Task switching is associated with temporal delays in Temnothorax rugatulus ants

Lay Summary

Switching between tasks may incur time costs and select for the evolution of individual specialization. Individuals in animal societies often specialize on particular tasks, though the benefits of such division of labor are not always clear. We tested whether there were temporal costs of switching between types of work in a species of ant. We found that individuals are faster in starting a new bout of work when repeating the same type of work, regardless of which task workers had specialized on. Workers may therefore specialize on certain tasks to avoid this cost of switching.

Twitter: @GMcLean0011

Abstract

The major evolutionary transitions often result in reorganization of biological systems, and a component of such reorganization is that individuals within the system specialize on performing certain tasks, resulting in a division of labor. Although the traditional benefit of division of labor is thought to be a gain in work efficiency, one alternative benefit of specialization is avoiding temporal delays associated with switching tasks. While models have demonstrated that costs of task switching can drive the evolution of division of labor, little empirical support exists for this hypothesis. We tested whether there were task-switching costs in Temnothorax rugatulus. We recorded the behavior of every individual in 44 colonies and used this dataset to identify each instance where an individual performed a task, spent time in the interval (i.e., inactive, wandering inside, and self-grooming), and then performed a task again. We compared the interval time where an individual switched task type between that first and second bout of work to instances where an individual performed the same type of work in both bouts. In certain cases, we find that the interval time was significantly shorter if individuals repeated the same task. We find this time cost for switching to a new behavior in all active worker groups, that is, independently of worker specialization. These results suggest that task-switching costs may select for behavioral specialization.

INTRODUCTION

Division of labor is a fundamental concept in evolution because it is often a component of the major transitions in evolution (Maynard Smith and Szathmáry 1995). Indeed, division of labor has been documented in multicellular groups (Michod 2007) and numerous animal societies, for example, shrimp (Duffy 1996), birds (Bednarz 1988; Arnold et al. 2005; Coulson and Coulson 2013), and dolphins (Gazda et al. 2005). In each of these cases, there are individuals within the group that carry out a specific subset of behaviors different from others in the group, rather than all individuals carrying out all behaviors at the same frequency. In eusocial animal societies such as social insect colonies, specialization, and thus division of labor, is particularly prevalent (Wilson 1975; Bourke 2011).

In certain transitions, for example, multicellularity (Fisher et al. 2013) and eusociality (Hölldobler and Wilson 1990), specialists have sometimes evolved distinct morphologies that facilitate the completion of certain tasks (Alvarado et al. 2015). Indeed, morphological specialists are often most efficient at performing their caste’s subset of behaviors (Wilson 1980; Powell and Franks 2005; Powell 2008). However, the evolution of morphological specialists can be constrained, for example, by small colony size (Bourke and Franks 1995); and the majority of ant species do not have morphological specialization among workers. Behavioral specialization, however, is prevalent (Hölldobler and Wilson 1990), and this suggests that the evolution of behavioral specialization precedes morphological specialization among workers (although, see Perrichot et al. [2016] for evidence of early morphological specialization in a basal ant). Given the presence of division of labor in species with a monomorphic work force, there are likely benefits that lead to task specialization in these groups even when morphological differentiation is not (yet) present.

One speculated benefit of division of labor is that increased task specialization leads to increased individual efficiency when performing that task due to learning. This hypothesis was proposed by Smith (1776) to explain the benefits of division of labor in human societies. Indeed, work in honeybees (Apis mellifera) has found that efficiency, that is, work completed per time, improves over the first several bouts of a corpse removal (Trumbo and Robinson 1997) and that Foragers improve almost 2-fold in their net foraging uptake over the course of their life (Dukas and Visscher 1994). Similarly, ants (Acromyrmex versicolor) also improve their undertaking efficiency as they perform the behavior more (Julian and Cahan 1999). Similarly, Temnothorax albipennis colonies emigrate to new locations faster as more emigrations are performed (Langridge et al. 2007). However, learning to perform a task better over a few bouts does not necessarily lead to more efficient specialists: both specialists and generalists may benefit from learning enough to reach near-maximal efficiency. Indeed, Dornhaus (2008) examined the worker efficiency across 4 tasks important to colony function in T. albipennis and found no association between the degree of specialization of a worker and the amount of work completed per time in the respective task. Thus, although in some species the efficiency gains with experience differentially benefit specialists, this may not be a universal phenomenon in social insects.

Another factor that could select for division of labor is the benefit of avoiding the costs of switching tasks (Duarte et al. 2012; Goldsby et al. 2012). Several mechanisms could lead to task switching costs. For instance, in the swarm-founding wasp, Polybia occidentalis, individuals in small colonies must sequentially switch between foraging for water, foraging for pulp, and building the nest (Jeanne 1999). P. occidentalis individuals that switch tasks do not carry the maximal loads of building material when returning to the nest, as they also have to acquire water for nest building. In larger colonies, individuals carry the maximum, or near-maximum, load of building material and then transfer this to building individuals that subdivide the building material (Jeanne 1996). As a result, small colonies with a higher rate of task switching likely expend more energy per capita due to extra foraging flights and longer cueing delays than larger colonies that divide labor (Jeanne 1986). In addition to production inefficiencies associated with switching, there may be other task switching costs. For example, individuals switching to a different behavior may delay work in the new task because of the time needed for cognitive retrieval of the motor patterns for the new task (Chittka et al. 1997). A third possibility that may increase switching times is a time cost of having to find (“forage for”) work. Specifically, individuals in certain societies, for example, honeybees (Johnson 2009), quit tasks and spatially explore the nest for new tasks. If workers face a choice between repeatedly performing a particular task where they are already informed about where the next work is required, or exploring the nest for work in new tasks after completing a previous task, these workers might save time by avoiding switching. Repeatedly switching tasks may therefore incur temporal delays that could reduce colony productivity.

As there is a lack of information on why division of labor evolved in certain systems (West et al. 2015), we investigated temporal task switching costs in Temnothorax rugatulus colonies. To do so, we isolated instances where individuals perform 1 task, spend an unconstrained amount of time not performing a (work) task (i.e., instead they were wandering around the nest, stationary, or self-grooming), and then perform a different task after the interval. We compared the length of these time intervals with those of instances where individuals performed 1 (work) task, spent some amount of time not performing a task as above, and then performed the same task again. In both cases, we refer to the time in-between task bouts as the “interval time”. We refer to the first task as the “original task” and the subsequent task as the “second task”. We tested the main hypothesis that individuals would have significantly shorter interval times when repeating the same task compared to individuals that switched to a new task; if this were the case, it would imply that switching between task types is costly and, thus, might be avoided by specialization. We were also interested in what processes may induce workers to switch tasks. For example, individuals may be interrupted and recruited to food sources (von Frisch 1946) or other tasks (Franks and Richardson 2006). We therefore also investigated interactions during the interval time to see if this influenced task switching.

METHODS

Study species

We used the ant T. rugatulus as a study species because entire colonies can be collected, maintained in the lab, and individuals in a colony can be marked, allowing us to collect individual-specific behavioral data. T. rugatulus colonies show behavioral specialization (Charbonneau and Dornhaus 2015) and their behavior is similar between the lab and the field (Charbonneau et al. 2015). T. rugatulus has medium-sized colonies (typically ~50 to 250 workers [Bengston and Dornhaus 2013]) and monomorphic workers (though, see Westling et al. [2014]), both of which are typical for most ant species (Dornhaus et al. 2012). T. rugatulus colonies live in rock crevices in the field, while in the lab the colonies are housed between 2 glass slides and cardboard siding on 1 or 3 sides (Pinter-Wollman et al. 2012; Charbonneau and Dornhaus 2015). This setup allows us to unambiguously record behavior of individuals in the colony. We collected 44 colonies of T. rugatulus from the Santa Catalina Mountains near Tucson, AZ, between 2012 and 2014, and we brought the colonies into the lab for observation. Seventeen of the colonies were provided with cardboard on 3 sides of the nest while the remaining colonies had built walls using small stones before video analysis began. We uniquely paint-marked each individual in every colony (Sendova-Franks and Franks 1993) and we recorded the behavior of paint-marked individuals in T. rugatulus colonies by recording a 5-min video of each colony (165 videos recorded in total). The number of recording intervals varied, but between 3 and 5 recordings of each colony were captured within a 21-day time span. All colonies were filmed within 3 months of being in the lab, which avoids long-term effects of lab housing on the colonies, for example, altered age distributions. We then scored the behavior of each individual at each second in each video. We generated unique R (R Development Team 2015), version 3.2.2, scripts to isolate all task switches or repeated tasks (see below for behaviors) along with other variables associated with task intervals (see Data accessibility).

Colonies in the laboratory were provided with ad libitum water and food (10 frozen Drosophila adults and 2-mL tube of honey water each week), maintained on a 12-h light cycle and at a constant temperature of ~21 °C. These colonies were also part of a set of studies on inactive individuals; however, they were not manipulated in any way other than the artificial nesting in the lab as described above (Charbonneau and Dornhaus 2015; Charbonneau et al. 2015). Colony size ranged from 26 to 138 workers; each colony contained at least 1 queen (maximum number of queens = 6) and between 0 and 132 brood.

Tasks and worker groups

Individuals were classified as performing 1 of the following behaviors at each second: nest building, brood care, inactivity, feeding, foraging, grooming others, self-grooming, trophallaxis, or wandering inside the nest (see Supplementary Material for task descriptions, adapted from Charbonneau and Dornhaus [2015]). We also classified individuals into 4 “worker groups” based on their overall profile of task performance using hierarchical clustering analysis (Charbonneau and Dornhaus 2015). We term these groups Inactives, Nurses, Wanderers, and Foragers by the activity that workers in the group perform more often compared to other workers. Importantly, although individuals from these groups tend to perform the work associated with their group name, individuals from these groups still perform other tasks as there is moderate specialization in this species (see below). For example, Wanderers tend to wander but are also observed performing brood care. It is also important to notice that each task does not have a specialized worker group; the number of groups (or clusters), while somewhat arbitrary, was determined by standard clustering methods (Charbonneau and Dornhaus 2015) and reflects the overall main divisions among workers in this species. Note that when referring to the worker groups as determined by the cluster analysis, the task name is capitalized (e.g., Nurses).

Identifying task switching

We isolated task switches in 2 different ways. First, we labeled as a “task switch” any instance where an individual performs 1 task, then is inactive, wanders inside, or self-grooms, and then performs a task different from the first. Any case where the second task was identical to the first we labeled as a “task stay.” Our rationale was that inactivity, wandering around in the nest, or self-grooming are all frequent behaviors that do not seem to fulfill any pressing colony needs, and thus may be considered “not working,” and may reflect ants waiting or looking for another task to do. However, we also repeated all analyses with the different assumption that self-grooming is a work task similar to other tasks (like foraging, brood care, etc.). The results of the second set of analyses are presented in the Supplementary Material. For each set of “original task-interval-subsequent task,” that is, for all “task switches” and “task stays,” we collected the following associated data: individual, colony, date of observation, time of day, colony size, and the types of original and subsequent task.

To isolate task switches and task stays from our original data, we collapsed sequential instances of wandering, self-grooming, or inactivity into a single string of nontask behavior. We then compared the behavior before (original task) and after (second task) the non-task-bout to label the sequence as a task “switch” or “stay.”

Statistics

Unless specified otherwise, we built linear models, or linear mixed models, with the R package lme4 (Bates et al. 2014) and generated approximate P-values with Satterthwaite approximations in the R package lmerTest for linear mixed models (Kuznetsova et al. 2012). We report R ² values based on calculations described by Nakagawa and Schielzeth (2013) and implemented by Lefcheck and Cassallas (2014). For mixed models, there are 2 possible R ² values: marginal and conditional. Marginal R ² represents the variation explained by fixed factors while conditional R ² represents the variance explained by both fixed and random factors in the model (Nakagawa and Schielzeth 2013). For multiple comparisons, we use a post hoc Tukey test with a Bonferroni correction implemented in the R package multcomp (Hothorn et al. 2008).

There was considerable skew in all of the time variables so we performed log-transformations to normalize the data following (Ives and Freckleton 2015) to satisfy the assumptions of the general linear models. We included individual and colony as random factors in any model where we were not directly estimating the effect of these 2 variables. In all statistical models, we include observation date and the time of day the when video was recorded as random factors. We used the rptR package (Nakagawa and Schielzeth 2010) to calculate repeatability with each video constituting 1 observation. We either estimated repeatability in the classical manner with a linear model, or using a Poisson model as described by Nakagawa and Schielzeth (2010). We estimated the repeatability for the number of switches an individual performs, the number of stays an individual performs, the proportion of switches versus stays, and the length of the interval for both switches and stays. We calculated the frequencies of all behavior transitions and used Pearson’s χ² goodness-of-fit tests to test whether individuals randomly chose a second task after performing the first task. We utilized the package “survival” (Therneau 2015) to employ a failure-time analysis of interval time. Failure-time analysis is a second method of assessing the length of interval times and allowed us to visualize the survival probability of interval times. We employed failure-time analysis to compare interval times across task switches and task repetitions. We used the Cox proportional hazards model in all of the failure-time analyses. All of the analyses were conducted in R version 3.1.2 (R Development Team 2015).

RESULTS

Interval time is longer when individuals switch tasks

In total, we identified 888 task switches and 1554 task stays for 1144 separate ant workers. Across all behavioral sequences, we find that the median time to complete the first task is 25 s, the overall median interval time is 29 s, and the median time to complete the second task is 25 s. As predicted by the “switching cost hypothesis,” we find that switching to a new task is associated with a significantly longer interval (25 s for “stays,” 35 s for “switches,” t ₂₄₃₃ = 5.5, P < 0.001, Figure 1). The fixed factors in this model are whether or not an individual switched, and an individual’s worker group; the random factors in this model are individual, observation date, and colony ID. Intriguingly, we find that the time to complete the first task (t ₂₃₅₄ = −5.2, P < 0.001) and the time to complete the second task (t ₂₃₀₃ = −3.0, P = 0.002) were both significantly shorter if individuals switched task (Figure 1). Overall, the time from starting the first task bout to finishing the second did not significantly differ between switches and stays (t ₂₄₀₉ = −0.9, P = 0.349). Similarly, in the failure-time analysis, we find that switching to a new task significantly reduces the hazard ratio, that is, interval times are significantly longer when an individual switches to a new task (P < 0.001, hazard ratio = −0.22, SE = 0.05, z = −4.66). There is a steady decay in length of interval time for both task switches and task stays (Figure 2). We find that workers from different worker groups (see Methods, clustering after Charbonneau and Dornhaus [2015]) do not differ significantly in their overall interval time (F _3,753 = 2.58, P = 0.06) and find that across all groups task repetitions result in shorter interval times (Figure 3). Although Foragers, Nurses, and Patrollers show significantly reduced interval time when switching tasks (F > 4.5, P < 0.05 for these 3 comparisons), the interval time for Inactives was not significantly different when comparing task repetitions to task switches (F _1,282 = 0.96, P = 0.36).

Figure 1.

Comparison of each of the components of the isolated sequences of task switches and task stays. The vertical axis is the z-score (centered data) of the log-transformed time data. The data were converted to z-scores to allow for comparisons between components. The interval time was significantly shorter if individuals repeated the same task (**P < 0.01, ***P < 0.001).

Figure 2.

Survival probability over time of interval time for task switches (gray line) and task stays (black line). Task stays have a significantly higher hazard ratio (P < 0.001) which results in a faster decay. Dashed lines represent the 95% confidence interval around their respective line.

Figure 3.

Comparison of interval times between different groups of specialized workers. Whether or not an individual switches tasks significantly influences the interval time (P < 0.05), with Foragers, Nurses, and Patrollers having significantly longer interval time after switching tasks. Inactives have longer, but not significantly longer, interval time after switching tasks. Points above 250 not plotted as they represent extreme outliers (>6 SDs from the mean).

Switching frequency between tasks

We quantified how often individual workers switched between particular combinations of tasks (Figure 4). We found that individuals did not randomly choose second tasks; individuals preferentially returned to the same task (P < 0.001 for all tasks, degree of freedom for all χ² tests = 6, all χ² values > 48), thus confirming the presence of specialization in this species.

Figure 4.

Relative frequency of changing to a different or the same task within each task. Points are scaled continuously by size, with empty cells indicating no switch from the original task to the second task was observed. Transitions between tasks is not random as individuals tend to repeat the same task rather than switching to other tasks (P < 0.001 for all goodness-of-fit tests).

Switching times do not vary among tasks

We investigated whether the time interval when switching tasks was influenced by the task that the individual switched from or the task the individual switched to. Neither the task an individual switched from (F _6,857 = 1.33, P = 0.24) nor the task an individual switched to (F _6,843 = 1.41, P = 0.22) predicted switching time (Table 1, Figure 5).

Table 1.

Model comparisons of models predicting differences in task switching times

Model	Fixed factors	Random factors	Marginal R 2	Conditional R 2	AIC
Full, no interaction	SwitchFrom + SwitchTo	Colony ID, Date, Individual	0.03	0.22	2518
Only SwitchTo	SwitchTo	Colony ID, Date, Individual	0.01	0.22	2516
Only SwitchFrom	SwitchFrom	Colony ID, Date, Individual	0.01	0.23	2522
Only random effects	None	Colony ID, Date, Individual	0	0.23	2518

The random factors did not vary between models, and often explained a considerable portion of the variation as estimated by the conditional R ². Neither factors, the task an individual switched from and the task an individual switched to, explained a significant amount of variation according to marginal R ² values. As there were multiple combinations of switching from one behavior to another, an interaction term is not considered. The AIC values prefer the model without the interaction, suggesting that the extra variable is not justified given the small increase in fit. AIC = Akaike Information Criterion.

Figure 5.

The average interval time when individuals switch from a task (a) or to a task (b). The interval time is plotted on an untransformed scale though significance tests were performed on log-transformed data. Points above 250 not plotted as they represent extreme outliers (>6.5 SDs above the mean).

Interval time when repeating same task

We also investigated the interval time when individuals performed the same task twice in a row, and compared the interval time among task types. We found that the type of task an individual performs twice in a row significantly influenced interval time (F _6,1067 = 9.8, P < 0.001, Figure 6), Specifically, we found that when ants repeated brood care, building, trophallaxis, or receiving allogrooming, they had longer interval times (Figure 6), and ants that repeated foraging or allogrooming others had shorter interval times. Although feeding has a low median interval time, the high variance in the group results in it not being significantly different from other tasks. Although interval times differ depending on the type of task that is repeated, task explained a small amount of the variance in interval time (Marginal R ² = 4%).

Figure 6.

The average interval time when individuals perform the same task twice in sequence. The interval time for foraging and allogrooming others are significantly shorter than the interval time for building, brood care, receiving allogrooming, and trophollaxis. The interval time is plotted on an untransformed scale but significant differences represent comparisons among log-transformed data. Points above 200 not plotted as they represent extreme outliers (>5 SDs above the mean). a and b represent significantly different groups.

Task switches not affected by interactions before or during interval time

To assess if new information from nest mates was influencing task and interval dynamics, we randomly selected 75 behavioral sequences and determined if individuals switched tasks in response to antennation from nest mates during the last 10 s of the first task in the sequence. We defined antennation as any contact of the antennae from the non-focal individual towards the focal individual. Of the 75 sequences, 29 were task stays and 46 were task switches. In 20.7% (6/29) of the task stays, an interaction was observed in the 10 s before terminating the first task, and, in 21.7% (10/49) of the task switches, we observed an interaction in those 10 s. We built a logistic model to determine if interactions influence whether an individual switches to a new task. We found that switching to a new task is not associated with an interaction with a nest mate when performing the first task (z ₃₆ = 0.0, P = 1.0). Thus, switches were not triggered by antennation from nest mates during task performance. The length of the interval between task bouts affected by such behavioral interactions was not different between task switches and task stays (t = 0.47, P = 0.64). Given the global effect of task switching leading to increased interval times, interactions may therefore change interval time; however, we may not have had an adequate sample size in the targeted analysis to recover the effect of task switching on interval time. We also randomly selected another 37 behavioral sequences to test if interactions during the interval time affected switching. We found that antennations during the interval time did not influence whether an individual switched to a new task (13.3%) versus repeated the same task (37.5%, z ₃₆ = 1.27, P = 0.21).

Colony differences

In general, we found that colonies slightly differed in the structure of task sequences. The number of switches per individual per 5 min of observation did not vary significantly between colonies (F _43,127 = 1.1, P = 0.40), though the number of stays per individual per 5 min of observation differed between colonies (F _43,127 = 1.4, P = 0.01). Similarly, the total number of switches and stays per 5 min of observation significantly differed between colonies (F _43,127 = 1.80, P = 0.01). However, the proportion of switches out of the total of switches and stays did not differ among colonies (F _43,127 = 1.2, P = 0.22). We also find significant among-colony differences in the length of the interval time (F _43,127 = 1.48, P = 0.05, = 0.05). Colony size did not predict the length of the interval time for switches, stays, or when all intervals are combined (P > 0.4 for all tests). Colony size did not explain the number of switches, the number of stays, the proportion of switches, or the total number of switches and stays (P > 0.1 for all tests).

Individual repeatability

We calculated individual repeatability for several variables to test for individual ant worker consistency in switching tasks and in repeating tasks. We find significant repeatability for the proportion of switches and the interval length for both repeating the same task and switching to a new task. The maximum repeatability we found was 0.22 while the minimum was 0 (Table 2).

Table 2.

Repeatability of multiple behaviors for individual ants

Behavior	Repeatability	95% confidence interval	Method
Number of switches	0.03	0–0.08	Poisson GLM
Number of stays	0.0	0–0.01	Poisson GLM
Proportion of switches	0.1	0.06–0.19	LM with MCMC
Length of interbout interval for stay	0.18	0.08–0.28	LM with MCMC
Length of interbout interval for switch	0.22	0.11–0.36	LM with MCMC

Listed in the table is the estimate of repeatability (on the original scale where applicable), r, the 95% confidence interval around the estimate, and the method for estimating repeatability. GLM = Generalized Linear Model; LM = Linear Model; MCMC = Markov chain Monte Carlo.

DISCUSSION

Our study detected increased interval time for task switches compared to “stays” (returning to the same activity after a break, Figure 1). Therefore, our results provide support for the hypothesis that task switching is associated with a time cost in T. rugatulus. This, in turn, provides evidence that temporal switching costs could select for division of labor (Smith 1776; Goldsby et al. 2012). Because, in this genus, individuals also do not appear to achieve efficiency gains in the long term through specialization (Dornhaus 2008), the evolution of worker specialization may be due to task switching costs. The time costs of switching tasks may be non-trivial in these colonies. On average there were 5.35 switches observed per 5-min recording, which would imply ~64 switches per hour per colony. Given the average difference in interval between switching and repeating behavior is ~10 s, this results in over 10 ant-minutes lost through increased interval time per hour as compared to a scenario where individuals do not switch tasks at all. Although inactive individuals are likely beneficial for the colony (Charbonneau and Dornhaus 2015; Hasegawa et al. 2016), we do not suspect the extra interval time is adaptive due to the difference in length between extended interval times and typical inactive periods. Specifically, the median interval time is 29 s whereas the median bout of inactivity is 150 s. This suggests that individuals performing tasks are not entering typical periods of inactivity in the interval.

Switching to a new task appears to result in longer interval time across all worker groups (Figure 3), although this is only significant for groups of active individuals (Foragers, Nurses, and Patrollers). One reason why Inactives may not have significantly lower interval times during task repetition is because they may not be looking to start a new task as soon as possible, while the other groups of workers are possibly trying to minimize the delay before starting the next bout of work. One piece of evidence that supports the idea that Inactives may not be readily looking for tasks is that they perform the lowest number of tasks after controlling for group size (1.79 tasks per worker for Inactives vs. 2.49 tasks per worker for Nurses). In sum, the presence of task switching costs across active worker groups suggests that this is a general phenomenon regardless of the behavior being performed.

However, our results are not robust to all definitions of interval time. Specifically, we conceptualized the “interval” between task bouts as the time needed to find, decide, or get started on the next task, and we assumed that ants may not only be standing or wandering around during this time, but may also be self-grooming while unsure what to do next. If self-grooming is not included as an aspect of interval time we do not detect a cost to task switches, and the first and second task bouts are significantly shorter during task repetitions compared to switches (Supplementary Figure 1). Our results show that there are many brief bouts of self-grooming that are often repeated multiple times before performing another behavior (Supplementary Figure 2). We, therefore, contend that much of self-grooming is due to individuals not finding tasks to perform. Indeed, studies that examined the dynamics of self-grooming suggest that self-grooming behavior is often performed when demand for other behaviors is low. For instance, when honeybees (A. mellifera) are presented with a colony stressor, self-grooming behavior is significantly reduced (Johnson 2002). In mice (Mus musculus), experimental selection for increased activity found correlated reductions in inactivity and self-grooming in the line selected for increased activity (Waters et al. 2013). These studies, and the short duration (median = 9 s, medians of all other tasks in this study = 14–61 s) of typical self-grooming bouts, are indicative of the behavior being a “filler” during times of inactivity rather than a task deliberately switched to in response to need for the task. Additionally, if self-grooming is not included in the interval time then individuals most often switch to self-grooming as a second task, regardless of starting task (Supplementary Figure 4). In this scenario, individuals would specialize on switching to self-grooming, and the majority of the behavioral sequences would end with self-grooming. In contrast, when self-grooming is included in interval time we recover signatures of specialization on tasks such as building and foraging (Figure 4). Self-grooming behavior is therefore appropriately included in the “interval.”

Although we find a temporal cost to switching tasks, we do not definitively measure whether a task was completed before the interval. One reason is that measuring completed tasks often necessitates focusing on a subset of behaviors such as nest building, foraging for food, or undertaking where task completion is unambiguous (Julian and Cahan 1999; Dornhaus 2008). While we do not estimate task completion, we still uncover a temporal cost of switching to a new task (whether or not the original task was completed) and also find that individuals that switch tasks spend less time overall performing any type of work across the entire sequence of first task, interval, and second task (Figure 1).

Given that we find evidence of task-switching costs, we investigated what may prompt individuals to switch tasks. One reason to switch tasks could be receiving information from nest mates signaling need for work in another area, as has been suggested for harvester ants (Pogonomyrmex barbatus) and other social insects (Gordon 1989; Cook and Breed 2013). We do not find that interactions (e.g., antennation) among nest mates more often precede task switches relative to task stays (21.7% vs. 20.7% of behavioral sequences); nor do we find that interactions during the interval time influence whether or not an individual switches to a new task versus repeats the same task (13.3% vs. 37.5%). We therefore do not suspect that individuals are switching tasks because of incoming tactile information from other nest mates. Individuals could therefore be switching tasks for several other reasons. Although we measured antennation, we did not measure chemical composition within the nest and thus chemical signals may be a factor when switching tasks. Individuals may switch tasks because all of the work in the first task is now completed. Alternatively, signals or cues in other modalities could cause individuals to switch to a task that has higher demand (Sasaki et al. 2014; Vander Meer et al. 1998). Individuals may even interrupt a current task to assess the need of the colony, as seen in honeybees (Johnson 2008, 2009); or they may be spontaneously ending tasks, as seen in T. albipennis individuals when building (Franks and Deneubourg 1997), to wander inside the nest to assess demand for other tasks.

In general, we find little evidence that the task an individual switched to or switched from influences the interval time (Figure 5). We do find statistical differences in interval time by task when individuals repeat a task (Figure 6). Both allogrooming others and foraging have shorter intervals when repeating these behaviors. With respect to foraging, individuals are likely transferring food to the colony and immediately departing for another foraging foray. For all of the behaviors where we estimate repeatability, we find that the repeatability is similar (Table 2) to other published estimates of repeatability in the literature (~0.24 for other tests of repeatability in the lab) (Bell et al. 2009). Our highest estimates of repeatability are those for the length of interval time for task switches and task stays, suggesting that workers may differ in how or how effectively they find a new task; the same may also explain the individual differences among individuals in probability of switching versus staying with the same task. At the level of the colony, we find that colonies differ significantly in the proportion of switches and may thus differ in in task allocation strategy. The presence of between-colony differences suggests that task switches may be a component of colony-level personalities described in T. rugatulus (Bengston and Dornhaus 2014; Jandt et al. 2014).

What are ants doing during the “interval time,” and why is this longer when switching tasks? There are several, nonmutually exclusive possibilities that lead to increased interval time. First, they may be looking for another task to perform, that is, assessing demand for work in other tasks (Johnson 2008). If “wandering in the nest” serves the purpose of finding a new task to perform, it may require time to assess the colony needs for multiple tasks. However, there is evidence that some individuals in T. rugatulus colonies specialize in wandering (Charbonneau and Dornhaus 2015); thus, it is not clear that this behavior is always motivated by the need to find a next task. Indeed, previous research has combined wandering with inactivity (Lachaud and Fresneau 1987). If the interval time is driven by the search for a new task, it might be predicted to be longer in species with larger nests, as individuals will need to traverse larger distances to assess colony needs (Jeanson and Lachaud 2015). Colonies with nests that are more spatially expansive may therefore suffer larger task-switching costs. In our study, nest enclosures were always the same size, and we did not find any effect of colony size (worker number) on interval time. Second, longer interval time may be caused by the need to move to a different work area, as tasks are generally spatially segregated in ant colonies (as shown specifically in this genus, Sendova-Franks and Franks 1995). One might predict that if switching to a new task is delayed by the travel time to a new work site, this would cause higher switching costs in more crowded colonies; we did not find this to be the case, though this needs experimental verification in future studies. Third, a cost associated with task switching may be working memory constraints. For instance, if the memory of a different motor pattern has to be retrieved when switching between types of tasks, and 2 such patterns cannot be active in working memory, this could cause a delay before starting the new task, as seen in bumblebees (Chittka et al. 1997). Careful experimentation will be needed to conclusively resolve which of these 3 hypotheses, if any, are driving increased interval time.

There are several benefits associated with division of labor; in larger societies with morphological specialization among workers, the differences in worker morphology allow certain workers to better perform certain tasks (Bourke 1999). In societies without polymorphic workers, other benefits likely select for division of labor. For instance, Smith (1776) suggested that specialization could be beneficial if individuals avoided delays when switching between tasks. Indeed, Jeanne (1986) found that workers in P. occidentalis were able to complete more work as a colony when individuals specialized on certain tasks. In this study, we find another benefit of specialization. Specifically, we find that when self-grooming is included in interval time that switching tasks increases the interval time between tasks in T. rugatulus. As task allocation in T. rugatulus is similar between the lab and the field (Charbonneau et al. 2015), we contend that these results are relevant to field populations in this species and possibly to other eusocial systems. Avoiding temporal task-switching costs may be a meaningful selective force maintaining specialization in both small societies with monomorphic work forces as well as in larger colonies. Task switching costs may therefore be a general force that maintains specialization in eusocial societies.

SUPPLEMENTARY MATERIAL

Supplementary material can be found at http://www.beheco.oxfordjournals.org/.

FUNDING

This work was supported by the National Institutes of Health (grant number: 2K12GM000708-16) for fellowship funding provided to the Center for Insect Science provided to G.M.L. This work was also supported by the National Science Foundation for funding provided to A.D. (grant numbers: DEB-1262292 and IOS-1455983).

Data accessibility: Analyses reported in this article can be reproduced using the data provided by Leighton et al (2016). Included with the data is the code to isolate instances of task switching and task repetition.