Given the weak support for two simple explanations (i.e., similar habitat and resource limitation hypotheses) that require no individual benefit for the existence of MSA, we argue that this unusual form of group living evolved and is maintained in harvestmen because it provides benefits. Here we discuss three physiological benefits: protection against stressful abiotic conditions, reduction of metabolic rates, and thermoregulation (Table 2).

Gregariousness may act as a behavioral mechanism to reduce water loss in many arthropods (isopods: e.g., Friedlander, 1965; millipedes: e.g., Dangerfield, 1993; insects: e.g., Danks, 2002). In harvestmen, the close body contact and the intertwining of legs may reduce airflow and thus individual water loss (Machado and Macías-Ordóñez, 2007). This may be the case for the dense SSA observed in some Eupnoi from xeric regions composed of thousands of quiescent individuals (e.g., Wagner, 1954; Figures 1A,B). However, all cases of MSA in harvestmen are loose (Table 1), which is unlikely to protect the individuals from dehydration because they are not so densely packed together to prevent water loss. Thus, although loose aggregations may also promote changes in microclimatic conditions, we anticipate that the potential benefits they provide in terms of protection against dehydration should be lower than in dense aggregations.

Aggregations have also been found to reduce the resting metabolic rates of some insects (reviewed in Chown and Nicolson, 2004). Such reduction may be beneficial because it decreases energy expenditure and spiracular water loss in tracheate arthropods, including arachnids (Hadley, 1994). There is one single study showing that Vonones ornatus (Cosmetidae) harvestmen in small and loose aggregations have reduced metabolic rates (Anderson, 1993). The exact mechanism linking gregariousness and metabolic rates in harvestmen has not been explored yet and certainly deserves attention. Another important question is whether the reduction in metabolic rates reported for loose SSA of V. ornata also occurs in loose MSA of other harvestman species. Therefore, this potential benefit cannot be discarded and deserves further consideration.

Finally, gregariousness may favor thermoregulation by improving heat production and/or conservation, protecting the individuals from cold conditions. This benefit is particularly important among endothermic vertebrates living in high latitudes, where the temperature may reach extremely low values (Gilbert et al., 2010). Among insects that live in less harsh habitats, gregariousness may also play an important role in their thermal ecology (Danks, 2002; Chown and Nicolson, 2004). For instance, caterpillars maintain higher and more stable body temperatures when aggregated (e.g., Casey et al., 1988; Joos et al., 1988; Casey, 1993; Fitzgerald, 1993), which leads to higher growth rates (e.g., Scriber and Lederhouse, 1983; Knapp and Casey, 1986). Similarly, temperatures are higher and buffered in aggregations of the lady beetle Coleomegilla maculata (Benton and Crump, 1979). Although there is no information on the temperature inside harvestman aggregations, we argue that dense aggregations may conserve heat and buffer temperature variations, at least in its core. Heat control and temperature buffering would be particularly beneficial for individuals in overwintering aggregations from cold regions (e.g., Holmberg et al., 1984; Novak et al., 2004) and in diurnal aggregations from hot xeric regions (e.g., Wagner, 1954). Nonetheless, since all cases of MSA in harvestmen are loose (Table 1), we consider that a thermoregulation benefit is unlikely.

Aggregations in harvestmen may provide several defense benefits against predators. At least among Laniatores, scent gland secretions released by the individuals are a powerful chemical deterrent that repel different types of predators, including invertebrates and small vertebrates (e.g., Eisner et al., 2004; Machado et al., 2005). Thus, a first defense benefit of gregariousness is a higher amount of secretion released upon disturbance — a collective retaliation that may improve the efficiency of the chemical defenses (Machado and Macías-Ordóñez, 2007; Table 2). In the case of MSA, behavioral observations show that some species rarely release scent gland secretions (Elpino-Campos et al., 2001; Pereira et al., 2004). In one report, the most common species in the aggregation does not even produce scent gland secretions (Machado and Vasconcelos, 1998). This finding can be explained by the “protector species hypothesis,” according to which individuals of a poorly defended species benefit by associating with individuals of a well-defended species that aggressively repels potential predators of both species (Sullivan, 1984; Quinn et al., 2003; Goodale et al., 2014). The protector species hypothesis assumes that the benefits of MSA in terms of collective retaliation are necessarily asymmetric because not all species contribute to repel the predators (Table 2).

In known cases of MSA in harvestmen, the poorly defended species (e.g., Mischonyx cuspidatus and Encheiridium montanum) comprise most part of the individuals in the aggregation, whereas the well-defended species (e.g., Discocyrtanus oliverioi and Holoversia nigra) account for only a small number of individuals (Machado and Vasconcelos, 1998; Elpino-Campos et al., 2001; Pereira et al., 2004; Table 1). It is possible that the well-defended species are not even gregarious. In this case, well-defended species may serve as attractors (i.e., initiators) to poorly defended species (i.e., followers) that would aggregate around the former. This hypothesis does not imply that initiators are deliberately attracting followers using recruitment pheromones, although it may be the case. If initiators have some benefit when followers aggregate around them, it would be advantageous to release recruitment pheromones. An obvious advantage for the initiators of attracting followers is a decreased probability of being singled out by predators (Sridhar et al., 2009). This “dilution effect” (Turner and Pitcher, 1986; Table 2) is another defense benefit attributed to harvestman aggregations (Machado and Macías-Ordóñez, 2007). In MSA, both initiators and followers may enjoy this defense benefit; if the benefit is similar to all species in the aggregation, is an open question that deserves future studies.

Harvestmen also obtain the benefit of having an early alarm signal in aggregations. The scent gland secretions released by aggregated individuals and their movement to escape from an attack work as an alarm — one chemically and the other mechanically mediated (Machado et al., 2002). Experimental field evidence shows that aggregations of Serracutisoma gnaspinii (Gonyleptidae) with a greater number of individuals disperse faster to the chemical alarm promoted by the emission of scent gland secretions (Machado et al., 2002). Large harvestman aggregations have more sensory legs full of chemoreceptors to perceive the scent gland secretions released by other individuals. Due to the alarm communication, aggregations may increase both the escape capability of the individuals and their probability to survive a predatory attack (Machado and Macías-Ordóñez, 2007). In MSA, there is observational evidence showing that scent gland secretions released by one species may work as a chemical alarm to individuals of other species that do not release scent gland secretions (Machado and Vasconcelos, 1998). This suggests that the benefits promoted by the alarm communication are asymmetric: while individuals of species that do not release scent gland secretions benefit from the chemical alarm released by other species, the opposite does not happen (Table 2). Regarding the mechanically mediated alarm, however, all species in MSA may be favored because it only requires that individuals bump each other while they are dispersing from the aggregation (i.e., “Trafalgar effect”; Treherne and Foster, 1981).

Finally, it is possible that the fast dispersion of aggregated individuals causes a confusion effect in the search image of visually oriented predators (Goodale et al., 2019; Table 2). Most examples of confusion effect have been reported for vertebrate prey, but experiments with aggregations of the water flea Daphnia magna showed that stickleback attack rates decline when aggregated individuals are in larger aggregations, are closer together, have similar body size, and are moving parallel to each other (Ohguchi, 1981). After disturbance, individuals in harvestman aggregations quickly disperse, running away or falling from the roosting site (Newman, 1917; Holmberg et al., 1984). Certain species of Eupnoi also exhibit bobbing behavior (Holmberg et al., 1984), which is a fast up and down body movement (Gnaspini and Hara, 2007). The collective movement of dozens to thousands of individuals bobbing probably confuses the identification and precise location of a potential prey’s body by the predator (Escalante et al., 2019). This confusion effect may be more accentuated in the MSA of sclerosomatids in Central America because individuals of different species show great variation in body color (Figures 1D, 3), which may disrupt even more the search image of potential predators (Table 2).

Aggregations during or immediately before the beginning of the breeding season may ensure that individuals of both sexes will be in close proximity. Thus, aggregations may provide mating benefits by increasing the opportunities to gain access to mates and by reducing the mating search costs (Ward and Webster, 2016). However, there is no evidence that harvestman aggregations are related to reproduction (Machado and Macías-Ordóñez, 2007). In many cases, aggregations occur during the winter, when no breeding activity occurs (e.g., Holmberg et al., 1984; Machado et al., 2000; Novak et al., 2004; Willemart and Gnaspini, 2004; Chelini et al., 2012). When aggregations occur during the breeding season, sexual interactions have not been observed close to them (e.g., Edgar, 1971; Cockerill, 1988; Grether et al., 2014a). In the case of MSA, the possibility of mating benefits is non-existent because individuals derive no reproductive advantage from aggregating with heterospecifics.

Gregariousness may increase the chances of detection by predators because aggregations may be visually or chemically more conspicuous than isolated individuals (Vulinec, 1990; Table 3), especially if individuals are colorful, as it is the case of some Prionostemma species (Table 1 and Figures 1D, 3). To date, we lack information about whether harvestman aggregations (SSA or MSA) increase conspicuousness and consequently predatory attacks. However, if aggregations increase conspicuousness and the risk of predatory attacks, we predict that individuals of cryptic species should avoid joining MSA with conspicuous species. In this situation, individuals of cryptic species would derive more benefits by aggregating exclusively with conspecifics because they would make the aggregation less conspicuous to visually oriented predators. On the contrary, individuals of conspicuous species should infiltrate aggregations of cryptic species where they would be more protected from visually oriented predators than in aggregations composed only of conspecifics. By doing so, individuals of conspicuous species would ‘break’ the aposematic value of the aggregations. This conflict between individuals of different species is an interesting, yet unexplored, idea that may have important implications for the patterns of species co-occurrence in MSA.

One of the most obvious costs of gregariousness is the increased transmission of internal and external pathogens and parasites (Côté and Poulin, 1995; Kappeler et al., 2015). Harvestmen are exposed to a great sort of pathogens and parasites (reviewed in Cokendolpher and Mitov, 2007). The group of endoparasites most frequently reported for harvestmen are gregarines, which are apicomplexans that infect the digestive tract of their hosts. A harvestman infects itself when the small oocysts, present in the feces of other individuals, attach to the tip of its legs and then are ingested during leg grooming (Cokendolpher and Mitov, 2007). Considering that aggregated individuals share the same roosting sites and probably defecate in or close to these sites, it is likely that harvestman aggregations increase the chances of infection by gregarines. Unfortunately, there is no information on host specificity for the gregarines that infect harvestmen. If the host-parasite relationship is not species-specific (e.g., Cokendolpher, 1993), the risk of contamination may be equally high in SSA and MSA. On the contrary, if there is some degree of specialization in the host-parasite relationship, the risk of infection should be lower in MSA than in SSA (Table 3).

Harvestmen in aggregations are also exposed to external parasites, including many species of Leptus mites (Cokendolpher and Mitov, 2007). These mites deposit a cone of cementing material at the attachment site that, along with their mouthparts, form a tight anchorage on the tegument of the host (Åbro, 1988). Thus, while attached, it is very unlikely that one harvestman carrying mites will infect others. However, we observed mites moving around on the body of Prionostemma harvestmen that form MSA (see section “Costs of gregariousness” below). This indicates that mites do not attach their mouthparts immediately to their hosts, and thus may move from one host to another. Currently, there is no information on how long it takes for a Leptus mite to climb the host and attach its mouthparts. We know, however, that many Leptus mites prefer certain body parts on their harvestman hosts (e.g., McAloon and Durden, 2000; Townsend et al., 2008), which implies that they move around some time until they find a proper attachment site. The main implication of this movement is that horizontal transmission of mites among aggregated harvestmen is possible. It seems that Leptus mites do not show great host specialization because the same species may parasitize numerous harvestman species (Cokendolpher and Mitov, 2007). Therefore, the costs of infection are likely to be similar in SSA and MSA.

When dozens, hundreds or even thousands of individuals aggregate in the same site, they may compete for resources, mainly food. However, when harvestman aggregations disperse at night, individuals visit different places and can move long distances (e.g., Grether and Donaldson, 2007). When individuals of Acutisoma longipes and Serracutisoma spelaeum leave the cave at night, each one follows a specific trail to forage on substrates as different as the external cave walls and the canopy of the forest surrounding the cave (Gnaspini, 1996; Machado et al., 2000). Since the foraging area shows little overlap between individuals, it is unlikely that they compete for food. We do not know if this also happens in large aggregations of Eupnoi that may have between 70,000 (Wagner, 1954) and 300,000 individuals (Mukherjee et al., 2010) in habitats with low productivity, such as deserts and cold forests. It is important to note, however, that harvestmen are highly generalist feeders and forage on many different items, including live and dead animals, as well as vegetal and fungi matter (Acosta and Machado, 2007). Thus, competition for food is not expected to have an important role in harvestmen ecology (Curtis and Machado, 2007). Supporting this claim, a long-term study on harvestman communities in deciduous woods from England found no evidence for resource competition among species (Adams, 1984). If competition is important, joining MSA may attenuate it, considering that differences in size, morphology, and microhabitat use between species somehow reflect differences in their food niche. This hypothesis has been originally proposed for MSA of birds that forage together (Powell, 1985; Sridhar et al., 2009), and to our knowledge, there is no demonstration that it also applies to invertebrates. Harvestmen offer an opportunity to test this hypothesis because of the numerous cases of SSA and MSA, which allow exploring how the intensity of competition for food varies according to the type of aggregation while controlling for aggregation size.

The research was conducted in Las Cruces Biological Station, Puntarenas, Costa Rica (8°47′ N; 82°57′W; 1,200 m. a.s.l.). The temperature in the study site ranges from 17 to 24°C, and the annual precipitation is about 3,600 mm, with a well-marked seasonality: a wet season between July and October, and a dry season between February and June. Data were collected in the understory of primary and secondary forests along the Jungle, Java, and Water trails.

We searched for Prionostemma harvestmen on three sampling occasions: January 2013, July 2013, and February 2014. Although the individuals were not marked, it is unlikely that they were resampled in different sampling occasions because Eupnoi harvestmen live only a few months as adults (Gnaspini, 2007). Moreover, we searched the trails only once and each individual and aggregation was sampled only once in the three sampling occasions. In all sampling occasions we searched for harvestmen during the day (08:00 to 14:00 h), when individuals are roosting motionless. On both sides of the trails, we looked for Prionostemma individuals roosting on the low vegetation (from 0 up to 2–3 m). For each individual, we recorded the species, which can be easily recognized based on the dorsal coloration (Figure 3), and the roosting status: (a) solitary, (b) in SSA, or (c) in MSA. For the purpose of this study, we defined aggregation as a group of two or more individuals, regardless of the species, resting in close proximity (<5 cm), with or without leg contact (Figure 1D). When we found isolated or aggregated individuals, we grabbed them by hand and quickly placed them inside plastic containers (30 cm diameter × 12 cm height). This allowed us to process one individual at a time while the others were caged in low density (2–4 individuals per container). After collecting the individuals, we carefully inspected each of them to record the number of missing legs, as well as the presence and number of mites on the body (including legs, dorsum, and venter).

We used the data collected in the three sampling occasions to describe the frequency of individuals belonging to each species that roost solitarily or in aggregations (SSA or MSA), the natural variation in the number of individuals in SSA and MSA, the species composition in MSA, and the patterns of species co-occurrence in MSA. Moreover, we used that data to test our hypotheses on benefits and costs of group living in harvestmen (see sections “Data analyses: benefits of gregariousness” and “Data analyses: cost of gregariousness” below).

In February-March 2014, we delimited a plot 1,100 m long and 6 m wide (3 m on each side of the above-mentioned trails). Inside this plot, we found 46 plants regularly used as roosting sites by Prionostemma individuals. These roosting sites were individually marked with numbered flags and inspected once every day (between 08:00 and 14:00 h) over 10.8 ± 1.3 days (range: 9–14 days). During each inspection, we recorded the number of individuals of each species in the roosting sites. We only used these data to describe the temporal variation in aggregation size and species composition.

To test our predictions on the defense benefits of gregariousness, we ran generalized linear models (GLMs) in which leg loss (yes or no) was the predictor variable with multinomial distribution of errors and logit link function. The response variable was the roosting status with three levels: solitary, SSA, and MSA. This analysis was performed for each species separately, so that we could evaluate if the potential benefits of gregariousness are similar between them. Given that the number of individuals of Prionostemma sp.9, sp.10, and sp.11 was very low (15, 9, and 5, respectively), we could not perform the analysis for these species. Moreover, for Prionostemma sp.8 only MSA were found in the field, so that the response variable had only two levels (solitary and MSA). In this case, the GLM had a quasibinomial distribution of errors (to deal with data overdispersion) and the link function was logit. Based on the model for each species, we estimated the marginal means to calculate the necessary contrasts to test our predictions. We also ran models in which the predictor variable was the number of missing legs (ranging from 0 to 4). Given that the explanatory power of this model was similar to that of the leg loss model (Supplementary Table 1), we present here only the results obtained with the leg loss model.

Finally, we evaluated whether the number of individuals in the aggregations (i.e., aggregation size) and the type of aggregation (SSA or MSA) influenced the proportion of autotomized individuals in the aggregations. For this, we ran a GLM in which the response variable (i.e., the proportion of autotomized individuals in the aggregation) had a quasibinomial distribution of errors (to deal with data overdispersion) with logit link function. The full model included an interaction between aggregation size and type of aggregation.

To test our predictions on the costs of gregariousness, we ran two GLMs. In the first model, the response variable was the presence of mites (yes or no), with quasibinomial distribution of errors (to deal with data overdispersion) and logit link function. In the second model, the response variable was the number of mites per individual, with negative binomial distribution of errors (to deal with overdispersion of the data) and log link function. In both models, the predictor variable was the roosting status with three levels: solitary, SSA, and MSA. These analyses were performed for each species separately, so that we could evaluate if the potential costs of gregariousness are similar between them. Again, we excluded Prionostemma sp.9, sp.10, and sp.11 from the analyses because the number of individuals was very low. Based on the models for Prionostemma sp.5, sp.6, sp.7, and sp.8, we estimated the marginal means to calculate the necessary contrasts to test our predictions.

All statistical analyses were performed in the software R version 4.1.0 (R Core Team, 2021). We used the package nnet (Ripley et al., 2016) for the multinomial models, the package Stats (R Core Team, 2021) for the binomial models, the package MASS for the GLM with negative binomial distribution of errors (Venables and Ripley, 2002), and the package emmeans (Lenth, 2019) to calculate the contrasts. The complete datasets and the scripts used in the analyses are available on Dryad: https://datadryad.org/stash/share/TMMKprXz1Iji8hvzbumzaVDQi0g0jqxzLPfdl2btxpA.

Taking together the three sampling occasions, we found 390 Prionostemma individuals, from which 73 (18.7%) were roosting solitarily and 317 (81.3%) in aggregations. The total number of aggregations was 78, being 56 (71.8%) MSA and 22 (28.2%) SSA. The aggregation size ranged from 2 to 9 in MSA and from 2 to 15 in SSA (Figure 4A). The number of species in MSA ranged from 2 to 5, with 92.3% of the aggregations containing 2 or 3 species.

The number of individuals was not evenly distributed among the seven Prionostemma species (Chi-square goodness-of-fit: X² = 419.8, df = 6, P < 0.001, Figure 4B). The two most common species were Prionostemma sp.5 (42.2% of all individuals) and Prionostemma sp.6 (31.1%); the other five species comprised together 26.7% of all individuals (Figure 4B). The species differed in the proportion of individuals that were found roosting solitarily or in aggregations (Chi-square goodness-of-fit: X² = 23.6, df = 6, P < 0.001, Figure 4B). For almost all species, individuals were more frequently found in aggregations than solitarily (Figure 4B). The only exception was Prionostemma sp.11, for which we collected only five individuals, one aggregated and four solitary (Figure 4B). When roosting in aggregations, individuals of Prionostemma sp.8, sp. 10, and sp. 11 were found only in MSA. In contrast, individuals of Prionostemma sp.5, sp.6, sp.7, and sp.9 were found both in MSA and SSA (Figure 4B).

To explore the co-occurrence of species in MSA we constructed an association matrix. This matrix quantified the number and percentage of times each species was found together with the other species in the same aggregation (Table 4). The three most common species (Prionostemma sp.5, sp.6, and sp.7) were commonly found with each other (Table 4). Each of the four other species (Prionostemma sp.8, sp.9, sp.10, and sp.11) were associated in roughly similar proportions with all species (Table 4). Additionally, three species combinations represented 54% (n = 30) of all species combinations observed in the field: (a) Prionostemma sp.5 + sp.6 (23%), (b) Prionostemma sp.5 + sp.6 + sp.7 (23%), and (c) Prionostemma sp.5 + sp. 7 (7%).

The number of individuals in the roosting sites was highly variable over time, ranging from 1 to 16. The mean (±SD) number of individuals was 3.3 ± 2.6 (n = 46), and the mean (±SD) coefficient of variation (CV) of the number of individuals across inspections was 55.2 ± 15.8% (range: 0–100%). The number of species in the roosting sites was also highly variable over time, ranging from 1 to 5. The mean (±SD) number of species was 1.8 ± 0.9 (n = 46), and the mean (±SD) CV across inspections was 43.3 ± 12.1% (range: 0–73.9%). Between two consecutive inspections, we recorded frequent transitions between all three categories of roosting status (Figure 5). The most frequent transition was from MSA to MSA, followed by MSA to solitary, and solitary to solitary (Figure 5). The least frequent transition was from solitary to SSA, followed by SSA to solitary, and MSA to SSA (Figure 5).

We recorded leg loss for 390 Prionostemma individuals. A total of 194 individuals (all species together) were found missing at least one leg. Most of the autotomized individuals in all seven species were missing only one (63.9%) or two legs (25.3%), but some of them were missing three (8.3%) or even four legs (0.5%); for four individuals (2.1%) we do not have information on the number of missing legs. The frequency of autotomized individuals differed between species (Chi-square goodness-of-fit: X² = 232.7, df = 6, P < 0.001), but the percentage of individuals missing at least one leg was always higher than 34% (Figure 6A).

Leg loss was not associated with the roosting status of the individuals in any of the four species analyzed (i.e., Prionostemma sp.5, sp.6, sp.7, and sp.8). In Prionostemma sp.5, sp.6, and sp.7, individuals were more likely found in MSA than in SSA or solitary, but this pattern did not differ between intact (i.e., eight-legged) and autotomized individuals (Figures 7A–C, see also contrasts 1–3 in Figure 7E). In Prionostemma sp.8, individuals were more likely found solitary than in MSA, but again this pattern did not differ between intact and autotomized individuals (Figure 7D, see also contrasts 1-3 in Figure 7E).

For Prionostemma sp.5 and sp.6, the probability of finding intact individuals in MSA was higher than finding intact individuals solitarily (Figure 7, contrast 5 in Figure 7E). Moreover, for Prionostemma sp.6 and sp.7, the probability of finding intact individuals in MSA was higher than in SSA (Figure 7, contrast 6 in Figure 7E). For Prionostemma sp.5, sp.6, and sp.8, the probability of finding autotomized individuals solitary, in SSA, and in MSA was similar (Figure 7, contrasts 7–8 in Figure 7E). For Prionostemma sp.7, however, the probability of finding autotomized individuals was higher in MSA than in SSA (Figure 7, contrasts 9 in Figure 7E). Lastly, the proportion of autotomized individuals in the aggregations was not associated with aggregation size or type of aggregation (Table 5).

We recorded the presence of parasitic mites for 390 Prionostemma individuals. A total of 115 individuals (all species together) had mites, and for 111 we have information on the number of mites they were carrying. Mites were present in different body parts (legs, dorsum, and venter), either motionless (probably sucking on the host’s hemolymph) or moving around (n = 12 observations). In 69.6% of all aggregations there was at least one individual carrying mites. The number of mites per individual ranged from 1 to 8, with the following relative frequencies: 1 = 57.7%, 2 = 25.2%, 3 = 9.9%, 4 = 2.7%, 5 = 2.7%, 6 = 0.9%, and 8 = 0.9%. The frequency of individuals carrying mites differed between species (Chi-square goodness-of-fit: X² = 181.3, df = 6, P < 0.001, Figure 6B). However, for the four species we analyzed (i.e., Prionostemma sp.5, sp.6, sp.7, and sp.8), the roosting status had no effect on the probability of having mites (Figure 8) or the number of mites per individual (Figure 9).

When compared with other species of Eupnoi, the Prionostemma aggregations described here are small, with no more than 16 individuals. Among sclerosomatids from temperate regions (e.g., Gyas, Leiobunum, and Nelima), aggregations are composed of hundreds to thousands of individuals, usually packed in high density (e.g., Wagner, 1954; Holmberg et al., 1984; Novak et al., 2004). One possible reason for this difference in the number and density of individuals may be related to the habitat: while Prionostemma species inhabit tropical forests, temperate sclerosomatid species inhabit cold and/or xeric places (Machado and Macías-Ordóñez, 2007). In tropical forests, where temperature is warm and pluviosity is high, aggregations may have no relevant role in protecting the individuals against harsh abiotic conditions. In turn, in cold and/or xeric places, a high number of individuals packed together may confer physiological benefits, buffering stressful abiotic conditions (Table 2; see also Danks, 2002).

The Prionostemma aggregations described here show the largest number of species among all MSA reported so far for harvestmen (Table 1), despite being relatively small aggregations compared to other harvestman species. Additionally, for six of the seven species we studied, individuals were found mainly aggregated (instead of solitary), and more frequently in MSA than in SSA. These findings suggest that Prionostemma harvestmen are mainly gregarious and highly tolerant to heterospecifics. How and why individuals from different species are brought together remains to be explored. Harvestmen use pheromones for different purposes (Raspotnig, 2012), and there is evidence for one species of Prionostemma from Nicaragua that individuals mark roosting sites with chemical compounds (which might function as pheromones) and come back to those roosting sites (Donaldson and Grether, 2007; Grether et al., 2014b). However, the attraction of heterospecifics to MSA requires that the compounds present in the chemical signal (i.e., infochemicals) to be recognized by all aggregating species. It is currently unknown if all the Prionostemma species studied here produce and deposit recruitment infochemicals. Perhaps Prionostemma sp.5 and sp.6 – the two most common species in the study site – may act as ‘initiators’ of the MSA depositing recruitment pheromones. The other species, which are rarer, may act as ‘followers’ and join the MSA by eavesdropping on the infochemical. Although speculative, a similar mechanism has already been reported for the formation of mixed-species flocks in birds by means of differential production of vocalizations (Sridhar et al., 2009; Magrath et al., 2015; Goodale et al., 2019, 2020). Future work on Prionostemma can provide novel insights on the function of chemical compounds in harvestman, as well as the role of chemical communication in group formation.

Three combinations of species in MSA represented almost 54% of the field records (Table 4). This is surprising as seven species can generate 120 possible combinations of species. Additionally, the three most common species tended to occur together. For visually oriented predators, two of these common species may be regarded as cryptic (Prionostemma sp.6: dark gray body; and Prionostemma sp. 7: green body), whereas one of them may be regarded as conspicuous (Prionostemma sp.5: red body) (Figure 3). In all three most frequent associations, there was a mixture of a conspicuous species and at least one cryptic species. This difference in conspicuity may promote asymmetries in the defense benefits obtained by individuals of each species. Two possible scenarios of these between-species asymmetries are plausible. First, individuals of conspicuous species may be safer in MSA composed mainly of individuals of cryptic species, because SSA of conspicuous species may attract more attention of visually oriented predators than SSA of cryptic species (Vulinec, 1990). Second, assuming that all Prionostemma are chemically defended, individuals of cryptic species may be safer in MSA composed mainly of individuals of conspicuous species if predators avoid attacking aggregations of aposematic prey (e.g., Rippi et al., 2001). Alternatively, individuals of both cryptic and conspicuous species may enjoy similarly higher defense benefits in MSA if the confusion effect they cause on potential predators is more intense than in SSA (see section ‘Benefits of mixed-species aggregations in harvestmen’ below).

The number of individuals and species in the roosting sites showed marked variation over time. All possible transitions between the categories of roosting status (solitary, SSA, and MSA) were frequently recorded (Figure 5). To our knowledge, this is the first time that the temporal dynamic of mixed-species aggregations is explored in harvestmen. Our data reinforces previous suggestions that Prionostemma individuals are constantly moving between aggregations (Donaldson and Grether, 2007). Why individuals do that is an open question that deserves investigation. One possibility is that suitable roosting sites are very common in the study site. After leaving a roosting site at dusk, an individual may go far away searching for food or mating partners. At dawn, instead of paying the costs of returning to the original roosting site, these individuals may simply go to the best roosting site nearby. An important implication of the constant movement of individuals is that one roosting site may have an aggregation in 1 day and a solitary individual in the next day. Moreover, frequent transitions between SSA and MSA suggest that the benefits and costs associated with these two types of aggregations are similar. Otherwise, we would expect higher temporal stability and low frequency of transitions.

We found a high frequency of leg loss in all Prionostemma species studied here (Figure 6A). This is consistent with previous reports for many Eupnoi harvestmen (e.g., Guffey, 1999; Houghton et al., 2011; Escalante et al., 2013, 2020, 2021; Domínguez et al., 2016; Escalante and Elias, 2021; Powell et al., 2021a,b). However, leg loss had no effect on the roosting status of the individuals, i.e., whether they were found solitary or in aggregations (SSA or MSA). Our hypothesis relating leg loss to the defense benefits of gregariousness assumed that autotomized individuals are more exposed to predation than intact individuals. Therefore, a first explanation for the lack of relationship between leg loss and roosting status is that our assumption does not hold. Although leg loss hampers the locomotor performance of sclerosomatid harvestmen (Escalante et al., 2013, 2020, 2021; Domínguez et al., 2016), a recent study showed that the recapture rates of autotomized and intact individuals of Prionostemma sp.5 are similar, suggesting that autotomy does not negatively impact individual survival (Escalante and Elias, 2021). That study also showed that autotomy changed micro-habitat use so that individuals missing legs were more frequently recaptured roosting among moss than on exposed tree trunks (Escalante and Elias, 2021). This finding suggests that autotomized individuals seek protection in concealed micro-habitats and perhaps this behavior is more effective in preventing predation than joining an aggregation.

If autotomized individuals are indeed more exposed to predation and find protection against predators in aggregations, we hypothesized that MSA would confer higher defense benefits due to the confusion effect, and predicted the probability of finding autotomized individuals roosting in MSA would be higher than in SSA. Our results support this prediction only for Prionostemma sp.7: the percentage of autotomized individuals was higher in MSA than in SSA. For Prionostemma sp.5 and sp.6, although the results we found are in the expected direction (i.e., a positive value of contrast 9 in Figure 7E), they were not statistically significant. This finding suggests that any possible defense benefit caused by the presence of multiple species in the aggregation is asymmetric. Whereas individuals of Prionostemma sp.7 may have higher defense benefits in joining MSA, the same is probably not true for Prionostemma sp.5 and sp.6. The reasons for this asymmetry remain to be better understood, but are unlikely to be explained by the confusion effect. As we mentioned above, Prionostemma sp.5, sp.6, and sp. 7 are commonly found together and show marked difference in body color. Thus, if the confusion effect emerges because phenotypic variation among aggregated species disrupts the search image of predators (Table 2), autotomized individuals of the three species should prefer to join MSA, which was not found in the field.

Finally, we expected that autotomized individuals of all species would prefer to join larger aggregations, where they would be more protected (Vulinec, 1990). This prediction was not supported by our data as aggregation size was not associated with the percentage of autotomized individuals. Although gregariousness may decrease the individual chance of predation via the dilution effect, it may also increase the chances of detection by predators because aggregations may be visually or chemically more conspicuous than isolated individuals (Tables 2, 3). In sclerosomatid harvestmen, the main predators during daytime (when individuals are roosting) are probably birds (Cokendolpher and Mitov, 2007), which are visually oriented and may detect aggregations more easily than solitary individuals. Therefore, the encounter and dilution effects may operate in different directions (see Figure 1 in Wrona and Dixon, 1991), and both need to be considered simultaneously in future studies (i.e., the attack-abatement effect sensu Turner and Pitcher, 1986) to fully assess the net fitness benefit of gregariousness.

Mite prevalence for most Prionostemma species studied here (Figure 6B) is within the range already reported for other sclerosomatids worldwide, which varies from 16.7% (Townsend et al., 2008) to 61% (McAloon and Durden, 2000). In a comparative study with six harvestman species from Trinidad, Townsend et al. (2008) suggested that differences between species in mite prevalence are mainly related to habitat use: species that live or forage on the leaf litter are more likely to be parasitized because erythraeid mites lay their eggs in the soil. In a similar way, a study on two gregarious Prionostemma species in Nicaragua showed that mite prevalence varied between roosting sites (Grether et al., 2014a). In some spiny palms, no mites were found infesting the aggregated individuals, but in one of them, mite prevalence was 31.1%. Here, we investigated whether the roosting status of the individuals affects mite parasitism in the four most common Prionostemma species in our study site. We found that mite prevalence differed between species (Figure 6B), which suggests some level of host specificity or preference by the erythraeid mites. However, within species mite prevalence was similar for solitary and aggregated individuals – regardless of whether they were in SSA or MSA (Figure 8). This finding refutes our hypothesis that gregariousness increases the chance of horizontal transmission of ectoparasites.

The intensity of parasitism in the seven Prionostemma species studied here ranged from 1 to 8 mites per individual. As reported in previous studies with sclerosomatids from the United States (McAloon and Durden, 2000) and phalangiids from Bulgaria (Mitov, 2000), most individuals carried few mites, whereas a few individuals were heavily parasitized. Clumped parasite distributions on hosts are a widespread pattern in animals (Leung, 1998). In the case of erythraeid mites, only the larval stages are parasitic, so that the high parasite load of few Prionostemma individuals may result from a host being exposed to an aggregation of larvae just after they hatch (McAloon and Durden, 2000). Despite the great inter-individual variation, we showed that the intensity of parasitism was not affected by the roosting status (Figure 9). This finding reinforces the conclusion that aggregating with conspecific or heterospecifics does not bring different costs in terms of mite parasitism in Prionostemma harvestmen.

The fluid membership of the Prionostemma aggregations, with marked daily turnovers in the number of individuals and species composition at the roosting sites (Figure 5), can have important implications for the transmission of mites between individuals. Given that individuals are likely changing their roosting status over time, a snapshot sample (as the one conducted here) does not provide a precise picture of their risk of being parasitized. A recent mathematical model has shown that aggregation size and roost site fidelity are key factors influencing pathogen spreading in populations of gregarious species (Laughlin et al., 2019). According to this model, pathogens spread faster among roosting sites when (i) individuals are distributed among a large number of small aggregations and (ii) exhibit low roost site fidelity. These two conditions apply to Prionostemma harvestmen (Figures 4A, 5), which may explain why nearly 70% of all aggregations had at least one individual carrying mites. However, to better understand the factors that influence the individual risk of being parasitized, future research should be conducted with individually marked harvestmen. The question to be explored is whether individuals with higher tendency of being found in aggregations are more likely to be parasitized than those with higher tendency of being found solitary. There is increasing evidence that individuals vary widely in their probability of contracting and spreading parasites (Barron et al., 2015), and Prionostemma harvestmen offer an opportunity to explore this subject in species that form both SSA and MSA.