Mechanisms promoting stable coexistence allow multiple species to persist in the same trophic level of a given network of species interactions. One of the most common stabilizing mechanisms of coexistence is niche differentiation, such as temporal and spatial patchiness. To understand the limits of coexistence between species we have to understand the limits of competitive interactions which translate in species exclusion or patterns of non-co-occurrence.

We evaluated spatiotemporal niche-based mechanisms that could promote stable coexistence between ants and spiders which forage on extrafloral nectary (EFN)-bearing plants. We observed co-occurrence and overlapping patterns between ants and spiders in a temporal and spatial scale in nine different EFN-bearing plant species in a Neotropical savanna, using both community- and species-level approach.

Ants and spiders showed asynchrony of their abundances over the year with low temporal overlapping patterns between them (temporal niche specialization). Greater abundance of ants occurred between September and March, whereas greater abundance of spiders occurred between March and August, exactly at the time when the abundance of ants decreases on plants. However, there might also be some levels of temporal overlapping, but then individual ants and spiders occupy different branches (spatial segregation). Finally, we also observed a spatial negative effect of the abundance of ants on the presence of spiders.

Synthesis. Our results suggest that spatiotemporal partitioning between ants and spiders may be one of the potential mechanisms behind a stable coexistence between these two groups of organisms that forage on EFN-bearing plants in the Brazilian savanna. 

Study area and plant species

Field study was conducted in a private natural savanna reserve (Clube Caça e Pesca Itororó de Uberlândia – 48º18’27”W; 18º59’27”S) in Uberlândia, Minas Gerais State, south-eastern Brazil. The vegetation is dominated by cerrado stricto sensu, consisting of trees 2–8 m height, with an understory dominated by shrubs, grasses and scattered perennial herbs (Appolinario & Schiavini, 2002; Del-Claro et al., 2019). The climate in the region is rainy from October to March and dry from April to September (Alvares et al., 2013; Novaes et al., 2020). In 2009, the total precipitation during rainy season was 2295mm, while in dry season was 597mm; and in 2010, the total precipitation during rainy season was 1110mm, while in dry season was 174mm. These both years represent the years of data collection.

We selected nine EFN-bearing tree species in our study site with the highest importance value index (Appolinario & Schiavini, 2002), namely: Caryocar brasiliense (Cambess) (Caryocaraceae), Lafoensia pacari (A. St.-Hil.) (Lythraceae), Ouratea spectabilis (Mart.) Engl., O. hexasperma (A. St.-Hil.) Baill (Ochnaceae), Qualea grandiflora(Mart.), Q. multiflora (Mart.), Q. parviflora (Mart.) (Vochysiaceae), Stryphnodendron adstringens (Mart.) Coville and S. polyphyllum (Mart.) (Fabaceae). 

Experimental manipulation

In January 2009, we selected two similar and opposite branches (80cm long and with similar number of leaves within species) of 30 individuals of each species (270 plant individuals in total; 540 branches in total). Our study followed the manipulation of Lange and Del-Claro (2014) which consisted of ant exclusion in one of the branches (Treatment) and free foraging of ants in the other branch (Control). For ant exclusion in Treatment branches, we added a band of non-toxic ant repellent resin (Tanglefoot®, Rapids, Michigan) at the base of the branch. It is important to mention that Tanglefoot does not exclude spiders, since they can jump or reach branches by using web threads. For the Control, we also applied the resin, but in this case, we covered only half part of branch circumference, allowing ants to forage freely. We also added the resin to the Control branches to control possible effects of the resin on the foraging of spiders or even of ants. At the beginning of the experiment, and biweekly and manually, we removed all ants and all structures that could act as bridges giving access to ant-excluded branches. We monthly checked the integrity of resin barrier in all branches. During 19 months (starting in January 2009), we monthly counted all ants and spiders in control branches, and only spiders in ant-excluded branches, between 8:00 h and 11:30 h. When monitoring ant excluded branches (Treatment), we did not see any ant, showing that our ant resin barrier worked. Furthermore, one may suggest that there is a temporal segregation during the day between ants and spiders, i.e., ants occurring during the day and spiders at night. However, both groups are commonly present in EFN-bearing plants in both periods of the day (Anjos et al., 2017; Calixto, Novaes, et al., 2020; Calixto, Lange, et al., 2020; Lange et al., 2017; Nahas et al., 2016).

Data analysis

We conducted all analyses using the software R 4.0.0 (R Core Team, 2020). Below we described the packages used for each analysis.

Ant-spider relationship

Since spiders can detect and avoid ant cues (Mestre et al., 2020), we first evaluated whether ant presence can impact on the frequency of spider presence, and whether this impact changes over time. If we find a significant result (higher frequency of spiders in branches without ants and over time), we confirm that ants can influence spider foraging over time, and then we can suggest spatiotemporal niche segregation between them. For that, we compared the frequency of spider presence over the year and between treatment branches (Control and Treatment) at plant community level (data of all plant species together). We conducted a Generalized Linear Mixed Model using Template Model Builder (GLMM) with binomial error followed by Wald’s test, in which we used the interaction between months and experimental treatments (Month*Treatment) as predictor variable, and the frequency of spider presence as response variable. We also added plant nested within month as random factor (1|Month/Plant) to control temporal and spatial repeated measures (Crawley, 2007). The model was fit using the package “glmmTMB” (Brooks et al., 2017) and Wald’s test was conducted in the package “car” (Fox & Weisberg, 2011). In this analysis, we considered data from all months observed (19 months).

Spatial scale analyses

To evaluate spatial-scale patterns of ants and spiders, we conducted co-occurrence analysis. We used this analysis to compare the patterns of co-occurrence between ants and spiders at plant community and species level. Two clarifications are needed regarding this analysis: 1) first we need to use only Control branches. Ants are not present in Treatment branches and adding these branches in our analyses would bias the results by increasing non-co-occurrence patterns; 2) second, we need to use Control branches that have at least one of the predators, since analyzing co-occurrence patterns from the absence of both would also bias the data by increasing non-co-occurrence patterns. Using these two approaches, we are testing how ants and spiders are spatially distributed on the same plant individual. Then, we used the C-score index from Stone and Roberts (1990) and observed non-random patterns of species co-occurrence based on a matrix of presence/absence of ants and/or spider on branches by randomizing (1000 randomizations) the original matrix (Gotelli, 2000; Ribas & Schoereder, 2002; Sanders et al., 2007). The C-score quantifies the number of checkerboard units that can be produced for each pair of species, i.e., it represents the mean number of checkerboard units between pair of species. By randomizing 1000 times the original matrix, we can compare the observed C-score to the average C-score generated by randomization. We used the function cooc_null_model and the algorithm sim2 in the “EcoSimR” package (Gotelli et al. 2015). This algorithm assumes that all “sites” are equiprobable, and it would be appropriate for quadrat censuses in a relatively homogeneous environment (Gotelli, 2000). In addition, sim2 has the best overall performance for species co-occurrence analysis (Gotelli, 2000). In this analysis, we considered the data from all months observed (19 months). Here, if we find a C-score larger than the average C-score produced from null distribution (P-value < 0.05), we assume that there is a pairwise species co-occurrence than expected by chance, i.e., there are no co-occurrence patterns between ants and spiders, suggesting spatial segregation.

In addition, we conducted a GLMM with binomial error followed by Wald’s test to verify whether the abundance of ants interferes with the presence of spiders on branches at both plant community and species level. For this, we used Control branches that had at least one ant and/or spider present, as detailed previously. We fit our model with the presence of spiders as response variable, abundance of ants as explanatory variable, and plants as random factor. In this analysis, we also considered data from all months observed (19 months).

Temporal scale analyses

To check for temporal-scale patterns (asynchrony) between ants and spiders at plant community and species level, we conducted two sets of analyses: one considering overlapping patterns and another considering seasonal patterns. To verify the overlapping patterns over the year between ants and spiders, we used two temporal abundance matrices (based only on 12 months of field collection). Then we fit niche overlap null models using Pianka’s metric and two randomization algorithms (1000 simulated null assemblies), RA3 and RA4, through the package “EcoSimR” (Gotelli et al., 2015). RA3 reshuffles the row (month) values, and RA4 reshuffles the non-zero row values; however, both retain the observed “niche breadth” of each species. In overlapping analysis, if we find P-values greater than 0.05 it means that there is no niche overlap between ants and spiders, and therefore both groups are not present on plants at similar times throughout the year (temporal niche partitioning). 

We used circular statistical analyses to evaluate whether there is seasonality of the presence of ants and spiders over the year. Months were converted into angles (1 month = 30 degrees), and the abundance of ants and spiders were used to calculate the mean vector (μ), length of mean vector (r), median, circular standard deviation, Rayleigh test Z, and Rayleigh test p. Rayleigh test p with value less than 0.05 and mean vector length (r) close to 1 indicate seasonality of the data, that is, phenological activity is concentrated around one single date or mean angle (Morellato et al., 2010). The mean month for ants and spiders abundance was obtained by converting the mean angular directions into corresponding mean months, if analyses were significant (Morellato et al., 2000; Morellato et al., 2010; Vilela et al., 2017). We performed Watson’s goodness of fit test before conducting circular analyses to verify data unimodality. Next, we compared the activity peak between ants and spiders performing a Watson-two test (U²) and evaluating circular diagrams and the mean vector (µ). Whether our results show that there is a significant difference between the mean vectors of ants and spiders, then we can infer that there is temporal niche partition. Circular analyses were conducted with the package “circular” (Agostinelli & Lund, 2017). In these overlapping and circular analyses, we only used twelve months of observation as both approaches work with cycles and therefore the analyses and results are more consistent with a closed cycle (e.g. annual analysis) than working with open ones.