Task specialization in two social spiders, Stegodyphus sarasinorum (Eresidae) and Anelosimus eximius (Theridiidae)


Correspondence: Virginia Settepani, Department of Bioscience, Ecology and Genetics, Aarhus University, Ny Munkegade 116, building 1535, office 216, 8000, Aarhus, Denmark.

Tel.: +45 87156597; fax: +45 87154326;

e-mail: virginia.settepani@biology.au.dk


Understanding the social organization of group-living organisms is crucial for the comprehension of the underlying selective mechanisms involved in the evolution of cooperation. Division of labour and caste formation is restricted to eusocial organisms, but behavioural asymmetries and reproductive skew is common in other group-living animals. Permanently, social spiders form highly related groups with reproductive skew and communal brood care. We investigated task differentiation in nonreproductive tasks in two permanently and independently derived social spider species asking the following questions: Do individual spiders vary consistently in their propensity to engage in prey attack? Are individual spiders' propensities to engage in web maintenance behaviour influenced by their previous engagement in prey attack? Interestingly, we found that both species showed some degree of task specialization, but in distinctly different ways: Stegodyphus sarasinorum showed behavioural asymmetries at the individual level, that is, individual spiders that had attacked prey once were more likely to attack prey again, independent of their body size or hunger level. In contrast, Anelosimus eximius showed no individual specialization, but showed differentiation according to instar, where adult and subadult females were more likely to engage in prey attack than were juveniles. We found no evidence for division of labour between prey attack and web maintenance. Different solutions to achieve task differentiation in prey attack for the two species studied here suggest an adaptive value of task specialization in foraging for social spiders.


Individual behavioural variation within groups of social animals can be highly advantageous in terms of overall increase in group fitness (Oster & Wilson, 1978; Wallace, 1982). In particular, task specialization is expected to result in enhanced colony efficiency and productivity, decreased individual energy expenditure to reach a communal outcome and improved resource allocation between colony members (Wilson, 1987; Schmid-Hempel, 1991; Rypstra, 1993; Ridley & Raihani, 2008). Task specialization can take many forms, from behavioural asymmetry in task performance to strict division of labour as seen in the eusocial insects, where behaviourally or morphologically specialized castes determine the limited behavioural repertoire of individuals. Eusocial insects are characterized by extreme reproductive skew, where the majority of individuals forego their own reproduction to perform other colony tasks, often coupled with division of labour where morphological and/or behavioural castes specialize in different colony tasks (Oster & Wilson, 1978; Beshers & Fewell, 2001). This high level of specialization has been recognized as one of the main explanations for the worldwide ecological success of social arthropods (Wilson, 1987). Facultative cooperative behaviour, when individuals differ in their propensity to cooperate, is important for group performance in other social organisms and has been described for several different taxa ranging from microbes to humans (Bergmuller et al., 2010). For example, individual differences in social tendencies have recently been found in social spiders (Pruitt et al., 2011), a group that is not commonly associated with task differentiation (Ward & Enders, 1985; Wickler & Seibt, 1993; Vakanas & Krafft, 2001; Ainsworth et al., 2002). Pruitt & Riechert (2011) found that colonies of the social theridiid spider Anelosimus studiosus increase their overall reproductive output when composed by a mixture of ‘social’ and ‘asocial’ phenotypes, suggesting a potential adaptive value of behavioural asymmetries in social spiders.

The great majority of spiders are solitary and highly aggressive even towards conspecifics, and only about 23 of more than 42.700 described spider species (Platnick, 2012) are permanently social (Agnarsson et al., 2006). Social spiders are characterized by a highly female-biased sex ratio (Avilés, 1997) and lack of pre-mating dispersal, which results in intracolony mating and extremely high levels of inbreeding (Lubin & Bilde, 2007). Similar to the eusocial insects, individual spiders spend their entire lifetime in communal nests where they cooperate in colony activities such as brood care, prey capture, nest building and web maintenance (Lubin & Bilde, 2007). However, unlike the eusocial insects, social spiders do not have overlapping generations, defined as the overlap of mother and offspring adulthood (Wilson, 1971), and females are seemingly totipotent, that is, potentially able to perform all tasks in the colony, including reproduction (Avilés, 1997). The occurrence of a discrete caste system has never been demonstrated in spiders (Lubin & Bilde, 2007); however, age-related behavioural differentiation has been shown for two social species in the family Theridiidae (Christenson, 1984; Lubin, 1995).

Cooperation allows social spiders to capture prey much larger than their own body size (Nentwig, 1985; Yip et al., 2008; Holm, 2010), with several individuals within a group feeding on the prey simultaneously. Whereas this allows a broadening of the dietary niche to meet increasing resource demands of the group, group living also induces competition and a risk of conflict over the distribution of resources (Ulbrich & Henschel, 1999; Whitehouse & Lubin, 1999). Resource distribution is an ecological key factor influencing group dynamics (Packer & Ruttan, 1988). Within-group competition for limited and concentrated resources might result in asymmetrical access to food and differential feeding success, which in turn determines body size variation among colony members (Vollrath & Rohde-Arndt, 1983; Christenson, 1984). Body size is an important predictor of spiders' competitive ability (Riechert, 1986), and uneven resource distribution could therefore influence behavioural asymmetries in social spider colonies. Hence, smaller individuals without sufficient resources for reproduction might specialize in nonreproductive tasks, leaving reproduction to larger group members (Vollrath & Rohde-Arndt, 1983; Rypstra, 1993; Ebert, 1998). Alternatively, task differentiation can be obtained through age-related differences, where individuals accrue or switch tasks with age and development (Christenson, 1984; Lubin, 1995).

Kin selection theory predicts that the amount of help provided is strictly linked to genetic relatedness between helpers and receivers, with cooperation preferentially directed towards close kin (Hamilton, 1964a,b; Gardner et al., 2011). In social spiders, the lack of pre-mating dispersal and the within-colony mating system create extreme levels of inbreeding (Riechert & Roeloffs, 1993; Smith & Engel, 1994; Avilés, 1997), with relatedness coefficients that can be much higher than that of full-sibs from unrelated parents (Lubin & Bilde, 2007). Thus, a helper that does not reproduce itself but enhances reproductive success of close relatives can gain indirect fitness benefits. Importantly, helpers have no future reproductive chance as both reproducers and helpers die following the reproductive season; therefore, direct benefits of helping should be negligible. Individual specialization in colony tasks is expected to be evolutionarily advantageous in such a system, if specialization leads to increased colony productivity and inclusive fitness benefits.

We investigated individual-level task differentiation in two nonreproductive colony tasks, that is, prey capture and web maintenance, in two independently derived social spider species, the old world Stegodyphus sarasinorum Karsch (family Eresidae) and new world Anelosimus eximius Keyserling (family Theridiidae) (Kullmann, 1972). These tasks involve potentially high costs of resource use in terms of materials such as digestive enzymes, poison and silk, and energy required to perform the activities and to produce the materials (Mommsen, 1978; Pasquet et al., 1999), as well as potential risk of injury from prey and increased risk of predation (Rayor & Uetz, 1990; Willey & Jackson, 1993). We examined task specialization by marking spiders involved in prey attack (or web maintenance) over three consecutive days and analysing their propensity to engage in repeated prey attack. If prey attack co-varies with hunger level (Nakamura, 1987), satiated spiders might be expected to be risk-averse and avoid attacking whereas starved spiders should be risk-prone (Pasquet et al., 1999 and references within; Scharf et al., 2010). Hence, if the spiders do not have task differentiation, and individual behaviour is state-dependent, marked individuals would attack prey or repair the web within the same day or the following days either less frequently than expected by chance due to satiation or at the same frequency as expected by chance. If spiders engage in task differentiation, the same individuals are expected to perform the same task repeatedly, irrespective of hunger level. Additionally, we examined instar and body size of spiders that performed a task relative to the average colony instar/body size composition to examine whether individual age or unbalanced distribution of resources (reflected in body size) were associated with behavioural differentiation.

We found evidence for task specialization in both species, but achieved in distinctly different ways: in Stegodyphus sarasinorum, individual spiders that attacked prey once were more likely to attack prey again, whereas in Anelosimus eximius specialization occurred according to instar when adult and subadult females were more likely to engage in prey attack than juveniles. These data from two social species that represent independent origins of sociality suggest that task differentiation may be a common solution to maximizing benefits of group living, although the adaptive benefits of task specialization remain to be quantified.

Materials and methods

Study organisms

Stegodyphus sarasinorum Karsh (Eresidae) is a social spider that lives in arid and semi-arid habitats throughout India and some adjacent countries (Jacson & Joseph, 1973; Platnick, 2012) in colonies that contain up to several hundred individuals. This genus contains two additional African social species, and the social species are generally found in seasonal habitats which determine their strongly seasonal developmental cycle (Seibt & Wickler, 1988). Hence, egg hatching and gerontophagy, that is, when all colony females including nonreproducing helpers are consumed by the offspring, occur roughly at the same time within colonies so that all the individuals in a colony approximately belong to the same instar (Crouch & Lubin, 2000; Lubin et al., 2009). The nests are compact silky masses with prey remains, leafs and exuvia incorporated and built on spiny shrubs. Each nest is surrounded by two-dimensional capture webs (Kaston, 1965; Bradoo, 1980). Usually, several nests located close to each other are interconnected by common capture webs [‘colony clusters’ (Avilés, 1997) or ‘polydomous’ colonies (Henschel, 1998)]. Stegodyphus sarasinorum shows dichotomy in its activity patterns: during the heat of the day, spiders hide inside the nest unless a prey is caught in the web, whereas most spiders can be found active outside the nest at dusk and dawn performing web maintenance by enlarging, cleaning and repairing the capture webs (Jacson & Joseph, 1973; Christenson, 1984; Pasquet & Krafft, 1992). Stegodyphus sarasinorum was studied in Andhra Pradesh, southern India (78°15′24″E, 12°49′35″N; 820 m elevation), from September to November 2010.

Anelosimus eximius Keyserling (Theridiidae) is a social species that inhabit rain forests of Central and South America ranging from Panama to southern Brazil (Levi, 1963; Platnick, 2012), and this genus contains the majority of described social species (Agnarsson et al., 2006). Anelosimus eximius create the largest colonies of the social species, reaching sizes of up to several metres in length and width and containing up to several thousand individuals (Avilés, 1997). A colony can have several reproductive cycles through the year (Avilés, 1986), which result in different instars occupying a nest at the same time, but with no overlap between females and adult daughters. The nests are basket-like structures with incorporated leaves, and long capture threads connected with the vegetation create a three-dimensional capture web extending from the top of the basket (Christenson, 1984; Nentwig, 1985; Avilés, 1997). Nests are built on bushes or trees along roads, rivers or in vegetation gaps in rain forest habitats (Avilés, 1997; Ebert, 1998). Anelosimus eximius shows the same bimodal activity pattern as described for S. sarasinorum regarding web maintenance activity (Ebert, 1998). We studied A. eximius in Ecuador, Napo province (77°48′57″W, 0°59′20″N, 420 m elevation) in June–July 2011.

Experimental setup

We studied 30 S. sarasinorum nests with colony sizes ranging from 20 to 265 individuals and 22 A. eximius nests with colony sizes ranging from 44 to 881 individuals. The experiments were performed during three consecutive non-rainy days for each nest. The experimental set-up can be divided into three parts:


To examine the propensity of spiders to engage in repeated prey attack, we performed marking experiments over three consecutive days. We collected grasshoppers and crickets (Orthoptera), from the area surrounding the nests and used them as prey. Prey length was measured to ±0.01 mm with a digital calliper and placed in the capture web within 10 cm of the nest retreat being careful not to shade or shake the nest. We used prey that was abundant in the surrounding environment and likely to be part of the natural diet of both spider species. All spiders that left the retreat to attack the prey within five minutes from when the first spider attacked the prey were marked with water-based paint on the abdomen using a laboratory pipette. We used a different colour for each of the 3 consecutive days.

Due to the relatively larger body size of Stegodyphus spiders and the two-dimensional shape of their capture web, it was possible to apply the colours directly onto these spiders while they were attacking, seemingly without affecting their task performance. For the smaller Anelosimus spiders attacking in three-dimensional webs, it was necessary to collect the spiders with forceps to mark attacking individuals. Marked spiders were immediately returned to their colony. To assess the effect of marking, we closely observed whether changes in behaviour were detected upon marking prior to data collection. Stegodyphus sarasinorum individuals that were marked in the process of feeding continued to feed during marking, without any obvious sign of disturbance. Similarly, A. eximius individuals that were collected, marked and returned to their colony often subsequently returned to feed on the prey. Observations of web maintenance behaviour confirmed that the marking technique did not appear to affect the spiders' behaviour. Of the total number of spiders observed attacking prey on a given day, an average of 45% (S. sarasinorum) and 30% (A. eximius) of them engaged in web maintenance on the same day. Hence, marking did not appear to alter the focal behaviours of spiders in these experiments.

We asked the following two questions:

  1. Does a spider attack for two days in a row? To answer this question, we combined data on spiders marked on Day 1 that also attacked on Day 2 and spiders marked on Day 2 that also attacked on Day 3 (Fig. S1 in Supporting Information). We analysed whether individual spiders attacked two days in a row more frequently than would be expected by chance, given the colony size and overall number of attackers.
  2. Does a spider attack twice within three days? To address this question, we analysed data on spiders marked on Day 1 or 2 that also attacked on Day 3 (Fig. S1). We analysed whether individual spiders attacked more frequently twice across three days than would be expected by chance, given the colony size and overall number of attackers.

For both questions, we expected marked and unmarked individuals to attack with the same probability in the absence of task differentiation. In the case of task differentiation, we expected to observe marked individuals attacking more frequently than unmarked individuals. If S. sarasinorum that attacked also fed on the prey, and hunger state rather than task differentiation determines who attacks, we would expect marked (and therefore satiated) spiders to be risk-averse and avoid attacking in consecutive days.

Web maintenance

Observations of spiders performing web maintenance were made each day around the time of dusk (from 5 to 7 p.m.). For each colony, we recorded the total number of spiders outside the nest retreat engaged in repairing the web and counted the number of marked individuals among them. This was done three times in five-minute intervals.

We asked two questions:

  1. Does a spider perform both attacking behaviour and web maintenance? To answer this question, we used the data from Day 3, as the last experimental day contained the majority of information because spiders in the colony at that point were marked through three consecutive days (Fig. S1). For each colony, we calculated the probability of marked spiders to engage in web maintenance using the maximum number of marked spiders observed in prey attack and the maximum number of spiders performing web maintenance.
  2. Does a spider that attacks also perform web maintenance in the same day? To address this question, we used data from Day 1, as a marked spider performing web maintenance on the first experimental day necessarily had been an attacker on the same day (Fig. S1).

As marked individuals represent attackers, we expected them to engage in web repairing significantly less than unmarked individuals if there is task specialization. Without division of labour, we expected to observe marked and unmarked individuals maintaining the web with the same probability.

Body size and age

At the end of the three experimental days, the colonies were collected and dissected, all spiders were counted and marked individuals identified. In S. sarasinorum where each colony is inhabited by individuals belonging approximately to the same instar, we focused on body size as a measure of differences amongst colony members. Body size in S. sarasinorum was measured as prosoma (head) width to ±0.01 mm. The prosoma is a useful measure of body size being a sclerotized structure that indicates structural size (Foelix, 1996; Jakob et al., 1996). We used a digital calliper to measure all the marked individuals as well as for 15 randomly chosen unmarked individuals. We asked the question: Do spiders that engage in prey attack differ in body size from that of a random sample of unmarked colony members?

For A. eximius where different instars live together in the same colony, we focused on individual instar as a measure of differences amongst colony members. We asked the question: Is the group of attackers composed of the same proportion of different instars as that of the entire colony? To address this question, we recorded the instar of each single spider in each colony and categorized them as follows: (i) juveniles (including all four juvenile instars), (ii) pre-subadult females (i.e. the second last instar before adulthood), (iii) subadult females (i.e. the penultimate instar) and (iv) adult females. In social spiders, males do not participate in any of the cooperative activities in the colony (Lubin & Bilde, 2007) and were therefore excluded from the analysis. Instars in A. eximius are discrete (Avilés, 1986) and recognizable on visual inspection. However, to avoid potential mistakes in the limited sample of spiders engaged in attacking, we verified the instar of all marked individuals. To be comparable with the previous studies in this species, we did not use prosoma width as proxy for A. eximius instars, but we examined the epigyne (genital plate) under a stereo microscope and measured the combined length of tibia and patella on leg pair I according to the description provided by Avilés (1986). One colony of A. eximius was destroyed by rain just before collection and was not considered in the instar composition analysis because marked individuals were no longer present or recognizable.

Statistical analysis

Attack and web maintenance

Observations were modelled following an extended hypergeometric distribution (Johnson et al., 2005). In the following description of the methods, we consider attacking behaviour, but the model is the same for web maintenance, only ‘attacking’ should be substituted with ‘performing web maintenance’. In our model, each unmarked spider has a probability pu of attacking on a subsequent occasion, and each marked spider has a probability Pm of attacking on a subsequent occasion. The relation between the two probabilities is explained in terms of their odds: Pm/(1 − Pm) = ωPu/(1 − Pu), where ω is a positive parameter. If ω = 1, marked and unmarked spiders have the same probability of attacking on a subsequent occasion; if ω < 1, marked spiders have a smaller probability than unmarked spiders of attacking on a subsequent occasion; and if ω > 1, marked spiders have a larger probability than unmarked spiders of attacking on a subsequent occasion. We call ω the propensity parameter because it describes the propensity of marked spiders to attack on following days.

The conditional distribution of marked spiders given the total number of attacking spiders on subsequent days follows the extended hypergeometric distribution. The propensity parameter ω was estimated for each colony using the r-package BiasedUrn (Fog, 2011) in r (R-DevelopmentCoreTeam, 2011), where relevant functions for the extended hypergeometric distribution are given under the name of Fisher's noncentral hypergeometric distribution.

We used the same modelling approach to analyse spider's propensity to (i) attack two days in a row, (ii) attack twice in three days, (iii) perform both attack and web maintenance during three days and (iv) perform both behaviours in the same day (see the earlier subsections Attack and Web maintenance for a detailed description of the questions studied).

We observed a substantial variation in the estimated values of the propensity ω for each colony and hence modelled the variation by letting log(ω) follow a normal distribution with mean μ and standard deviation σ. The contribution to the likelihood from each colony is found by numerical integration with respect to ω. This model provides estimates of μ and σ, confidence regions for μ and σ, approximate tests of the hypotheses that μ = 0 and σ = 0.

To summarize, we tested the null hypothesis H0 = 0, σ = 0), that is, individuals come out of the nest at random; none has a significantly higher propensity to perform a task more often than others (nonpreferential population hypothesis), and this is true for all colonies tested, that is, there is no significant variation across colonies (equal propensity hypothesis).

Presentation of results

We give a representation of the information in the data about the propensity parameter ω in Figs 2-5: the figures show contour plots of the log-likelihood function for μ and σ which are the mean and the standard deviation of the normal distribution of log(ω).

Figure 1.

Proportion of individuals of social spider colonies participating in colony tasks (a) prey attack and (b) web maintenance on Day 3. Dark bars: Marked spiders (i.e. individuals that had previously engaged in prey attack) engaging in activity on Day 3, Light bars: Unmarked spiders engaging in activity on Day 3. The total length of bars is the proportion of the colony engaging in activity on Day 3. Numbers on the x-axis indicate specific colony size.

Figure 2.

Confidence regions for the mean and standard deviation of the colony-specific propensity ω for attacking on consecutive days. (a) Stegodyphus sarasinorum. The point (μ,σ) = (0,0) is far outside the 95% confidence interval depicted by the level curve A, and the lines at μ = 0 and σ = 0 do not intercept the level curves B and C, respectively, which means that spiders that engage in prey attack on one occasion have a higher propensity to attack the next day compared to a random expectation; however, the propensity ω varies amongst colonies. (b) Anelosimus eximius. The point (μ,σ) = (0,0) is on the border of the A level curve, and the line at μ = 0 intersects the level curve B, that is, there is no significant tendency for individuals to attack two days in a row. However, the line at σ = 0 does not intersect the level curve C, indicating significant variation amongst colonies in their propensity to attack. For a detailed description of graph interpretation, see Presentation of results in Materials and methods.

Figure 3.

Confidence regions for the mean and standard deviation of the colony-specific propensity ω for individual spiders to repeatedly attack over three days for (a) Stegodyphus sarasinorum and (b) Anelosimus eximius. See Fig. 2 for the detailed interpretation.

Figure 4.

Confidence regions for the mean and standard deviation of the colony-specific propensity ω for performing web maintenance after participation in prey attack over three days for (a) Stegodyphus sarasinorum and (b) Anelosimus eximius. In both species, the vertical line at 0 intersects the contour labelled B, which means that attackers did not have a higher or lower propensity to perform web maintenance than expected by random. However, there is a significant variation amongst colonies in their propensity to perform web maintenance as the horizontal line at σ = 0 does not intersect the contour labelled C.

Figure 5.

Confidence regions for the mean and standard deviation of the colony-specific propensity ω for performing web maintenance after attacking within the same day. In (a) Stegodyphus sarasinorum, the vertical line at 0 intersects the contour labelled B, and in A. eximius (b), the vertical line at μ = 0 is close to the level curve B, but it does not cross it, that is, attackers did not have a higher or lower propensity to perform web maintenance after participation in prey attack than expected by random. However, there is a significant variation amongst colonies in their propensity to perform web maintenance as the horizontal line at σ = 0 does not intersect the contour labelled C.

The level curve labelled A is the boundary of an approximate 95% confidence interval of (μ,σ). The area bounded by this contour is a measure of the information in the data: A small area means that the confidence interval is small, that is, the information in the data is high. The confidence region can be used to test the hypothesis H0 = 0, σ = 0) described above. If (0,0) is outside the confidence interval, the hypothesis is rejected. To analyse independently μ = 0 and σ = 0, we can look at the level curves B and C.

The level curve B can be used to evaluate the hypothesis that the mean of the log propensity is zero (μ = 0), that is, spiders come out of the nest at random; none has a higher propensity to perform at a task more often than others. We call this the hypothesis of a nonpreferential population. If a vertical line through μ = 0 in the x-axis does not intersect the level curve B, the hypothesis of a nonpreferential population is rejected; if it does intersect the contour, the P-value is larger than 0.05 and the hypothesis cannot be rejected.

Similarly, the level curve C can be used to test the hypothesis that the standard deviation of the log propensity is zero (σ = 0), which states that an equal propensity is relevant for all colonies, that is, that there are no differences amongst colonies in their propensity to perform a task. If the horizontal line through σ = 0 in the y-axis does not intersect the level curve C, the hypothesis of an equal propensity is rejected on the 5% level.

The numerical labels on the level curves in the figures are the distances to the maximum value of the log-likelihood function. The level curves labelled A, B and C correspond to the distances -1.35, -1.92 and -3.00, respectively.

Body size and age

To test for size-dependent task differentiation, body size for the marked and unmarked S. sarasinorum in each colony, respectively, was compared with a Wilcoxon paired test using the software r (R-DevelopmentCoreTeam, 2011).

To examine whether instar influenced the probability of an individual A. eximius to attack, we used a generalized linear mixed model (GLMM) with a Poisson distribution. We tested whether attackers belonged to specific instars or represented a random sample of the entire colony. We included instar as a fixed effect and colony size as a random effect to control for colony size effects. As an overall test of the effect of age on attacker group composition, we used a likelihood ratio test (χ2) comparing the full model with a reduced model including the same random effect structure but without the fixed effect term. Analyses were performed with r (R-DevelopmentCoreTeam, 2011) using the function lmer of the R-package lme4 (Bates et al., 2011).



The two species studied showed different tendencies in task differentiation of attacking behaviour. In S. sarasinorum, marked individuals were observed attacking in consecutive days more frequently than expected by chance. In A. eximius, marked individuals attacked at a similar frequency as expected by chance, indicating a lack of task specialization (Fig. 1a). In both species, there was significant variation in propensity to repeatedly attack amongst colonies.

  • 1.Does a spider attack for 2 days in a row?

The analysis of whether individual spiders attacked two days in a row more frequently than would be expected by chance showed striking differences between species (Fig. 2). Individual S. sarasinorum that attacked on one day were significantly more likely to also attack on the subsequent day, rejecting the hypothesis of a nonpreferential population (μ = 0: < 0.001). Colonies varied in their propensity to attack, strongly rejecting the hypothesis of an equal propensity amongst colonies (σ = 0: = 0.0041). In contrast, neither the hypothesis of a nonpreferential population nor that of equal propensity could be rejected in A. eximius (μ = 0: = 0.16, σ = 0: = 0.5).

  • 2.Does a spider attack twice within three days?

The same pattern was found when analysing whether individual spiders attacked more frequently over three days than would be expected by chance (Fig. 3). S. sarasinorum were more likely to engage in attack twice in three days compared with a nonpreferential population (μ = 0: < 0.001) and had significantly higher variation in propensity amongst colonies compared with the equal propensity hypothesis (σ = 0: = 0.0098). Anelosimus eximius conformed to the hypotheses of a nonpreferential population (μ = 0: = 0.88) with a tendency for variation in propensity amongst colonies (σ = 0: = 0.034). Overall, there was variation amongst colonies in both species in the propensity for individual spiders to attack repeatedly, and S. sarasinorum showed a strong tendency for some individuals to attack more frequently than others whilst this was not found in A. eximius.

Web maintenance

In both S. sarasinorum and A. eximius, we found no evidence that the propensity of an individual to maintain the web was influenced by its participation in previous prey attacks during the three experimental days (Fig. 1b). Marked individuals (i.e. previous attackers) and unmarked individuals were observed repairing the web with equal probability, indicating lack of division of labour into a prey attack caste and a web maintenance caste.

  • 1.Does a spider perform both attacking behaviour and web maintenance?

Both species conformed to the hypothesis of a nonpreferential population in relation to web maintenance behaviour during three days (μ = 0: S. sarasinorum P = 0.18 and A. eximius P = 0.36, Fig. 4). However, the hypothesis of equal propensity was rejected (σ = 0: S. sarasinorum P = 0.0019 and A. eximius P = 0.0047, Fig. 4), indicating variation in propensity for attackers to perform web maintenance amongst colonies.

  • 2.Does a spider that attack also perform web maintenance on the same day?

The analysis confirmed the above pattern, that is, spiders of both species conformed to the hypothesis of a nonpreferential population (μ = 0: S. sarasinorum P = 0.91 and A. eximius P = 0.037, Fig. 5) and showed significant variation amongst colonies in propensity (σ = 0, S. sarasinorum P < 0.001 and A. eximius P < 0.001, Fig. 5). Thus, although colonies within species varied in their propensity for attackers to engage in web maintenance, none of the two species showed deviations from random participation in web maintenance.

Body size and age

When sorting the colonies, we were able to recover the majority of the individuals that had been marked along the three days experiment. We recovered an average of 88% of the marked individuals in S. sarasinorum and 71% in A. eximius. We found no significant difference in body size between S. sarasinorum females that engaged in prey attack and that of a random sample of colony members, that is, prosoma width did not differ statistically between marked and unmarked individuals (Wilcoxon signed ranks test: V = 295, = 0.205). Among colonies, prosoma width (representing body size) ranged from 0.5 mm to 1.73 mm in the colony with the smallest individuals and from 1.37 mm to 3.25 mm in the colony with the largest individuals, with an overall average of 1.52 ± 0.01 mm.

In contrast, instar composition of the group of attackers within A. eximius nests was significantly different from the instar composition of their respective colony (χ2 = 805.53, d.f. = 3, < 0.001, Fig. 6). Adult and subadult females constituted a larger proportion of attacking spiders relative to their proportion within a random sample of colony members. The opposite pattern was found for juveniles, which were underrepresented amongst attackers relative to their representation in the colony.

Figure 6.

Percentage (means ± SE) of different instars of Anelosimus eximius in the group of attacking spiders (dark bars) relative to that of the entire colony (light bars), Juv = juveniles, Pre-sub = pre-subadult instar, Sub = subadults instar 2, Ad = adults.


We studied task specialization in two traits, prey attack and web maintenance, in two permanently social spider species and found evidence for task differentiation in prey attack in both species. The Indian Stegodyphus sarasinorum showed task specialization at the individual level, where some individuals within each colony attacked prey significantly more frequently than other colony members (individual task specialization), independent of hunger state, body size and instar. The New World Anelosimus eximius showed task specialization according to instars, where older individuals were more likely to attack prey than younger spiders (instar task specialization). Our data confirm that social spiders do not show strict division of labour; in neither species did colony members specialize in one of the two tasks studied. Individual spiders' engagement in web maintenance behaviour was not influenced by their previous engagement in prey attack behaviour. Our study did not directly address the question of whether individual spiders vary in their propensity to engage in web maintenance; however, observations revealed no consistent pattern in participation in maintaining the colony capture web, indicating that individual or instar specialization in this particular task may be lacking. The finding that task differentiation occur in prey attack behaviour but not in web maintenance behaviour in both species suggests that specialization in foraging tasks might be associated with group advantages in cooperative spiders.

In social spiders, individuals are hypothesized to be totipotent, that is, able to take on all colony tasks, and it has been debated whether and to which degree task differentiation occurs (Avilés, 1997; Lubin & Bilde, 2007). Task specialization was not previously reported in any social Stegodyphus species. A study showing variation amongst individuals in their tendency to approach prey observed in the African social S. mimosarum was attributed to variation in hunger state (Ainsworth et al., 2002); however, our study took hunger state into consideration and clearly suggests that social S. sarasinorum may show individual task differentiation in prey-attacking behaviour independent of their hunger state. Our data on A. eximius support findings of behavioural asymmetries in social species of the family Theridiidae (Christenson, 1984; Lubin, 1995), where previous studies attributed behavioural asymmetries to unbalanced resources distribution (Vollrath & Rohde-Arndt, 1983; Rypstra, 1993; Ebert, 1998), or absence of sperm in the receptacles of some females in a colony (Vollrath, 1986). We found that subadult and adult females more frequently engaged in prey attack, confirming similar age-related task acquisition in A. eximius (Christenson, 1984) and in the social theridiid Achaearanea wau (Lubin, 1995). In this system of age-related task differentiation, individual spiders accrue tasks with age as a gradual process. This differs from temporal polyethism found in many social insects, where individuals switch task with age in a predictable sequence (Wilson, 1971). However, in the bumble bee Bombus impatient, most of the workers perform a suite of tasks rather than specialize in only one, and even though they switch tasks frequently, they tend to perform the same task for a number of consecutive days (Jandt et al., 2009). We studied task specialization over a period of only three days; hence, studies over longer periods of time are needed to establish whether the observed task specialization patterns are transient or permanent.

Task specialization should enhance colony efficiency and productivity (Oster & Wilson, 1978; Wallace, 1982), whether it occurs as complete division of labour as in the eusocial insects or as facultative cooperative behaviour, when individuals differ in their propensity to cooperate (Bergmuller et al., 2010). Specialization is assumed to result in improved performance of the associated task, so groups with higher specialization perform better than groups with less specialized members, given that groups consists of kin or are interdependent in other ways (Roberts, 2005). Kin selection predicts that the amount of help provided is positively correlated with the relatedness between the helper and the recipient (Hamilton, 1964a,b) and is widely recognized as a main force in the evolution and maintenance of cooperation in different taxa from microorganisms to vertebrates (Benton & Foster, 1992; Queller & Strassmann, 1998; Russell & Hatchwell, 2001; Griffin & West, 2003; Griffin et al., 2004; Schneider & Bilde, 2008). Cooperation in spiders is expected to evolve by kin selection (Schneider & Bilde, 2008; Bilde & Lubin, 2011), as within-colony genetic relatedness is extraordinarily high as mating occurs strictly among colony members (Lubin & Bilde, 2007; Lubin et al., 2009). Indeed, kin discrimination (Schneider, 1995; Bilde & Lubin, 2001) and kin-mediated benefits of cooperation have been demonstrated in two Stegodyphus species (Schneider & Bilde, 2008; Ruch et al., 2009). The social spiders show specialization also in reproductive tasks, with less than half of the females reproducing whereas the remaining females provide allomaternal care (Vollrath & Rohde-Arndt, 1983; Vollrath, 1986; Salomon & Lubin, 2007). Reproductive skew can arise by random contest competition (Rypstra, 1993; Lubin, 1995; Ulbrich & Henschel, 1999) or may be adaptive when there are reproductive constraints on individual breeding and under levels of high relatedness (Keller & Reeve, 1994; Cant, 1998). Although determination of reproductive roles in social spiders remains poorly understood, there is evidence to suggest they may be predetermined in the juvenile stage of social Stegodyphus dumicola (L. Grinsted and T. Bilde, pers. obs). One possibility is that the individuals that specialize in prey attack are those females that are more likely to become reproducers. In our study, S. sarasinorum attackers were observed feeding on the prey during the entire feeding period, indicating that the benefit of attacking is better access to food (Willey & Jackson, 1993). Over longer periods of time, this should translate into relatively larger body size of attackers and better reproductive potential (Schneider & Lubin, 1998). Specialization in foraging and reproducing tasks is likely to reduce competitive interactions and ensure colony reproductive output under ecological constraints, and the costs of cooperation by nonreproductive helpers should be balanced by inclusive fitness benefits (Salomon & Lubin, 2007). A similar scenario may apply to ambrosia beetles that lack morphological differentiation and have regular inbreeding; these beetles show division of labour and age polyethism between larvae and adults in relation to colony tasks, and cooperative breeding in reproduction (Biedermann & Taborsky, 2011).

Our finding that the two study species differ in showing individual and instar specialization, respectively, could perhaps be explained by the markedly different colony composition of the two species. Stegodyphus colonies are highly synchronous and contain only one generation per year, that is, each colony is inhabited by individuals belonging approximately to the same instar (Lubin & Bilde, 2007). In contrast, Anelosimus eximius shows a less seasonal and synchronized breeding season, and colonies are simultaneously inhabited by more than one instar (Avilés, 1986). The different colony compositions in S. sarasinorum and A. eximius, respectively, may shape the evolution of task differentiation in different ways. The absence of a mixture of discrete instars within colonies in S. sarasinorum suggests that task specialization should evolve at the individual level. Indeed, our results support this hypothesis. In A. eximius, where individuals of different instars are present that might be more or less prone to perform different tasks, instar specialization where tasks are accrued with age is more likely. The same final output (i.e. a subset of individuals in a colony are more likely to perform a specific task) can hence be achieved through different trajectories to meet the same goal of optimizing colony efficiency.

Our study supports two main conclusions: (i) social spiders show task differentiation in prey-attacking behaviour but not necessarily in web maintenance behaviour, at least over defined periods of time and (ii) two independently derived social species, S. sarasinorum and A. eximius, achieve task specialization in two different ways: through individual and instar task specialization, respectively. Although our data are collected over a limited time period, these results suggest that social spider colonies optimize efficiency in colony productivity in similar ways as the social insects and that food acquisition may be a trait that shapes task differentiation as this trait is involved in determining future reproductive roles and also in reducing conflict over resources. We propose that adaptive behavioural solutions to increase colony efficiency in prey attack may vary among species with different life histories, so synchronous colonies evolve individual specialization, whereas mixed instar colonies evolve instar specialization.


This research was supported by a grant from The Danish Council for Independent Research to TB, the Faculty of Science and Technology at Aarhus University, Proprietær Niels Christensen og Hustru Oline Christensens Mindelegat, Oticon Fonden, and the Anne and Torkel Weis-Fogh Trust Fund from the University of Cambridge. We wish to thank Gösta Nachman for preliminary statistical advice, Rohini Balakrishnan and the Agastya International Foundation for support provided in India, as well as Leticia Avilés and collaborators for help and expertise in Ecuador. We thank the Spiderlab at Aarhus University for being a continuous source of suggestions, support and smiles and Yael Lubin for constructive criticism of previous versions of the manuscript.