Study region and groups
We run this study from January to July 2019 in the rural region of Viamão, Rio Grande do Sul state, Brazil, near the southern limit of the distribution of brown howler monkeys (Culot et al. 2019). The landscape of the region is composed of a mosaic of forest fragments with varying levels of disturbance, crops, pastures and rural and suburban human settlements. At the patch-scale, we analyzed fecal samples of 60 groups of howlers (2-9 individuals, mean=5, SD=2) that inhabited 60 isolated forest fragments (1.2-257 ha, mean=25.8, SD=50.5, median=6.7, Fig. 1a). We analyzed independent landscapes surrounding 32 of these forest fragments for the analysis at the patch-landscape-scale (Fig. 1b). Howler group size in this subsample ranged from two to seven individuals (mean=5, SD=2, N=32).
Fecal sample collection and parasitological analysis
We collected 295 fecal samples from all individuals of the 60 groups once for the analysis at the patch-scale. We used the subsample of 32 groups above for the analysis at the patch-landscape-scale. Our sampling was approved by the Brazilian System of Authorization and Information on Biodiversity (SISBIO license nr 66648-1), although the Brazilian legislation (Article 10 of IBAMA’s Normative Instruction nr. 154, 1st March 2007) does not require a license for the collection of fecal samples outside of legally protected nature reserves.
We collected ca. 2 g of material from the center of each stool to avoid contamination with larvae, eggs and oocysts found on the forest floor (Gillespie 2006) using disposable wooden spatulas. We pooled all individual samples of each howler group for assessing their parasite richness and preserved them in 10% formalin. This pooling increases the likelihood of detecting the group’s parasites because the release of eggs, oocysts and larvae is not continuous; that is, while a parasite of a given host may lay eggs in a given day, a conspecific parasite in another host individual from the same species may not (Gillespie 2006). Therefore, the likelihood of sampling all parasite taxa may increase with an increase in the number of stools composing a group’s fecal pool. We included the number of fecal samples (=howler group size) per patch or patch-landscape in the modelling to assess its potential effect on the patterns of parasite richness. We transported the fecal samples in ice within 8 h of collection and stored them in a refrigerator at ca. 2ºC until analysis, which took place after one to eight months of collection.
We used the flotation and the centrifuge-sedimentation in formalin ethyl-acetate techniques (De Carli 2001) to separate eggs, oocysts, cysts, larvae and adult parasites from the fecal remains of 4 g of each group’s fecal pool. We analyzed the slides under an Olympus CH30 stereoscopic microscope using 200x magnification lenses.
We classified the parasite richness (number of parasite species) of each fecal pool into four categories: (a) overall richness, and richness of species with (b) direct cycle and transmission via ingestion of the infective stage in the arboreal milieu (hereafter direct-arboreal), (c) direct cycle and transmission via ingestion of the infective stage on the ground (hereafter direct-soil) and (d) indirect cycle and transmission via ingestion of the intermediate host (hereafter indirect-IH). We took advantage of adult helminth specimens recovered in necropsies of howler monkeys that died in conflicts with the study region’s anthropogenic environment (Jesus et al. submitted) to identify the helminths at the species or genus levels because the taxonomic identification of eggs is unreliable (Gillespie 2006; Solórzano-García and Pérez-Ponce de Léon 2017).
We treated each forest fragment as a sampling unit in the patch-scale approach. We estimated fragment area (size) using polygons created in Google Earth Pro version 7.1.8 (Google Inc. 2017). For the patch-landscape-scale approach, we estimated forest cover, matrix permeability, patch density and Euclidean mean distance to the nearest fragment (Table 1) in radii from the center of the focal fragment (Arroyo-Rodríguez and Fahrig 2014) of each of the 32 independent patch-landscape sampling units.
We quantified the types of land cover in each landscape using satellite images with 30-m spatial resolution made available by the Brazilian Annual Land Use and Land Cover Mapping Project (MapBiomas, collection 4). We classified the land cover types following MapBiomas: forest formation (including dense, open and mixed ombrophilous forests, semideciduous and deciduous seasonal forests, and secondary forest), planted forest of commercial tree species, grasslands, farming (including annual and perennial crops and pasture), wetlands, water (rivers and lakes) and urban infrastructure (urban areas with a predominance of non-vegetated surfaces, including buildings and roads and other transportation infrastructure). The mapping of the MapBiomas Project has an accuracy of 85.8% for the Atlantic Forest biome. We used ArcGis 10.3 (Esri 2014) for the GIS processing and Fragstats (McGarigal et al. 2012) to calculate the landscape metrics described below.
The proportion of the patch-landscape covered by forest is the main metric of habitat availability for arboreal primates such as howler monkeys. A larger forest cover may promote a lower richness of direct-soil parasites because howlers will be less likely to descend to the ground to cross non-forest matrix elements. It may also promote a lower richness of direct-arboreal parasites because howlers will be able to use larger home ranges, thereby reducing the risk of reinfection (see Bicca-Marques and Calegaro-Marques 2016).
The type of matrix influences the effectiveness of fragment isolation via its permeability to species dispersal (Metzger and Décamps 1997). A permeable matrix that allows howlers to move between forest fragments increases their risk of infection with direct-soil parasites. We classified the permeability of land cover types in a gradient from low (weight 1) to high (weight 10) by calculating the mean of their weights in the literature (Galán-Acedo et al. 2019b; Rabelo et al. 2019; Jardim et al. in prep.; see Supplementary Material Table A1).
Patch density is a measure of the fragmentation of the patch-landscape. A highly fragmented patch-landscape may reflect a greater presence of people and domestic animals in the landscape and inside the target forest fragment. This presence increases the howlers’ risk of contact with generalist direct-soil parasites shared with these hosts.
The mean Euclidean distance to the nearest forest fragment is a measure of between-fragment isolation in the patch-landscape. The higher the isolation between fragments in a given patch-landscape, the longer the distance that howlers have to cross in the matrix to move between habitat patches. Consequently, the higher the risk of infection with direct-soil parasites.
We identified the spatial scale with the greatest explanatory power (scale of effect) of the categories of parasite richness in the analysis at the patch-landscape-scale (Jackson and Fahrig 2012). We built buffers with radii of 250, 500, 750 and 1,000 m from the center of the target forest fragment of each patch-landscape (Fig. 2). We used 250 m as the smallest radius because the likelihood of successful howler dispersal through a non-forest matrix between discrete habitat patches decreases significantly at distances longer than 200 m (Mandujano and Estrada 2005). We calculated the effect of each patch-landscape metric (Table 1) for each category of parasite richness inside each buffer. The 750-m buffer, for instance, showed the greatest effect of forest cover on overall parasite richness. Then, we generated an equation containing all patch-landscape metrics and their scales with greatest effects to model their potential as predictors of parasite richness. The equation for the modelling of overall parasite richness was:
where is the forest cover inside the 750-m buffer, is the patch density inside the 1,000-m buffer, is the mean Euclidean distance to the nearest forest fragment inside the 1,000-m and is the permeability of the matrix inside the 1,000-m buffer.
We used the variance inflation factor (VIF) to check for multicollinearity between variables at the patch-landscape-scale. We excluded matrix permeability from all equations because it was strongly collinear with forest cover (VIF>4; Supplementary Material Table A2) in all models. The remaining three metrics were not colinear (all VIF<4; Supplementary Material Table A3). Therefore, we modelled the effect of forest cover, patch density and mean Euclidean distance to the nearest forest fragment on the four categories of parasite richness.
We used generalized linear mixed models (GLMMs) to assess the relationship between habitat patch (forest fragment) size or patch-landscape metrics and the four categories of parasite richness. We checked the normality, homoscedasticity and autocorrelation of residuals to validate the models. We built the models with the Gaussian family because these assumptions were met. Moreover, we used the logit family to build binomial models with binary variables. We used fragment size and howler group size as fixed factors and season of fecal sample collection as random factor in the global model of the analysis at the patch-scale. Similarly, we used group size and the three patch-landscape metrics as fixed factors and season of fecal sample collection as random factor in the global model of the analysis at the patch-landscape-scale. We included the season of fecal sample collection because it may influence the dynamics of parasitic infections due to seasonal fluctuations in climatic conditions (Altizer et al. 2006). We used the function dredge of the MuMln package of R (Barton 2016) to assess the influence of all predictor combinations on the four categories of parasite richness.
We used the Akaike Information Criterion (AIC) to select the model(s) with the greatest explanatory power of the predictor effects on parasite richness. Specifically, we used the AICc as recommended for small samples (Burnham and Anderson 2003). Although the model with the lowest AICc has the best adjustment, all models with ΔAICc<6 are equally parsimonious (Richards 2015). We considered that a given patch-landscape metric explains the parasite richness of howler monkeys if it is included in the best model or in many parsimonious models (Richards 2011) and if its relationship with parasite richness is significant. We run all analyses in R 3.5.1 (R Core Team 2018) using the lme4, car and MuMln packages (Bates et al. 2015; Barton 2016; Fox and Weisberg 2019). We set a level of significance of 0.05 in all analyses.
All associated data will be available in a data repository when the paper is published.