Phenological modularity in amphibian calling behaviour: Geographic trends and local determinants

Introduction

The need for a better understanding of the interplay between environmental conditions and the architecture and functioning of biodiversity is a pressing challenge in ecology (McCann 2007; Naeem et al. 2012; Rohr et al. 2014). The temporal dimension has been early recognised as an axis of community structuring (Schoener 1974a,b; Jaksic 1982; Kronfeld-Schor & Dayan 2003). In this vein, a central component of the biodiversity architecture is the phenological structure (Visser et al. 2010). Phenology of species involves the temporal coupling of vital activities – that is reproduction, migration – to environmental conditions (Visser et al. 2010). Phenology is closely associated with climatic variables, which makes it vulnerable to global climate change processes (Scheffers et al. 2016; Kronfeld-Schor et al. 2017; Vázquez et al. 2017). Temperature has a main role in the functioning of all biological systems (Angilletta 2009). The thermal dependence of metabolic rate determines changes in behaviour of individuals and physiology that scale up to community structure and function (e.g. Gillooly et al. 2001; Allen et al. 2002; Dell et al. 2011). Consequently, seasonal trends in temperature are the main determinants of phenological patterns (Angilletta 2009; Steen et al. 2013; Tonkin et al. 2017). Therefore, unravelling the connections among thermal dependence and community structure is emerging as a key objective in different areas of ecology (Dell et al. 2011, 2014; Araújo et al. 2013; Pawar et al. 2016).

The thermal dependence of amphibian species phenology was recently associated with environmental gradients. The activities of individuals become more dependent on the temperature at lower latitudes, with higher local productivity, and are also related to local species richness (Canavero et al. 2018). These results and the evidence for an association between the phylogenetic composition of communities and thermal dependence indicate its connection with other features of community structure. The representation of phenology as a bipartite network of species registered at different times was proposed as a fruitful approach for the analysis of the time dimension of community architecture (e.g. Canavero et al. 2009; Aizen & Rovere 2010; Forrest & Miller-Rushing 2010; Biella et al. 2017). A modular organisation of ecological networks has been identified as a recurring feature of mutualistic, host-parasite, trophic and phenological networks, closely related with the stability of communities (Olesen et al. 2007; Tur et al. 2015; Biella et al. 2017). Modular structure implies the existence of link-dense regions representing a group or module of species that interact more strongly among themselves than with the rest of the network. The degree of this organisation type is named the modularity of the network (Olesen et al. 2007). The opposite pattern, antimodular, could be also expected when the interactions are more frequent between groups than within groups (Hintze & Adami 2010; Grilli et al. 2016). A modular organisation in amphibian phenology would indicate the existence of groups of species with a higher tendency to be active at the same time in comparison with the other species in the community (Canavero et al. 2009). Niche conservatism and phylogenetic signal in the breeding period could determine that related species are active under similar conditions (Canavero et al. 2018). Consequently, the phylogenetic composition of communities may be also be related to the strength and nature of its modular organisation (e.g. Rezende et al. 2009; Krasnov et al. 2012; Peralta 2016).

In this context, it is expected that thermal dependence, modularity and phylogenetic structure should be three main and interrelated components of the phenology of amphibian species (Peralta 2016; CaraDonna et al. 2017; Canavero et al. 2018). Specifically, the following hypotheses were considered: latitudinal gradients in climatic conditions and community structure are pervasive in nature (Brown 2014), and despite multiple attempts to relate single environmental variables with biological trends, large-scale geographic gradients typically involve changes on several abiotic and biotic variables, mutually correlated (Storch 2012).

Consequently, the emerging prediction from this hypothesis is that community structure may be related to this general latitudinal gradient of abiotic and biotic variables (Fig. 1 H₁) (Canavero et al. 2009). Statistically, this covariation can be represented as a ‘latent variable’, a variable that is not directly measured but it is associated to several observed variables – similar to the axis of a principal component analysis (Shipley 2016). Phenological modularity may be part of this general gradient and/or be directly affected by independent variables.

The next hypotheses refer to these direct connections. The second hypothesis points to the expected association between community species richness and modularity (Fig. 1 H₂). Since the seminal works of May (1972), it was stated that the stability of communities that increase in species richness might depend on the increase in modularity (see also Stouffer & Bascompte 2011).

The third hypothesis predicts an association between the phylogenetic structure of communities and phenological modularity (Fig. 1 H₃). The phylogenetic similarity among species may reflect trait similarity and consequently similar response to environmental variables and species interactions (Cadotte & Tucker 2017). This may determine the temporal aggregation of species activity reflected in a modular organisation of community phenology (Canavero et al. 2009).

The fourth hypothesis predicts an association between available energy and modularity (Fig. 1 H₄). Available energy may foster or preclude species coexistence (Hawkins et al. 2003; Steen et al. 2013; Leibold & Chase 2017) and also affects the calling behaviour amphibians (Ziegler et al. 2016). These effects may determine aggregation or overdispersion of species activity along months, and consequently, the strength of the modular organisation of amphibian phenology.

Finally, the fifth hypothesis predicts the effect of thermal dependence on modularity (Fig. 1 H₅). In the phenology of amphibian species, thermal dependence on species richness mainly describes the metabolic response to the seasonal changes in temperature (Allen et al. 2002; Dell et al. 2014; Canavero et al. 2018). This response involves a combination of nested diversity and species replacement between months (Canavero et al. 2009). Both nestedness and species turnover are directly connected with the strength of modularity (Fortuna et al. 2009; Ortiz & Arim 2016).

These hypotheses were translated to casual determinants of phenological modularity on amphibians’ calling activity (Fig. 1 H₆). Alternative causal connections were evaluated with structural equation models using 52 local communities from tropical to temperate locations (see Appendix S1: Fig. S1). A single structure that outperforms the entire set of alternative hypotheses was identified.

We focus on amphibian phenologies, which present a number of advantages to approaching this aim (Wiens et al. 2010; Schleuning et al. 2014). First, herpetologists have reported annual trends in activity for several locations, covering a wide biogeographic range in the Neotropics (see Canavero et al. 2009, 2018). Second, the evolutionary history of amphibians in this region is relatively well-known (Pyron & Wiens 2011) allowing for quantification and comparison of the phylogenetic community structure (e.g. Wiens et al. 2006). Third, amphibian activity and performance are closely related to environmental conditions. Specifically, amphibians are ectotherms with moist and naked skin, and aquatic reproduction, which is consequently coupled with weather conditions (Wells 2007); calling behaviour of anurans has a thermal dependence, an order of magnitude larger than that reported for other activities or taxonomic groups (Canavero et al. 2018); and the global decline in amphibian species was related to local and global environmental changes more than other taxa (Wake & Vredenburg 2008; Grant et al. 2016).

Methods

Phenological modularity and its relationship with environmental conditions and community structure were analysed across 52 local communities in the Neotropics (compiled in Canavero et al. 2009, 2018). The latitude, potential evapotranspiration, total species richness (the sum of all species that called at least once during the study period in a local community), an index of phylogenetic community structure and a normalised modularity index were calculated for each local community.

We considered two local environmental variables: latitude and potential evapotranspiration (PET). The first could better represent climatic conditions and biotic variables than any other single variable (Rohde 1992; Willig et al. 2003). We obtained the latitude directly from the papers of each database (Table 1). The PET is considered as a proxy of the environmental available energy in the context of species-energy theory (e.g. Hawkins et al. 2003); it represents the local relationship between heat and relative humidity (Currie 1991). We obtained the local values of PET using the database presented with the software Spatial Analysis in Macroecology (Hijmans et al. 2005; Rangel et al. 2006).

To measure the phylogenetic distances within communities, we reconstructed a phylogenetic tree of the 361 species of Neotropical anurans found in our community datasets. We began by checking which species were in common between our data and the molecular matrix of Pyron and Wiens (2011). When we did not find a species in the molecular matrix, we used ‘surrogate’ species of the same genus, or of the family. With these sequences, we reconstructed a tree of 164 species, where Homo sapiens was the out-group, and using the same concatenated sequences and models of Pyron and Wiens (2011). We used the program RAXML. To this new tree, we added the remaining taxa that did not have genetic information using a taxonomic criterion (n = 197), without modifying the known relationships between other species already in the tree. We include eight nodes shared with the tree of Wiens (2011, Fig. 2) to estimate the node ages, and the software r8s 1.8 (Sanderson 2012) (for more details, see Canavero et al. 2018, Appendix S1). Next, the phylogenetic structure of local communities (e.g. phylogenetic attraction or repulsion) was estimated with the net relatedness index (NRI):

where MPD_sample (mean pairwise distance) is the average phylogenetic distance between each taxon in a sample (i.e. local community) and all other taxa from the same sample. Random communities were constructed with a null model that shuffles species labels across the phylogeny, randomising phylogenetic relationships among species. This null model allowed measuring the geographic representation of the regional phylogeny at the local level. A positive NRI value indicates that the MPD is lower than expected, suggesting phylogenetic attraction, and a negative NRI means phylogenetic repulsion (Webb et al. 2002, 2008). This analysis was performed using the ses.mpd function from the ‘picante’ R-package (Webb et al. 2008; Kembel et al. 2010) in R version 3.03.

We constructed presence (1) – absence (0) matrices for each anuran community with months in columns and species in rows. These matrices represent bipartite networks wherein one level is represented by months – temporal dimension – and the other by anuran species that present calling activity (Canavero et al. 2009). At the temporal dimension, modularity identifies the existence of subsets of species whose probability of co-occurring across months is higher when compared with the probability of co-occurring with other species of the community (Olesen et al. 2007). In terms of the performance of the index, it was identified as a novel and robust method for estimating modularity in different networks, and its use has expanded in theoretical and applied ecology (e.g. Thébault & Fontaine 2010; Scheffer et al. 2012; Clune et al. 2013; Borthagaray et al. 2014; Takemoto et al. 2014; Grilli et al. 2016).

Phenological modularity was estimated for each of the communities using the spinglass.community function in the ‘igraph’ R-Package (Csardi & Nepusz 2006), including an algorithm based on simulated annealing which is a stochastic optimisation technique that identifies modules in each of the bipartite networks of species vs months by maximising a modularity function (Newman & Girvan 2004; Guimerà & Amaral 2005; Reichardt & Bornholdt 2006). The modularity (M) function applied was as follows:

where m is the number of connections, A_ij is the element of the A adjacency matrix in row i and column j, k_i is the number of connections of i, k_j is the number of connections of j, c_i is the module of i, c_j is the module of j and ∆(x, y) is 1 if x = y and 0 otherwise (see Borthagaray et al. 2014). The standardised deviations from a null expectation in modularity (ZMod) were estimated with a null model that randomised presences, but retained the total number of species per month and the total number of months in which species were observed. This null model was developed using the nullmodel function in the ‘bipartite’ R-Package (Dormann et al. 2008) and applying the ‘shuffle.web’ null model method for binary data. Null expectations in modularity and standard deviation were estimated from 10 000 simulations. ZMod values below −2 or above 2 indicate statistical significance (P < 0.05 error level, two-tailed Z-test) and allow comparing communities with different richness (Ulrich et al. 2009; Takemoto et al. 2014). To visualise an example of a modular community, we used the gplot function in the ‘sna’ R-Package (Butts 2016).

Results

Modularity is observed as a recurrent feature in the organisation of amphibians’ phenology (Fig. 2). It was detected in 22 communities but covering a range from antimodular in three communities to significantly modular (see Table 1). This type of range from negative to positive deviation indicates that even non-significant modularities are part of a biological gradient, which requires attention (Ulrich & Almeida-Neto 2012). Similarly, of the 34 time series of calling anurans with significant thermal dependence, 17 yielded a ZMod modularity index that deviated from null expectations (P < 0.05), 16 on the positive side (i.e. significant phenological modularity) and one on the negative side (Table 1). Four communities presented significant deviations (P < 0.05) from null expectations of NRI and six presented marginal deviations (P < 0.10). Of the ten communities that presented marginal and significant deviations, seven showed phylogenetic attraction and three showed repulsion (Table 1). High variability is detected covering the possible extremes of indices, which is well accounted for by the path analysis. In fact, along with the 16 path analysis models considered (see Appendix S1: Fig. S1a,b), one presented a significantly better performance and no discrepancy with the observed covariation matrix (w_iAIC_C = 0.571). When we only consider the models supported by data, the weighted w_iAIC_C was 0.82 for the best model (Fig. 3). This model considered a latent variable directly related with a positive connection with latitude (r = 0.81) and NRI (r = 0.52), and a negative connection with E (r = −0.82) and PET (r = −0.70) (Fig. 3). This result partially supports H1 (Fig. 1) in terms of the expected covariation of several components of the environment, biotic and abiotic. Further, the path model also indicated an effect over ZMod from S_loc (r = 0.36) (Fig. 1 H₂, a connection between complexity and community structure), PET (r = 0.56) (Fig. 1 H₄, the energy availability foster the time partitioning) and E (r = −0.60) (Fig. 1 H₅, communities with less thermal dependence will structure time dimension, species could use a wider time window along the year without restricting its activity to the warm period). This model accounted for 27% of the variation in phylogenetic relatedness (NRI), 42% of the ZMod variance and 68% of the variation in thermal dependence of the activity in amphibian communities (Fig. 3).

Discussion

Phenological modularity was identified here as a main feature of amphibians’ phenology. The anuran species may be interacting, adapting and evolving, modularly using the time as structuring axis on which the complexity of the system is expressed (Canavero et al. 2009). A temporally segmented organisation of amphibians’ activity is expected from classic and recent theories (Schoener 1974a,b; Jaksic 1982; Kronfeld-Schor et al. 2017). Specifically, the increase in modularity in interaction networks with species richness – that is complexity – is a long-standing prediction in ecology (May 1972; Stouffer & Bascompte 2011). Recent results further indicate that modularity may enhance species adaptability, improving long-term community stability (Scheffer et al. 2012; Clune et al. 2013). However, these theoretical expectations were not previously supported on empirical patterns of amphibians’ phenology. Here, we report the pervasive occurrence of modularity in the study system within a range from negative to positive deviation, which could indicate that even non-significant modularities are part of a biological gradient (Ulrich & Almeida-Neto 2012), and also its dependence on environmental conditions (Takemoto et al. 2014).

We advance on the connection between phenological modularity with other components of community organisation, such as available energy, thermal dependence of calling activity and species richness. The increase in modularity with available energy may be related with an increase in the number of active species with similar environmental requirements – modules – and/or an expansion of the time window of activity allowing different modules to be represented along the year (Currie 1991; Hawkins et al. 2003; Leibold & Chase 2017). We also found that thermal dependence of phenology – the rate of change in diversity with temperature along months – is related to the modular organisation of the amphibians’ phenology. Communities with large thermal dependence have lower phenological modularity. This finding may connect community phenology with climate change (Kronfeld-Schor et al. 2017; Vázquez et al. 2017). In this sense, it is noted that at higher latitudes, communities have lower thermal dependence but are more segmented in time use. Consequently, it may be expected that a rise in thermal variability and average temperature will produce a larger impact across lower latitudes. It may be highlighted that, on the basis of individuals’ thermal dependence, previous studies suggested a similar trend in the vulnerability of amphibian communities with latitude (Dillon et al. 2010). If modularity is positively related to stability (May 1972), the lower modularity associated with high thermal dependence may exacerbate the effect of climate change. However, the direct connection between modularity and stability may demand more theoretical and empirical analyses (Thébault & Fontaine 2010). The final direct connection with modularity involves a direct effect of species richness. The increase in species richness is associated with larger segregation of species along months, promoting a rise in phenological modularity (May 1972; Kronfeld-Schor & Dayan 2003; Canavero et al. 2009; Leibold & Chase 2017). This, in turn, may be related with the species segregation in time likely reducing interspecific competition (Schoener 1974a,b; Jaksic 1982; Scheffer & van Nes 2006).

Understanding the interplay among the organisms’ energetics, their thermal dependence and their consequences at higher levels of organisation is a pressing issue for ecology in a global warming scenario (Dillon et al. 2010; Dell et al. 2011, 2014; Vázquez et al. 2017; Canavero et al. 2018). The interpretation of activation energy in thermal biology is not restricted to a direct connection between temperature, metabolism and species coexistence (Allen et al. 2002; Dell et al. 2014; Segura et al. 2015; Canavero et al. 2018). Indeed, it represents a useful description of the association between temperature and community structure. This association was here related to phenological modularity, indicating that understanding the connection between environmental temperature organisms’ physiology and community assembly has to also consider the effect of temperature on the modular organisation of communities.

Although we did not find a direct connection between communities’ phylogenetic structure and phenological modularity, there is an indirect association probably mediated by the common response to a latitudinal trend in other variables. The existence of a non-random phylogenetic, as well as a modular organisation of local communities, is emerging as frequent properties of natural communities (e.g. Rezende et al. 2009; Krasnov et al. 2012; Peralta 2016). However, the connection between these two structural properties is just starting to be explored and deserves further attention. On the other hand, we found that communities with phylogenetically related species are observed at high latitudes where the environment is less productive and thermal dependence is lower. This result could be showing a connection between these species with the ability to be active with low activation energy and in environments that provide low energy (Wiens et al. 2006, 2010; Peralta 2016). Untangling which species are these and which roles they play on the phenological community structure is a relevant issue to develop conservation measures of community processes.

Phenology is a key component of community structure, shaping the strength and nature of biotic interactions, biodiversity and ecosystem functioning (Forrest & Miller-Rushing 2010; Biella et al. 2017). In a scenario of global change in climatic conditions and species diversity, understanding the interplay between environmental temperature, phenology, species richness and community structure has become a pressing issue (Dillon et al. 2010; Scheffers et al. 2016; Kronfeld-Schor et al. 2017; Vázquez et al. 2017). Here, we found a significant association among three main components of community structure: species richness, thermal dependence and phenological modularity. As is the case for most elements of community structure, available energy and thermal dependence directly foster the modular organisation of amphibians’ activity. Indeed, thermal dependence was found to be a parameter that can be used to track the effects of climate change on community structure (Dillon et al. 2010; Vázquez et al. 2017). If anything, our results reinforce the idea that the connection between modularity and environmental conditions is a ‘key area for active investigation’ (Takemoto et al. 2014).