006). Some nutrients that are ignored by foragers at low levels of starvation are collected and distributed within colonies after a lengthy period of starvation (Chong et al., 2002). This reveals a lower food quality acceptability threshold of highly starved ants. During trophallaxis, the ant receiving food contacts the donor's head and mouth with its forelegs. At low levels of deprivation, this rate of contact and the probability that receivers accept trophallaxis increases with food concentration. This effect of food concentration is not observed at high levels of starvation (Mc Cabe et al., 2006). Therefore, we hypothesized that after a long starvation duration, the total number and the structure of the networks of trophallaxis, the dynamics of food accumulation and the heterogeneity (distribution) of the trophallactic activity among individuals (Cassill, 2003; Sendova-Franks et al., 2010) should be similar regardless of food concentration. To test this hypothesis and to provide a baseline model, we compared the food dissemination activity for two food sources differing in their sucrose concentration (0.1 versus 1 mol l⁻¹) in highly starved (5 days) groups of Camponotus cruentatus workers.

From five large mother colonies of the ant Camponotus cruentatus (Latreille 1802) (collected in Rochefort du Gard, France, September 2016), we created 10 subcolonies of 25 randomly picked individuals, kept in a plastic box (175×125×50 mm) containing a circular nest (95×6 mm), with access to water and sucrose solution (0.3 mol l⁻¹). All subcolonies were maintained in a controlled environment, with a temperature of 21±1°C, a relative humidity of 60±5% and a constant 12 h/12 h photoperiod. Ants were individually labelled with an Aruco tag (https://sites.google.com/site/usetrackerac/) allowing automatic identification of ants. Each tag was attached to the abdomen, had a side length of 1.25 mm and weighed less than 1% of the average mass of the ant. After 5 days of starvation, ants had access to a 0.1 mol l⁻¹ (low concentration) or 1 mol l⁻¹ (high concentration) sucrose solution feeder placed 40 mm from the nest entrance (Fig. S1A) and were filmed for a period of 120 min. Each colony was randomly tested three times at low and high concentration (N=60), with a 5 day break between the two trials.

Experiments were filmed using Panasonic GH4 cameras at a resolution of 4K. Based on snapshots taken every 4 min, in each experiment, we manually counted the number of trophallactic events. Using this footage, trophallactic events and the respective role of both partners (donor or receiver), were identified on the basis of body posture and the mandible position (Cassill and Tschinkel, 1999a; Greenwald et al., 2015; see also Fig. S1B). A Mann–Whitney U-test (MW) and a Kolmogorov–Smirnov test (KS) were respectively used to compare the total (Fig. 1A) and the temporal cumulative number of trophallactic events (Fig. 1B) between all experiments at high versus low food source concentration. Additionally, in a sample of four colonies, each tested once at high and low concentration, we recorded the identity of the donor and the receiver of each trophallactic event in the nest based on body posture and the position of the mandibles (Greenwald et al., 2015). The complete trophallaxis network was then built (see Fig. S2). Each node corresponds to one ant having performed at least one trophallactic event. Pairs of distinct nodes were connected with a directed and weighted edge, from the donor to the receiver. We calculated the ‘betweenness’, the closeness, the eigenvector and the clustering coefficient of each node of the network. Betweenness is an estimate of how important an individual ant is for promoting connectivity across the entire colony and is measured by the number of times an individual acts as a bridge along the shortest path between two other ants (Dell et al., 2014). Closeness is based on the shortest paths from an individual to every other individual: the more central an ant is, the lower its total distance from all other ants (Wey et al., 2008). The eigenvector is a value accounting for the centrality of a node's neighbours (Butts, 2008). The clustering coefficient determines the existence of ‘communities’ in a network, such as nodes with many more edges between these nodes than with the others (Saramäki et al., 2007). To test for an effect of food concentration on the structure of the trophallaxis network and on the distribution of the trophallaxis activity, for each metric, we applied a Kruskall–Wallis test (KW) of homogeneity on the N=8 (4×2) experiments (Table 1). If the KW indicated an inhomogeneity (P<0.05), we compared the two distributions (respectively for 0.1 and 1 mol l⁻¹), for each colony, with a MW.

We statistically evaluated the heterogeneity of the distribution of the trophallaxis activity among all the workers of each of N=8 experiments, using the Lorenz curve (Fig. 2; Fig. S3A) and the Gini coefficient (Fig. S3B). The Lorenz curve displays the share of trophallaxis activity (Y-axis) accounted for by X% of workers (sorted by the number of trophallactic events per individual) in the colony. A perfectly equitable distribution of foraging activity would correspond to the line Y=X. The Gini coefficient is known as the ratio between the area below the experimental Lorenz curve and the triangular area below the perfect equality case Y=X, and provides a measure of the degree of inequality in the distribution of trophallaxis activity, ranging from 0 (perfect equality) to 1 (perfect inequality).

To estimate whether the observed Gini coefficients and the social network metrics were different from random expectation, we compared each empirical network against two ensembles of N=1000 randomized networks we created by (1) randomly rewiring all edges between all nodes, destroying all features of the original network (full random network, FR); (2) rewiring the edges of the original network while maintaining the distribution of the number of trophallactic events (given/received) by each individual (degree random network, DR) (Holme and Saramäki, 2012). We then performed a Z-test (ZT) to evaluate the significance of the differences between the observed and random metrics. All analysis was conducted in Python 3.6 with NetworkX 2.1, PyGraphViz 1.4, Numpy 1.14, Scipy 1.0.0 and MatPlotLib 2.2.2 packages.

The total number of trophallactic events observed (Fig. 1A, MW: P>0.33 in each case) and the dynamics of food accumulation (Fig. 1B, KS, D=0.07, P=1.0) were independent of the current and the previous (whatever the current concentration) food source concentration.

The proportion of ants involved in trophallaxis activity was independent of the food concentration: ∼90% of ants performed at least one trophallactic event and ∼70% of ants gave and received food in each experiment (Fisher exact test, P>0.40 in each case). All eight experiments were homogeneous in terms of the distribution of the number of trophallactic events individually performed and of social network metrics of each individual (Table 1, KW, P>0.05), showing a robustness of the food dissemination and trophallaxis network against food concentration. The closeness was, however, significantly different between the high and low food concentration in two colonies, but without a clear trend, as closeness was higher at high concentration in one of the colonies and at low concentration in the other (MW: P<0.05). All trials, regardless of food concentration, had a significant skew in the distribution of the trophallaxis activity: 25% of the workers accounted for more than 50% of the total number of trophallactic events over the course of each experiment (Fig. 2). The observed Gini coefficients were systematically higher than the corresponding theoretical ones obtained from the FR networks (Fig. S3A,B; ZT: P<0.0001). The comparison of each experimentally measured social network metric revealed a significant difference from the corresponding FR network (Figs S4, S5; ZT: P<0.05), except for the clustering coefficient of two experiments at 1 mol l⁻¹. Finally, no difference occurred between experimentally measured metrics and those from DR networks (Figs S4, S5; ZT: P>0.05), except for the closeness coefficient of an experiment and the eigenvector coefficient of two experiments. The highlighted heterogeneity of the trophallaxis activity seems therefore to be the main factor shaping the trophallaxis network as the experimental networks were not different from the theoretical DR ones. Crop content modulates the individual trophallaxis rate (Greenwald et al., 2018) but we observed a skew in the total number of trophallactic events performed by each ant. This suggests that a high proportion of ants that receive food in a quantity that exceeds their individual nutritional need redistribute it. Therefore, trophallaxis allows them not only to meet their individual needs but also to contribute to food dissemination throughout the colony (Quevillon et al., 2014). Division of labour is assumed to be based on a distribution of the threshold responses among the workers (Pinter-Wollman et al., 2012). In contrast, starvation lowers thresholds (Mailleux et al., 2011) and potentially allows for identical responses to very different food concentrations. Despite this phenomenon, our results suggest that interindividual differences in activity are still maintained.

Our study is, to our knowledge, the first one concerned with the effects of food source concentration on food dissemination activity in ants. Foraging and recruitment processes are affected by food concentration (Cassill, 2003; Josens et al., 1998; Mailleux et al., 2006; Reid et al., 2012). However, our results suggest that the food dissemination network and dynamics depend mainly on the nutritional needs rather than on the concentration of food for long starvation durations (at least for the two different concentrations tested here). This is in agreement with the literature showing that insects consume a large range of food types/quality when starved (Jackson et al., 1998; Mayor et al., 1987; Scharf, 2016). Note that this study concerns subgroups of workers and that further experiments are needed to generalize our results by testing more natural ant colonies (in the presence of queen and brood).

Further investigations are also required to link the mechanisms underlying the observed interindividual variability in trophallaxis activity, colony needs and food dissemination. For example, testing the effect of shorter starvation durations would be important as less hungry individuals with non-empty crop content (Greenwald et al., 2018) exchange only highly energy-rich food and ignore food at low concentration (Cassill, 2003). In contrast, the starvation duration we tested could have led to a high level of excitation that may prevent any effect of food concentration on the dynamics and the structure of the intranidal trophallaxis network.

We thank two anonymous referees for their helpful comments.