Cacti fruits are key resources to many frugivorous animals in Neotropical arid and semiarid regions. However, most studies have focused on a particular animal group or cacti species, but few have explored the overall interactions of such species at the community level. Here we monitored frugivory on five cacti species using camera traps that sampled diurnal and nocturnal interactions. We investigated the structure of interactions with bird, mammal, and reptile frugivores in the Brazilian Caatinga dry forest. We hypothesized that the height of cacti limit interactions with different types of frugivores, which would result in highly structured and modular interaction networks. In 2,929 camera-days, we recorded 23 vertebrate species feeding on cacti fruits, including seven new records, all determined to be primary seed dispersers. As predicted, the cacti-frugivore network was modular and non-nested, with the two shortest cacti species grouped in a module dominated by interactions with reptiles and non-flying mammals. The tallest cacti species were dominated by frugivory interactions with birds and had comparatively less interaction diversity than shorter cacti species. Our results support the contention that cacti are keystone species in semiarid ecosystems where they produce small-seeded fleshy fruits year-round.

Fruiting phenology

We collected monthly data on fruit availability on the cacti from September 2018 to July 2019 (except in June). To collect phenological data, we established five transects in the study area based on pre-existing trails, covering a total area of 5,250 m². All transects had 5 m width but different extensions (130 m, 120 m, 100 m, 300 m, and 400 m) because the trails’ lengths were different. We then quantified and summarized the number of reproductive individuals (or agglomerations in the case of T. inamoena) per cacti species along transects (Table S2). Every month, we recorded the presence/absence of the fruiting phenophase and quantified the number of individuals with ripe or unripe fruits (d’Eça-Neves and Morellato 2004). We then calculated the monthly percentage of fruiting individuals for every species by dividing the number of fruiting plants (sum of all plots) by the total number of adult plants (sum of the individuals’ number at all plots).

Frugivory records

We monitored vertebrate frugivory events in five cacti species for 11 months, from September 2018 to July 2019, with camera traps (Bushnell 8 MP Trophy Cam). We searched for cactus individuals of every focal species with ripe fruits over the study area, and the individuals chosen to be monitored with cameras could include, but were not restricted to, those observed for phenology. The sampling effort of camera-days was balanced between the cacti species by monitoring an equivalent number of individuals of every species every month, intensifying the sampling in their respective fruiting peaking. Our sample effort was camera-days of sampling time to M. zehntneri (711), T. inamoena (450), X. gounellei (621), C. jamacaru (699), and P. pachycladus (448), totalling 2,929 camera-days of monitoring and an average of four individuals per species monitored each month (Table S2).

We installed the cameras over the study area, tied in trees or in wooden stakes when necessary to reach the fruits of tall columnar cacti (> 4 m height), and kept a distance of at least 70 m between cactus of the same species monitored at the same time. We set the cameras close (2–5 m) and toward a cactus fruit ensuring that all fruits available were framed by the camera. We programmed them to take two pictures followed by a 10 seconds video, in minimal intervals of one minute when triggered. This setting was appropriate to detect fruit consumers and the videos helped to ensure the detection of frugivory events; this time interval aimed to prevent the cameras from taking many redundant registers. We left the cameras in the field working 24 hr per day to maintain an equal daytime and nighttime sampling effort for both diurnal and nocturnal frugivores (Blanco et al., 2019). 

Every month we removed the cameras to collect their memory cards and installed them in other cacti because this time was enough for the fruit to ripen and fall or to be entirely consumed. At the laboratory, we analyzed all the content recorded in the cameras to quantify the fruit consumption events. We considered one event of frugivory when: there was a photo/video of an animal eating one or more fruits; when there was a photo/video of an animal standing in front of the fruit and in the next photo/video, if less than five minutes after the previous, the animal and the fruit were gone; when there was a sequence of photos/videos of the animal eating the fruit continuously without leaving.

We considered two or more events of frugivory when an animal was seen in a photo/video eating and then leaving the fruit and in the next photo/video coming back and eating again, or when there were two individuals eating from the same or different fruits in the same photo/video. For each frugivory event, we recorded: the cactus species, animal species, date, time of the day, and fruit height. We estimated the fruit height visually in the field when installing the cameras for all cacti species except M. zehntneri, for which we used the mean height from the literature (Menezes et al., 2013). We identified every animal at the species level when possible using field guides and checking with specialists. To access whether our records represented well the vertebrate community that utilizes cacti fruits, we performed species accumulation curves for every cacti species in software EstimateS 9 (Colwell 2013).

Network structure

We calculated nestedness and modularity in order to describe the organization of the cacti-frugivore network (Bascompte et al., 2003, Olesen et al., 2007). For that, we built a quantitative interaction matrix where the cacti species were set in the rows (lower level) and the vertebrate species were set in the columns (higher level). Cells were filled with the frequency of interactions by vertebrate frugivore species on a cactus species divided by the sampling effort (camera-days) and multiplied by 1000. With this we built a network graph and performed network metrics analyses. The graph was drawn in Pajek 5.15 (Batagelj and Mrvar 2003) using the ‘Kamada-Kawai – separate components’ method, in which vertices (i.e. species) with a larger number of connections or connecting modules are drawn closer to the center.

Nestedness quantifies the degree to which interactions of specialized species are subsets of interactions of the more generalist species in the network and was quantified by the wNODF index (Almeida-Neto and Ulrich 2011). The wNODF has values ranging from 1 (not nested) to 100 (perfectly nested). We analyzed network modularity that quantifies the prevalence of interactions within subsets of species in the community and was calculated using the DIRTLPAwb + algorithm (Beckett 2016) using the computeModules function in the Bipartite package. In this case, modularity varies from 0 (no modules) to 1 (maximum degree of modularity). To test the significance of wNODF and Q value, we used 1000 randomizations for the observed network with the null model proposed by Vázquez et al. (2007), which preserves marginal totals (takes account of interaction abundance) and keeps network connectance constant (Vázquez et al., 2007, Dormann et al., 2008). A null model was generated with the function vaznull with the R-package bipartite (Vázquez et al., 2007, Dormann et al., 2008). We estimated a 95% confidence interval for modularity (DIRTLPAwb +) and Nestedness (wNODF) metrics from the 1000 simulated values and a metric value was considered significant if it did not overlap with the confidence interval.

Frugivory patterns

We used Bray-Curtis (B-C) dissimilarity index to compare frugivore composition among cacti species. Values of B-C index range from 0 to 1, where lower values indicate that cacti species interact with a different set of animals. We expected that cacti that have similar heights would share higher values of B-C, which means they share a similar set of interacting species. We performed the analyses on software PAST version 3.0. In addition, to examine if fruit height influences frugivory by birds, mammals, and reptiles, we used generalized linear mixed-effect models (GLMM) with a Poisson error family to model the number of visits as function of fruit height and animal group as a fixed effect, and cacti species (n = 4) as a random effect. We conducted the analyses using the package “lme4” in software R version 4.2.2.

Excel, EstimateS 9, R Studio, PAST.