The study was carried out in the tropical region of Lacandona, south of the state of Chiapas, Mexico. The original vegetation consists mainly on lowland tropical rain forests. Deforestation of the region began in the 1970’s, resulting in the reduction of old-growth continuous forest from 95% in 1976 to 56% in 1996 (De Jong et al., 2000); only 36% of the original old-growth continuous forest remains today (Carabias et al., 2012). Currently, the main land-use practices in the region consist of cattle pastures, the cultivation of maize and other crops (De Jong et al., 2000; Zermeño-Hernández et al., 2015). The resulting landscape comprise a mosaic of human-modified habitats that include semi-urban settlements, agricultural land, open pastures for cattle, riparian zones, patches of secondary and old-growth forests of various sizes. The region has a mean annual temperature of 24°C; average annual rainfall is 3,000 mm with June to October as the wettest months (551 mm month^–1) and February to April as the driest months (<100 mm month^–1) (Comisión Federal de Electricidad, 2006; van Breugel et al., 2006).

Four different habitat types were selected for this study including the following: (i) riparian habitat within old-growth continuous forest (RM); (ii) riparian habitat in active cattle pastures (RP); (iii) old-growth continuous forest 1,000 m away from riparian vegetation (MF); and (iv) active cattle pastures (P) 1,000 m away from riparian vegetation. For a total of 12 sampling sites. Each habitat type was replicated three times and study sites were at least 1.5 km away from each other (Figure 1). Streams were all permanent (although with variable amounts of running water throughout the year) while stream width varied from 2 to 8 m. Study sites in pastures were active cattle pastures, although these sites were predominantly devoid of a tree cover, all sites had isolated trees and a few trees that serve as live fences. It is common that isolated trees are left standing to provide shade for cattle, firewood for cooking, or for aesthetic reasons (Galindo-González et al., 2000). The cattle pastures were located in the Marqués de Comillas municipality, on the south side of the Lacantún River. Old-growth continuous forest sites were located in the 330,000 ha Montes Azules Biosphere Reserve (MABR) on the north side of the river (16°04′ N to 90°45′ W; INE, 2000, Figure 1).

Sampling of bats was conducted twice during the dry season (December to May) and twice during the wet season (June to November) for 3 consecutive years (2011, 2012, and 2013), using a standardized method of four nights per site. A previous study in the same locations showed that over 70 sampling nights, 34 during the dry season and 36 during the rainy season, a total of 1,752 individuals belonging to 28 species of Phyllostomidae were captured (de la Peña-Cuéllar et al., 2015). For this study however, we considered the six most abundant species of phyllostomid bats in the region, including the following: Artibeus jamaicensis, A. lituratus, Sturnira lilium, Carollia perspicillata, Glossophaga soricina, and Platyrrhinus helleri, the minimum number of captures needed to be considered in the study were one capture per habitat per season, species abundance cut off of n = 60.

Five nets (12 m long × 2.6 m high) were set at ground level and were opened at dusk (1800–1830) for four consecutive hours, which corresponds to the peak foraging time for most phyllostomid species (La Val, 1970). The bat sampling nets were arranged according to habitat type: (1) in the riparian habitats (including continuous forest and pastures), nets were located parallel and/or diagonally across the stream, depending on site characteristics; (2) in continuous forest, nets were positioned across natural flying corridors; (3) in active pastures, nets were located under the canopy of isolated trees. In all sites we searched for similar physical characteristics that allow the same mist net arrangement, this is one individual net and two pairs of nets in an “L” shape (two nets connected perpendicularly). Nets were located ca. 50 m apart. Nights with a full moon or heavy rain were avoided during bat sampling in order to prevent variation in capture success associated with these conditions (Morrison, 1978). Captured individuals were temporarily stored in cloth bags and identified to species following Medellín et al. (2011). For all captured bats, we determined sex by inspecting genitalia (Racey and Speakman, 1987). In females, we detected pregnancy by palpation (Racey and Speakman, 1987), and lactation by the occurrence of enlarged nipples surrounded by a hairless skin area and by extruding milk with a gentle finger pressure on the nipple base. Sex ratio was calculated as the ratio of males to females in each site (Russo et al., 2010).

There is a risk of exposure to some significant zoonotic agents for any person handling bats in the field, following biosecurity recommendations (Newman et al., 2011). For this, all participants that sampled bats were appropriately trained to handle bats; bites and scratches were avoided by using leather gloves, previous anti-rabies vaccination and hand washing before and after bat manipulation.

To determine the influence of vegetation structure on the abundance of male and female bats, for each site we recorded all trees ≥10 cm diameter at breast height (dbh) within a 0.1 ha (20 × 50 m) plot (Gentry, 1982). Transects were located along streams in riparian continuous forest and riparian pasture habitats and randomly located in continuous forest and pasture habitats. We considered the following vegetation attributes: density of individuals, species richness, forest basal area, and tree height.

First, we compared capture rates for each sex among habitat types by using a standardized capture rate (captures/1,000 mist net hour) that compensated for differences in number of nets, size of nets and length of time nets were open (Perry et al., 2010). We compared capture rates using analysis of variance on ranks (ANOVA). Data were checked to meet assumption of homoscedasticity and normality. We performed non-metric multidimensional scaling analyses (NMDS) based on the identity and abundance of tree species occurring in the sampling sites, to obtain a continuous synthetic variable summarizing dissimilarity patterns among vegetation species composition. The matrix used in the analysis was built using the Bray-Curtis index (Magurran, 2004). This iterative method of ordination has the advantage of properly handling non-linear species response of any shape (Oksanen, 2013) and has a good performance even when beta diversity is high (McCune and Grace, 2002). It is one of the preferred ordination methods for analyzing community data (McCune and Grace, 2002). The scores of axis 1 were used as an explanatory variable for evaluating differential sex response to tree species composition. Second, we fitted a general linear mixed model (GLMM) for each species separately with binomial error distribution and logit link function to determine the influence of vegetation traits on male and female abundance. In order to assess the relative effects of the explanatory variables on males in comparison to females, the model was run with the proportion of females to males per night (n = 70) as the response variable, with “site” as a random factor (Lintott et al., 2014). We considered the following explanatory variables: habitat type; season (dry and rainy), Vba, forest basal area per site; Vab, total number of trees; Vrich, trees species richness; Vh, average tree height; Vspcomp, scores of NMDS axis 1 (see Supplementary Tables 1, 2). For each model, we calculated Akaike’s information criterion (AICc) corrected for small sample size following (Burnham and Anderson, 2004). This approach allowed us to select the most plausible models from a set of models. The set of models considered for every response variable, at each scale, included the null model (without explanatory power) and other models that considered each explanatory variable independently. We compared the model using Δi, which is the difference of AICc between a given model and the best (lowest AICc) model. We also calculate the AIC weights (wi) for each model. The wi represents the weight of the evidence that a certain model is the best model given the data and the set of candidate models. The 95% confidence set of the best models was defined by summing the wi, from the largest to the smallest, until the sum is = 0.95. Only models with an AICc lower than the null model were considered to define the 95% confidence set of plausible models.

All analyses were performed with R v. 1.0.136 (R Core Team, 2019).

We completed 70 nights of capture effort, 34 during the dry season, and 36 during the rainy season, resulting in a total capture effort of 180 net hours in RM, RP, and P, and 140 net hours in MF habitats. The total number of captures of the six most common bat species was 1,365 individuals (78% of all captures) including the following: Artibeus jamaicensis (n = 199, 11%), Artibeus lituratus (n = 396, 23%), Sturnira lilium (n = 521, 30%), Carollia perspicillata (n = 60, 3%), Glossophaga soricina (n = 109, 6%) and Platyrrhinus helleri (n = 81, 5%). These species were also present at all study sites. Overall, 43% of the sampled individuals from the species considered were males, while 57% were females, 1.131 proportion of females to males. The capture rates were significantly skewed toward females (F = 5.282, P < 0.001) (Figure 2). Riparian pasture, was the habitat with the highest female capture rates (433 individuals, 2.14 capture rate), followed by riparian continuous forest (152, 0.75) and active pasture (141, 0.69). Continuous forest was the habitat with the lowest female capture rates (50, 0.28) (Figure 2).

Riparian habitats in mature forests (RM) and pastures (RP) showed higher vegetation structural complexity than non-riparian habitats. Despite of the fact that RM contained more trees than RP, these habitats showed similar average canopy height (Table 1), whereas the lowest canopy height was recorded in pastures. On the other hand, basal area was greater in riparian habitats that in non-riparian habitats, in both old-growth forests and pastures (Table 1). The bioplot resulting of NMDS ordination is in Supplementary Figure 1.

Seasonality and habitat type were the best explanatory variables describing the presence of A. lituratus females; while for S. lilium, tree species composition and seasonality appeared as the most important variables explaining the incidence of females. In the case of G. soricina we found that the presence of females was positively related to vegetation structure including forest basal area and average tree height (Table 2). The results for all species are in Supplementary Table 3.

Contrary to our hypothesis, we found that females of A. lituratus did not exhibit relationship with habitat quality. The largest species of Stenodermatinae is A. lituratus, and as in other mammals, larger species could be more sensitive to human habitat disturbance (Lande, 1987). Nevertheless, the high number of females in this bat species, reflects selective foraging in a resource rich environment and higher roost availability in forested habitats. Even though Artibeus species may cross inhospitable matrix areas in fragmented landscapes, covering different vegetation types and flying distances ranging from 5 to 10 km (Galindo-González, 1998), females of A. lituratus might locally depend on temporal foliar roosts, and prefer larger trees within the dense and shaded mature forest that can provide energetic and thermal requirements to leave the young while foraging (Evelyn and Stiles, 2003; Bianconi et al., 2006; Arnone et al., 2016). For the same study area, A. lituratus has shown to select roosts with high humidity located in trees with the greatest basal areas (Ortiz-Ramírez et al., 2006).

Furthermore, some studies have argued that Glossophaginae are resilient to land use change (Willig et al., 2007). Our data indicate that the presence of G. soricina females is significantly and positively correlated to vegetation attributes such as tree height and forest basal area, supporting the idea that the species is an habitat specialist (Aguiar et al., 2014). The association between females of G. soricina with large trees could be due to the fact that large trees provide greater availability of roosts and foraging opportunities (Evelyn and Stiles, 2003; Ortiz-Ramírez et al., 2006). There is evidence showing food differentiation between specimens of G. soricina in the same areas where females preferred a plant food item different from males (Alvarez and Sánchez-Casas, 1999). During pregnancy and breeding seasons females might feed on the nearest available resource, whereas males fly larger distances in search of other feeding areas (Sosa et al., 1996). This foraging behavior might reduce the activity of G. soricina to habitats with high resource availability limiting its activity to small home ranges increasing its susceptibility to local extinction (Arita and Santos-del-Prado, 1999). This increases the importance of vegetation traits associated to the reproductive energetic demand (pregnancy, lactation and roost defense) (Charles-Dominique, 1991; Klingbeil and Willig, 2010).

Tree species composition was the strongest predictor variable for S. lilium, this frugivorous bat is known for its preference for understory shrubs and pioneer tree species, females of S. lilium are able to forage among forest elements as a result of non-random distribution of resources across the landscape (Loayza and Loiselle, 2008). Moreover, frugivorous bats like S. lilium which can visit different vegetation types can be considered as indicator taxa of habitat change in riparian vegetation, rather than highly specialized taxa in which population decline rapidly under environmental changes (de la Peña-Cuéllar et al., 2015).

The importance of seasonality for A. lituratus and S. lilium females might reside on differences in abundance and diversity of food resources between the wet and dry seasons. Seasonal fluctuations in rainfall influence phenology of fruiting plants and affects productivity in tropical forests (Ramos Pereira et al., 2010). In tropical regions, usually the rainy season corresponds to greatest fruit abundance than the dry season (Smythe, 1986). Even, many tropical bats timing their reproductive phenology to match periods of peak food availability, female bats may be constrained by the energetic requirements associates with reproduction, which might force them to alter their foraging behavior (Lintott et al., 2014). Resource availability due to seasonality can result in shortened flights during the exploration for food and shelter, whereas during the dry season when food resources are often scarce, females respond to local-scale vegetation structure increasing foraging movements into resource-rich pioneer fruiting plant species areas like secondary forests, while males tend to select areas close to old-growth forests (Rocha et al., 2017).

Our results show that responses of bat to human disturbance are sex-specific. Taking into account sex ratios of bat populations may help to a better understanding of the pervasive consequences of habitat loss and fragmentation. Sex-specific studies are important for bat conservation practices in order to promote habitat conditions favorable for both, females and males (Perry et al., 2007). Vegetation attributes like three height and basal area reflect the age and vertical complexity of forest, and more varied niche opportunities for bats, enhancing greater taxonomic and phylogenetic biodiversity (Martins et al., 2017). Our results suggest that the structural complexity of the vegetation and large trees influence the presence of females of G. soricina. In this sense, management efforts should promote riparian vegetation cover with large forest basal areas, important for the conservation of the entire bat community. Even though our analysis was restricted to six common bat species, we assumed that the maintenance of habitats that favor habitat generalist species should also benefit bat species that are habitat specialists (Istvanko et al., 2016; Rocha et al., 2018). Therefore, we encourage the inclusion of species of sensitive trophic guilds (gleaning insectivores and carnivores) and particularly roosting habits emphasizing adequate protection of females in conservation plans. Conservation actions toward female protection are particularly important due to their high level of parental investment associated with rearing pups (Istvanko et al., 2016). Management decisions that do not guarantee the protection of the habitat frequently used by female bats would likely have detrimental long term consequences on their reproduction, jeopardizing the dynamics and long-term persistence (van Toor et al., 2011; Frank et al., 2016).

In human dominated landscapes, the presence of isolated trees in pasture promote bat flights across pasture and increase bat detectability (Galindo-González et al., 2000), based on foraging behavior frugivorous bats travel across pastures and visit isolated trees while foraging, and use canopies for roosts or to decrease predation risks (Galindo-González, 1998), also isolated trees in pastures act as stepping stones for traveling across fragmented landscapes to different forest remnants (Guevara et al., 1989). However, pastures cannot sustain the same species richness of bats as old-growth forest and riparian vegetation (de la Peña-Cuéllar et al., 2015). Moreover, land use change in tropical landscapes seems to have considerable effects on bat population dynamics, for instance evidence suggests that forest-adapted insectivorous species are particularly sensitive to habitat conversion (Medellín et al., 2000; Williams-Guillén and Perfecto, 2010). Furthermore, frugivorous bats respond to matrix quality in different manners, whereas studies have found that frugivorous abundance was positively associated with the proportion of high quality habitats (Pinto and Keitt, 2008; Avila-Cabadilla et al., 2012; de la Peña-Cuéllar et al., 2015), some other studies have shown that frugivorous bat richness and abundance are higher in moderately fragmented landscapes than in old-growth forest (Willig et al., 2007; Klingbeil and Willig, 2009). Overall agricultural intensification may cause detrimental effects on bats and thus presumably on the ecosystem services they provide (Williams-Guillén et al., 2015). Additional research is needed to directly examine the effects of pregnancy and lactation on habitat selection by bats. We encourage radio-tracking studies that can show specific habitat use of males and females (roost and foraging areas) and if there is temporal segregation between sexes; this could provide information about how different habitats have an impact in the demography of bat populations.