1. Ecology
Download icon

A small number of workers with specific personality traits perform tool use in ants

  1. István Maák
  2. Garyk Roelandt
  3. Patrizia d'Ettorre  Is a corresponding author
  1. Department of Ecology, University of Szeged, Hungary
  2. Museum and Institute of Zoology, Polish Academy of Science, Poland
  3. Laboratory of Experimental and Comparative Ethology UR 4443, University Sorbonne Paris Nord, France
  4. Institut Universitaire de France (IUF), France
Research Article
  • Cited 1
  • Views 1,272
  • Annotations
Cite this article as: eLife 2020;9:e61298 doi: 10.7554/eLife.61298

Abstract

Ants use debris as tools to collect and transport liquid food to the nest. Previous studies showed that this behaviour is flexible whereby ants learn to use artificial material that is novel to them and select tools with optimal soaking properties. However, the process of tool use has not been studied at the individual level. We investigated whether workers specialise in tool use and whether there is a link between individual personality traits and tool use in the ant Aphaenogaster senilis. Only a small number of workers performed tool use and they did it repeatedly, although they also collected solid food. Personality predicted the probability to perform tool use: ants that showed higher exploratory activity and were more attracted to a prey in the personality tests became the new tool users when previous tool users were removed from the group. This suggests that, instead of extreme task specialisation, variation in personality traits within the colony may improve division of labour.

Introduction

Tool use is a widespread phenomenon within the animal kingdom (Shumaker et al., 2011; Sanz et al., 2013) and new examples of animal tool use are regularly discovered, such as recently in pigs (Root-Bernstein et al., 2019) and seabirds (Fayet et al., 2020). Tool use is defined as "the external employment of an unattached or manipulable attached environmental object to alter more efficiently the form, position, or condition of another object, another organism, or the user itself, when the user holds and directly manipulates the tool during or prior to use and is responsible for the proper and effective orientation of the tool" (Shumaker et al., 2011, p. 5). Most of the reports concern vertebrates, particularly primates and birds, which can manufacture tools to solve specific tasks (Hunt and Gray, 2002; Sanz et al., 2013; Auersperg et al., 2014), use multiple tools sequentially (Martin-Ordas et al., 2012), and choose effective tools based on their functional properties (Visalberghi et al., 2009). We know that capuchin monkeys have been using stone tools to process food for at least 3000 years (Falótico et al., 2019) but presumably the use of tools appeared even earlier in invertebrates. Smith and Bentley-Condit, 2010 reported about 50 cases of tool use in insects, encompassing 30 different genera. Among the best described examples is the use of debris to transport liquid food by some species of ants (Morrill, 1972; Barber et al., 1989), in particular, several species of the genus Aphaenogaster (Fellers and Fellers, 1976; Tanaka and Ono, 1978; McDonald, 1984; Agbogba, 1985). Workers of these species are characterised by the lack of a distensible crop and by a chitinous gaster, preventing the transportation of large amounts of liquid food inside their bodies, a feature common in many other ant species (Hölldobler and Wilson, 1990; Davidson et al., 2004). Moreover, unlike most ant species, Aphaenogaster workers do not perform mouth-to-mouth exchange of liquid food (i.e. trophallaxis, Delage and Jaisson, 1969); therefore, foragers cannot use this form of food transmission among colony members. Workers forage individually mostly on the ground level and can cover large areas in habitats with scarce food sources (Cerdá et al., 1998). When Aphaenogaster foragers discover a source of liquid food, such as fruit pulp or body fluids of dead insects, they collect debris (pieces of leaves, soil, sand grains), drop them into the liquid food, and then transport these soaked debris to their nest. This behaviour qualifies as tool use, namely tool-assisted food transport. Indeed, these ants do not drop debris in non-food substances (Banschbach et al., 2006).

In a previous study (Maák et al., 2017), we have investigated whether Aphaenogaster ants are selective in the choice of material to be used as tools and we demonstrated that ant workers prefer materials that are easy to handle and with good soaking capacity. Furthermore, ants can learn to use artificial material that is novel to them and select the material with optimal soaking properties, thus showing that tool use is not behaviourally fixed in ants (Maák et al., 2017). Tool selection also depends on the foraging environment and varies with food type (viscosity), distance, and availability of tools (Lőrinczi et al., 2018).

The process of tool use in ants has not been studied at the individual level. In particular, we do not know whether any forager with sufficient information about the location of the food and the availability of the tools would perform tool use or whether there are specialised workers which perform tool use repeatedly. In A. rudis, it was observed that the tool use behaviour is carried out by a small subset of individuals within the group of foragers and only a small number of workers perform the debris dropping task (Banschbach et al., 2006). In A. subterranea as well, the proportion of workers observed to use tools was only a small fraction of the total number of foragers. The number of tool users did not increase with colony size, while the number of total foragers did. Moreover, there was no significant relationship between the number of ants working at the debris dropping task and the number of debris pieces dropped, indicating that a small number of workers can perform this task very efficiently (Lőrinczi, 2014).

We asked the question of what makes a good tool user in A. senilis ants. It is indeed unknown whether some individual characteristics of workers, such as personality traits, would predict tool use behaviour. When inter-individual differences in behavioural traits are consistent across time and/or context they are considered personality traits (Réale and Dingemanse, 2012). In ants, both workers and colonies can show personalities (Pinter-Wollman, 2012) and there is a link between individual and collective behaviour (Carere et al., 2018). There is some evidence that the allocation of workers to a certain task may be influenced by individual personality, for instance in the ant Leptothorax acervorum (Kühbandner et al., 2014), where more exploratory and aggressive workers were also more active in the nest (rested less). This inter-individual difference could enhance division of labour (Jeanson and Weidenmüller, 2014), the phenomenon in which different individuals perform different tasks. Division of labour is beneficial for colony organisation and it is believed to be one of the principal factors explaining the extraordinary ecological success of social insects (Hölldobler and Wilson, 1990). A broadly acknowledged model explaining the emergence of division of labour within a colony is the ‘response threshold model’, which posits that individual workers differ in their sensory perception and/or in their behavioural responses to stimuli associated with specific tasks (Robinson, 1992; Beshers and Fewell, 2001). However, the possible interplay between consistent individual differences and division of labour has not been explored (Jeanson and Weidenmüller, 2014).

We also know that there is an association between personality traits and cognitive traits in ants. Consistent individual differences in exploratory activity predicted learning performance of individual carpenter ants, with ‘active-explorers’ being slower in learning than ‘inactive-explorers’ (Udino et al., 2017) and individual differences in exploratory activity were linked to cognitive judgement bias, the propensity to anticipate either positive or negative consequences in response to ambiguous information (d'Ettorre et al., 2017). Therefore, we expected a link between individual personality traits and tool use.

We performed three different experiments to study the tool use behaviour at the individual level in the ant A. senilis. In Experiment 1, we observed the tool use process in whole colonies to characterise the behaviour of ants during the two parts of the process: the transport of tools to the bait and the transport of tools from the bait to the nest. In social insects, task partitioning -a phenomenon in which a piece of work is divided among two or more workers- has been reported in a wide variety of foraging tasks in several species of ants, bees, wasps and termites (Ratnieks and Anderson, 1999). However, partitioning may reduce task efficiency and reliability unless the number of workers involved in the task is high or a given subtask is carried out by morphologically specialised workers (Ratnieks and Anderson, 1999). In Aphaenogaster, workers are morphologically similar, therefore we predicted that we would not observe task partitioning and that the same worker could perform both parts of the task.

In Experiment 2, we created sub-colonies to investigate whether there is an individual specialisation in tool use. Are the same ants using tools repeatedly? Are ants that use tools to collect liquid food also transporting solid food items? We predicted that ants would specialise in tool use only, under the general assumption that specialised individuals work more efficiently than less specialised ones (Robinson, 1992). In this experiment, we also asked whether the simple fact of observing nestmates using tools during one trial would facilitate the task performance of naive workers by social learning. The involvement of social learning in solving complex tasks has been shown in social insects, for instance, bumblebees (Loukola et al., 2017).

In Experiment 3, we studied the possible link between individual personality traits and tool use. Previous studies demonstrated the existence of colony-level personality in A. senilis (Blight et al., 2016a; Blight et al., 2016b), we thus predicted that consistent behavioural differences will be found also at the individual level in this species (worker personality). Using workers that were individually characterised for personality traits, namely exploratory activity and reaction to prey, we created sub-colonies that were tested in tool use trials. After each trial, the ants that used tools were removed from the sub-colony. Thus, we could test whether personality would predict which ants will become tool users in the next trial.

Results

Experiment 1: tool use process in whole colonies

To characterise the entire tool using process, the plastic box housing the ant colony was connected to a foraging arena with a detachable bridge. At the beginning of the experiment, we placed 10 tools (small pieces of sponge) and then liquid food (diluted honey) in the foraging arena. Next, we removed the bridge so that ants in the foraging arena could not go back in the ant colony. This way we could quantify the number of workers transporting tools to the bait over the total number of workers that were present in the foraging arena. After 30 min, the connecting bridge was replaced, giving the possibility to the ants to transport tools from the bait to the nest (see Materials and methods). We used three colonies and five replicates each (one replicate = 1 trial with food and tools). The results show that the number of workers present in the foraging arena before replacing the bridge had a positive effect on the latency to drop the first tools into the bait (LMM t = −2.22, N = 15, p=0.047), meaning that the higher the number of workers present, the shorter the latency. Also, the shorter was the latency to drop the first tool into the bait, the shorter the total time needed to transport all the tools to the bait (t = 2.26, N = 15, p=0.043). However, the number of workers involved in tool use did not have any effect on the latency to drop the first tool to the bait (t = 1.34, N = 15, p=0.20) nor on the total time of tool transport to the bait (t = −0.46, N = 15, p=0.65).

The latency to transport the first tool inside the nest did not depend on the dynamics of the tool transport to the bait. In particular, the latency for the first tool transport to the nest was not influenced by the latency to drop the first tool into the bait (LMM t = 0.99, N = 15, p=0.34), nor by the total time to transport all tools to the bait (t = −1.35, N = 15, p=0.21), nor by the number of workers involved in tool transport to the bait (t = −0.44, N = 15, p=0.67). The transport to the nest started always well after (124.4 ± 15.3 min, mean ± SE; Table 1) the completion of the tool transport to the bait.

Table 1
Experiment 1: tool use process in whole colonies.

Summary table showing the number of workers present in the arena (# workers in arena), the latency to drop the first tool on the bait (First tool on bait), the total time devoted to tool transport to the bait (Tot. time tool transport), the number of tools transported to the bait (# tools on bait), the latency to transport the first tool to the nest from the start of the experiment (First tool to nest) and the number of workers involved in tool transport to the bait (# workers transp. tools to bait). Five replicates (R1-R5) for each colony are shown. The last column shows the number of tools transported by each worker to the bait (# tools transp. by each worker); for instance, in R1 there were two tool users, one transported nine tools and the other 1.

Colony# Workers in arenaFirst tool on bait
(min)
Total time tool transport
(min)
# Tools on baitFirst tool to nest (min)# Workers transp. tools to bait# Tools transp. by each worker
 1 R1199211017429; 1
 1 R243191910-25; 5
 1 R31623151020131; 8; 1
 1 R42313379-27; 2
 1 R52942310-25; 5
 2 R1626221011151; 2; 1; 4; 2
 2 R214720109622; 8
 2 R3401158-33; 1; 4
 2 R4303251014425; 5
 2 R5712101012637; 2; 1
 3 R116452010-31; 4; 5
 3 R29761017228; 2
 3 R3237710222110
 3 R44182410178110
 3 R5712031921425; 4

We then investigated further the workers that used tools. Compared to the number of workers present in the foraging arena (23.33 ± 6.04 workers, mean ± SE), only a few workers performed the tool use behaviour (2.33 ± 0.35 workers, mean ± SE; Table 1). We observed workers repeatedly transporting tools within the same trial and between trials (Table 2). Some workers participated in both parts of the task (transport of tools to the bait and also from the bait to the nests) and repeated transports by the same worker were observed in both parts of the tool use process (Table 2). In the next step, we asked whether the tool users were simply those workers that were the first to discover the location of the food and the tools. This was not the case. Indeed, an average of 11.35 ± 6.49 workers (mean ± SE, min = 3, max = 33 workers; see Table 3 for further details) contacted the tools and the food before the first tool user dropped the first tool into the bait. Paint marked tool users (N = 18) took on average 388 ± 134.16 sec (mean ± SE) to locate the tools after contacting the food for the first time. Once they had the information about the location of both the food and tools, the latency to drop the first tool into the bait was 385.67 ± 128.30 sec (mean ± SE). However, these two latencies were not correlated (rs = −0.006, N = 18, p=0.98), meaning that those workers that located the tools earlier did not necessarily start the transport of tools to the bait faster.

Table 2
Experiment 1: tool use process in whole colonies.

The tool use behaviour is composed of two parts: transport of tools to the food source (bait) and transport of imbibed tools inside the nest. The table shows the total number of tool users that participated (Particip.) in both parts of the tool use process and those that transported tools to the bait (# tool users), tool users that were marked (Marked tool users), number of workers that transported more than one tool within a trial (>1 tool within trial), and that transported more than one tool among trials (>1 tool among trials). Shown is the total for the three experimental colonies (sum of five trials each).

Transport to baitTransport to nest
Colony# particip. in both parts# Tool usersMarked tool users>1 tool within same trial>1
tool across trials
Marked tool users>1
tool within same trial
111155131
211533151
32988091
Table 3
Experiment 1: tool use process in whole colonies.

The tool users were not the first workers obtaining information about the presence of food and tools. The table shows the number of workers that contacted both the tools and the food before the first tool was dropped into the bait; the latency (Lat.) for the first worker to obtain information (info.) about the presence of both the tools and the food and the latency for the first tool user to obtain information about the tools and the food.

Colony# Workers contacting tools and food before the first tool was dropped on the baitLat. first worker having info. about both tools and food (min)Lat. first tool user having info. about both tools and food (min)
 1 R1102.617
 1 R2170.1717
 1 R3110.5523
 1 R4180.382
 1 R580.73
 2 R1518.0325
 2 R250.987
 2 R330.751
 2 R440.872
 2 R550.181
 3 R1270.7545
 3 R241.927
 3 R3111.057
 3 R441.855
 3 R5331.75106

Experiment 2: is there specialisation in tool use?

In this experiment, we investigated whether ants specialise in tool use or also transport solid food items (here cricket legs). We also tested the effect of a familiarisation-trial (pre-trial) with food and tools on naive workers. In this experiment, we focused only on the transport of tools to the bait. For the pre-trial, a group of sub-colonies (sub-colonies 1) was familiarised with the food (diluted honey) and the tools (10 pieces of sponge) separately, while the other group (sub-colonies 2) was familiarised with the tools and the food simultaneously. Therefore, workers of sub-colonies two could perform the tool use behaviour while workers of sub-colonies one could not. Immediately after this pre-trial, workers that performed tool use in sub-colonies two were removed so that we could see whether the fact of observing nestmates using tools during this pre-trial would facilitate tool use in naive workers by social learning.

The day after, each sub-colony was tested with liquid food and 10 tools, then with five cricket legs, then again with liquid food and 10 tools. The procedure was repeated the following day and, after 2 days of rest, each sub-colony received liquid food and 20 tools (Figure 1A). This procedure to test for specialisation in tool use was the same for sub-colonies 1 and sub-colonies 2.

Procedure followed in experiment 2.

Each sub-colony received honey and tools in steps 1 and 3. The yellow items represent the tools (small pieces of sponge,~1 mm3), below is shown the plate with diluted honey (0.25 ml). In step 2, the ants received five cricket legs (Acheta domestica). On day 5, the ants received 20 tools and the honey bait (A). For individual workers that transported four consecutive tools to the bait, the time needed to transport one tool significantly decreased within one trial (B). For individual workers that transported at least two consecutive tools in different trials, the time needed to transport one tool to the bait did not change among trials (C). Box plots show medians, quartiles, min-max values, outliers (black dots) and individual data points (empty circles). NS – non-significant, *: p<0.05, ****: p<0.0001.

Figure 1—source data 1

Experiment 2 - the time (sec) needed to transport one tool by the same individual that performed consecutive tool transports to the bait within a single trial.

The time required for the transport of each tool during consecutive transports by the same worker within a trail was analysed with LMM (Gaussian error, maximum likelihood fit). Only workers that performed four consecutive transports were included in the analysis (N = 54; second-third tool: p=0.03, second-fourth tool: p<0.001).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig1-data1-v1.xlsx
Figure 1—source data 2

Experiment 2 - the time (sec) needed to transport one tool for individual workers that transported at least two consecutive tools in different trials.

The average time needed to transport one tool by the same worker between the trials was analysed with LMM (Gaussian error, maximum likelihood fit). Up to three trials were included (N = 29; first-second trial: p=0.99; first-third trials: p=0.32).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig1-data2-v1.xlsx

Of the 40 individually marked workers per sub-colony involved in this experiment, about 20% performed tool use during at least one of the four trials with 10 tools (Table 4). The average number of individuals per sub-colony showing tool use was 8.25 (CI95% [6.47, 10.03]). This was notably lower than 10.2 (CI95% [10.18, 10.22]), which is the average number of tool users obtained by randomly assigning the same number of observed tool use events to a simulated ant population based on the same number of individuals, sub-colonies and trials (see Simulation data of experiment 2, Supplementary file 1), and the confidence intervals do not overlap. This indicates that the individual distribution of tool use events was not random in our experiments. Of the 99 tool users in total, 64.7% performed several tool transports within the same trial and 33.3% participated repetitively in more than one trial (Table 4). The majority of these ants that performed multiple tool transport across trials (26 over 31 workers) participated in two trials (2.89 ± 0.42 per trial, mean ± SE), three workers participated in three trials and two workers in four trials (Table 5). It is important to note that the occurrences of tool use were repeatable across trials at the individual level (RICC = 0.22, N = 1880, p=0.001). Half of the workers (50.38%) that performed tool use in the final trial with 20 tools, also participated in at least one of the previous trials with 10 tools (Table 6), thus confirming that the same ants use tools over and over again, including across trials.

Table 4
Experiment 2: Is there specialisation in tool use?

The total number of tool users and cricket leg transporters (transp.) and the percentage of workers that performed repeated tool use within or between trials or that participated also in the transport of cricket legs. The last column shows the number of very active workers, which participated in at least two tool use trials and one leg transport.

Colony/
subcolony
Total tool usersRepeats within (%)Repeats between (%)Transporting also legs (%)Total leg transp.% tool users among leg transp.≥2 trials,
≥1 leg (%)
1/164 (66.7)1 (16.7)2 (33.3)633.31 (16.7)
1/243 (75)1 (25)1 (25)5201 (25)
2/1127 (58.3)2 (16.7)1 (8.3)1100-
2/2108 (80)4 (40)2 (20)5402 (20)
3/166 (100)5 (83.3)1 (16.7)714.31 (16.7)
3/2128 (66.7)3 (25)3 (25)31002 (16.7)
4/185 (62.5)2 (25)3 (37.5)6501 (12.5)
4/2107 (70)4 (40)3 (30)6502 (20)
5/163 (50)3 (50)2 (33.3)4502 (33.33)
5/252 (40)2 (40)1 (20)425-
6/1148 (57.1)3 (21.4)4 (28.6)757.12 (14.29)
6/263 (50)1 (16.7)2 (33.3)5401 (16.7)
Average8.255.33 (64.7)2.58 (33.3)2.08 (25.9)4.9248.311.25 (15.15)
Table 5
Experiment 2: is there specialisation in tool use?

Number of workers participating in more than one trial and the average number of tools they transported.

Colony/
subcolony
Two trials# ToolsThree trials# ToolsFour trials# Tools
1/119.5
1/213.75
2/124
2/243.12
3/153.3
3/221.7513.75
4/123
4/241.75
5/133
5/223.25
6/131.5
6/218.33
Average2.893.461.505.671.003.75
Table 6
Experiment 2: is there specialisation in tool use?

The total number of workers using tools in the last trial (trial with 20 tools) and the number of workers that performed tool use also in previous trials (10 tools).

Colony/
subcolony
Total # tool users# Using tools in previous trials (%)
1/142 (50%)
1/241 (25%)
2/153 (60%)
2/251 (20%)
3/143 (75%)
3/232 (66.7%)
4/121 (50%)
4/221 (50%)
5/1--
5/253 (60%)
6/153 (60%)
6/283 (37.5%)
Average4.272.09 (50.38%)

Workers that performed tool use were not necessarily specialised in foraging for liquid food; they also transported cricket legs to the nest. About half (48.31%) of the workers that transported cricket legs were also tool users in at least one trial (Table 4). Among the 99 tool users, 15 workers (15.15%) showed a particularly high activity by participating in at least two tool use trials and at least one cricket leg transport (Table 4).

The consecutive transport of several tools within one trial enhanced the efficiency of a tool using worker. When workers consecutively transported four tools, the time to transport one tool decreased with the number of tools transported (second-third tool: LMM t = −2.25, N = 54, p=0.03, second-fourth tool: t = −3.75, p<0.001; Figure 1B, Source code 1). However, the average time to transport one tool did not change significantly between different trials (first-second: t = −0.01, N = 29, p=0.99; first-third t = −1.02, p=0.32; Figure 1C, Source code 2). It was possible that while a worker was involved in repeated tool transports to the bait, another worker started using tools. The interference of this new tool user did not alter the efficiency of the previous tool user (LMM t = 0.5, N = 66, p=0.62), meaning that the time to transport a tool to the bait by the first worker did not change before and after interference, and there was no significant effect of the sub-colony (t = −0.52, N = 66, p=0.6).

We analysed the effect of the single pre-trial -in which the sub-colonies two received liquid food and tools simultaneously, while the sub-colonies one received first food and then tools (in absence of the food)- on the performance of the ants in the subsequent trial (with food and tools at the same time). The latency to bring the first tool to the bait was not different between the two types of sub-colonies (LMM t = −0.5, N = 12, p=0.64; Figure 2A, Source code 1) as well as the total time of tool transport (t = −0.88, N = 12, p=0.42; Figure 2B, Source code 2). The number of workers involved in tool use was also not different between the two groups (LMM t = −0.46, N = 12, p=0.67), indicating that there is no social learning in these conditions.

Effect of the pre-trial in experiment 2.

There was no significant difference between the different groups of sub-colonies in the latency to transport the first tool to the bait (A) and the total time needed to transport the 10 tools to the bait (B). Sub-colonies 2: had a pre-trial (liquid food and tools simultaneously); Sub-colonies 1: without a pre-trial (received first food and then tools in absence of the food). Workers manipulating tools during the pre-trial were removed (see Materials and methods). Box plots show median, quartiles, min-max values and outliers (black dot).

Figure 2—source data 1

Experiment 2 - the latency (sec) to transport the first tool to the bait by the two subcolonies with and without a pre-trial.

The effect of the pre-trial on the latency to bring the first tool to the bait in the first trial was analysed with LMM (Gaussian error, maximum likelihood fit). All colonies (N = 6) and subcolonies (N = 12) were included (p=0.64).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig2-data1-v1.xlsx
Figure 2—source data 2

Experiment 2 - the total time (sec) needed to transport the ten tools to the bait by the two subcolonies with and without a pre-trial.

The effect of the pre-trial on the total transport time observed during the subsequent first trial was analysed with LMM (Gaussian error, maximum likelihood fit). All colonies (N = 6) and subcolonies (N = 12) were included (p=0.42).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig2-data2-v1.xlsx

Experiment 3: does worker personality predict tool use?

We created eight sub-colonies each with 20 individually marked workers, which were characterised for personality traits using two tests, open-field and reaction to prey (Figure 3A,B), repeated after two days (see Materials and methods). Ants showed significant consistency over time in their behavioural responses. In the open-field test, the time spent walking in the periphery (RICC = 0.27, N = 154, CI95% [0.11, 0.41], p<0.001) and the total time spent in the central area were significantly repeatable across the two sessions (RICC = 0.43, N = 154, CI95% [0.31, 0.56], p<0.0001; Figure 3C, Source code 1). In the reaction to prey test, the time spent in contact with the prey was also highly repeatable (RICC = 0.61, N = 154, CI95% [0.49, 0.71], p<0.0001; Figure 3D, Source code 2). The time spent in contact with the prey was negatively correlated with the time spent walking in the periphery (rs = −0.30, N = 154, p<0.001) and positively correlated with the total time spent in the central area (rs = 0.33, N = 154, p<0.001), indicating that the more exploratory ants were also those more interested in the prey.

Experimental set-up used for the open-field (A) and the reaction to prey tests (B).

The ant is in the acclimatisation tube for 1 min and the test starts when the tube is removed (see Materials and methods). The consistency across the two sessions of the open-field test regarding the time spent in the central area (C) and the consistency across the two sessions of the reaction to prey test (D). The black line with confidence band (grey) is plotted based on the Pearson correlation of the two variables.

Figure 3—source data 1

Experiment 3 - correlation between the two sessions (repeats) for the time (sec) spent walking in the central area (open-field test) by individual ants.

Repeatability across the two sessions, for each individual ant, concerning the time spent walking in the central area was assessed with intra-class correlation (Lessells and Boag, 1987) by using LMM (Nakagawa and Schielzeth, 2010). The total time spent in the central area were significantly repeatable across the two sessions (N = 154; p<0.0001). Six ants died between the two sessions of the personality tests, therefore the sample size is 154 ants instead of 160 (20 ants in each of the eight sub-colonies).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig3-data1-v1.xlsx
Figure 3—source data 2

Experiment 3 - correlation between the two sessions (repeats) for the time (sec) spent in contact with the prey (reaction to prey test).

Repeatability across the two sessions, for each individual ant, concerning the time spent in contact with the prey was assessed with intra-class correlation (Lessells and Boag, 1987) by using LMM (Nakagawa and Schielzeth, 2010). The time spent in contact with the prey was highly repeatable (N = 154; p<0.0001). Six ants died between the two sessions of the personality tests, therefore the sample size is 154 ants instead of 160 (20 ants in each of the eight sub-colonies).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig3-data2-v1.xlsx

Ants were observed in seven tool use trials (each with liquid food and 10 tools) over 4 consecutive days (Figure 4). After each trial, the ants that used tools were removed from the sub-colony. In this way, we could test whether personality would predict which ants will become tool users in the next trial. We used Principal Component Analysis based on the variables of the personality tests to calculate a ‘personality score’ (see Materials and methods and Figure 5A,B, Source code 1).

Procedure followed in experiment 3.

Test-sub-colonies received honey and tools in every trial. The yellow items represent the tools (small pieces of sponge,~1 mm3), below is shown the plate with diluted honey (0.25 ml). On days 1–3, two trials were performed (10 tools offered), and after each trial, the ants that transported tools to the bait were removed. On day 4, one trial (10 tools offered) was performed and then the previously removed workers were returned to their test-sub-colony. All the workers were used in Trial 8, in which 20 tools were offered.

Plots of the first two dimensions of the Principal Component Analysis.

Correlation circle of the variables: TWalk.Out (time spent walking in the periphery), TotT.Center (total time spent in the central area); TContact.Prey (time in contact with the prey) (A). Projection of individuals on the PCA factorial space: red dots refer to ants that used tools, black dots refer to ants that did not use tools; the confidence ellipses representing individuals using tools (red arrow) or not (green arrow) show significant difference (no overlap) (B). The probability of using tools is linked to the individual personality score. The mean and CI of the personality score and the individual data points (empty circles) for the non-tool users (mean = −0.33, CI [−0.56,–0.09]) and tool users (mean = 0.57, CI [0.26, 0.88]) is plotted (C). The ants’ personality score significantly predicted the probability of using tools (***: p<0.001).

Figure 5—source data 1

Experiment 3 - the variables used in the Principal Component Analysis.

A ‘personality score’ for each ant was calculated with a Principal Component Analysis based on (a) total time spent in the central area of the open field, (b) time spent walking in the periphery, (c) time spent in contact with the prey (average of the two sessions for each variable). The figures show the average values of the two sessions used to study the behaviours observed in the two personality tests: open-field and reaction to prey (N = 154). Six ants died between the two sessions of the personality tests, therefore the sample size is 154 ants instead of 160 (20 ants in each of the eight sub-colonies).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig5-data1-v1.xlsx
Figure 5—source data 2

Experiment 3 - the differences in personality score between tool user and non-tool user workers.

The personality score of a given ant is represented by its individual value related to the first principal component, accounting for the 59.14% of the total variance, normalised by subtracting the mean value for its colony. The link between personality score and tool use behaviour (tool user or not) was analysed with a GLMM, binomial error structure (logit-link). The ants’ personality score significantly predicted the probability to use tools in at least one trial (N = 154, p<0.001). Six ants died between the two sessions of the personality tests, therefore the sample size is 154 ants instead of 160 (20 ants in each of the eight sub-colonies).

https://cdn.elifesciences.org/articles/61298/elife-61298-fig5-data2-v1.xlsx

The ants’ personality score significantly predicted the probability of using tools in at least one trial (GLMM z = 3.97, N = 154, p<0.001). Moreover, the confidence intervals of the personality score of tool users (mean = 0.57, CI95% [0.26, 0.88]) and non-tool users (mean = −0.33, CI95% [−0.56,–0.09]) did not overlap (Figure 5C, Source code 2). A positive personality score characterised those ants that spent more time exploring the central area in the open field test and that spent more time in contact with the prey (Figure 5A,B).

Given that tool users were more exploratory than non-tool users, one may wonder whether the observed relationship between personality and tool use behaviour is merely the result of the fact that more exploratory ants are more likely to encounter the tools and possibly use them up before other workers can find them. We thus analysed a randomly chosen subset of the tool use trials of experiment three to investigate this. First, it should be noticed that the tool use process did not start immediately after the tools and the food were offered to the ants. The latency to bring the first tool to the bait was 894.71 ± 153.49 sec (mean ± SE, N = 21 trials; see ‘Contact_with_tools’ in data of experiment 3, Supplementary file 1), which gives plenty of time for all the ants to explore and find the tools. From a total of 21 trials, only in five occasions, the tool user was the first worker to investigate the tools. In the majority of the trials (16 out of 21) an average of about five workers investigated the tools by antennation before the first tool user but did not pick them up (4.94 ± 0.52 (mean ± SE) workers, min = 1 worker, max = 8 workers). This indicates that the tool use behaviour is not merely prompted by the fact of finding the tools or the food before other workers. If this would be the case, in the reaction to prey test we should expect a negative correlation between the latency to find prey and the time spent in contact with it (the shorter latency, the longer contact time). In contrast, we found a positive correlation between the latency to find the prey and the time spent in contact with the prey: the longer the latency, the longer the time spent with the prey (session 1: rs = 0.27, N = 154, p<0.001; session 2: rs = 0.42, p<0.001).

Tool users typically transported more than one tool to the bait. In most of the cases, the transport of all the 10 tools was performed by one worker, in 23% of the cases by two workers and only in one case by three workers. The removal of the active workers after each trial did not influence the characteristics of the tool transport to the bait compared to the first trial (latency of first tool to bait: LMM −0.7 < t < 0.94, N = 28, p>0.36; total transport time: LMM −0.7 < t < 1.02, p>0.32; number of workers involved: GLMM−0.66 < z < −0.07, N = 28, p>0.51). Therefore, the workers that took over the tool use task were not significantly less efficient than the removed tool users.

In the final trial with 20 tools, a total of 18 workers used tools. Of these, 12 workers (66.6%) previously performed tool transport and they brought 106 (66.25%) of the 160 tools that were transported to the bait in this last trial. Of the 34 workers removed during the first 3 days of the experiment, 11 resumed tool using in this last trial, and 4 of these workers were active during the first trial with 10 tools 7 days earlier.

Discussion

Workers of some ant species use debris as tools to collect and transport liquid food to the nest. We studied tool use behaviour at the individual level in Aphaenogaster senilis to determine whether any forager that knows the location of the food and the tools puts the two together, or if it is an attribute of a subset of workers. Three experiments explored this question. Experiments 1 and 3 showed that several workers contacted the food and the tools before the first tool user started transporting tools to the bait. Therefore, tool users were not simply the first workers that learned about the location of the food and the tools. Experiment 2 showed that only a small subset of foragers (about 8%) used tools and that the observed individual distribution of tool use events was not random. Moreover, tool use was a significantly repeatable behaviour across trials at the individual level. Finally, in experiment 3, we found that workers show consistent inter-individual behavioural variation in exploratory activity and reaction to prey. These highly repeatable behavioural traits were used to calculate a personality score that significantly predicted the probability of using tools. Based solely on these behavioural traits we could distinguish two groups of workers: the tool users and the non-tool users. In sum, our data indicate that tool use is not a stochastic phenomenon but it is performed by a subset of workers with specific behavioural traits.

In both experiments 1 and 2, we observed tool users performing repeated tool transport within the same trial and also between trials. In experiment 2, more than 60% of the tool users performed repeated tool transport to the bait and their efficiency increased during the process: for a given ant, the time to transport one tool decreased with the number of tools transported and was not affected by the presence of another worker transporting tools. This improved efficiency is likely explained by the fact that ants learned the location of the tools with respect to the bait and could return quickly to pick up the next tool. In several ant species, foraging speed is enhanced by route learning (Czaczkes et al., 2011; Pasquier and Grüter, 2016), which appears to be a general phenomenon. More than 30% of the tool users participated in more than one trial, nevertheless, the performance of these workers did not improve along with the trials as the average time to transport a tool did not change significantly across trials. In other social insect species, however, an improvement over consecutive trials has been observed. For instance, bees become more efficient in foraging with experience (Dukas and Visscher, 1994; Klein et al., 2019) and, in the context of nest emigration, Temnothorax albipennis ant workers improve their performance over successive emigrations (Langridge et al., 2008). This suggests that in A. senilis the limiting factor to speed up the process is memorising the location of the tools, which is different in every trial.

In experiment 1, we observed that the same ant could perform both parts of the tool use task, that is transport of tools to the bait and from the bait to the nest. Therefore, as we predicted, there was no clear evidence of task partitioning. Nevertheless, our results regarding the lack of task partitioning are not conclusive due to the relatively small total number of ants that were observed participating in both parts of tool use process. To the best of our knowledge, the observations reported in the literature also lack conclusive evidence. In A. rudis, similarly to A. senilis, no evidence of task partitioning was found (Banschbach et al., 2006), although task partitioning was suggested in earlier observations of A. famelica (Tanaka and Ono, 1978). This suggests that this behavioural aspect might be species-specific but ideally requires further research.

The results of experiment one also showed that the dynamics of tool transport to the bait does not influence the transport of food-imbibed tools to the nest, which starts on average 2 hrs after completion of the first part. In natural conditions, covering the food quickly with debris gives an advantage to Aphaenogaster ants in the competition with more dominant ant species, which cannot exploit the food once it is covered (Fellers and Fellers, 1976), thus leaving plenty of time for Aphaenogaster to bring the food-imbibed tools to the nest. Therefore, it appears to be advantageous to first completely cover the food with debris and then to start transporting them to the nest. During the first part of the process, the results of experiment one showed that the higher the number of workers present in the foraging area the shorter was the latency to start the tool use process. A shorter latency resulted in a shorter total time to complete the task, but interestingly, this did not depend on the number of workers involved in tool use. Enhanced efficiency related to shorter latency was also found in the ant Temnothorax albipennis, where shorter recruitment latencies to high quality nest sites, compared to low-quality ones, improved the nest emigration process (Mallon et al., 2001; but see Robinson et al., 2009). Similarly, in the ant Formica fusca, shorter latencies to return to the nest after exploration of a novel arena characterised larger colonies and indicated efficient exploration of the surroundings (Somogyi et al., 2020).

Tool use events were not randomly distributed among workers, however, the results of experiment two also indicate that tool users were not necessarily specialised only in this task. About half of the workers that transported solid food (cricket legs) were also tool users. This may not hinder efficiency since it has been shown that task specialisation does not translate into higher performance in ant species without morphologically differentiated workers (Dornhaus, 2008). Some workers exhibited very high activity by participating in several tool use trials and transporting more than one cricket leg. These resemble ‘elite’ workers that show high performance in several tasks, as observed in some ant species (Robson and Traniello, 1999; Pinter-Wollman et al., 2012). It would be interesting to test whether these very active workers act as key individuals that increase the activity level of group members (Robson and Traniello, 1999).

In experiment 2, we also tested whether the simple fact of observing nestmates transporting tools during one trial would facilitate the task performance of naïve workers via social learning. We did not find a difference in the performance of workers that were given the possibility to observe nestmates performing tool use and workers that did not have this possibility, suggesting that social learning might not be involved in the ontogeny of the tool use behaviour in A. senilis. Our finding is in agreement with the observation by Tanaka and Ono, 1978 that naïve foragers can carry out tool use in A. famelica. Spontaneous tool use and manufacture has been also shown in vertebrates, such as naive juvenile New Caledonian crows (Kenward et al., 2005), while social learning appears to be essential when the task is complex and non-natural, as observed in bumblebees (Alem et al., 2016; Loukola et al., 2017). Nevertheless, we should acknowledge that in the present study we used only one type of tool. We know that Aphaenogaster ants are able to select among different types of tools according to their soaking capacity and the environmental context (Maák et al., 2017; Lőrinczi et al., 2018) and that they learn to choose the optimal tools over successive trials (Maák et al., 2017), therefore we cannot exclude that social learning could play a role in the process of tool selection, a possibility that awaits formal testing.

Experiment 3 clearly showed that workers of A. senilis are characterised by consistent behavioural variation at the individual level concerning exploratory activity and reaction to prey. Consistent inter-individual behavioural variation has also been shown in other ant species (e.g. Kühbandner et al., 2014; Udino et al., 2017) and it might be a general characteristic of workers. What is new in our work is the discovery that individual behavioural variability is directly linked to the probability to perform a certain task, in this case tool use. Workers with a positive personality score, characterised by high explorative behaviour and prey-attraction, were more likely to be involved in tool use than workers with a less positive or negative personality score. Despite finding relatively small effect sizes in the analysis of the personality data, our results reveal the importance of even slight individual differences in behavioural traits, if they are consistent, in the organisation of social life. Inter-individual variability is the basis for division of labour, the phenomenon in which different individuals perform different tasks. According to the ‘response threshold model’ (Robinson, 1992; Beshers and Fewell, 2001), workers characterised by a low threshold will respond to low stimulus intensity (high responsiveness), while workers with high threshold will respond to high stimulus intensity (low responsiveness) (Bonabeau et al., 1996). Several proximate mechanisms underlying inter-individual behavioural variability have been described, including genetic diversity, ontogeny, nutrition, experience and learning (reviewed in Jeanson and Weidenmüller, 2014). Experience may act as ‘self-reinforcement’ and directly modulate the individual response threshold. It was hypothesised that the simple fact of performing a task would lower the corresponding stimulus threshold, while not performing a task would further increase the individual threshold (Theraulaz et al., 1998). Indeed, a study using clonal ants showed that, all else being equal, foragers that were successful had a higher probability to perform the foraging task again, compared to unsuccessful foragers, which specialised in brood care (Ravary et al., 2007).

Individual personality differences within a given behavioural group (foragers, nurses) may contribute to a fine-tuned division of labour but this possibility is relatively unexplored. A study in honeybees found evidence for life-long personality differences in workers and suggested that the response thresholds to some stimuli could be related to personality type, thus contributing to more robust inter-individual behavioural differences leading to division of labour even when individuals age together or share similar experiences (Walton and Toth, 2016). In carpenter ants, there is evidence for inter-individual variability in sucrose responsiveness and learning success in different behavioural groups of workers performing different tasks (Perez et al., 2013). In the ant Myrmica rubra, individual personality differences are connected to spatial fidelity (the position in the nest) and ants located in a given position show low thresholds to perform tasks associated with that position, thus generating division of labour (Pamminger et al., 2014). However, examples of interplay between personality and task performance in social insects are generally scarce.

Individual behavioural flexibility is another important aspect guaranteeing division of labour in social insects, particularly following changes in the social environment, such as changes in colony demography (Jeanson, 2019). We found that A. senilis colonies could cope well with the removal of active tool users, which were immediately replaced by individuals that were previously less active. Similarly, in the red harvester ant, Pogonomyrmex barbatus, when the most active foragers were experimentally removed they were replaced by other individuals (Beverly et al., 2009). In the ant Temnothorax rugatulus as well, when the most active nurses and foragers were removed they were quickly replaced by workers from the reserve pool of inactive individuals (Charbonneau et al., 2017). In T. albipennis, in the context of nest emigration, 20% of ants were more active and performed transports repeatedly in successive experimental emigration trials (Pinter-Wollman et al., 2012). Pinter-Wollman et al., 2012 performed an experiment in which they removed these active ants during some emigration trials and then they put them back. Removed active ants were replaced by previously less active individuals, but when they were returned to the colony they did not resume their active role, they were thus permanently replaced. The authors suggest that modifications in the social context and experience can cause a long-lasting change in the response threshold of workers (Pinter-Wollman et al., 2012; Jeanson, 2019). This is not what we observed in A. senilis, in which many of the removed active workers resumed tool use when they were returned to their sub-colony in the last trial with 20 tools, including workers that were removed 10 days earlier (after the very first trial). This is not surprising if we consider that the probability to perform tool use in A. senilis is linked to personality, which, by definition, is consistent over time.

In conclusion, our work shows that instead of extreme task specialisation, the involvement of workers with appropriate personality ensures high efficiency in tool use. Given the scarcity of examples linking individual personality and task performance in social insects, our study provides new insight into the interplay between personality and division of labour and should encourage further theoretical and empirical studies in this direction.

Materials and methods

Study species and housing

Request a detailed protocol

Twelve colonies of the monogynous Mediterranean ant species, Aphaenogaster senilis, were used. This species occurs in open and sunny locations (e.g. forest edges, lawns, fields, sand dune areas), and colony size varies between a few hundred to a few thousand workers (Boulay et al., 2007). Ants were collected in the Doñana National Park (Spain) in March 2019 and kept at the Laboratory of Experimental and Comparative Ethology (University Sorbonne Paris Nord, France) under standard conditions (temperature 24 ± 4°C; relative humidity 50–60%; natural light cycle). Each colony was housed in a circular plastic box (14.5 cm diameter) with a regularly moistened plaster floor (representing the nest) placed inside a larger plastic box (18 × 25.5 × 7.7 cm) constituting the foraging arena. The standard diet consisted of dead crickets (Acheta domestica) and apple/honey mix provided twice a week. Two weeks before the experiments, to increase the motivation for carbohydrates, the standard diet was reduced to crickets only; water was always provided ad libitum.

Experiment 1: tool use process in whole colonies

Request a detailed protocol

The tool use process is composed of two parts: the transport of tools to the bait and the transport of tools from the bait to the nest (Maák et al., 2017). The aim of this experiment was to characterise the behaviour of ants during the entire process. We used three colonies containing the queen, brood, and around 1000 workers and we marked individually 100 workers in each colony with small dots of enamel paint. The day before the experiment, the plastic box housing the colony was connected to a separate foraging arena of the same size (18 × 25.5 × 7.7 cm) with the help of a detachable bridge allowing the workers to circulate. On the day of the experiment, 10 tools (small pieces of firm sponge, ~1 mm3) were placed in this foraging arena. After 10 min, the food bait (0.25 ml of diluted honey) was placed in the arena at 5 cm distance from the tools and the bridge was removed, so that the ants present in the foraging arena with the food and the tools could not go back to the nest. This was considered as the start of the experiment. The number of ants in the foraging arena was counted and the presence of paint marked workers noted, we could thus quantify the number of workers transporting tools to the bait over the total number of workers that were present in the foraging arena. After 30 min, the connecting bridge was replaced, giving the possibility to the ants to circulate both ways and to transport tools from the bait to the nest. The experiment was repeated five times for each colony, with an interval of 1 day between two trials. Each trial was video recorded for 4 hr. We noted the latency (from the start of the experiment) of every tool dropped into the bait, the latency (from the start of the experiment) of the first tool transported from the bait to the nest (in most of the cases, the transport of all tools inside the nest was not completed in 4 hr) and the individual identity of workers performing the task if these were marked.

Experiment 2: is there specialisation in tool use?

Request a detailed protocol

In this experiment, we focused on the first part of the tool use process, the transport of tools to the bait, and we investigated whether ants specialise in tool use or whether ants that use tools to collect liquid food are also involved in the transport of solid food items (here cricket legs). We also tested the effect of a familiarisation-trial with food (diluted honey) and tools on naive workers. From each of the six source colonies (different from those used in Experiment 1), we created two sub-colonies, each with 10 larvae and 40 workers (30 foragers from outside and 10 nurses from inside the nest), individually marked with dots of enamel paint. Each sub-colony was housed in a separate plastic box (28 × 28 × 8 cm) containing a dark plastic tube with a water reservoir acting as a nest. Workers readily transported the larvae inside the nest and were left undisturbed for 1 day.

For the pre-trial, sub-colony one was familiarised with the food and the tools separately: workers received 0.25 ml of diluted honey and were allowed to feed for 15 min, then the bait was removed and, 30 min later, workers received 10 tools (pieces of sponge, ~1 mm3). Sub-colony two was familiarised with the tools and the food simultaneously: workers received 0.25 ml of diluted honey and 10 tools at the same time (the tools were located 4 cm away from the honey bait). Workers were observed until all the tools were deposited on the honey bait. The two workers that interacted (i.e. antennation, grasping) most with the tools in sub-colony one and those that transported the highest number of tools in sub-colony two were removed. If observing nestmates using tools has an effect, we expected workers from sub-colonies two to be more efficient than workers of sub-colonies one in the subsequent trial.

The day after, each sub-colony was tested in three steps (Figure 1). Step 1: presentation of 0.25 ml of diluted honey and 10 tools simultaneously. Workers were observed for 45 min (during this time usually all the tools were placed on the honey bait) in which we noted the latency to transport each tool to the bait and the identity of workers using tools. Step 2: 2 hr later, presentation of five cricket legs (placed on the opposite side of the nest). Workers were observed for 30 min and latency to pick up each leg was noted, as well as the identity of the workers carrying the legs inside the nest. Step 3: 2 hr later, Step 1 was repeated (honey bait and tools).

The following day, each sub-colony was tested again with the three steps procedure. After 2 days of rest, each sub-colony received honey and 20 tools (instead of 10), to give the possibility to more workers to transport tools (Figure 1). We noted the latency to transport each tool to the bait and the identity of the transporting workers.

Experiment 3: does worker personality predict tool use?

In this experiment, we investigated the possible link between personality traits and tool use. Workers were tested twice for two different personality traits and observed in tool use trials. Each sub-colony underwent seven tool use trials (Figure 4). After each trial, the ants that used tools were removed from the sub-colony. In this way, we could test whether personality would predict which ants will become tool users in the next trial. All tests were video-recorded. From each of the eight source colonies, we created two sub-colonies: (1) one ‘test-sub-colony’ with 20 individually marked workers (15 outside and five inside workers) and six larvae placed in a plastic box (25 × 18 × 9 cm) containing a dark plastic tube with a water reservoir acting as a nest; (2) one ‘host-sub-colony’, with 10 workers and five larvae placed in a plastic box (16 × 12 × 9 cm) with a nest tube. This host-sub-colony (2) housed the workers removed for the test-sub-colony (1) during the experiments (see below).

All the 20 workers of the test-sub-colonies (1) were individually characterised for their personality traits using two tests (open-field and reaction to prey), each repeated after a 2-day interval to assess individual consistency over time.

Open-field

Request a detailed protocol

This is an adaptation of the classical open-field test developed to test exploratory behaviour and anxiety in rodents (e.g., Prut and Belzung, 2003) and already used with ants (d'Ettorre et al., 2017; Udino et al., 2017; Carere et al., 2018). An ant was individually placed in an acclimatisation tube (Ø 2.7 cm) for 1 min at the centre of a circular arena (Ø 11.5 cm) with a floor of clean filter paper (replaced after each trial), in which an area of 9.5 cm diameter was considered as the central zone and the external part as the periphery (Figure 3A). Then, the tube was removed and the behaviour of the ant was observed for 3 min. More exploratory ants are expected to spend more time in the central area, while less exploratory ants will spend more time walking along the edges of the arena, where they are protected by the walls. We measured the time spent walking and resting in the central area (total time in the central area) and the time spent walking in the peripheral area with the help of a behavioural transcription software (Ethoc version 1.2, CNRS Research Centre on Animal Cognition, Toulouse).

Reaction to prey

Request a detailed protocol

A similar circular arena with clean filter paper as a floor (but without the delimitation between central and peripheral zone) was used for this test. An ant was placed in the acclimatisation tube for 1 min and one fruit fly (Drosophila hydei) freshly killed by freezing was placed 4 cm away from the ant (Figure 3B). After the removal of the tube, the ant could interact with the prey for 3 min. We recorded the duration of the contact with the prey using the software Ethoc.

Tool use

Request a detailed protocol

In each trial, the test-sub-colony (1) received 0.25 ml of diluted honey and 10 tools. We recorded the latency to transport each tool to the bait and the identity of workers using tools for a maximum of 45 min. Each test-sub-colony underwent seven trials: two trials on day 1 (one in the morning, one in the afternoon), two trials on day 2, two trials on day 3 and one trial on day 4 (Figure 4). After each trial, the ants that transported tools to the bait were removed from the test-sub-colony and placed in the host-sub-colony. In the afternoon of day 4, those workers were returned to the test-sub-colony, which had a rest of 2 days. Afterwards, a last trial was performed with 20 tools instead of 10 (Figure 4).

Statistical analysis

If necessary, the variables were log-transformed prior to the analyses to meet the normality and homogeneity of residuals. Statistical analyses were carried out in R (version 3.6.1) Statistical Environment (R Development Core Team, 2019). In all models, the colony (or sub-colony) identity (ID) was included as a random factor.

Experiment 1: tool use process in whole colonies

Request a detailed protocol

We used linear mixed-effects models (LMM, Gaussian error, maximum likelihood fit) to analyse: (i) the effect of the number of workers in the arena (before reconnecting the bridge) and of the number of workers involved in tool use on the latency of the first tool to the bait (sec); (ii) the effect of the latency of the first tool to the bait and of the number of workers involved in tool use on the total time of tool transport to the bait. In both models, the number of workers involved in tool use was included as explanatory variable, whereas the number of workers in the arena or the latency of the first tool to the bait were included as covariates. A similar model (LMM, Gaussian error, maximum likelihood fit) was built to analyse the effect of (a) latency of the first tool to the bait, (b) total transport time, (c) number of workers involved in tool use on the latency of transport the first tool to the nest. Trial number was included as a random factor. The correlation between the latency to locate the tools after contacting the food and the latency to drop the first tool into the bait was tested with Spearman rank correlation.

Experiment 2: is there specialisation in tool use?

Request a detailed protocol

Number of ants using tools: to compare our empirical results to randomised data, we tested them against the null-model (any ant with sufficient information has the same probability of performing tool use) by randomly assigning the same number of observed tool use events to a simulated ant population based on the same number of individuals, sub-colonies, and trials as in experiment 2. The 95% confidence interval (CI95%) was calculated based on the random model with 1000 simulations performed with the help of a Macro function of Microsoft Excel (see Supplementary file 1).

We used the structure of the original data set of 12 sub-colonies with 40 individuals each. For 11 sub-colonies, there were four repeated measurements per individual, and in one sub-colony, there were only three repeated measurements per individual, resulting in a total of 1880 cases (see Supplementary file 1). To these cases, we (a) randomly assigned 136 ‘tool use events’, as this was the number observed in our original data set. Then, we (b) checked for each sub-colony whether each individual was assigned as a ‘tool user’ (one or more tool use events during the three or four repeated measurements) or not (0 tool use events). Finally (c), the number of tool users was averaged over all 12 sub-colonies, and the result was saved to a list. The whole procedure (a-c) was repeated for 1000 times and the mean number of tool users over all simulation runs (with CI95%) was calculated. We could then compare the observed number of tool users (mean and CI95%) with that generated by the model. The lack of CI overlap between observed and simulated data supports non-random distribution of the tool use behaviour.

Repeatability of tool use behaviour was calculated with intra-class correlations (Lessells and Boag, 1987) using GLMM for binomial data (tool users and non-tool users) of the R package rptR (Stoffel et al., 2017).

The time required for the transport of each tool during consecutive transports by the same worker within a trail was analysed with LMM (Gaussian error, maximum likelihood fit). Only workers that performed at least four consecutive transports were included. A similar model was built to compare the average time needed to transport one tool by the same worker between the trials. The order of the tools (1 to 4) or the order of the trials (up to three trials) was included as a fixed factor.

Sometimes another worker started transporting tools to the bait while the first worker was performing multiple tool transports, we thus wondered whether this ‘interference’ might influence the worker performance. The time needed to the first worker to transport one tool to the bait before and after interference by another worker was analysed with LMM (Gaussian error, maximum likelihood fit). The order of the tools (before and after interference) and the sub-colony ID were included as fixed factors.

The effect of the pre-trial on the (a) latency of the first tool to the bait, (b) total transport time, (c) number of workers involved in tool use in the first trial was analysed with LMMs (Gaussian error, maximum likelihood fit). The type of sub-colony was included as a fixed factor, whereas colony ID as a random factor.

Experiment 3: does worker personality predict tool use?

Request a detailed protocol

Six ants died between the two sessions of the personality tests, therefore the sample size is 154 ants instead of 160 (20 ants in each of the eight sub-colonies). Repeatability across the two sessions of the open-field and the reaction to prey test was assessed with intra-class correlations (Lessells and Boag, 1987) by using LMM calculations (Nakagawa and Schielzeth, 2010) with individual identity (ID) as a random factor. The correlations between the time spent in contact with the prey (average of the two tests) and the two traits measured during the open-field test (average of the total time spent in the central area and time spent walking in the periphery) were performed with Spearman rank correlation because of the non-normal distribution of the reaction to prey data.

A ‘personality score’ for each ant was calculated with a Principal Component Analysis (package Factominer) based on (a) total time spent in the central area of the open field, (b) time spent walking in the periphery, (c) time spent in contact with the prey (average of the two tests for each variable). The personality score of a given ant is represented by its individual value of first principal component, accounting for the 59.14% of the total variance, normalised by subtracting the mean value for its colony.

The link between personality score and tool use behaviour (tool user or not) was analysed by using a GLMM with a binomial error structure (logit-link). The correlation between the latency to find the prey and the time spent in contact with the prey measured during the reaction to prey test (separately for the two sessions) was performed with Spearman rank test (data not normally distributed).

The effect of the removal of the workers on (a) latency of the first tool on bait, (b) total transport time, (c) number of workers involved in tool use was analysed with LMMs (Gaussian error, maximum likelihood fit) and GLMM (Poisson error, maximum likelihood fit), respectively, with the order of the trials as a fixed factor. In this analysis, we used data up to the first four trials because only few sub-colonies performed more than four trials.

General details on function and packages

Request a detailed protocol

GLMs were performed using the ‘glm’ function from the Stats package. LMMs and GLMMs were performed using lmer and glmer functions from the lme4 package (Bates and Maechler, 2013). Poisson models were checked for overdispersion. The repeatability was calculated with the rpt function (rptR package, Stoffel et al., 2017). We assessed 95% confidence intervals (CI95%) by 1000 bootstraps and p values by 1000 permutations (alpha level = 0.05). For the PCA, the confidence ellipses around the categories of two factors (tool user or not) were drawn with plotellipses function from the FactoMineR package (Le et al., 2008).

Data availability

All data generated during this study are included in the manuscript and supporting files. Source data files have been provided for all Figures.

References

    1. Agbogba C
    (1985)
    Observations on foraging of liquid sugar and insect body fluids by two species of Aphaenogaster: A. senilis and A. subterranea (Hym Formicidae)
    Insect Soc 32:427–434.
    1. Delage B
    2. Jaisson P
    (1969)
    Social relations among ants of the genus Aphaenogaster
    Sciences Naturelles 268:701–703.
  1. Book
    1. Hölldobler B
    2. Wilson EO
    (1990)
    The Ants
     Cambridge, USA: Belknap Press of Harvard University.
    1. Lőrinczi G
    (2014)
    Some notes on the tool-using behaviour of the ant, Aphaenogaster subterranea (Hymenoptera: formicidae)
    Tiscia 40:17–24.
    1. McDonald P
    (1984)
    Tool use by the ant, Novomessor albisetosus (Mayr)
    J New York Entomol S 92:156–161.
    1. Robinson EJH
    2. Smith FD
    3. Sullivan KME
    4. Franks NR
    (2009) Do ants make direct comparisons?
    Proceedings of the Royal Society B: Biological Sciences 276:2635–2641.
    https://doi.org/10.1098/rspb.2009.0350
  2. Book
    1. Robson SK
    2. Traniello JFA
    (1999)
    Key individuals and the organization of labor in ants
    In: Detrain C, Deneubourg J. L, Pasteels J. M, editors. Information Processing in Social Insects. Basel: Birkhauser. pp. 239–260.
  3. Book
    1. Sanz C
    2. Call J
    3. Boesch C
    (2013)
    Tool Use in Animals: Cognition and Ecology
    New York, USA: Cambridge University Press.
  4. Book
    1. Shumaker RW
    2. Walkup KR
    3. Beck BB
    (2011)
    Animal Tool Behavior: The Use and Manufacture of Tool by Animals
    Baltimore, USA: Johns Hopkins University Press.

Decision letter

  1. Christian Rutz
    Senior Editor; University of St Andrews, United Kingdom
  2. Ammie K Kalan
    Reviewing Editor; Max Planck Institute for Evolutionary Anthropology, Germany
  3. Tomer J Czaczkes
    Reviewer; University of Regensburg, Germany

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This study uses multiple experimental paradigms to investigate tool use in an ant species at both the individual and colony level. The authors show that tool use is performed by only a fraction of the non-specialized workers present. However, workers with specific personality traits, namely those that are more explorative and attracted to prey, were most likely to become tool users in the colony.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for submitting your work entitled "A small number of workers with specific personality traits perform tool use in ants" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by a Senior Editor.

Our decision has been reached after consultation between the reviewers. Although all reviewers found your manuscript interesting, they raised a large number of substantive concerns about the methodology and analyses. Additional work is required to address these points (including a comprehensive personality assay), which may fundamentally change the conclusions drawn. Based on the reviewers' extensive discussions, and their individual reviews appended below, we regret to inform you that your work, in its present form, will not be considered further for publication in eLife. We are willing to allow submission of a new manuscript that contains new experiments addressing the shortcomings of the present study, but please note that resubmission does not guarantee another round of in-depth review, let alone eventual acceptance.

This is a summary of the main concerns:

1) Personality claim: While this finding was the most novel, we found it to be poorly supported by the evidence presented. The two assays used to assess personality essentially only test for exploration activity, since both assays measure search/exploration time (since prey contact time is likely to be dependent on exploration, those ants who explore more will find the prey faster). In order to assess personality, we would expect more than one facet of personality to be measured, such as sociability, aggression etc. (see Udino et al., 2017). Such an approach would warrant reference to personality in this manuscript, and be much more convincing.

2) Tool-use assay: Related to the previous point, the fact that the experimental assay is also in itself an exploration assay is problematic. Without controlling for discovery or tool encounter rates, the finding can simply be explained as "those ants which are more likely to explore more are more likely to use tools". Given this confound, linking tool use to any aspect of personality is problematic.

3) Lack of clear hypotheses being presented: There is no mention of null hypotheses and predictions for the different experiments, which makes it difficult to assess results. Moreover, alternative hypotheses have not been sufficiently addressed in the manuscript (see detailed reviewer reports for more information).

4) The statistical analyses need improvement: This includes revising models where critical random or control effects appear to have been left out to ensure results are robust, adding effect sizes for all main results as well as other concerns (see reviewer reports). The authors should also include full code for all analyses rather than only the model specifications.

5) Additional concerns regarding the contextualization of the manuscript, clarity of the Materials and methods and reporting of results were brought up and can be found in the detailed reviewer reports below.

Reviewer #1:

The author's present an excellent set of experiments investigating task partitioning and specialization in one species of ants that are well known for using tools to transport food. The novel aspect of this manuscript specifically tests whether individual personality characteristics are associated with tool-use propensity, which if I understand correctly, goes beyond the usual colony approach to such questions. Overall I found the paper well written, the experiments clearly described and well designed, and the results easy to understand. The paper is likely to be of interest to a wide range of animal behavioural ecologists and I think warrants publication in eLife. However, I do have a couple major concerns that would need to be addressed in a revision.

1) The Introduction is generally well written but there is a lack of integration for introducing the experiments and aims of the study. The paragraph that begins “ We performed three different experiments to study the tool use behaviour…” needs to be more detailed rather than simply outlining each experiment. Can the authors include more about the overall hypotheses of the study and the predictions for the experiments? What is the aim of each experiment with respect to your overall research questions? This will also better prepare the reader for what and why particular analyses will be done for each experiment.

Essentially I think the authors can do a much better job to present the results of their experiments in a more unified and compelling way with respect to specific research questions and hypotheses.

2) The above issue, of the experiments being presented distinct and separate rather than contributing to an overall research goal needs improvement also in the Discussion. In fact parts of the Discussion would be better suited (and would be very helpful) in the Introduction. For example theoretical framing and predictions. The first two paragraphs are highly repetitive of the results but around the middle of paragraph three this improves considerably. Rather, it would be nice for the Discussion to start by briefly summarize findings across experiments with their relative contribution to the goals/hypotheses of the study.

3) My last concern includes missing random or control fixed effects in some of the mixed models used in the statistical analysis. I outline these below in detail but essentially without controlling for this additional substructure in the data the authors run the risk of producing erroneous results. Please have a look at the statistical suggestions and modify and re-run the relevant models where necessary.

Reviewer #2:

This study asks an interesting and timely question ("which individual ants perform tool use, and is tool use a personality trait?"). The experiments are performed carefully. The writing is clear. The Introduction introduces the topic well. The full data are provided in a clear manner, which is fantastic. However, I have several concerns about the conclusions drawn from the data. Firstly, some of the results on which conclusions are based are not compared against a null expectation. Secondly, and most critically, the “tool use” personality assay may not reflect anything more than tool discovery, and may thus conceptually identical to the exploration assay. The Discussion is somewhat unfocussed. In terms of subjective interest, I judge the results of experiments 1 and 2 to be not exciting enough for eLife. The conclusions arising from experiment 3 are exciting, but may not be supported by the data. Addressing my major concerns may require more video and statistical analysis, but will not require new experiments to be performed.

1) "Only a small number of workers performed tool use" this is technically true, but throughout the manuscript there is the implication that the tool users are in some way special (Results: "only a few workers performed the tool use behaviour"). The “null model” here would be that any ant with sufficient information has the same probability of performing tool use. Specifically, an ant must encounter both the tools and the food to perform tool use. As the results presented show, once tool use starts, almost all the tools are “used up” by one or two individuals very quickly, not giving other ants a chance to perform tool use, even if they could. The proportion of ants performing tool use is given. To put this in context, we need to know the proportion of ants which encountered tools and food at all. This “null hypothesis” needs to be explicitly presented, and refuted or not, given the data.

2) All the results presented in paragraph five are given without context of a null model, and are thus meaningless. I.e., if you randomly distributed the same number of transport events and tool use events over the same number of individuals and trials, would you get similar proportions? The empirical results can be compared to randomised data in order to clarify whether these results represent more than probabilistic behaviour amongst homogenous ants.

3) "The consecutive transport of several tools within one trial enhanced the efficiency of a tool using worker." This is interpretation, not results, and alternative null hypotheses are not excluded. It could be that when an ant “feels” like being efficient, it transports many tools and does so quickly. In other words, the “second tool” group (Figure 1B) may contain both slow ants which only manage 2 trips, and fast ants which manage 4. The “fourth tool” group only contains ants which manage at least 4 trips. Moreover, and more critically, it makes simple sense that when an ant works faster, it can manage more tools in a given time. To truly demonstrate improvement over time, only ants which manage 4 transport events should be considered, and the rank of the transport event used to predict transport speed. This may or may not have been done, the description of the analysis was not clear. The raw data is not separated by individual ant ID, so I could not check this myself.

4) The effect size of the key finding is not given. Examining the data, the normalized personality score ranges from -4 to 3 – so 7 in total. The non-tool users have an average score of -0.33. The tool users have a mean score of 0.57. I managed to recreate the analysis, and found an effect estimate of 0.6174. Using emmeans, we get a prob estimate of 0.349. This means that while the effect is highly significant, I would not say the effect size is very large.

5) I am not convinced that the “tool use” assay in experiment 3 is showing anything more than tool and food discovery probability. It stands to reason that ants which spend more time in the centre of the arena and less time on the periphery (the main components of the personality score) are more likely to encounter both the tools and the food first, and thus be in a position to being doing tool use, perhaps using up all the tools before the other workers found them, but definitely reducing the probability of other workers finding the tools (as fewer tools remain). The authors need to demonstrate that ants with a higher personality score are more likely to perform tool use even when controlling for opportunity. Otherwise, the seemingly interesting results boil down to "ants which spend more time in the centre of the arena are more likely to find things in the centre of the arena", which is not a very exciting result.

Reviewer #3:

In this manuscript, the authors perform three experiments of Aphaenogaster senilis using tools (e.g. debris) to soak up a liquid food resource to be carried to their nest. The primary conclusion is that a small subset of workers uses these tools, and they do so because of specific personality traits. I found the study to be quite interesting and the paper to be well-written. The authors have strong support for most of their conclusions, and indeed for the first part of their main conclusion, that only a small subset of workers perform the task, but I am less confident in the authors' evidence for the second part of that main conclusion, that these differences are due to specific personality traits.

My primary concern is in how the authors measured personality traits in workers. They performed each of two tests on two occasions: an open field test and test in which individuals could interact with prey. They found that certain behavioral traits (e.g. time spent in central arena in open field and time spent interacting with prey) were significant repeatable across the two timepoints, and conclude that these are personality traits. They then use a PCA to show that these behavioral traits correlate with tool use (ants that explored more were more likely to use tools). This consistency over time is necessary for personality, but as far as I can tell the authors did not test whether these behaviors were consistent across contexts, which is also important (I might also prefer to see more than 2 timepoints tested, though this is a minor concern). So there is a behavioral trait that is correlated across time but not contexts; I do not find this to be rigorous evidence of personality. In addition to the possibility of it being "personality" driving tool use, a non-mutually-exclusive possibility is that ants who explore more are more likely to find the tools sooner, thus becoming the tool-users, which I'm glad that the authors also discuss as a likely explanation.

I must admit that I am generally a little sceptical of the special attention sometimes placed on the idea of animal personality. To me, it is not inherently more interesting if tool use is due to a behavioral trait that is or is not consistent across contexts. Indeed, I would find the study just as interesting if the authors did not discuss personality, per se (though it is possible this would make it less broadly appealing). And I think the conclusion would be better supported and more compelling.

Other than this concern related to personality, I found the conclusions to be generally well supported. I thought it was interesting that workers improve in tool-use performance within but not across trials, and I think the authors suggestion that this relates to route-learning is a good one.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "A small number of workers with specific personality traits performs tool use in ants" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Christian Rutz as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Tomer J Czaczkes (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this decision letter to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to accept without delay manuscripts, like yours, that they judge can stand as eLife papers without additional data, even if they feel that they would make the manuscript stronger. Thus the revisions requested below only address clarity and presentation.

Summary:

Tool use is relatively rare in the animal kingdom and therefore of widespread academic interest. In addition to vertebrates, some social insects also demonstrate flexible tool use behaviours. In this manuscript, Maák et al. investigate task specialization, social learning and personality effects relating to tool use in the ant species Aphaenogaster senilis. These ants engage in tool-assisted foraging whereby debris is used as a tool to soak up liquid food, which is then transported to the nest. Previous research has shown that workers of this species can learn to use artificial materials as tools and will select the best tools based on optimal soaking characteristics.

Using a collection of three experiments, the authors demonstrate that only a small number of workers in A. senilis perform the tool use task but that these workers are not specialized in tool use, in that they will also carry and transport solid food items to the nest without tools. The authors found no support for social learning, specifically workers who had observed successful tool users were not more likely to become tool users themselves in subsequent trials. Remarkably, the authors did show that workers with certain personality traits were more likely to become tool users if previous tool users were removed from the colony.

Although the absolute number of workers demonstrating these phenomena was small, and further research is needed to address the repeatability and longer-term associations of worker personality and tool use behaviour, the overall contribution of this manuscript is valuable for its novel insight into tool-use behaviour in social insects at the individual level, going beyond colony dynamics.

Revisions:

The reviewers were all very content with the revised submission and congratulate the authors on substantially improving the manuscript by seriously taking into account our initial feedback. We had only a handful of major comments left that should be addressed before publication.

1) Please explain why exploring the periphery of the plate in your open field test would not in fact be ants that are the most exploratory and how this affects your inference about tool users being more explorative. Those ants who explore only in the center could actually be argued to be less explorative. Perhaps this trait needs rephrasing?

2) Can you strengthen the claim that tool users are the same individuals across trials (i.e., repeatability across trials) or at least make this more clear and transparent? A corollary of the personality-tool use conclusion is that you should have more individuals repeating tool use across multiple trials than expected by chance. This should mean that the same ants use tools over and over again. We are convinced that this is true at least within a trial, but do we see it across trials, as we should? The authors say that we do and provide evidence from experiment 1, in Table 2, and from experiment 2. However the evidence for experiments 1 is weak if we are interpreting Table 2 correctly, in total for the 3 colonies they saw only 2 ants transport tools to the bait in multiple trials (out of 19 marked tool users), and only 3 ants transport tools to the nest in multiple trials (out of 17 marker tool users). Is this actually more than would be expected by chance?

Indeed, experiment 3 provides the best evidence of personality predicting tool use across trials but given the low absolute numbers in repeat tool users across trials for Experiments 1 and 2 we would like the authors to discuss this aspect more critically in the manuscript.

3) Perhaps related to the above point, some of the results should be discussed with more transparency and caution. Specifically, can you acknowledge that effect sizes are small for the personality results and add why this small effect is still important. Please also address in the manuscript that your results for a lack of task partitioning are still not conclusive (due to the low number of ants that complete both parts of the tool use task, i.e. Table 2) and ideally requires further research.

4) Please add the latency to find prey and contact time correlation that was in the revisions letter to the reviewers but was not added to the manuscript. It would also be important to provide the correlation for all individuals, irrespective of whether they are tool users or not. Please also ensure the raw data for this correlation is included in the excel files.

https://doi.org/10.7554/eLife.61298.sa1

Author response

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

This is a summary of the main concerns:

1) Personality claim: While this finding was the most novel, we found it to be poorly supported by the evidence presented. The two assays used to assess personality essentially only test for exploration activity, since both assays measure search/exploration time (since prey contact time is likely to be dependent on exploration, those ants who explore more will find the prey faster). In order to assess personality, we would expect more than one facet of personality to be measured, such as sociability, aggression etc. (see Udino et al., 2017). Such an approach would warrant reference to personality in this manuscript, and be much more convincing.

We have clear evidence that the two personality assays do not test both for exploratory activity since it is not true that those ants who find the prey faster spend more time in contact with the prey. Under your hypothesis, we would expect that the shorter the latency to find the prey, the longer should be the contact time, i.e. a negative correlation between latency to find the prey and contact time with the prey. We had the data about latency to find the prey, so we could test these correlations. When we analyze the behavior of tool users, there is no significant correlation between latency and contact in both repeats (test 1: rs = 0.153, p = 0.259; test 2: rs = 0.201, p = 0.136, N = 56); when we analyze the non-tool users, there is a positive correlation (test 1: rs = 0.363, p = 0.0002; test 2: rs = 0.569, p = 0.001, N = 98). Therefore, we never find a correlation in the direction expected under your hypothesis. Indeed, many ants found the prey very quickly but did not show interest in it. We believe that the time spent in contact with the prey measures the interest of the ant for the prey (i.e. the motivation to bring the prey to the nest) and not the simple exploratory behavior. We did not add these data to the manuscript, but we are happy to include them if you judge it necessary.

Concerning the second point, we agree that personality has several facets and we have investigated that in a different study (Udino et al., 2017). However, for this study we believe that the two facets that we considered are sufficient and we now include a new analysis showing that there is consistency also across contexts (see reply to reviewer 3). It should be noted that the majority of the vertebrate literature considers only one trait: exploratory behavior as predictor of an individual’s personality, see for instance one of the most well-known papers in this field (more than 800 citations, google scholar) Dingemanse et al. Fitness consequences of avian personalities in a fluctuating environment. Proc R Soc B (2004). This is because exploratory tendency is often part of a behavioral syndrome including aggressiveness, neophilia, and boldness (e.g., Sih et al. Behavioural syndromes: An ecological and evolutionary overview. Trends Ecol Evol (2004)). Describing a behavioral syndrome in these ants was beyond the scope of our study.

2) Tool-use assay: Related to the previous point, the fact that the experimental assay is also in itself an exploration assay is problematic. Without controlling for discovery or tool encounter rates, the finding can simply be explained as "those ants which are more likely to explore more are more likely to use tools". Given this confound, linking tool use to any aspect of personality is problematic.

We are grateful for this comment because it forced us to analyze new data and made our study much more convincing. Tool users are not simply those ants that explore more, and thus are more likely to find the tools and the food first. We have now quantified this.

– In Experiment 1, we show that on average of about 11 workers contacted the food and the tool before the first tool user started using tools. Moreover, the first tool user was never the first worker having information about the food and the tools (see Table 3).

– In Experiment 2, we compared our data with randomized data (as suggested by reviewer 2) and we show that the observed average number of workers using tools 8.25 (CI95% [6.47, 10.03]) is lower than the average number of tool users obtained by randomly assigning the same number of observed tool use events to a simulated ant population 10.2 (CI95% [10.18, 10.22]). Therefore, the individual distribution of tool use events was not random in our experiments.

– In Experiment 3, we re-analyzed a subset of trials and we show that only in 5 out of 21 trials the tool user was the first worker to investigate the tools. On average, about 5 workers contacted the tools before the first tool user but did not pick them up.

More details are given in our reply to the reviewers.

3) Lack of clear hypotheses being presented: There is no mention of null hypotheses and predictions for the different experiments, which makes it difficult to assess results. Moreover, alternative hypotheses have not been sufficiently addressed in the manuscript (see detailed reviewer reports for more information).

We have clarified our hypothesis and predictions (see also reply to the reviewers).

4) The statistical analyses need improvement: This includes revising models where critical random or control effects appear to have been left out to ensure results are robust, adding effect sizes for all main results as well as other concerns (see reviewer reports). The authors should also include full code for all analyses rather than only the model specifications.

We have reanalyzed the data as suggested by the reviewers (see reply to reviewers for details) and the results did not change. Therefore, our results are robust. We had included the colony (or sub-colony) identity as a random factor, but this was not clear probably because it was written at the end of the Materials and methods section.

We now include the full codes and model outputs for all the analysis (Supplementary file 1) using the template suggested by reviewer 2.

5) Additional concerns regarding the contextualization of the manuscript, clarity of the Materials and methods and reporting of results were brought up and can be found in the detailed reviewer reports below.

We have followed all the suggestions and we modified the Introduction and Discussion. We also clarified the Materials and methods and reported more details about the results.

Reviewer #1:

The author's present an excellent set of experiments investigating task partitioning and specialization in one species of ants that are well known for using tools to transport food. The novel aspect of this manuscript specifically tests whether individual personality characteristics are associated with tool-use propensity, which if I understand correctly, goes beyond the usual colony approach to such questions. Overall I found the paper well written, the experiments clearly described and well designed, and the results easy to understand. The paper is likely to be of interest to a wide range of animal behavioural ecologists and I think warrants publication in eLife. However, I do have a couple major concerns that would need to be addressed in a revision.

1) The Introduction is generally well written but there is a lack of integration for introducing the experiments and aims of the study. The paragraph that begins “ We performed three different experiments to study the tool use behaviour…” needs to be more detailed rather than simply outlining each experiment. Can the authors include more about the overall hypotheses of the study and the predictions for the experiments? What is the aim of each experiment with respect to your overall research questions? This will also better prepare the reader for what and why particular analyses will be done for each experiment.

Essentially I think the authors can do a much better job to present the results of their experiments in a more unified and compelling way with respect to specific research questions and hypotheses.

2) The above issue, of the experiments being presented distinct and separate rather than contributing to an overall research goal needs improvement also in the Discussion. In fact parts of the Discussion would be better suited (and would be very helpful) in the Introduction. For example theoretical framing and predictions. The first two paragraphs are highly repetitive of the results but around the middle of paragraph three this improves considerably. Rather, it would be nice for the Discussion to start by briefly summarize findings across experiments with their relative contribution to the goals/hypotheses of the study.

We re-wrote parts of the Introduction, which now includes the theoretical framing and predictions, as suggested. We have clarified the aims of each experiment and linked them to the overall hypotheses. We have substantially modified the Discussion as suggested and we removed repetitive elements.

3) My last concern includes missing random or control fixed effects in some of the mixed models used in the statistical analysis. I outline these below in detail but essentially without controlling for this additional substructure in the data the authors run the risk of producing erroneous results. Please have a look at the statistical suggestions and modify and re-run the relevant models where necessary.

Thank you. We followed the suggestions.

Reviewer #2:

This study asks an interesting and timely question ("which individual ants perform tool use, and is tool use a personality trait?"). The experiments are performed carefully. The writing is clear. The Introduction introduces the topic well. The full data are provided in a clear manner, which is fantastic. However, I have several concerns about the conclusions drawn from the data. Firstly, some of the results on which conclusions are based are not compared against a null expectation. Secondly, and most critically, the “tool use” personality assay may not reflect anything more than tool discovery, and may thus conceptually identical to the exploration assay. The Discussion is somewhat unfocussed. In terms of subjective interest, I judge the results of experiments 1 and 2 to be not exciting enough for eLife. The conclusions arising from experiment 3 are exciting, but may not be supported by the data. Addressing my major concerns may require more video and statistical analysis, but will not require new experiments to be performed.

1) "Only a small number of workers performed tool use" this is technically true, but throughout the manuscript there is the implication that the tool users are in some way special (Results: "only a few workers performed the tool use behaviour"). The “null model” here would be that any ant with sufficient information has the same probability of performing tool use. Specifically, an ant must encounter both the tools and the food to perform tool use. As the results presented show, once tool use starts, almost all the tools are “used up” by one or two individuals very quickly, not giving other ants a chance to perform tool use, even if they could. The proportion of ants performing tool use is given. To put this in context, we need to know the proportion of ants which encountered tools and food at all. This “null hypothesis” needs to be explicitly presented, and refuted or not, given the data.

We tested the hypothesis that “any ant with sufficient information has the same probability of performing tool use” by reanalysing the videos and checking whether the tool users were the first workers to obtain information about the location of both the food and the tools in Experiment 1. We found that an average of 11.35 ± 6.49 workers (mean ± SE; min = 3, max = 33 workers), contacted the food and the tools before the first tool user dropped the first tool into bait (see Table 3 for further details). These support the conclusion that only a few workers perform the tool use behaviour. Further evidence is given below.

2) All the results presented in paragraph five are given without context of a null model, and are thus meaningless. I.e., if you randomly distributed the same number of transport events and tool use events over the same number of individuals and trials, would you get similar proportions? The empirical results can be compared to randomised data in order to clarify whether these results represent more than probabilistic behaviour amongst homogenous ants.

We compared our empirical results to randomised data. The observed average number of individuals per sub-colony showing tool use was 8.25 (CI [6.47, 10.03]). This was notably lower than 10.2 (CI [10.18, 10.22]), which is the average number of tool users obtained by randomly assigning the same number of observed tool use events to a simulated ant population based on the same number of individuals, sub-colonies, and trials, and the confidence intervals do not overlap. This indicates that the individual distribution of tool use events was not random in our experiments and therefore our results represent more than probabilistic behaviour among homogeneous ants. Moreover, the occurrences of tool use were repeatable at the individual level (RICC = 0.218, p < 0.001).

3) "The consecutive transport of several tools within one trial enhanced the efficiency of a tool using worker." This is interpretation, not results, and alternative null hypotheses are not excluded. It could be that when an ant “feels” like being efficient, it transports many tools and does so quickly. In other words, the “second tool” group (Figure 1B) may contain both slow ants which only manage 2 trips, and fast ants which manage 4. The “fourth tool” group only contains ants which manage at least 4 trips. Moreover, and more critically, it makes simple sense that when an ant works faster, it can manage more tools in a given time. To truly demonstrate improvement over time, only ants which manage 4 transport events should be considered, and the rank of the transport event used to predict transport speed. This may or may not have been done, the description of the analysis was not clear. The raw data is not separated by individual ant ID, so I could not check this myself.

We agree with the reviewer and we have redone the analysis by including only those ants that performed 4 consecutive transports. Indeed, the majority of ants in the previous dataset transported 4 tools (18 individuals out of 21, 54 vs. 60 tool transports). The results were similar to the previous ones: within trial, the time to transport one tool decreased with the number of tools transported (second-third tool: LMM t = -2.25, N = 54, p = 0.03, secondfourth tool: t = -3.75, p < 0.001). However, the average time to transport one tool did not change significantly between different trials (first-second: t = -0.015, N = 29, p = 0.99; first-third t = 1.02, p = 0.32).

We have provided the individual ant ID in the raw data.

4) The effect size of the key finding is not given. Examining the data, the normalized personality score ranges from -4 to 3 – so 7 in total. The non-tool users have an average score of -0.33. The tool users have a mean score of 0.57. I managed to recreate the analysis, and found an effect estimate of 0.6174. Using emmeans, we get a prob estimate of 0.349. This means that while the effect is highly significant, I would not say the effect size is very large.

We now give the full output of the models (Supplementary file 1). The

CI95% of the personality score of tool users (mean = 0.57, CI [0.26, 0.88]) and non-tool users (mean = -0.33, CI [-0.56, -0.09]) does not overlap. Even if the effect size is not extremely large, we believe that our results show a clear division of personality scores between the two groups. This is also evident from the new Figure 5C, which we plotted as the reviewer suggested.

5) I am not convinced that the “tool use” assay in experiment 3 is showing anything more than tool and food discovery probability. It stands to reason that ants which spend more time in the centre of the arena and less time on the periphery (the main components of the personality score) are more likely to encounter both the tools and the food first, and thus be in a position to being doing tool use, perhaps using up all the tools before the other workers found them, but definitely reducing the probability of other workers finding the tools (as fewer tools remain). The authors need to demonstrate that ants with a higher personality score are more likely to perform tool use even when controlling for opportunity. Otherwise, the seemingly interesting results boil down to "ants which spend more time in the centre of the arena are more likely to find things in the centre of the arena", which is not a very exciting result.

We give several lines of evidence that engaging in tool use is not simply the consequence of encountering the tools. From Experiments 1 and 2 (see above) our new analyses show that tool use is not a random process performed by those workers who were the first to obtain the information about the location of the food and the tools. We also obtained new data from the videos of Experiment 3. Our results point in the direction that being more explorative is not enough for a worker to become a tool user. First, it should be noticed that the tool use process did not start immediately after the tools and the food were offered to the ants. The latency to bring the first tool to the bait was 894.71 ± 153.49 seconds (mean ± SE, N = 21 trials), which gives plenty of time for all the ants to explore and find the tools. From a total of 21 trials, only in 5 occasions the tool user was the first worker to investigate the tools. In the majority of the trials (16 out of 21) an average of about five workers investigated the tools by antennation before the first tool user but did not pick them up (4.94 ± 0.52 (mean ± SE) workers, min = 1 worker, max = 8 workers). This indicates that tool use behaviour is not merely prompted by the fact of finding the tools before other workers. As for the personality traits, we clarified that the time spent in contact with the prey is not merely linked to the exploratory activity (see our reply to the first Editors’ concern) and that the ant behaviour is consistent across contexts (see reply to reviewer 3). Moreover, tool use is repeatable at the individual level (data Experiment 2). Therefore, we believe that the relationship between tool use and personality is genuine and informative.

Reviewer #3:

In this manuscript, the authors perform three experiments of Aphaenogaster senilis using tools (e.g. debris) to soak up a liquid food resource to be carried to their nest. The primary conclusion is that a small subset of workers uses these tools, and they do so because of specific personality traits. I found the study to be quite interesting and the paper to be well-written. The authors have strong support for most of their conclusions, and indeed for the first part of their main conclusion, that only a small subset of workers perform the task, but I am less confident in the authors' evidence for the second part of that main conclusion, that these differences are due to specific personality traits.

My primary concern is in how the authors measured personality traits in workers. They performed each of two tests on two occasions: an open field test and test in which individuals could interact with prey. They found that certain behavioral traits (e.g. time spent in central arena in open field and time spent interacting with prey) were significant repeatable across the two timepoints, and conclude that these are personality traits. They then use a PCA to show that these behavioral traits correlate with tool use (ants that explored more were more likely to use tools). This consistency over time is necessary for personality, but as far as I can tell the authors did not test whether these behaviors were consistent across contexts, which is also important (I might also prefer to see more than 2 timepoints tested, though this is a minor concern). So there is a behavioral trait that is correlated across time but not contexts; I do not find this to be rigorous evidence of personality. In addition to the possibility of it being "personality" driving tool use, a non-mutually-exclusive possibility is that ants who explore more are more likely to find the tools sooner, thus becoming the tool-users, which I'm glad that the authors also discuss as a likely explanation.

I must admit that I am generally a little sceptical of the special attention sometimes placed on the idea of animal personality. To me, it is not inherently more interesting if tool use is due to a behavioral trait that is or is not consistent across contexts. Indeed, I would find the study just as interesting if the authors did not discuss personality, per se (though it is possible this would make it less broadly appealing). And I think the conclusion would be better supported and more compelling.

Other than this concern related to personality, I found the conclusions to be generally well supported. I thought it was interesting that workers improve in tool-use performance within but not across trials, and I think the authors suggestion that this relates to route-learning is a good one.

We would like to thank the reviewer for the general appreciation of our work and for the helpful comments. We formally tested the correlation between the variables collected in the exploratory behaviour assay and the reaction to prey assay and we show that they are correlated. The time spent in contact with the prey was negatively correlated with the time spent walking in the periphery (rs = -0.3, N = 154, p < 0.001) and positively correlated with the total time spent in the central area (rs = 0.33, N = 154, p < 0.001), indicating that the more exploratory ants were also those more interested in the prey. This shows that there is consistency across contexts. Therefore, we believe that we can talk about personality traits.

[Editors’ note: what follows is the authors’ response to the second round of review.]

Revisions:

The reviewers were all very content with the revised submission and congratulate the authors on substantially improving the manuscript by seriously taking into account our initial feedback. We had only a handful of major comments left that should be addressed before publication.

1) Please explain why exploring the periphery of the plate in your open field test would not in fact be ants that are the most exploratory and how this affects your inference about tool users being more explorative. Those ants who explore only in the center could actually be argued to be less explorative. Perhaps this trait needs rephrasing?

The open-field test is a common measure of exploratory behavior and anxiety in rodents (e.g., Prut and Belzung, 2003). The test used in our study is an adaptation of the classical open-field test. We use it to quantify exploratory activity, namely the tendency of an individual ant to thoroughly explore a novel environment. We found that this behavior is usually significantly repeatable in ants at the individual level. We used our ant open-field test in other studies, involving different ant species, to tackle different research questions. For instance, learning performance (Udino et al., 2017), cognitive judgment bias (d’Ettorre et al., 2017), individual behavioral type, and group performance (Carere et al., 2018), all cited in our manuscript. In this last study, we found that individual exploratory activity in an open-field arena was associated with the performance at the group level in cocoon recovery, thus suggesting a linear link between individual and collective behavior.

In contrast to individual ants that spend more time in the central area, ants that preferentially walk along the periphery of the open-field arena are typically looking for protection (walls) and perhaps for the possibility to escape. These ants cross the central area rapidly and therefore do not thoroughly explore the novel environment. Their possible motivation to explore is affected by their “fear” of the novel environment (please note that I use anthropomorphism here, only in order to make the picture clearer). We might even interpret this as an indication of “anxiety” in ants, analogous to rodents, but we do not (yet) dare to do so. It is also very incautious to use words such as “stress” referred to ants, but the concept is similar. Some ants are more “stressed” than others in the novel environment, therefore they explore less and spend more time at the periphery, where it is safer to walk. It is not judicious to use this phrasing in the manuscript. To be able to openly use such wording for ants (anxiety, stress, etc.), several years of research are still needed.

In the present case, A. senilis ants that spend more time in the central area were typically less “stressed/anxious” and showed a slow and detailed exploration of the open-field arena. These “bolder” and more explorative ants did not explore only in the center. Typically, they spent some time walking in the periphery as well, and thus exploring also this part of the arena. The so-called less explorative ants, instead, spent most of their time repeatedly walking along the edges at the periphery. Such individual differences between ants (i.e. being more explorative and “curious” towards novel objects/environments) may play a key role in relation to tool use. More explorative ants will find and approach novel tools easily, which -in turn- may result in a higher probability of using tools than less explorative ants.

We added a short explanation about the open-field in the Material and Methods:

“Open-field. This is an adaptation of the classical open-field test developed to test exploratory behaviour and anxiety in rodents (e.g., Prut and Belzung, 2003) and already used with ants (d’Ettorre et al., 2017; Udino et al., 2017; Carere et al., 2018). An ant was individually placed in an acclimatization tube (Ø 2.7 cm) for 1 min at the centre of a circular arena (Ø 11.5 cm) with a floor of clean filter paper (replaced after each trial), in which an area of 9.5 cm diameter was considered as the central zone and the external part as the periphery (Figure 3A). Then, the tube was removed and the behaviour of the ant was observed for 3 min. More exploratory ants are expected to spend more time in the central area, while less exploratory ants will spend more time walking along the edges of the arena, where they are protected by the walls.”

2) Can you strengthen the claim that tool users are the same individuals across trials (i.e., repeatability across trials) or at least make this more clear and transparent? A corollary of the personality-tool use conclusion is that you should have more individuals repeating tool use across multiple trials than expected by chance. This should mean that the same ants use tools over and over again. We are convinced that this is true at least within a trial, but do we see it across trials, as we should? The authors say that we do and provide evidence from experiment 1, in Table 2, and from experiment 2. However the evidence for experiments 1 is weak if we are interpreting Table 2 correctly, in total for the 3 colonies they saw only 2 ants transport tools to the bait in multiple trials (out of 19 marked tool users), and only 3 ants transport tools to the nest in multiple trials (out of 17 marker tool users). Is this actually more than would be expected by chance?

Indeed, experiment 3 provides the best evidence of personality predicting tool use across trials but given the low absolute numbers in repeat tool users across trials for Experiments 1 and 2 we would like the authors to discuss this aspect more critically in the manuscript.

We understand the concerns of the reviewers but we do not fully agree with the above.

We believe that both Experiments 2 and 3 provide multiple and convincing evidence of repeated tool use by the same individuals across multiple trials. Experiment 2 was indeed designed to test the occurrence of repeated tool use within and across trials. We found that 33.3% of the tool using workers participated in more than one consecutive trial: “Of the 99 tool users in total, 64.7% performed repeated tool transports within the same trial and 33.3% participated repeatedly in more than one trial (Table 4). The majority of these ants (26 over 31 workers) participated in 2 trials (2.89 ± 0.42 per trial, mean ± SE), 3 workers participated in 3 trials and 2 workers in 4 trials (Table 5).” This percentage is even higher if we consider the final trial with 20 tools. In this trial, performed after one week, more than 50% of the workers showed repeated tool use across trials: “Half of the workers (50.38%) that performed tool use in the final trial with 20 tools, also participated in at least one of the previous trials with 10 tools (Table 6), thus confirming that the same ants use tools over and over again, including across trials.” In Experiment 2, we showed statistically that tool use is repeatable across trials at individual level, as stated in the previous version of the manuscript: “The occurrences of tool use were repeatable at the individual level (RICC = 0.218, p = 0.001).” The formulation of this statement was not incisive, we changed this sentence as follows: “It is important to note that the occurrences of tool use were repeatable across trials at the individual level (RICC= 0.22, N = 1880, p = 0.001).” We also specified “across trials” in the corresponding sentence in the Discussion: “Moreover, tool use was a significantly repeatable behaviour across trials at the individual level.”

Repeated tool use across trials is further supported by the findings of Experiment 3, where 66.6% of workers used tools repeatedly across trials, despite the fact that they had the chance to do so only once, in the final trial with 20 tools: “In the final trial with 20 tools, a total of 18 workers used tools. Of these, 12 workers (66.6%) previously performed tool transport and they brought 106 (66.25%) of the 160 tools that were transported to the bait in this last trial. Of the 34 workers removed during the first 3 days of the experiment, 11 resumed tool using in this last trial, and 4 of these workers were active during the first trial with 10 tools seven days earlier.”

Now, to address the specific criticism (the evidence for experiments 1 is weak): Experiment 1 was an exploratory experiment in which we observed entire colonies where workers were not all individually marked. These ants have spines on the thorax and the paintmark goes away easily. Because of the presence of the spines, other methods for marking ants are difficult to use. Marking with numbers/plates on the gaster (abdomen), instead of the thorax, is not optimal since it may impede movements, particularly in this ant genus. Aphaenogaster (meaning: not showing the abdomen, from Greek: a, α privative + phaino, appearing/shining/coming to view + gaster, abdomen) is very different from Camponotus or Formica, or even Temnothorax, where the marking is relatively easy, also when applied on the top of the gaster. Therefore, we had to do the best of a bad job concerning individual marking in A. senilis. We will improve in this technical endeavor.

Contrary to Experiments 2 and 3, Experiment 1 intended to characterize the ant behavior during the two parts of the tool use process and to study the possible relationship between the number of workers using tools and the number of foragers in the arena. Observing repeated participation of the workers in the tool use process was accessory because the paint mark of the individual workers could be removed by self- and allo-grooming (as explained above). The fading of the paint marks contributed to the low number of repeated observations of tool use across trials (see Table 2). The reviewers rightly ask whether the values observed in relation to the repeated tool transport performed by the same workers across trials are higher than what would be expected by chance (by chance: 4.22). This answer is no, because these observations in Experiment 1 were affected by the fading of our pant marking. However, Table 2 was probably not easy to read, and it has been misinterpreted. In Experiment 1, there was a total of 16 marked tool users transporting tools to the bait (Table 2, column 4), and no tool users were observed to transport tools inside the nest multiple times. The number of ants transporting tools to the nest across multiple trials was confounded with the number of tool users that transported more than one tool to the nest within a single trial (column 8). We slightly modified the headings in the table to improve clarity.

3) Perhaps related to the above point, some of the results should be discussed with more transparency and caution. Specifically, can you acknowledge that effect sizes are small for the personality results and add why this small effect is still important. Please also address in the manuscript that your results for a lack of task partitioning are still not conclusive (due to the low number of ants that complete both parts of the tool use task, i.e. Table 2) and ideally requires further research.

We agree with the reviewers that Results should always be discussed with transparency and we would like to emphasize that any ambiguity that may appear in the manuscript is unintentional. We refer to the small effect sizes and their importance in the Discussion: “Despite we found relatively small effect sizes in the analysis of the personality data, our results reveal the importance of even slight individual differences in behavioural traits, if they are consistent, in the organization of social life.” We also addressed the lack of conclusive evidence regarding task partitioning in tool use and stated that this requires further research: “Therefore, as we predicted, there was no clear evidence of task partitioning. Nevertheless, our results regarding the lack of task partitioning are not conclusive due to the relatively small total number of ants that were observed participating in both parts of tool use process. To the best of our knowledge, the observations reported in the literature also lack conclusive evidence. In A. rudis, similarly to A. senilis, no evidence of task partitioning was found (Banschbach et al., 2006), although task partitioning was suggested in earlier observations of A. famelica (Tanaka and Ono, 1978). This suggests that this behavioural aspect might be species-specific but ideally requires further research.”

4) Please add the latency to find prey and contact time correlation that was in the revisions letter to the reviewers but was not added to the manuscript. It would also be important to provide the correlation for all individuals, irrespective of whether they are tool users or not. Please also ensure the raw data for this correlation is included in the excel files.

The required information was added to the manuscript: “If this would be the case, in the reaction to prey test we should expect a negative correlation between the latency to find prey and the time spent in contact with it (the shorter latency, the longer contact time). In contrast, we found a positive correlation between the latency to find the prey and the time spent in contact with the prey: the longer the latency, the longer the time spent with the prey (session 1: rs = 0.27, N = 154, p < 0.001; session 2: rs = 0.42, p < 0.001).”

The description of the analysis was also added to the Materials and methods section (Statistical analysis): “The correlation between the latency to find the prey and the time spent in contact with the prey measured during the reaction to prey test (separately for the two sessions) was performed with Spearman rank test (data not normally distributed)”. The data were included in the excel file.

https://doi.org/10.7554/eLife.61298.sa2

Article and author information

Author details

  1. István Maák

    1. Department of Ecology, University of Szeged, Szeged, Hungary
    2. Museum and Institute of Zoology, Polish Academy of Science, Warsaw, Poland
    Contribution
    Conceptualization, Formal analysis, Supervision, Investigation, Methodology, Writing - original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0999-4916
  2. Garyk Roelandt

    Laboratory of Experimental and Comparative Ethology UR 4443, University Sorbonne Paris Nord, Villetaneuse, France
    Contribution
    Formal analysis, Investigation, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Patrizia d'Ettorre

    1. Laboratory of Experimental and Comparative Ethology UR 4443, University Sorbonne Paris Nord, Villetaneuse, France
    2. Institut Universitaire de France (IUF), Paris, France
    Contribution
    Conceptualization, Formal analysis, Supervision, Funding acquisition, Investigation, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    d-ettorre@univ-paris13.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8712-5719

Funding

Institut Universitaire de France (Senior Member Grant)

  • Patrizia d' Ettorre

Washington University in St. Louis (Vistiting Professor Fellowship)

  • Patrizia d' Ettorre

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Xim Cerdá and A de Fouchier for collecting ant colonies in Doñana National Park, Paul Devienne for technical assistance. We are grateful to Heiko G Rödel for help with the statistical simulation. Data of Experiment three are part of the Master thesis of GR. Many thanks to the Editors and three Reviewers for their very helpful comments. The study was funded by a grant of Institut Universitaire de France (IUF) to PdE. The first substantial revision was written during her semester residency at Washington University in St Louis as Clark Way Harrison Visiting Professor, hosted by the inspiring lab of Joan E Strassmann and David C Queller.

Senior Editor

  1. Christian Rutz, University of St Andrews, United Kingdom

Reviewing Editor

  1. Ammie K Kalan, Max Planck Institute for Evolutionary Anthropology, Germany

Reviewer

  1. Tomer J Czaczkes, University of Regensburg, Germany

Publication history

  1. Received: July 21, 2020
  2. Accepted: November 20, 2020
  3. Version of Record published: December 9, 2020 (version 1)

Copyright

© 2020, Maák et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,272
    Page views
  • 112
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Ecology
    2. Evolutionary Biology
    Hanna M Bensch et al.
    Research Article

    Living with relatives can be highly beneficial, enhancing reproduction and survival. High relatedness can, however, increase susceptibility to pathogens. Here, we examine whether the benefits of living with relatives offset the harm caused by pathogens, and if this depends on whether species typically live with kin. Using comparative meta-analysis of plants, animals, and a bacterium (nspecies = 56), we show that high within-group relatedness increases mortality when pathogens are present. In contrast, mortality decreased with relatedness when pathogens were rare, particularly in species that live with kin. Furthermore, across groups variation in mortality was lower when relatedness was high, but abundances of pathogens were more variable. The effects of within-group relatedness were only evident when pathogens were experimentally manipulated, suggesting that the harm caused by pathogens is masked by the benefits of living with relatives in nature. These results highlight the importance of kin selection for understanding disease spread in natural populations.

    1. Ecology
    2. Microbiology and Infectious Disease
    Silvia Brochet et al.
    Research Article

    Ecological processes underlying bacterial coexistence in the gut are not well understood. Here, we disentangled the effect of the host and the diet on the coexistence of four closely related Lactobacillus species colonizing the honey bee gut. We serially passaged the four species through gnotobiotic bees and in liquid cultures in the presence of either pollen (bee diet) or simple sugars. Although the four species engaged in negative interactions, they were able to stably coexist, both in vivo and in vitro. However, coexistence was only possible in the presence of pollen, and not in simple sugars, independent of the environment. Using metatranscriptomics and metabolomics, we found that the four species utilize different pollen-derived carbohydrate substrates indicating resource partitioning as the basis of coexistence. Our results show that despite longstanding host association, gut bacterial interactions can be recapitulated in vitro providing insights about bacterial coexistence when combined with in vivo experiments.