Behavioral Ecology and Sociobiology

, Volume 69, Issue 9, pp 1459–1472

Workers ‘specialized’ on inactivity: Behavioral consistency of inactive workers and their role in task allocation


    • Graduate Interdisciplinary Program in Entomology and Insect ScienceUniversity of Arizona
  • Anna Dornhaus
    • Department of Ecology and Evolutionary BiologyUniversity of Arizona
Original Paper

DOI: 10.1007/s00265-015-1958-1

Cite this article as:
Charbonneau, D. & Dornhaus, A. Behav Ecol Sociobiol (2015) 69: 1459. doi:10.1007/s00265-015-1958-1


Social insect colonies are often considered to be highly efficient collective systems, with division of labor at the root of their ecological success. However, in many species, a large proportion of a colony’s workers appear to spend their time completely inactive. The role of this inactivity for colony function remains unclear. Here, we investigate how inactivity is distributed among workers and over time in the ant Temnothorax rugatulus. We show that the level of inactivity is consistent for individual workers, but differs significantly among workers, that is, some workers effectively specialize on ‘inactivity’. We also show that workers have circadian rhythms, although intra-nest tasks tend to be performed uniformly across the whole day. Differences in circadian rhythms, or workers taking turns resting (i.e., working in shifts), cannot explain the observation that some workers are consistently inactive. Using extensive individual-level data to describe the overall structure of division of labor, we show that ‘inactive workers’ form a group distinct from other task groups. Hierarchical clustering suggests that inactivity is the primary variable in differentiating both workers and tasks. Our results underline the importance of inactivity as a behavioral state and the need for further studies on its evolution.


Task allocationSpecializationInactivityColony organizationShift workCircadian rhythmSocial insectTemnothoraxDivision of labor


Social insects are among the most abundant, diverse, and widespread taxonomic groups (Wilson 1991; Samways 1993). Their ecological success is often attributed to division of labor (DOL) and worker specialization (Oster and Wilson 1978; Hölldobler and Wilson 1990). Social insects have evolved successful strategies for allocating tasks to workers while solving a set of specific problems and constraints (Charbonneau and Dornhaus In revision). Some of these strategies have inspired solutions to analogous problems in diverse fields such as computer sciences (Johnson 2012), robotics (Gerkey and Matarić 2004), logistics (Zhang and Chen 2011), sociology (Durkheim 1997), and economics (Becker and Murphy 1992).

A key feature of DOL is worker specialization, where subsets of workers tend to do one or a few tasks more than other workers. Most studies of specialization focus on a few prominent tasks such as foraging, building, and brood care (Beshers and Fewell 2001; Dornhaus 2008; Duarte et al. 2011), but rarely on more innocuous tasks such as grooming (though these can also be specialized, Moore et al. 1995). Much work has been done to show that workers within colonies can often be separated into discrete task groups, sometimes called behavioral castes (Wilson 1976; Mirenda and Vinson 1981; Herbers 1983; Lenoir and Ataya 1983; Corbara et al. 1989; Retana and Cerdá 1991; Pinter-Wollman et al. 2012; Mersch et al. 2013). There is also a wealth of empirical, theoretical, and conceptual work on potential task allocation strategies (e.g., reviewed in: Charbonneau and Dornhaus In revision; Gordon 1996; Robinson and Huang 1998; Beshers and Fewell 2001; Duarte et al. 2011). The different proposed task allocation mechanisms are not mutually exclusive, and different tasks may be governed by different mechanisms (Gordon 2002). However, we generally still know very little about the adaptive benefits/costs of each strategy (Charbonneau and Dornhaus In revision).

Despite high levels of worker inactivity being prevalent in social insects, inactivity is one of the least well understood behaviors that workers engage in. The literature is filled with reports of >50 % of workers in colonies being inactive across all social insect taxa (bees, Lindauer 1952; Jandt et al. 2009; wasps, Gadagkar and Joshi 1984; ants, Mirenda and Vinson 1981; Herbers 1983; Cole 1986; Retana and Cerdá 1990; Schmid-Hempel 1990; Dornhaus 2008; Dornhaus et al. 2009; and termites, Rosengaus and Traniello 1991), and yet, the role of inactivity (rest and quiescence) is rarely considered in understanding task allocation strategies or colony organization (for exceptions, see Herbers 1981; Fresneau 1984; Cole 1986; Corbara et al. 1989; Retana and Cerdá 1990; Retana and Cerdá 1991). Perhaps this is because inactivity is essentially a lack of doing anything else and so seems unimportant. However, because so many workers spend so much time doing what appears to be nothing at all, omitting inactivity in task allocation studies may skew our understanding of social insect DOL. For example, if inactivity does not serve a purpose per se, and is common for all workers, then activity in one task may not trade off with activity in another task simply because much time is spent being inactive anyway. On the other hand, if only a few workers are very inactive, then they may fulfill some unrecognized critical function. Finding adaptive explanations for inactivity is particularly relevant because high levels of inactivity are likely not an artifact of simplified living conditions in the lab (Charbonneau et al. 2015).

Potential adaptive functions of high levels of inactivity in social insects have been proposed but rarely tested (reviewed in Charbonneau and Dornhaus In revision). The most commonly cited explanation is that inactive workers are ‘reserves’ that become active when workload increases, and yet, the literature on this topic is surprisingly inconsistent. In fact, several studies that explicitly set out to test the reserve worker hypothesis fail to support it (Fewell and Winston 1992; Johnson 2002; Jandt et al. 2012). Moreover, there is evidence that when workload increases, workers other than inactive workers increase their activity (Mirenda and Vinson 1981; Johnson 2002), or that workers are incapable of reallocating workers to necessary tasks, even at the expense of losing half of their brood (Kwapich and Tschinkel 2013).

Additional ecological functions that have been proposed include a worker/colony conflict involving inactive workers acting selfishly by conserving their energy and minimizing their exposure to risk so that they may lay their own eggs (tested and supported, Hillis et al. In prep.; Jandt and Dornhaus 2011; tested, but not supported, Cole 1981; Cole 1986; Ishii and Hasgeawa 2013), and inactive workers performing an as-yet unidentified function in which they appear idle but providing a function (behaviorally idle rather than functionally idle) such as playing a role in communication (proposed explanation, not tested, O’Donnell and Bulova 2007) or acting as food reserves (evidence of food reserves, but link to inactivity not tested, Sendova-Franks et al. 2010). Alternatively, inactivity may be a constraint related to age where young workers may less active due to inexperience/physical vulnerability (tested and supported, Corbara et al. 1989; Klein et al. 2008) and/or old workers due to senescence (tested and supported immature, Corbara et al. 1989; Klein et al. 2008; immature and senescent, Fresneau 1984; evidence against, Johnson 2008; Retana and Cerdá 1990). Ultimately, the question of why colonies would produce so many inactive workers, in spite of potentially high production and maintenance costs, is still very much a mystery.

Perhaps one of the simplest explanations for inactivity is that, because of physiological need for rest (or sleep) even in insects (Klein et al. 2003; Klein et al. 2010), all workers may be spending a certain amount of time ‘inactive’. In such a case, we would expect all workers to have more or less similar needs and consequently for all workers to have consistent and comparable levels of inactivity. We know that workers can vary in activity levels over the course of the day (circadian rhythms; e.g., Klein and Seeley 2011) and between seasons (Fellers 1989). Complex activity patterns can arise when these interact (Pol and de Casenave 2004). In the case of honey bees and bumble bees, foragers have been shown to have diurnal rhythms, while in-nest workers, such as nurses, do not (Moore 2001; Yerushalmi et al. 2006).

There is also evidence that honey bee foragers employ ‘shift work’ during the day, where workers show fidelity to foraging either in the morning or in the afternoon (Kraus et al. 2011). This may be linked to sleep timing and resource availability (Klein and Seeley 2011). However, in-hive tasks, such as nursing and even food-storing behaviors associated with foraging, are performed at all times of the day by the same group of individuals, suggesting an absence of circadian rhythm for these internal workers (Moore et al. 1998). This pattern was consistent across all workers, suggesting an absence of ‘shift work’ among in-hive workers (Moore et al. 1998). In comparison to honey bees, ant circadian rhythms have rarely been studied (although see North 1987; North 1993; Sharma et al. 2004; Lone and Sharma 2011; Lone and Sharma 2011).

Shift work (individually different circadian rhythms in activity) could be an efficient way to allocate work in cases where tasks require constant attention (e.g., brood requiring constant attention from nurses) or to increase the amount of work that can be accomplished in a limited workspace (e.g., nurses caring for brood may need to move around the brood pile and the presence of additional workers would only serve to slow them down). Furthermore, if workers show consistency in their circadian rhythm with regard to when they are active and when they are resting, and individuals vary in the timing of activity phases (thus effectively working in staggered shifts), then monitoring worker behavior only at certain times (e.g., only during the day) will generate the appearance of consistently more or less active individuals. This would be the case even if, over a 24 h period, all workers essentially spend the same amount of time active. Because of this, it is critical to monitor workers on different time scales, including at different times of day, to establish whether individual differences in activity level exist.

Here, we investigate the structure of DOL, the degree of individual task specialization, and worker variation in the ant Temnothorax rugatulus. We pay special attention to the behavioral state/task of inactivity. Specifically, we quantify DOL and specialization, and describe how tasks relate to each other by determining how the likelihood of performing one task affects the likelihood of performing other tasks, including being inactive. We investigate temporal variation in worker activity levels, testing for worker circadian rhythms and whether individual consistency in activity level can be explained by workers taking turns resting or effectively doing shift work.


Colony collection and housing

We collected five colonies of T. rugatulus ants in the Santa Catalina Mountains near Tucson, Arizona, USA in pine forest at an altitude of approximately 2500 m (see Table S 1 for detailed information about collected colonies). Colonies were housed in artificial nests that emulate the small rock crevices they inhabit in the field (Charbonneau et al. 2015). The artificial nests consist of a 2-mm-thick piece of cardboard sandwiched between two glass slides (76.2 × 50.8 mm). The cardboard (38.1 × 50.8 mm, or half of the size of the glass slides) serves as a spacer between the glass slides and as a back wall. Additional walls are constructed by the ants from grains of green ceramic-coated sand (Fig. S 1a). In the field, colonies create similar walls from surrounding sand and dirt. Artificial nests are kept in open-top plastic containers (11.1 × 11.1 × 3.3 cm) whose walls were lined with ‘insect-a-slip’ (BioQuip product #2871A).

Colonies were given water and fed ad libitum (water-filled plastic test tubes, stoppered with cotton balls, semiweekly, and 2 mL Eppendorf tube of honey water and 10 frozen adult Drosophila flies weekly), kept on a 12 h light regimen (lights on at 8 a.m. and off at 8 p.m.), and at constant temperatures (approximately 21–24 °C) and humidity (approximately 20–25 % relative humidity).

Behavioral data collection

Workers were individually painted 3–7 days before filming with unique combinations of four paint spots, one on the head, one on the thorax, and two on the abdomen, so that they could be individually identified and tracked (Fig. S 1b). Videos (5-min long) of normal colony activity were taken with an HD camera (Nikon D7000 with 60 mm lens) at six time points throughout the day: 8 a.m., 12 p.m., 4 p.m., 8 p.m., 12 a.m., and 4 a.m., repeated on 3 days spread over a 3-week period for each colony (18 × 5 min videos per colony). During the night videos, minimal white CFL lighting was used (switched on 5–10 min before the start of recording and switched off 15–20 min after) because red light or infrared would have made reading the color coded ants impossible, thus preventing measurement of individual level activity. All videos were taken within 3 months of their collection to limit potential laboratory effects, such as artificial age structures due to increased forager age.

Video analysis

For each ant, the task it performed was recorded at every second by an observer analyzing the videos. A complete list of tasks and definitions can be found in Table 1. The tasks were broadly classified as being either ‘active’ (e.g., brood care), ‘undifferentiated’ (walking inside the nest with no clear task), or ‘inactive’ (completely immobile), comparable to the broad classification used by Cole (1986). If less than 10 s separated two events of brood care, feeding, foraging, or 20 s for building, the task was considered to be uninterrupted. Videos were analyzed by seven different observers. Data from each video were spot checked by a single person to ensure uniformity of behavioral observations.
Table 1

List of possible behaviors observed during video analysis, their broad class of activity, codes, and detailed descriptions. For every second of analyzed video, each ant has one of these behaviors attributed to it. (Similar to Charbonneau et al. 2015)





Nest building

Manipulating a stone in any way (moving, pushing, and pulling)


Located in feeding area or on water tube or wandering outside of the nest and not engaged in building. Also if returning to the colony from foraging areas and performing trophallaxis or returning with drosophila.

Brood care

Manipulating brood (feeding, grooming, and moving)


Grooming itself

Grooming other (giver)

Grooming another ant

Grooming other (receiver)

Be groomed by another ant


Receive or give liquid food to/from another adult ant


Feeding on drosophila inside nest (brought back by foragers)


Wandering inside nest

Anytime an ant is mobile inside the nest wall and not engaged in any ‘active’ task



Immobile and not engaged in any ‘active’ task

Statistical analyses

Statistical analyses were performed in R (Version 3.0.3) and consisted of mixed-effects models and Tukey’s post hoc tests (packages ‘nlme’ v3.1–115 and ‘multcomp’ v1.3–2), principal component analyses (PCA) (base ‘stats’ package and ‘prcomp’ function), hierarchical cluster analysis (base ‘stats’ package and ‘hclust function), Pearson’s correlations (base ‘stats’ package and ‘cor’ function), and repeatability measures (modified package ‘rptr’ v0.6.405). DOL was quantified using the index established in Gorelick et al. (2004); for updated definitions see also Gorelick and Bertram 2007; Dornhaus et al. 2009).

For the PCA, data from all five colonies were aggregated to increase sample size (PCAs on individual colonies show similar patterns to the pooled PCA–Fig. S 2). The aggregated data were centered and scaled by subtracting the mean times spent on a task and dividing by the standard deviation. Only daytime observations were used to avoid variation due to circadian rhythms, and workers with less than three observations (40 workers out of a total of 265) were excluded from the PCA in order to minimize random variation. This data set was also used in a hierarchical cluster analysis. Rotations were not necessary as only eight variables were used, and the spread between variable vectors was good. The first three components of the PCA were retained (Kaiser–Guttman stopping rule, Guttman 1954). These explain 62.2 % of inter-worker variation in time spent on tasks.

For balance in analyses of temporal variation in individual-level inactivity, we retained only workers identified at least once at each time period (day and night—across colonies, a mean of 68 % retained, SD 15 %). In all analyses, we used only worker data because queens are thought to have smaller task repertoires (Herbers 1983) and their sample size does not allow independent analyses.

Repeatability was calculated according to the methods described in Nakagawa and Schielzeth (2010) where fixed effects are used for systematic effects across individuals (e.g., time of day), while random effect is used for non-systematic variation (e.g., if the observer identity was not systematically related to repeated measurements taken) (see Nakagawa and Schielzeth 2010). Thus, our model includes individual worker IDs as the random effect (and grouping variable), while colony, which is expected to have a systematic effect on time budgets because colonies differ in their overall level of activity, is a fixed effect variable.


Overall structure of division of labor

Rank correlations of mean worker time spent on tasks (ranks calculated within colonies) showed that across ants, time spent inactive was negatively correlated to time spent on all tasks, except trophallaxis and brood care which are not correlated to inactivity (Fig. 1). Although this may not be surprising because inactivity is the lack of doing a task, a possible positive correlation might have emerged if highly inactive workers were at the same time highly specialized on another task, such as brood care; e.g., it is possible for a narrowly specialized ant to be doing nothing most of the time but if all its remaining time is spent on brood care, that ant could still be doing most of the brood care compared to its nestmates, particularly because it is the rank, not the absolute fraction of time spent on a task that is analyzed. This however was not the case, and highly inactive workers tended not to spend much time on any other specific tasks compared to their nestmates.
Fig. 1

Pearson’s ρ for correlations between worker ranks of time allocated to tasks. Bars in dark gray represent contrasts that were significant (p < 0.05 after Holms correction) and white bars are non-significant. Inactivity is negatively correlated with all other tasks, except for trophallaxis and brood care for which the correlation is non-significant

Brood care was negatively correlated to foraging and positively correlated to feeding on Drosophila flies brought back to the nest by foragers (i.e., ‘eating’). Foraging, building, grooming, trophallaxis, and wandering inside are all generally positively correlated. Building and foraging were positively correlated to eating, which is surprising since brood care is also positively correlated to eating, but foraging and brood care are negatively correlated.

A PCA performed on the amount of time workers spent on each observed task shows that workers vary widely along an inactivity-wandering inside axis that closely tracks PC1 (Fig. 2, left). Because inactivity is nearly orthogonal to both PC2 and PC3, the relationships between task vectors on these axes (PC2 and PC3) highlight relationships between tasks independently of inactivity. There appear to be three distinct groups of tasks branching out in different directions: (1) foraging/building/trophallaxis, (2) brood care/eating, and (3) grooming/wandering inside (Fig. 2, right). Vector loadings and eigenvalues for the principal components are shown in Table 2 (loadings and eigenvalues for PC1-8 can be found in Table S 2).
Fig. 2

Principal component analysis on mean worker time allocated to tasks shows that (left) workers strongly vary along an inactivity/wandering inside axis, approximately parallel to PC1. Also, as inactivity is nearly parallel to PC1 and orthogonal to PCs 2 and 3, the figure on the right (PC2 vs PC3) shows the approximate relationships between task time allocations, independently of inactivity. Specifically, there appear to be three groups of tasks (tasks parallel to each other and apart from other groups): (1) foraging/building/trophallaxis, (2) brood care/eating, and (3) grooming/wandering inside. Hierarchical clustering analysis shows four separate worker groups: + nurses (n = 24), *extra-nest workers (n = 27), black circle patrollers/groomers (n = 56), black triangle inactives (n = 118). Data were mean-centered and scaled (subtracted the mean activity in a task and divided by the standard deviation) before analysis. Analysis includes only daytime data for workers having a minimum of 3 observations

Table 2

Eigenvector loadings, eigenvalues, and proportion and cumulative proportion of variance for the principal components retained (PCs 1–3)













Brood care








Wandering inside
























Prop. of Variance




Cumul. Prop. Var.




A subsequent hierarchical cluster analysis (Euclidian distance and Ward’s linkage method) classified workers (from all colonies, data scaled, and centered) into four separate groups based on the proportion of time spent on each task (cluster number based on the four task vector groups from the PCA): nurses (34 workers), extra-nest workers (26 workers), generalists (62 workers), and inactive workers (103 workers; Fig. 3). In the analysis, workers were first separated into active and inactive workers (first branch of left dendrogram) and active workers are subsequently divided into generalists, nurses, and extra-nest workers. Tasks were also placed into a hierarchical dendrogram (Euclidian distance and Ward’s linkage method), but not explicitly clustered (top dendrogram). This shows a similar pattern of task relationships as the PCA where inactivity and other tasks are first separated, then nurse tasks (brood care and eating), foraging tasks (foraging and building), and generalist tasks (trophallaxis, wandering inside, and grooming). Interestingly, here trophallaxis was grouped with inactive workers rather than with foraging tasks as in the PCA.
Fig. 3

Hierarchical cluster analysis (Euclidian distance, Ward’s linkage method) of workers according to time spent on tasks (left dendrogram) shows that workers are first separated into active and inactive workers (first branch) and active workers are subsequently divided into patrollers/groomers, nurses and extra-nest workers. Clustering of tasks shows a similar pattern (top dendrogram) where inactivity and other tasks are first separated, then nurse tasks (brood care and eating), extra-nest tasks (foraging and building), and other tasks (trophallaxis, wandering inside and grooming). Z-scores indicate whether individual workers (represented by thin colored lines in the central graph) spend more time (positive z-score —red), less time (negative z-score—blue) or equal time (zero z-scorewhite) than the mean amount of time spent on that tasks for all workers

We found weak worker specialization (workers focusing on few tasks: median DOLind for all colonies 0.208), but high segregation in which workers perform which tasks (median DOLtask 0.838). Overall, we found a median DOLtotal 0.425, a measure that incorporates specialization and task segregation, for T. rugatulus. Because these indices of DOL used are normalized for number of tasks and number of individuals, they can be directly compared with other systems (Gorelick et al. 2004). Our DOLtotal values were higher than those reported for Temnothorax albipennis (0.38, Dornhaus et al. 2009), solitary and communal halictine bees (~0.08–0.21, Jeanson et al. 2007), Bombus impatiens (0.09–0.12, Jandt et al. 2009), and the ant Camponotus festinatus (0.15–0.25; Dornhaus A, Duffy K, unpublished data).

Inter-worker variation in worker time budgets

Linear-mixed effects models show significant variation between workers for all tasks, including inactivity, but not for trophallaxis (LMM p < 0.0001 for all tasks, except p = 0.038 for grooming, and p = 0.64 for trophallaxis; Random Effect: Colony). This indicates that individual workers differ in their propensity to engage in each task with the exception of trophallaxis (which is the way ants receive and give liquid food).

Workers differed significantly in their level of inactivity, and these differences were consistent over the 2-week period of observation (LMM F = 2.14 p < 0.0001, Table 3; see Fig. 4 for distributions of worker inactivity for each colony). Overall, 25.1 % of workers (of a total of 265 recorded) were never observed being anything but inactive over the 18 sample observations of 5 min each, while 2.6 % of workers were observed being constantly active over this period (i.e., engaged in ‘active’ tasks according to Table 1). Over the observation period, 71.9 % of workers were inactive at least 50 % of the time.
Table 3

Individual workers (Ind_Ant) consistently differ in the proportion of time they are inactive, and overall inactivity is lower during the day compared to night (timeperiod) the lack of interaction between these factors suggests the absence of ‘shift work’. Results from a linear mixed-effects model*





p value





















*Fixed-effects: Ind_Ant, timeperiod, and Ind_Ant x timeperiod.

Random effects: colony/date.
Fig. 4

Frequency distribution of individual inactivity for each colony measured across 3 days and 6 time points each day. The number of workers and the mean observation time (seconds/ant) is shown for each colony in the upper left corners

Typically specialized tasks (foraging, building, and brood care) are shown to be highly repeatable (i.e., repeatability test (Nakagawa and Schielzeth 2010) shows that intra-worker variation is much lower than inter-worker variation; Fig. 5), indicating that the amount of time workers spend on these tasks is highly consistent. Tasks such as eating, grooming, and wandering inside have low repeatability, suggesting that most workers perform them to some extent, with little consistent differences among workers. Most interestingly, inactivity is also highly repeatable, on par with tasks typically performed by specialized workers such as foraging, building, and brood care.
Fig. 5

Foraging, building, brood care, and inactivity have comparably high levels of repeatability. All task repeatability estimates are significant. Colony was added as a fixed effect to the calculation of adjusted repeatability to control for inter-colony variation. All tasks were significantly repeatable (p value <0.05)

Does individual inactivity depend on the time of day? Circadian rhythm

Workers spent significantly less time inactive during ‘daytime’ (i.e., observations with lights on; 8 a.m., 12 p.m., and 4 p.m.) than during ‘nighttime’ periods (i.e., with the lights off; 8 p.m., 12 a.m., and 4 a.m.) (LMM F = 7.24 p < 0.01; random effects: colony/date, Table 3). Extra-nest activity (‘foraging’ and ‘building’) and ‘wandering inside’ were each significantly higher during ‘daytime’ observations (p < 0.01, mean of 5.2 % of time during day vs. 3.6 % at night; p < 0.01, mean of 16.5 % of time during day vs. 14.0 % at night, respectively), while time spent on in-nest activities was not significantly different (p = 0.41, mean of 15.7 % of time during day vs. 16.5 % at night; Fig. 6).
Fig. 6

Significant difference in extra-nest tasks, in-nest tasks, and inactivity between day and night (p < 0.01, 0.41, and <0.01). Boxplots show the lower and upper quartiles (box), median (horizontal line within box), and extremes (whiskers) for worker inactivity at each time period (day is light grey, night is dark grey). n.s. p > 0.05, *p < 0.05, **p < 0.01, and ***p < 0.001

Overall worker inactivity also varies significantly at shorter timescales over the course of the day (i.e., among 4 h intervals at which the videos were recorded; LMM F = 7.36 p < 0.0001; random effects: colony/date, Table 4). However, the only significant contrast at this level is between 8 p.m. and 12 a.m. observations (Fig. 7).
Table 4

Individual workers (Ind_Ant) consistently differ in the proportion of time they are inactive (Ind_Ant) and overall inactivity varies over 4-h intervals (timepoint). The lack of interaction between these factors suggests the absence of ‘shift work’ at this timescale. Results from a linear mixed-effects model*




F value

p value





















*Fixed-effects: Ind_Ant, timepoint, and Ind_Ant x timeperiod.

Random effects: colony/date.
Fig. 7

Variation of worker inactivity over 4-h intervals. Boxplot shows the lower and upper quartiles (box), median (horizontal line within box), and extremes (whiskers) for worker inactivity at each time point

Do workers differ in circadian rhythm, employing ‘shift work’?

Above, we have discussed that the proportion of time spent in the inactive state in a sample is significantly affected by the identity of the ant and by the time of day (main effects in Table 2). If ants were working in ‘shifts’, i.e., taking turns resting, we would not necessarily expect these main effects; moreover, we would expect a significant interaction between worker identity and time, to reflect the fact that different individuals are inactive at different times. This is not the case: ant identity and time period (‘daytime’ vs. ‘nighttime’, and all 4-h interval time points compared to each other) show no significant interaction in their effect on level of inactivity (p = 0.21 and 0.66, respectively, Tables 3 and 4). This indicates that, although workers may differ in the proportion of time they are inactive, the differences do not depend on the time of day they were observed. Thus, workers do not have complementing or staggered circadian rhythms as would be expected if they were working in ‘shifts’, and all workers generally increase the proportion of time spent inactive from daytime to nighttime. This suggests that daytime inactivity is positively correlated to nighttime inactivity and the least inactive workers during the day will also be the least active at night (further confirmed by a Pearson correlation on within-colony worker inactivity ranks: Fig. 8, ρ 0.45, p < 0.0001, n = 265); however, there is considerable variation (see Fig. 9 for individual changes in inactivity).
Fig. 8

Within colony inactivity ranks are consistent between ‘daytime’ and ‘nighttime’ observations. Pearson correlation performed on within-colony worker inactivity ranks (Pearson’s ρ 0.45, p < 0.0001)
Fig. 9

Consistent differences are shown by workers (each represented by a very narrow box) having variable inactivity levels on each graph. Each worker is represented by a (very narrow) box, with the black line indicating the median inactivity. Individual ants in both daytime and nighttime graphs are sorted from left to right in decreasing median time spent inactive during the daytime. For daytime data (left graphs with light gray boxes), the median for workers on the left end of the graphs tends towards 100 % time spent inactive while on the right end it tends towards 0 % time inactive (as indicated by the overall black median line generally starting at 100 % and dropping to 0 %). If workers did not have significant differences, we would expect similar medians for all workers (a horizontal black line) and largely overlapping boxes. The comparison of the daytime (left graphs with light gray boxes) and nighttime (right graphs with dark grey boxes) data are helpful for understanding the lack of day/night shift-work. If all workers were either active during the day OR active during the night, we would expect graphs skewed in opposite directions. If activity level of workers was consistent during day and night, as is predicted by the rank correlation in Fig. 8, we should see a downward sloping median line on the right set of graphs that parallels the median line on the left. Although this pattern is not obvious, there is a general trend of decreasing inactivity from workers on the left to those on the right for the nighttime graphs. Boxplots show the lower and upper quartiles (box), median (horizontal line within box), 1.5 interquartile ranges (whiskers), and outliers (points) for each ant


Our results show that workers differ in the tasks or groups of tasks they allocate most time to. Workers can be grouped into external workers (who forage and build the nest), nurses, patrollers/groomers, and inactive workers. Perhaps the most surprising result of this study is that inactivity is highly repeatable and explains a large portion of inter-worker variation, on par with specialized tasks such as foraging, building, and brood care. This suggests that some workers are much more likely to be inactive than other workers, effectively ‘specializing’ on inactivity, and these differences appear to be stable over at least a 2-week period. Whether worker inactivity changes at longer timescales, such as seasonally or ontogenetically, remains to be tested. Thus, individual-level differences in activity are a real phenomenon which cannot be explained by temporal variation in inactivity over the course of the day. This suggests that inactivity is likely not the result of constraints, such as need for rest, delays between tasks, time for digestion, or other necessary side-effects of other activities, all of which would be expected to affect all workers more or less equally. This raises the question of whether high levels of inactivity could be an adaptive element of DOL in social insect colonies.

The level of inactivity of individual workers was shown here to fluctuate over the course of the day at short timescales (5-min intervals). Cyclical fluctuations in colony-level activity patterns had been shown in Temnothorax allardycei, where bursts of activity would occur approximately every 26 min (Cole 1991). Workers in isolation showed no such activity cycles, and as group size increased, the variation in cycle period decreased. As such, these cycles are thought to be the result of inter-individual interactions, rather than a circadian pattern of activity. Additionally, there is evidence that in-nest workers tend to have regular activity cycles, but that these can be disturbed by returning foragers (Boi et al. 1999). However, both Cole (1991) and Boi et al. (1999) define activity as movement rate, measured as pixel change rates. It is difficult to estimate how such measures of movement rate translate to measure of behavioral activity discussed herein. For instance, in our data, time spent on ‘wandering inside’ (7.4 % of mean worker time) is considered an undifferentiated activity and is excluded both from active time and inactive time, but likely accounts for much of the activity recorded by Cole (1991). Furthermore, tasks which account for large portions of active time in our data, such as brood care (4.5 % of mean worker time), involve very little movement and so would register little activity according to Cole’s measure of activity.

Nonetheless, our observations of short timescale variation in inactivity are consistent with the idea of colony-wide activity fluctuations, in that we find strong consistency within 5-min intervals and apparently stochastic variation across such intervals. However, our observation timescale (5 min videos at 4 h intervals) was not designed to test for a periodicity over intervals such as 26 min (as found in Cole 1991).

Observations of the temporal patterns of inactivity show higher levels of inactivity during the night, suggesting a circadian rhythm. Furthermore, more time is spent on extra-nest activities such as foraging activity during the day than during the night. However, in-nest tasks show no such periodicity. Earlier work on social synchronization in honey bees suggested the existence of a ‘colony-clock’, where worker activity cycles were coordinated within colonies (Frisch and Koeniger 1994). More recent work suggests that this may not the case. In fact, honey bee foragers show strong evidence of periodicity in their activity patterns where they are more active during the day, but in-nest workers, including workers involved in food storage, do not (see Moore 2001 for a full discussion on this topic). A similar phenomenon has been shown in bumble bees where foragers have been shown to have diurnal rhythms while in-nest workers do not (Yerushalmi et al. 2006) but, prior to this study, these patterns had not been shown in ants.

Why are foragers more active during the day? Honey bee and bumble bee foragers rely heavily on visual cues to navigate (Dyer and Could 1983) and cannot fly in total darkness. They also adjust their activity according to daily floral rhythms so as to maximize the amount of nectar and pollen collected during foraging trips (Moore et al. 1998), while in-nest workers do not have these constraints. T. rugatulus ants have been shown to use visual cues in navigation (Bowens et al. 2013), and so they may also benefit from foraging during the day, even if they do not need light simply to move. In addition, brood may need tending around the clock. Although we do not know the feeding frequencies of larvae in T. rugatulus, we do know that in the fire ant Solenopsis invicta, larvae are fed 2–50 times per hour, depending on larval size and satiation (Cassill and Tschinkel 1995) and larva feeding requirements are unlikely to change overnight. Thus, nurses may need to be equally active at night as during the day.

There is a long standing tradition of using multivariate analyses to study colony organization (Lenoir and Mardon 1978). Here, we use PCA to look at the relationships between tasks in a multivariate space. Our analyses show that inactivity and wandering inside are almost parallel to the first component. Although many active tasks correlate negatively with inactivity, they show a large amount of variation that is independent of it (their vectors in the PCA are not parallel to the inactivity vector). This suggests that the amount of time workers spend on a particular active task is not well predicted by their level of inactivity. Instead, a relative independence of inactivity from active tasks suggests that may be a motivation other than a lack of available work for workers to spend more time on inactivity.

Looking at the organization of task vectors along PC2 and PC3, we can look at the relationships between active tasks, independent of inactivity. There appear to be three groups of tasks (i.e., where task vectors within groups roughly parallel each other and are at an angle from other groups). This grouping is further supported by the hierarchical cluster analysis between tasks (top dendrogram) that first distinguishes between active and inactive tasks, and subsequently between nurse tasks, foraging tasks, trophallaxis, and patrolling/grooming. The main difference here is that trophallaxis seems to be its own group, while in the PCA trophallaxis is grouped with foraging tasks.

Overall, DOL in colonies of T. rugatulus appears to segregate work into task groups. Similarly to task clustering, workers can first be separated into active and inactive workers, and subsequently active workers can be further separated into three main task groups: external workers (building, foraging, and trophallaxis), nurses (brood care and in-nest feeding), and patrollers/groomers (wandering inside and feeding). That colonies might be organized in groups of workers with similar task profiles is not a new idea. Many studies have shown that monomorphic colonies could be grouped into a number of behavioral (temporal) castes in ants (Wilson 1976; Mirenda and Vinson 1981; Calabi 1988; Corbara et al. 1989; Retana and Cerdá 1990; Retana and Cerdá 1991). Workers tend to be clustered into either two groups (intra- and extra-nidal; Lenoir and Ataya 1983; Pamminger et al. 2014), or three groups (extra-nidal, nurse, and social/generalist; Mirenda and Vinson 1981; Herbers 1983; Mersch et al. 2013), though some have found many more (e.g., Fresneau 1984 found five groups; Corbara et al. 1989 found six groups). Generally, worker groups seem to represent the progression of workers from more central (e.g., nurses) to less central (e.g., foragers) specializations as predicted both by most interpretations of age/temporal polyethism (e.g., Seeley 1982; Gordon 1996) as well as the foraging-for-work task allocation strategy (Franks and Tofts 1994). Our results are broadly in agreement with this body of work; however, we also show the presence of an additional group composed of highly inactive workers. There are only a few studies where inactivity is considered in colony organization and a distinct ‘inactive’ group is found (Fresneau 1984; Corbara et al. 1989; Retana and Cerdá 1991). However, the data presented here is significantly more comprehensive (continuous tracking of complete colonies over 5 min spans at multiple time points across the day and over multiple days vs scan sampling of subsets of workers over multiple days) which allows us to show that worker inactivity levels are highly consistent at multiple timescales (daily and over 2–3 weeks). The presence of a patrolling/grooming groups is also somewhat interesting as this is also a rarely described group which may serve a similar function as that suggested by Johnson (2008) in honeybees of global information collecting.

The results from the PCA generally agree with the inter-task rank correlations, but provide slightly different insights. On the one hand, ranks are relative measures (to a worker’s nestmates) of likelihood to engage in tasks, while the PCA deals with absolute likelihood to engage in tasks. For example, a worker that does the most foraging and brood care in a colony (thus highly ranked) may nonetheless be doing very little foraging (i.e., spend little time foraging) as compared to workers in other colonies, or even to how much total available time for work. Because the PCA uses absolute measures and pools multiple colonies, observed trends could potentially be the result of inter-colony variation rather than inter-worker variation. This is likely not the case because PCAs done on individual colonies show similar patterns to the pooled PCA. Additionally, rank correlations that necessarily account for inter-colony variation generally agree with the PCA further suggesting that the patterns of worker variation are indeed real.

Highly repeatable tasks are tasks that workers tend to specialize on and thus highly visible and tractable. This might explain why they are often the focus of DOL studies (Gordon 1996; Beshers and Fewell 2001; Duarte et al. 2011). However, as suggested by the spread of workers along PC1 as well as the task and worker hierarchical cluster analyses, these tasks tend to only be done by active workers. Thus, studies that uniquely focus on these tasks are studying a biased subsample of the colony’s workers and consequently only looking at one dimension of worker variation. Tasks such as eating, grooming, wandering inside may not be very repeatable, and thus not very specialized, but most workers do them to some extent and are therefore very relevant to understanding worker time budgets (i.e., how workers allocated their time to tasks) because the time spent on these tasks is time that cannot be spent on other, more specialized tasks. Furthermore, the difference between specialized and generalized tasks may reflect a difference between self-serving and colony-serving tasks. Generalized tasks such as eating and grooming tend to be actions that most workers will need to do for themselves (self-serving) and contribute to individual function, while specialized tasks such as foraging and brood care tend to be activities thought to contribute to colony function (colony-serving). Additionally, self-serving tasks have been linked to inactivity and worker reproduction (Hillis et al. In prep.).

At first glance, inactivity may be thought of as the lack of activity and thus may seem trivial. However, we show that there is a subset of workers that effectively ‘specialize’ on inactivity. Studies that uniquely focus on active or specialized tasks are studying a biased subsample of the colony’s workers; by excluding inactivity, an entire dimension of task allocation is eliminated. This would be like using the left-hand graph of Fig. 2 to understand colony organization without knowing that there is a rich third dimension in which workers vary in inactivity, independently of active tasks. How can we expect to obtain a comprehensive understanding of the tradeoffs involved in allocating time to tasks when we ignore inactives, which compose close to half of the colony, or simply lump them in with other workers?

The question of consistent individual differences in behavior is central to the studies of personality and behavioral syndromes (Dingemanse et al. 2002; Bell et al. 2009; Pearish et al. 2013). In recent studies, personality is broadly defined as a behavior that is consistent through time and across situations (Bell et al. 2009). This definition can sometimes be problematic because the timescale over which behavior needs to be consistent is open-ended, and there are no clearly established criteria on what is a biologically meaningful timescale (Stamps and Groothuis 2010; Dall and Griffith 2014). Early personality literature stated that ‘temperament’, the foundation of personality, is reflected in early appearing tendencies that continue through the life of an individual (reviewed in Gosling 2001). Although it may seem that we could use the term ‘personality’ to describe observed consistent levels of worker inactivity, we hesitate to do so because we can easily imagine scenarios where the behaviors are consistent over the observed timescale (~weeks), but not over the lifetime of individuals. For example, inactivity could conceivably be the result of inexperience or immaturity in younger workers or senescence in older workers (Seid and Traniello 2006). The timescale of our experiment (a few weeks) would not have changed the status of Temnothorax workers, who are thought to live for years (which makes a longitudinal study difficult in this species), with regard to whether they are young or senescing workers.

In this study, we showed that colony organization in T. rugatulus ants resembles typically observed worker groups: extra-nest workers, nurses, and patrollers/groomers, which tracks predictions from age/temporal polyethism and/or foraging for work. We also showed the occurrence of a less commonly described group: the inactive workers. Indeed, inactivity was shown to be a highly consistent individual behavior on which a subset of workers effectively ‘specialize’. Incorporating inactivity in future studies of task allocation and DOL may be important as we show that inactives are a distinct group of workers with their own sets of behaviors and should likely not be either ignored for lack of undertaking ‘active’ tasks, or be counted as less efficient workers in typically described groups such as nurses and foragers.


We thank Alex Downs, Andrew Scott, Mary Levandowski, Matthew Velazquez, Neil Hillis, and Nicole Fischer for their help with ant painting and maintenance, and data collection. We also thank the entire Dornhaus lab for their ongoing feedback. Research supported through the GIDP-EIS and EEB Department at University of Arizona, as well as NSF grants no. IOS-1045239, IOS-0841756, and DBI-1262292 (to A.D.).

Supplementary material

265_2015_1958_MOESM1_ESM.docx (478 kb)
Fig S1(DOCX 477 kb).
265_2015_1958_MOESM2_ESM.docx (71 kb)
Fig S2(DOCX 71 kb).
265_2015_1958_MOESM3_ESM.docx (14 kb)
Table S1(DOCX 14 kb).
265_2015_1958_MOESM4_ESM.docx (17 kb)
Table S2(DOCX 16 kb).

Copyright information

© Springer-Verlag Berlin Heidelberg 2015