Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

The role of environment, dispersal and competition in explaining reduced co-occurrence among related species

  • Ben G. Weinstein ,

    Roles Conceptualization, Data curation, Writing – original draft, Writing – review & editing

    Affiliation Dept. of Ecology and Evolution, Stony Brook University, Stony Brook, New York; United States of America

  • Catherine H. Graham,

    Roles Conceptualization, Funding acquisition, Writing – review & editing

    Affiliation Swiss Federal Research Institute (WSL), Birmensdorf, Zurich, Switzerland

  • Juan Luis Parra

    Roles Funding acquisition, Investigation, Writing – review & editing

    Affiliation Instituto de Biología, Universidad de Antioquia, Medellín, Colombia

The role of environment, dispersal and competition in explaining reduced co-occurrence among related species

  • Ben G. Weinstein, 
  • Catherine H. Graham, 
  • Juan Luis Parra


The composition of ecological assemblages depends on a variety of factors including environmental filtering, biotic interactions and dispersal limitation. By evaluating the phylogenetic pattern of assemblages, we gain insight into the relative contribution of these mechanisms to generating observed assemblages. We address some limitations in the field of community phylogenetics by using simulations, biologically relevant null models, and cost distance analysis to evaluate simultaneous mechanisms leading to observed patterns of co-occurrence. Building from past studies of phylogenetic community structure, we applied our approach to hummingbird assemblages in the Northern Andes. We compared the relationship between relatedness and co-occurrence among predicted assemblages, based on estimates of suitable habitat and dispersal limitation, and observed assemblages. Hummingbird co-occurrence peaked at intermediate relatedness and decreased when a closely-related species was present. This result was most similar to simulations that included simultaneous effects of phylogenetic conservatism and repulsion. In addition, we found older sister taxa were only weakly more separated by geographic barriers, suggesting that time since dispersal is unlikely to be the sole factor influencing co-occurrence of closely related species. Our analysis highlights the role of multiple mechanisms acting simultaneously, and provides a hypothesis for the potential importance of competition at regional scales.


The composition of ecological assemblages depends on diversification of clades in a region, colonization of species from outside that region and the niche overlap among species. When species diverge from a common ancestor, they tend to have a similar set of biotic and abiotic requirements, such that their niches are conserved [1,2]. Niche conservatism can result in co-occurrence of closely related species in local assemblages if conserved traits permit existence in a given environment (i.e environmental filtering)[3]. However, co-occurring related species may compete for limited resources due to shared functional morphology or foraging strategies [4,5]. Limiting similarity predicts that if niche overlap between two species surpasses some threshold, interspecific competition will exceed intraspecific competition, leading to local extinction of one competitor [6,7]. In addition, species may not co-occur because they have not dispersed to all suitable habitats, either because of insufficient time for dispersal or insurmountable geographic barriers [8,9].

A common approach for integrating evolutionary history and community ecology is to compare the phylogenetic structure of observed assemblages to a biogeographic species pool using randomization tests [10,11]. This approach has been increasingly criticized with several recent calls for more critical use of these methods [1214]. Given the complexity of biogeographic patterns, and the potential influence of many mechanisms leading to similar patterns, it is often difficult to disentangle all possible causes in a single model.

Here we explore patterns of hummingbird co-occurrence and relatedness in the Northern Andes combining multiple analyses including: simulations, hierarchical models of co-occurrence and relatedness, predicted assemblages based on broad-scale environment, and analysis of habitat connectivity.

We simulated assemblages to simultaneously evaluate how different mechanisms influence the phylogenetic structure of assemblages. Simulated assemblages were created based on two mechanisms: phylogenetic conservatism of traits important to environmental filtering, and phylogenetic repulsion based on traits mediating competition [15,16]. Phylogenetic conservatism implies that the traits associated with species occurrence evolved with greater phylogenetic signal than by Brownian Motion, such that closely related species tend to co-occur given strong environmental filtering for the traits in question [1]. Phylogenetic repulsion implies that traits associated with species occurrence evolved with divergent selection, such that closely related species are less likely to co-occur if the traits in question lead to increased niche overlap resulting in increased competition. These simulations provide a basis of comparison as to how the simultaneous effects of phylogenetic conservatism and repulsion might influence patterns of phylogenetic co-occurrence in our empirical data.

We used a non-linear hierarchical model to simultaneously evaluate the effect of filtering and competition on patterns of co-occurrence, while allowing for species-level variation. This approach should improve upon previous studies that use randomization tests to identify assemblages as either underdispersed, overdispersed, or indistinguishable from a null expectation. Further, by using a hierarchical model we can evaluate the overall effect of relatedness on patterns of co-occurrence, while allowing for species-level variation. Rather than fitting a model to either an entire assemblage or species individually, a hierarchical model allows data-poor species to borrow strength from more commonly sampled species by allowing variance to be shared among all hummingbirds. This approach allows us to evaluate the shape of the relatedness and co-occurrence curve for each species and compare it to the curve based on predicted environmental suitability. In contrast, randomizations treat assemblages as the fundamental unit which implicitly assumes that all co-occurring species are influenced by the same mechanisms [17,18].

Evaluation of the phylogenetic structure of an assemblage requires a potential pool of species that could occur in a given location. While past studies have focused largely on geography in constructing species pools, we include information on environmental tolerances and dispersal limitation [19,20]. We used ensemble bioclimatic models based on current species distribution to evaluate environmental tolerances and included a dispersal filter based on elevational barriers between assemblages. We then compared these predicted assemblages to the observed assemblages based on the frequency of co-occurrence among related species. Further, we evaluated the possibility that early branching sister taxa are more likely to co-occur than late branching sister taxa due to increased time for dispersal across geographic barriers [21].

To explore our integrated approach, we used hummingbird assemblages from the Northern Andes. This system is relatively well-developed, with community phylogenetic studies finding support for both competition and environmental filtering as potential mechanisms influencing assemblage structure. Initial work using geographic species pools found evidence for phylogenetic clustering, which was attributed to environmental filtering in high elevation assemblages [22,23]. Acknowledging the importance of considering multiple mechanisms, subsequent analyses explored subsets of data including: sub-clades [24,25], environmentally restricted assemblages [26,27] and geographically constrained null models [28]. This most recent analysis revealed that once species pools were constrained by mean annual temperature, many assemblages were more evenly phylogenetically dispersed than previously appreciated. Our approach builds from this body of work by 1) explicitly accounting for environment using ensemble bioclimatic models, rather than single variables (as in [28]); 2) focusing on individual species, rather than assemblages, to evaluate phylogenetic structure; 3) moving beyond a dichotomous null-modelling framework that evaluates biotic interactions and environmental as opposing forces; and 4) comparing predictions of relatedness and co-occurrence to simulations built with known patterns of trait evolution. We hope that these improvements will provide increased rigor to the field of community phylogenetics as the community continues to evaluate what can be learned from pattern-based studies.


Simulation study to evaluate multiple mechanisms

To determine the patterns of relatedness and co-occurrence as a function of known mechanisms, we created a range of scenarios for species co-occurrence and phylogenetic relatedness based on the degree of environmental filtering and limiting similarity (Fig 1). These scenarios were generated by evolving hypothetical species using varying degrees of phylogenetic conservatism and repulsion in environmental tolerances (following [29], using the 'scape' function in R packge PEZ, [30]) (Table 1). As phylogenetic conservatism increases, related species tend to co-occur more often. As phylogenetic repulsion increases, related species tend to co-occur less often. We calculated the relationship between phylogenetic distance to the closest related species and the probability of co-occurrence. Each simulation was repeated fifty times with a 10X10 square landscape with two environmental gradients and a balanced phylogeny with 16 taxa.

Fig 1. Conceptual diagram of four simulated assemblage types given known patterns of trait evolution.

Each combination of phylogenetic conservatism and repulsion in the occurrence trait (a) would lead to a different pattern of co-occurrence among related species (b) leading to a different function representing the relationship between probability of presence and phylogenetic distance to the closest related species (c).

Table 1. Simulations used to evaluate the importance of environmental filtering and competition in structuring assemblages.

G parameter is the strength of phylogenetic conservatism ( and the strength of phylogenetic repulsion (g.repulse) as implemented in the R package PEZ [29]. When = 0, the environmental tolerances evolve according to Brownian motion.

Hummingbird and climate data

We used 230 hummingbird assemblage lists (presence/absence data) from published references, focusing on the Columbian and Ecuadorian Andes and adjacent lowlands [23,27]. These list included 133 species, with a mean occupancy of 18 sites. We excluded lists from eco-lodges with high density of hummingbird feeders and merged assemblages within 0.1° cells (~10km) because assemblages in close proximity may be non-independent. For the bioclimatic models, hummingbird point localities were collected from published literature, environmental organizations, and museum specimens [23,27]. To avoid circularity in predicted species presence, records from the assemblage data were not included in the localities used to create the bioclimatic models, and species with fewer than 20 occurrence records were removed from the analysis [31]. In total we had 22,046 records for 132 species, with an average of 166 localities per species. The phylogenetic relatedness to the closest related species was computed using the cophenetic distance in a regional phylogeny that included all 130 hummingbird species in the observed assemblages [32].

For climate covariates, we used the Worldclim database [33] at the 0.5 minute resolution (0.0083 degrees) and aggregated to 0.1 degrees to match the resolution of our assemblage data using the function ‘aggregate’ in the R package ‘raster’ [34]. We choose this relatively coarse resolution because we were interested in the broad climate drivers of distributions. While we acknowledge climate layers at this resolution for the Andes Mountains do not reflect climate variation along its steep environmental gradients, they should provide the general climate envelope we required for this study. Predicted habitat suitability does not automatically imply presence, rather we use the bioclimatic models to gauge the environmental suitability of a geographic location for each species assuming no dispersal limitation. We used principal component analysis and Pearson’s correlations on all 19 Worldclim variables to identify the following relatively non-correlated variables defining the environmental gradients in our region: mean annual temperature, mean annual precipitation, temperature seasonality (standard deviation of temperature across 12 months) and precipitation seasonality. These variables were highly correlated with the remaining 15 bioclimatic variables, but had correlation coefficients below 0.4 among each other. In addition, these variables have a biological basis and have been shown to predict hummingbird distributions in previous work [28,35,36]. Mean annual temperature is correlated with elevation and represents both energetic and flight (i.e., air density) constraints on hummingbirds [37]. Precipitation relates to the nectar availability and floral diversity of hummingbird-visited plants [38]. Temperature and precipitation seasonality broadly define important gradients in the Ecuadorian Andes ranging from dry Tumbesian regions in the southwest to the wet tropical forest in Amazonia.

Ensemble bioclimatic models

To predict habitat suitability, we built ensemble models using maximum entropy (MAXENT), boosted regression trees (GBM), and generalized linear models (GLM) (R package biomod2 [39]). We used ensemble bioclimatic models because they combine results from multiple algorithms and tend to perform better than individual techniques, since they circumvent any idiosyncratic weakness of each modeling method [40]. The GLM allowed for quadratic terms with no interactions among environmental variables. We chose MAXENT because it performed well in comparisons with other presence-only data approaches [41], but see [42], and in previous work in the Andes [4345]. GBM is a machine-learning method that estimates the relationship between a response variable and its predictors without a priori specification of a model by using cross-validation of smaller candidate trees [46]. For our GBM model, we used 2500 maximum trees with three-fold cross-validation. Output of all models scale from 0 (not suitable) to 1 (completely suitable) making it possible to generate an ensemble model (see below). For all modeling methods we used 2000 background points for each model using target sampling where points are drawn from the pool of geographic locations where we had sampled locality data [47].

For model evaluation we split the observed data into 80% training data and 20% testing data and computed area under the curve (AUC) and True Skill Statistic (TSS) as a measures of goodness of fit [48,49]. This split was performed once for each model for each species. To create an ensemble model per species, we combined models that had an AUC score greater than 0.75. We combined models for each species by calculating the mean habitat suitability predicted by each model.

Factors influencing dispersal

We evaluated dispersal limitation using both an elevation-weighted cost distance metric to infer the current potential for dispersal between localities and time-since-divergence to evaluate the influence of time for dispersal between current localities given historical isolation (i.e., allopatric speciation). Cost distance is an effective predictor of dispersal limitation and population genetic structure, and is a better predictor of hummingbird beta-diversity between localities than Euclidean distance [27]. Computing cost distances among locations requires constructing an environmentally-weighted cost surface based on change in elevation, such that points in close proximity, but separated by large mountains or deep valleys, are more costly to cross than points separated by flat terrain. This model of dispersal incorporates both physical geographic barriers, as well as environmental change, since large changes in elevation also result in drastic changes in temperature and precipitation regimes. For each assemblage we generated a unique cost distance surface based on the elevation of that assemblage and calculated the least cost path from the focal assemblage to all other assemblages. This metric should not be seen as a predicted dispersal path, but rather a quantitative measure of geographic connectedness. Cost paths were calculated using the R package gDistance [50] using the mean transition probability of moving from one cell to any of its eight neighboring cell.

Next, for each pair of sister taxa (n = 58), we evaluated the relationship between time since divergence and the mean cost distance between their localities. As described above we used the change in elevation cost surface to calculate the least cost path between each presence locality of one sister taxa to the closest presence locality of the other sister taxa. Assemblages where both taxa co-occur had zero cost distance. We then modeled cost distance between sister taxa and modeled it as a function of time since divergence using linear regression. Given that there are relatively few early branching splits in the hummingbird phylogeny (only 3 splits greater than 10myr) and older splits could present undue leverage, we also evaluated the relationship solely among sister taxa younger than 10 myr.

Biologically motivated species pools

To provide a null hypothesis for the observed pattern of phylogenetic relatedness in the hummingbird assemblages we used biologically motivated species pools. We created two types of null assemblages: environment assemblages and environment + dispersal assemblages. Environment assemblages were based on predicted habitat suitability from bioclimatic models for each species at each geographic location for which we had observed assemblage data. Defining predicted suitable areas required turning the index of suitability returned from the bioclimatic into a statement of predicted presence or absence. Due to differences in species prevalence, taking a fixed probability cutoff across all species would bias presence towards more common species [51]. We therefore applied a threshold to the suitability values for each species based on the distribution of suitability values from that species’ observed localities [45,52]. We used the probability of suitability that allowed 95% (quantile 0.05) of the known presences to be predicted present. The 0.05 quantile threshold maximized the number of true predicted presences in the observed data, while minimizing the number of false negatives.

In the environment + dispersal assemblages, predicted suitable locations were considered unavailable for species’ presence if the cost distance to the closest observed location was higher than a species-specific threshold. This threshold was determined by calculating the cost distance for every observed locality of that species to the closest observed locality. To avoid letting a single assemblage drive this value, we used the 95th quantile of the distribution as the maximum cost distance between suitable sites.

Statistical analysis

For each of the four simulations, the two predicted assemblages, and the observed data, we compared the relationship between phylogenetic distance to the closest related species and the probability of co-occurrence. The full model reads: the presence of hummingbird species (i) at site (Y) is a Bernoulli trial with a probability of occurrence ρ. The probability of occurrence is a logit transformed function of a species intercept (α), the linear effect of relatedness on occurrence (βi,1) and the non-linear effect of relatedness on occurrence (βi,2). We included a polynomial term because we believe multiple mechanisms may shape co-occurrence, leading to a non-linear relationship. We chose a hierarchical model that first fits a distribution to all species responding to the presence of closely related species (‘group level distribution’ (), but then allows each species () its individual response to relatedness (βi).

All model parameters had non-informative priors (Normal(mean = 0.001, precision = 0.001) with variances Gamma(shape = 0.001, rate = 0.001)). All posterior probabilities for model parameters were estimated using Markov chain Monte Carlo (MCMC) in JAGS. Models were run for 50,000 draws with two chains, thinning by five draws.


Our simulations showed that as phylogenetic conservatism in environmental tolerances increased, species were more likely to co-occur with closely-related species (Fig 2). When environmental tolerances evolved without either phylogenetic conservatism or repulsion, relatedness had little impact on co-occurrence. When environmental tolerances evolved without phylogenetic conservatism, but with phylogenetic repulsion, there was a strong positive relationship between distance to the closest related species and presence. Finally, when the occurrence trait evolved with both phylogenetic conservatism and repulsion, there was increasing co-occurrence when a moderately related species was present, followed by a decrease in co-occurrence when a closely related species was present.

Fig 2.

Probability of occurrence based on a thousand draws from the posterior distributions for each of the simulation assemblages (A) and for the observed and predicted assemblages (B). The simulated assemblages modeled hypothetical species occurrence based on a known model of trait evolution incorporating phylogenetic conservatism and repulsion among closely related species. The predicted assemblages are rearrangements of the true observed assemblages based on predicted habitat suitability (environment assemblages) and the addition of a dispersal filter (environment + dispersal assemblages). The hierarchical posteriors were plotted on the scale of the input data to show the estimated relationship between occurrence and the phylogenetic distance to the closest related species. The shaded region is the central 95th quantile of the posterior distribution. The observed pattern of co-occurrence at first increases with increasing relatedness to a co-occurring species and then decreases in co-occurrence when a very closely related species is present. The overall observed co-occurrence is less than expected given species abiotic tolerances and potential dispersal.

To create biologically meaningful null models of assemblage structure, we used bioclimatic models and cost path analysis to create predicted assemblage lists based on predicted habitat suitability and dispersal limitation. For each of the 132 species in the observed assemblage lists, we created an ensemble model from each of the GLM, MAXENT, and GBM models that had an AUC score greater than 0.75. This yielded a final count of 100 species occurring in 230 geographic assemblages. The AUC of these models was high across all individual model types (mean: GBM = 0.97, GLM = 0.87, MAXENT = 0.90). Although TSS scores were more variable across modeling types (mean: GBM = 0.87, GLM = 0.65, MAXENT = 0.65), the correlation between AUC and TSS scores was high across all models and species (GBM = .0.97, GLM = 0.96, MAXENT = 0.91), suggesting that our evaluation of model performance was consistent across metrics. Twenty-five of the 32 species excluded from the observed assemblage lists occurred in less than 10 localities, and five species occurred in just a single locality. On average observed assemblage lists had 91% completeness, and in 121 of the 230 assemblage lists all species were corrected predicted as present. We therefore limited our analysis of co-occurrence to species shared by all observed, environment and environment + dispersal assemblages.

In both the environment and environment + dispersal assemblages, the probability of a species’ occurrence initially increased with decreasing phylogenetic distance to a species in an assemblage, followed by a decrease among closely-related species. The environment + dispersal assemblages showed a lower co-occurrence than environment suggesting that dispersal between localities was a factor in shaping predicted species lists. We compared the environment and environment + dispersal assemblages to our observed assemblage data to determine whether we would have arrived at the observed pattern based solely on environmental tolerances. Similar to the predicted assemblages, the observed assemblages showed an initial increase in the probability of co-occurrence with increasing relatedness, followed by decreased co-occurrence among closely related species (Fig 2). Explorations of species-specific responses showed that data-poor species had wider credible intervals, and our observed pattern is present in many, but not all, species throughout the hummingbird phylogeny (Fig 3). Given the lack of overlap in the confidence intervals of the estimate among the observed and environment assemblages, we infer that co-occurrence is less than expected among closely related species. The confidence interval of the observed and environment + dispersal assemblages had minor overlap, suggesting that including a dispersal filter reduced co-occurrence of related species.

Fig 3. Example species level curves for probability of occurrence and phylogenetic distance to the closest related species.

The graph shows the presence (y = 1) or absence (y = 0) of a given species at each of the geographic assemblages.Since the nearest closest related species are often the same species in multiple assemblages, there are multiple points overlapping. The number of assemblage lists that define each point is shown by the size of the point. Species were selected to show a variety of co-occurrence curves and represent a diversity of phylogenetic lineages within the hummingbird clade.

To estimate the importance of time on the dispersal on co-occurrence of closely species, we looked at the relationship between time since divergence and the presence of geographic barriers (Fig 4). We found that older sister taxa were separated by smaller changes of elevation, but the relationship was only significant when including the two deepest sister taxa pairs (R2 = 0.32, df = 27, p = 0.01). For species pairs less than 10 myr, there was only a modest, non-significant, relationship between the time since divergence and average cost distance between localities of sister taxa (R2 = 0.01, df = 24, p = 0.6).

Fig 4. Average cost distance between localities for each pair of sister taxa.

The cost distances incorporates the change in elevation and Euclidean distance as a measure of geographic barriers between localities. We calculated the average cost distance between each presence locality of one sister taxa to the closest presence locality of the other sister taxa. Assemblages where both taxa co-occur had zero cost distance. The solid regression line is for all species pairs, and the dashed regression line is for only species pairs that diverged less than 10 myr ago. Two example sister pairs are shown. The Phaethornis sister taxa have high cost distance due to presence of the Andes, whereas the Lesbia sister taxa have relatively low cost distance due to their short distance along mountain slopes. In the insets, one sister taxa is shown as a solid triangle, and the other an open circle. The least cost path along the change in elevation frictional surface is shown in black.


The relative roles of environmental filtering, dispersal, and competition in promoting the co-occurrence of related species remains a central focus of biogeography. We found the relationship between co-occurrence and relatedness was non-linear; the probability of co-occurrence was greatest for moderately related species and decreased among both distantly and closely related species. The increase in co-occurrence among moderately related species as compared to distantly related species is likely due to environmental filtering based on shared environmental tolerances.

The pattern of reduced co-occurrence among closely-related species could be caused by interspecific competition, or allopatric speciation with insufficient time for re-expansion to secondary sympatry [21,36,53]. Because reduced co-occurrence was maintained when predicted suitable environments and dispersal filters were considered, our results suggest that interspecific competition may be a factor shaping distributions at a broad spatial scale [5355]. While these results are not conclusive proof that competition influences range dynamics and community structure, they present an interesting hypothesis worth investigating at local scales [56]. For example, future studies could conduct local scale observations or experiments to evaluate if resource competition increases among closely related species [14].

Environmental filtering, which results in co-occurrence of related species with similar traits, is well documented across multiple taxonomic groups [57], including hummingbirds [23]. In hummingbirds, only some lineages, notably the brilliants and coquettes, are adapted to high-elevation environments [22,58,59]. Species in these lineages have adapted physiological and biomechanical traits needed to thermoregulate in low temperatures, and sustain flight in low partial pressures [60]. For example, several high-elevation lineages have amino acid substitutions associated with hemoglobin adaptation to low oxygen environments [61]. In contrast, Phaethornis hermits are widespread throughout the lowlands, but their elevation ranges rarely extend into the highlands [62].

Hummingbirds compete for nectar resources and related species often use similar resources [63,64]. Biogeographic analyses of hummingbird-plant networks show that related species tend to have similar levels of niche breadth and specialization [65]. Historical constraints, either through phylogenetic tradeoffs or past climate influences may lead to constraints on foraging breadth and increased overlap among related species[66,67]. At a local scale, competition among hummingbird species appears to be mediated by morphological traits, such as bill length, that determine the degree of resource overlap [68,69]. Given that hummingbird traits exhibit weak trait lability [70], then closely related species may compete more for limited floral resources [22]. This mechanism would result in the lack of co-occurrence of closely related species at the biogeographic scale.

Biogeographic patterns suggestive of competition have been reported in several other groups [45,54]. Sylvia warblers with greater dispersal ability occupy a greater proportion of potential suitable habitat, especially where there are few congeneric species [55]. Similarly, Anolis lizards are less likely to occur in predicted suitable habitat when a morphologically similar species is present [53]. Recent physiological measurements of basal metabolic rate showed little variation across tropical birds along a wide altitudinal gradient [71], suggesting that biotic interactions may have a role in setting altitudinal range limits [72]. In addition, comparisons of the elevation ranges between the main Andes and less diverse isolated mountains showed that some birds and mammals have broader elevational ranges on isolated mountains [45,73]. These patterns were interpreted as potential evidence for competition limiting elevation ranges, but bioclimatic models suggest that in birds, they were at least partly due to differences in environment among similar elevation ranges on different mountain slopes [74]. In our case, we explicitly evaluated whether the lack of co-occurrence is due to environment by comparing our observed assemblages to those predicted based on bioclimatic models.

To investigate whether dispersal explains the lack of co-occurrence among very close relatives we tested the relationship between sister taxa and cost distance as a proxy for dispersal barriers. We found some evidence that older sister taxa are, on average, separated by smaller geographic barriers and tend to co-occur more frequently than younger sister taxa. This result could be interpreted as incomplete range filling [8], since allopatric speciation is common in hummingbirds [75,76]. Macroecological analysis of primates and songbirds indicate that the rate of range overlap of sister taxa increases as a function of time since divergence [21,77]. However, our data only show this pattern when including the oldest lowland sister divergences (2 sister taxa comparisons are greater than 10myr). Excluding these two taxa, we do not see any relationship for the rest of the hummingbird sister taxa (23 sister taxa comparisons are less than 10myr).

There are several reasons why we hesitate to ascribe the pattern of reduced co-occurrence of closely related species wholly to insufficient time for dispersal. If dispersal limitation over evolutionary time was a driving force in community assembly, we would expect clades to be highly localized within areas surrounded by geographic barriers of unsuitable habitat [78]. Comparisons across barriers should then show high phylogenetic beta diversity, but low phylogenetic alpha diversity [79]. However, when comparing lowland assemblages across the Andes we observed the opposite pattern with relatively low phylogenetic beta diversity and high phylogenetic alpha diversity [27]. In addition, we would expect younger clades to have a smaller extent of occurrence compared to older clades, since there has been less time to occupy potentially suitable habitat. Yet, the youngest hummingbird clade, the Bees, are less than 2.5 million years old and range from Alaska to the Tropics. Hummingbirds are strong fliers and many species make latitudinal or elevational migrations with well-established patterns of vagrancy. For instance, the American Ornithology checklist has records for 23 hummingbird species, even though only 10 species commonly breed north of the US-Mexico border [80]. Together, these lines of inference suggest that dispersal is unlikely to be the sole mechanism in occupying potentially suitable habitat.

The role of niche conservatism in promoting co-occurrence of related species through environmental filtering is well documented [57]. The more challenging question is whether the reduced co-occurrence of close relatives is related to dispersal limitation or competition [12]. Ultimately, these two factors are conflated in macroecological analysis, since the species for which we expect the greatest ecological overlap are the same species for which there has been less time since divergence. We hope that our approach, which considers the simultaneous effects of multiple mechanisms on biogeographic patterns of co-occurrence, will serve both as a model for similar biogeographic studies and a stimulus for more detailed local investigation.


This manuscript was improved by feedback and T. Ives and D. Futuyma. W. Pearse and M. Helmus made PEZ code available, and H. Lynch advised in developing the Bayesian approach. We thank M. Lentino and the Phelps collection for provided localities for Venezuela and M. Lim, A. Shankar and B. Tinoco assisted in checking hummingbird localities. Research was supported by NSF Dimensions of Biodiversity (DEB 1136705) and NASA (NNX11AO28G) grants to CHG and the NSF Graduate Research Fellowship to BGW. All code can be found on the github repo, and the output can be viewed at The authors declare no conflict of interests.


  1. 1. Losos JB. Phylogenetic niche conservatism, phylogenetic signal and the relationship between phylogenetic relatedness and ecological similarity among species. Ecol Lett. 2008;11: 995–1003. pmid:18673385
  2. 2. Wiens JJ, Graham CH. NICHE CONSERVATISM: Integrating Evolution, Ecology, and Conservation Biology. Annu Rev Ecol Evol Syst. 2005;36: 519–539.
  3. 3. Keddy P. Assembly and response rules: two goals for predictive community ecology. J Veg Sci. 1992;3: 157–164. Available:
  4. 4. McGill BJ, Enquist BJ, Weiher E, Westoby M. Rebuilding community ecology from functional traits. Trends Ecol Evol. 2006;21: 178–85. pmid:16701083
  5. 5. Macarthur RRH. Population ecology of some warblers of northeastern coniferous forests. Ecology. 1958;39: 599–619.
  6. 6. Levin S. Community equilibria and stability, and an extension of the competitive exclusion principle. Am Nat. 1970;104: 413–423. Available:
  7. 7. Gause GF. Experimental Analysis of Vito Volterra’S Mathematical Theory of the Struggle for Existence. Science. 1934;79: 16–17. pmid:17821472
  8. 8. Svenning J-C, Skov F. Limited filling of the potential range in European tree species. Ecol Lett. 2004;7: 565–573.
  9. 9. Losos JB, Glor RE. Phylogenetic comparative methods and the geography of speciation. 2003;18: 220–227.
  10. 10. Webb CO, Ackerly DD, McPeek M a., Donoghue MJ. Phylogenies and Community Ecology. Annu Rev Ecol Syst. 2002;33: 475–505.
  11. 11. Emerson BC, Gillespie RG. Phylogenetic analysis of community assembly and structure over space and time. Trends Ecol Evol. 2008;23: 619–30. pmid:18823678
  12. 12. Warren DL, Cardillo M, Rosauer DF, Bolnick DI. Mistaking geography for biology: inferring processes from species distributions. Trends Ecol Evol. Elsevier Ltd; 2014;29: 572–580. pmid:25172405
  13. 13. Godoy O, Kraft NJB, Levine JM. Phylogenetic relatedness and the determinants of competitive outcomes. Ecol Lett. 2014;17: 836–844. pmid:24766326
  14. 14. Kraft NJB, Adler PB, Godoy O, James EC, Fuller S, Levine JM. Community assembly, coexistence and the environmental filtering metaphor. Funct Ecol. 2015;29: 592–599.
  15. 15. Helmus MR, Savage K, Diebel MW, Maxted JT, Ives AR. Separating the determinants of phylogenetic community structure. Ecol Lett. 2007;10: 917–25. pmid:17845292
  16. 16. Ives A, Helmus M. Generalized linear mixed models for phylogenetic analyses of community structure. Ecol Monogr. 2011;81: 511–525. Available:
  17. 17. Gómez JP, Bravo G a, Brumfield RT, Tello JG, Cadena CD. A phylogenetic approach to disentangling the role of competition and habitat filtering in community assembly of Neotropical forest birds. J Anim Ecol. 2010;79: 1181–92. pmid:20642767
  18. 18. Barnagaud J-Y, Daniel Kissling W, Sandel B, Eiserhardt WL, Sekercioğlu CH, Enquist BJ, et al. Ecological traits influence the phylogenetic structure of bird species co-occurrences worldwide. Ecol Lett. 2014;17: 811–20. pmid:24754339
  19. 19. Borregaard MK, Amorim IR, Borges PA V, Cabral JS, Fernández-Palacios JM, Field R, et al. Oceanic island biogeography through the lens of the general dynamic model: Assessment and prospect. Biol Rev. 2016; pmid:26923215
  20. 20. Carstensen DW, Lessard JP, Holt BG, Krabbe Borregaard M, Rahbek C. Introducing the biogeographic species pool. Ecography (Cop). 2013;36: 1310–1318.
  21. 21. Pigot AL, Tobias J a. Species interactions constrain geographic range expansion over evolutionary time. Ecol Lett. 2013;16: 330–338. pmid:23231353
  22. 22. Stiles FG. Ecomorphology and phylogeny of Hummingbirds: Divergence and convergence to adaptations to high elevations. Neotrop Ornithol. 2008;19: 511–519.
  23. 23. Graham CH, Parra JL, Rahbek C, McGuire JA. Phylogenetic structure in tropical hummingbird communities. Proc Natl Acad Sci U S A. 2009; 19673–8. pmid:19805042
  24. 24. Parra JL, McGuire J a, Graham CH. Incorporating clade identity in analyses of phylogenetic community structure: an example with hummingbirds. Am Nat. 2010;176: 573–587. pmid:20849270
  25. 25. Parra JL, Rahbek C, Mcguire JA, Graham CH. Contrasting patterns of phylogenetic assemblage structure along the elevational gradient for major hummingbird clades. 2011; 2350–2361. 10.1111/j.1365-2699.2011.02558.x
  26. 26. Laube I, Graham CH, Böhning-Gaese K. Intra-generic species richness and dispersal ability interact to determine geographic ranges of birds. Glob Ecol Biogeogr. 2012;22: n/a-n/a.
  27. 27. Weinstein BG, Tinoco B, Parra JL, Brown LM, McGuire J a., Stiles FG, et al. Taxonomic, Phylogenetic, and Trait Beta Diversity in South American Hummingbirds. Am Nat. 2014;184: 211–224. pmid:25058281
  28. 28. Lessard J-P, Weinstein BG, Borregaard MK, Marske KA, Martin DR, McGuire JA, et al. Process-Based Species Pools Reveal the Hidden Signature of Biotic Interactions Amid the Influence of Temperature Filtering. Am Nat. 2016;187: 000–000. pmid:27277404
  29. 29. Pearse WD, Cadotte M, Cavender-Bares J, Ives a. R, Tucker C, Walker S, et al. pez: Phylogenetics for the Environmental Sciences. Bioinformatics. 2015;31: 2888–2890. pmid:25948716
  30. 30. Ives A, Helmus M. Generalized linear mixed models for phylogenetic analyses of community structure. Ecol Monogr. 2011;81: 511–525. Available:
  31. 31. Mateo RG, Felicísimo ÁM, Muñoz J. Effects of the number of presences on reliability and stability of MARS species distribution models: The importance of regional niche variation and ecological heterogeneity. J Veg Sci. 2010;21: 908–922.
  32. 32. McGuire JA, Witt CC, Remsen J V., Corl A, Rabosky DL, Altshuler DL, et al. Molecular phylogenetics and the diversification of hummingbirds. Curr Biol. 2014;24: 910–916. pmid:24704078
  33. 33. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A. Very high resolution interpolated climate surfaces for global land areas. Int J Climatol. 2005;25: 1965–1978.
  34. 34. Hijmans RJ. Introduction to the ‘ raster ‘ package (version 1. 9–81). 2012; 1–27.
  35. 35. Velásquez-Tibatá J, Salaman P, Graham CH. Effects of climate change on species distribution, community structure, and conservation of birds in protected areas in Colombia. Reg Environ Chang. 2012;13: 235–248.
  36. 36. Graham CH, Silva N, Velásquez-Tibatá J. Evaluating the potential causes of range limits of birds of the Colombian Andes. J Biogeogr. 2010;37: 1863–1875.
  37. 37. Altshuler D, Dudley R. The ecological and evolutionary interface of hummingbird flight physiology. J Exp Biol. 2002;205: 2325–2336. Available: pmid:12124359
  38. 38. Feldman RE, McGill BJ. How important is nectar in shaping spatial variation in the abundance of temperate breeding hummingbirds? J Biogeogr. 2013;41: 489–500.
  39. 39. Georges D, Thuiller W. Multi-species distribution modeling with biomod2. R-CRAN Proj. 2012;1: 1–12.
  40. 40. Araújo MB, New M. Ensemble forecasting of species distributions. Trends Ecol Evol. 2007;22: 42–7. pmid:17011070
  41. 41. Elith J, Graham* C, Anderson R. Novel methods improve prediction of species’ distributions from occurrence data. Ecography (Cop). 2006;29: 129–151. Available:
  42. 42. Yackulic CB, Chandler R, Zipkin EF, Royle JA, James D, Grant EHC, et al. Presence-only modelling using MAXENT: when can we trust the inferences? 2013; 236–243. 10.1111/2041-210x.12004
  43. 43. Parra JL, Graham CC, Freile JF. Evaluating Alternative Data Sets for Ecological Niche Models of Birds in the Andes alternative data sets for niche models of birds in Evaluating ecological the Andes. Ecography (Cop). 2012;27: 350–360.
  44. 44. Tinoco B a., Astudillo PX, Latta SC, Graham CH. Distribution, ecology and conservation of an endangered Andean hummingbird: the Violet-throated Metaltail (Metallura baroni). Bird Conserv Int. 2009;19: 63.
  45. 45. Gutiérrez EE, Boria R a., Anderson RP. Can biotic interactions cause allopatry? Niche models, competition, and distributions of South American mouse opossums. Ecography (Cop). 2014;37: 741–753.
  46. 46. Elith J, Leathwick JR, Hastie T. A working guide to boosted regression trees. J Anim Ecol. 2008;77: 802–13. pmid:18397250
  47. 47. Phillips SSJ, Dudík M, Elith J, Graham CH, Lehmann A, Leathwick J, et al. Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecol Appl. 2009;19: 181–97. Available: pmid:19323182
  48. 48. Guisan A, Guisan A, Zimmermann NE, Zimmermann NE. Predictive habitat distribution models in ecology. Ecol Modell. 2000;135: 147–186.
  49. 49. Nogués-Bravo D. Predicting the past distribution of species climatic niches. Glob Ecol Biogeogr. 2009;18: 521–531.
  50. 50. Etten J Van. R package gdistance: distances and routes on geographical grids. 2011;
  51. 51. Liu C, Berry PPM, Dawson TPT, Pearson RGR. Selecting thresholds of occurrence in the prediction of species distributions. Ecography (Cop). 2005;3: 385–393. Available:
  52. 52. Pearson R, Dawson T, Liu C. Modelling species distributions in Britain: a hierarchical integration of climate and land-cover data. Ecography (Cop). 2004;27: 285–298. Available:
  53. 53. Algar AC, Mahler DL, Glor RE, Losos JB. Niche incumbency, dispersal limitation and climate shape geographical distributions in a species-rich island adaptive radiation. Glob Ecol Biogeogr. 2013;22: 391–402.
  54. 54. Wisz MS, Pottier J, Kissling WD, Pellissier L, Lenoir J, Damgaard CF, et al. The role of biotic interactions in shaping distributions and realised assemblages of species: implications for species distribution modelling. Biol Rev Camb Philos Soc. 2012;88: 15–30. pmid:22686347
  55. 55. Laube I, Graham CH, Böhning-Gaese K. Intra-generic species richness and dispersal ability interact to determine geographic ranges of birds. Glob Ecol Biogeogr. 2013;22: 223–232.
  56. 56. Jankowski JE, Londo GA, Robinson SK, Chappell MA. Exploring the role of physiology and biotic interactions in determining elevational ranges of tropical animals. 2012; 1–12. 10.1111/j.1600-0587.2012.07785.x
  57. 57. Cavender-Bares J, Kozak KH, Fine P V, Kembel SW. The merging of community ecology and phylogenetic biology. Ecol Lett. 2009;12: 693–715. pmid:19473217
  58. 58. McGuire JA a, Witt CCC, Remsen JV V, Corl A, Rabosky DLL, Altshuler DLL, et al. Molecular Phylogenetics and the Diversification of Hummingbirds. Curr Biol. 2014;24: 910–916. pmid:24704078
  59. 59. McGuire J a, Witt CC, Altshuler DL, Remsen J V. Phylogenetic systematics and biogeography of hummingbirds: Bayesian and maximum likelihood analyses of partitioned data and selection of an appropriate partitioning strategy. Syst Biol. 2007;56: 837–56. pmid:17934998
  60. 60. Altshuler DL, Dudley R, McGuire J a. Resolution of a paradox: hummingbird flight at high elevation does not come without a cost. Proc Natl Acad Sci U S A. 2004;101: 17731–6. pmid:15598748
  61. 61. Projecto-Garcia J, Natarajan C, Moriyama H, Weber RE, Fago A, Cheviron Z a, et al. Repeated elevational transitions in hemoglobin function during the evolution of Andean hummingbirds. Proc Natl Acad Sci U S A. 2013;110: 20669–74. pmid:24297909
  62. 62. Stiles FG. Phylogenetic Constraints Upon Morphological and Ecological Adaptation in Hummingbirds (Trochilidae): Why Are There No Hermits in the Paramo? Ornitol Neotrop. 2004;15: 191–198.
  63. 63. Maglianesi MA, Bluthgen N, Bohning-Gaese K, Schleuning M. Morphological traits determine specialization and resource use in plant-hummingbird networks in the neotropics. Ecology. 2014;95: 3325–3334.
  64. 64. Feinsinger P. Organization of a tropical guild of nectarivorous birds. Ecol Monogr. 1976;46: 257–291. Available:
  65. 65. Martín González AM, Dalsgaard B, Nogués-Bravo D, Graham CH, Schleuning M, Maruyama PK, et al. The macroecology of phylogenetically structured hummingbird-plant networks. Glob Ecol Biogeogr. 2015;24: 1212–1224.
  66. 66. Dalsgaard B, Magård E, Fjeldså J, Martín González AM, Rahbek C, Olesen JM, et al. Specialization in plant-hummingbird networks is associated with species richness, contemporary precipitation and quaternary climate-change velocity. PLoS One. 2011;6: e25891. pmid:21998716
  67. 67. Dalsgaard B, Trøjelsgaard K, Martín González AM, Nogués-Bravo D, Ollerton J, Petanidou T, et al. Historical climate-change influences modularity and nestedness of pollination networks. Ecography (Cop). 2013;36: 1331–1340.
  68. 68. Vizentin-Bugoni J, Maruyama PK, Sazima M. Processes entangling interactions in communities: forbidden links are more important than abundance in a hummingbird-plant network. Proc R Soc B Biol Sci. 2014;281: 20132397–20132397. pmid:24552835
  69. 69. Feinsinger P, Colwell R. Community organization among neotropical nectar-feeding birds. Am Zool. 1978;18: 779–795. Available:
  70. 70. Graham CHC, Parra JLJJL, Tinoco BAB, Stiles FG, McGuire J a. Untangling the influence of ecological and evolutionary factors on trait variation across hummingbird assemblages. Ecology. 2012;93: 99–111.
  71. 71. Londono GA, Chappell MA, Castaneda M del R, Jankowski JE, Robinson SK. Basal metabolism in tropical birds: Latitude, altitude, and the “pace of life.” Funct Ecol. 2015;29: 338–346.
  72. 72. Jankowski J, Robinson S. Squeezed at the top: interspecific aggression may constrain elevational ranges in tropical birds. Ecology. 2010;91: 1877–1884. Available: pmid:20715605
  73. 73. Terborgh J, Weske J. The role of competition in the distribution of Andean birds. Ecology. 1975;56: 562–576. Available:
  74. 74. Cadena CD, Loiselle B. Limits to elevational distributions in two species of emberizine finches: disentangling the role of interspecific competition, autoecology, and geographic variation in the. Ecography (Cop). 2007; 491–504.
  75. 75. Bleiweiss R. Origin of hummingbird faunas. Biol J Linn Soc. 1998;65: 77–97.
  76. 76. Benham PM, Cuervo AM, McGuire J a., Witt CC. Biogeography of the Andean metaltail hummingbirds: contrasting evolutionary histories of tree line and habitat-generalist clades. J Biogeogr. 2015;42: 763–777.
  77. 77. Pigot a. L, Tobias J a. Dispersal and the transition to sympatry in vertebrates. Proc R Soc B Biol Sci. 2014;282: 20141929–20141929. pmid:25621326
  78. 78. Wiens JJ. The niche, biogeography and species interactions. 2011; 2336–2350. pmid:21768150
  79. 79. Graham CH, Fine PVA. Phylogenetic beta diversity: linking ecological and evolutionary processes across space in time. Ecol Lett. 2008;11: 1265–77. pmid:19046358
  80. 80. American Ornithologists’ Union. Check-list of North American Birds. 7th ed. Washington DC; 1998.