A recurrent question in animal contests is whether individuals adopt a self or mutual assessment rule to decide to withdraw from a contest. However, many empirical studies fail to find conclusive support for one of these two possibilities. A possible explanation is that assessment strategies vary between individuals. In the contests of the orb-web spider Trichonephila clavipes, males perform a vibrational display on webs that may escalate to physical contact. Since all individuals perform the vibrational phase and only some of them escalate, we proposed two hypotheses: 1) all individuals perform mutual assessment during the vibrational phase, or 2) some individuals that do not escalate adopt self-assessment, while individuals that escalated adopt mutual assessment. To evaluate these hypotheses, we investigated the relationship between the duration of the vibrational phase and frontal leg length (a proxy of male fight capacity) of loser and winner males in contests that escalated and did not escalate to the physical contact phase. We found a non-significant relationship between duration and losers leg length for both contests that escalate and did not escalate. While we found a positive relationship between duration and winners leg length, particularly in contests that did not escalate. These results do not provide support for mutual assessment or for a mix of different assessment rules among individuals. We suggest that in T. clavipes, the dynamics of the vibrational phase may be explained by two different contest strategies (opponent-only assessment or size-based aggressiveness) that are dependent on intruder motivation to escalate.

Study area

We conducted this study between December 2018 to March 2019 and between February 2021 and March 2021 in the Ecological Station of the Federal University of Minas Gerais (19° 52’ 25’’ S, 43° 58’ 21’’ W), Minas Gerais State, Brazil. The Ecological Station is a 114 ha area composed of a mixture of semi-deciduous Atlantic forest, Brazilian savanna (i.e., Cerrado), and invasive plant species (Antonini and Martins 2003). Individuals of T. clavipes are abundant in this area between December and June (Almeida and Peixoto 2021; Silva et al. 2021).

Field sampling

We searched for female webs along three trails of approximately 20 m, 300 m, and 400 m in length. We selected female webs with mature females or females in their last instar because males of T. clavipes defend females in both ontogenetic stages (Almeida and Peixoto 2021). We considered females that had a swollen and sclerotized epigynum with two distinct copulatory openings to be mature (Higgins 2000; Rittschof 2011), and females with an abdominal diameter greater than 6 mm (Moore 1977) and without a sclerotized epigynum to be in their last instar. Furthermore, for last instar females, we also observed them for seven days after the sample to confirm they did not make an additional molt. We sampled only female webs with at least two males established (16 webs with two males, 10 webs with three males, seven webs with four males and one web with five males). We defined the central male as the individual located up to 5 cm from the female's abdomen (Christenson and Goist 1979) and peripheral males as the other males that were established inside the female web up to 5 cm distant from the last web spiral (Christenson and Goist 1979; Brown 1985). We did not sample female webs in which males were all established between the central and the peripheral position, because this avoided the identification of the central male. In addition, all sampled males were sexually mature as assessed through dark brown coloration and sclerotized pedipalps (Myers and Christenson 1988).

In other empirical studies that reported male contests in T. clavipes, the rival with the larger body size typically wins the fight (Christenson and Goist 1979; Vollrath 1980; Rittschof 2010). In these studies, the measure chosen as a surrogate of body size is the cephalothorax width (Rittschof 2010; Constant et al. 2011). However, it is important to note that T. clavipes males do not use their cephalothorax directly in fights, and for this reason, it may be that cephalothorax width is related to winning chances due to its correlation with other traits that determine the outcome (Vieira and Peixoto 2013). A possible surrogate of fighting capacity that is directly used in T. clavipes fights is the male frontal legs (Christenson and Goist 1979; Constant et al. 2011). Males use the frontal legs to perform vibrations on the web and also during the physical contact phase. In fact, there is empirical evidence that central males have longer frontal leg length than peripheral males in webs of females (Silva et al. 2021 – Supplementary Material). Furthermore, in our data, we also found that longer frontal legs increased the chances of victory (Supplementary Material, Fig. S1). Therefore, we considered the frontal leg length as a reliable trait to estimate the fighting capacity of T. clavipes males. We estimated the leg length of males by photographing (using a Canon EOS Rebel T5) the male with a ruler positioned adjacent (but without touching the male). Then we used the ImageJ software to measure the leg length based on the scaled photograph (Schneider et al. 2012).

The agonistic interactions between T. clavipes males generally occur when the female is feeding on a prey (Farr 1976; Christenson and Goist 1979; Christenson et al. 1985; Cohn et al. 1988). Therefore, to induce interactions between males, we carefully placed a grasshopper on the female web (Silva et al. 2021). After placing the grasshopper, we waited for 10 min to observe if the female captured the prey and the males started to interact. To define the beginning of the interaction between males, we used the description provided in other studies on agonistic interactions of T. clavipes males (Christenson and Goist 1979; Rittschof 2010; Constant et al. 2011; Silva et al. 2021). The interaction started when the first male began vibrating the female web (i.e. when a male used the first pair of legs to pull and release web threads) and the other male started to vibrate in response (Silva et al. 2021). We considered the end of the interaction when one of the males fled to the periphery of the web or was knocked down from the female web. We recorded all the interactions between males using a voice recorder and registered the duration of each phase of each interaction. At the end of the interaction, we also registered which individual won (i.e. established in the central position) and lost (i.e. established in the peripheral position) the contest. To avoid pseudoreplication, we sampled only one interaction per female web. We also marked males and females with non-toxic paint on their abdomens after the end of the interactions to avoid resampling them (Christenson and Goist 1979; Constant et al. 2011; Rittschof 2011). We did not mark the males before they interacted to avoid stressing them before the observations. In addition, males of this species move very slowly during the interaction and it was possible to track the winner and the loser without marking them. Even in interactions that escalated to physical contact, males rarely changed positions on the web (the peripheral male won the interaction in 2 contests), which allowed us to correctly identify each one. In specific cases in which females disrupted the male-male interaction or males ran into other female webs before returning to the web in which they were fighting, we discarded the observation from our sample. We observed a total of 34 interactions between males. Half of these interactions occurred in webs of mature females and the other half in webs of females in the penultimate instar.

Statistical analysis

To evaluate whether all (or most) males perform mutual assessment or whether there is a mixture of self- and mutual assessors during the vibrational phase, we built two general linear mixed models (GLMMs) - one for winners and the other for losers. We considered the duration of the vibrational interaction as the response variable, and the male front leg length and escalation occurrence as explanatory variables. We also log-transformed the duration of the vibrational phase to reach homogeneity of variances. To control for a potential effect of female quality on the motivation of each individual to fight (Silva et al. 2021), we inserted female age (juvenile or mature) in both models as a random variable (the inclusion of female age as a fixed factor provided similar results). In each model, we inserted frontal leg length and escalation occurrence both as independent factors and an interaction term between them. Based on this, if all males perform mutual assessment during the vibrational interactions, we expect the interaction term to be non-significant, while the relationship between the duration of the vibrational interaction and frontal leg length should be positive for loser males and negative for winner males (Taylor and Elwood 2003). However, if some males perform mutual assessment while others perform self-assessment, we expected the interaction term to be significant. The relationship between the duration of the vibrational interaction and frontal leg length should be positive for loser males and negative for winner males in interactions that escalated to the physical aggression phase and positive for losers, but non-related to winner’s frontal leg length in interactions that did not escalate. To provide a measure of how much variation in the duration of the vibrational phase is explained by our predictor variables, we calculated the marginal and conditional pseudo-R² for mixed-effects models as a non-dimensional standardized measure of goodness of fit (Nakagawa and Schielzeth 2013).

As additional information to investigate whether all males performed mutual assessment or whether the population was composed of a mix of self and mutual assessors, we performed a second analysis. If the fights that do not escalate are composed of mutual and self-assessors, the variation in the contest duration explained by winner leg length should be higher when compared to escalated contests that are predicted to be composed of mutual assessors. To estimate this difference, we calculated the coefficient of determination R²_resid (Ives 2019) separately for two general linear mixed models (GLMMs) using the front leg length of winner males as an explanatory variable - one considering the initial contest phase of fights that escalated and the other the initial contest phase of fights that did not escalate to the physical contact phase. We also used the female age as a random factor to control possible differences in interactions for females in different reproductive stages. The R²_resid is used to report the variance explained by a fixed variable in GLMMs (Ives 2019). Values of R²_resid closer to 1 indicate a stronger correlation between the predictor variable and the response variable used in the model (Ives 2019). We chose R²_resid because it is appropriate to compare different data sets using the same model and is minimally affected by differences in sample size (Ives 2019). According to our first hypothesis, if all males perform mutual assessment, we expect contests that escalate and that do not escalate to have a similar relationship between initial phase duration and winners' fighting capacity. Therefore, we expect a similar R²_resid for the models testing the relationship between the duration of the vibrational phase and front leg length considering non-escalated fights and the same model considering only fights that escalated to the physical aggression phase. On the other hand, according to our second hypothesis, the vibrational phase in contests that do not escalate should be composed of a mix of self and mutual assessors, while contests that escalate should be composed of mutual assessors. This would lead to a smaller R²_resid value for the model testing the relationship between the duration of the vibrational phase and front leg length for non-escalated fights when compared to the same model considering only fights that escalated to the physical contact phase. To perform our analyses, we used the packages "lme4" (Bates et al. 2014), “MuMIn” (Barton 2022), “rr2” (Ives 2018) in the R software (R Core Team 2023).