Energy input, habitat heterogeneity and host specificity drive avian haemosporidian diversity at continental scales

Abstract

The correct identification of variables affecting parasite diversity and assemblage composition at different spatial scales is crucial for understanding how pathogen distribution responds to anthropogenic disturbance and climate change. Here, we used a database of avian haemosporidian parasites to test how the taxonomic and phylogenetic diversity and phylogenetic structure of the genera Plasmodium, Haemoproteus and Leucocytozoon from three zoogeographic regions are related to surrogate variables of Earth's energy input, habitat heterogeneity (climatic diversity, landscape heterogeneity, host richness and human disturbance) and ecological interactions (resource use), which was measured by a novel assemblage-level metric related to parasite niche overlap (degree of generalism). We found that different components of energy input explained variation in richness for each genus. We found that human disturbance influences the phylogenetic structure of Haemoproteus while the degree of generalism explained richness and phylogenetic structure of Plasmodium and Leucocytozoon genera. Furthermore, landscape attributes related to human disturbance (human footprint) can filter Haemoproteus assemblages by their phylogenetic relatedness. Finally, assembly processes related to resource use within parasite assemblages modify species richness and phylogenetic structure of Plasmodium and Leucocytozoon assemblages. Overall, our study highlighted the genus-specific patterns with the different components of Earth's energy budget, human disturbances and degree of generalism.

1. Background

Identifying biotic and abiotic drivers of parasite diversity and assemblage composition provides insights into how pathogen distribution may change in the context of anthropogenic disturbance and climate change, which is a key topic for wildlife conservation and public health [1–3]. One of the most notorious diversity patterns in macroecology is the latitudinal diversity gradient [4], where the highest diversity of major groups of organisms is found in low-latitude regions. Among the abiotic mechanisms proposed to explain latitudinal gradients is the species richness (SR)–energy hypothesis [5]. This hypothesis posits that the number of species of an assemblage is influenced by the available environmental energy, which results from a combination of solar radiation, temperature and water–energy dynamics. Another mechanism, the habitat heterogeneity hypothesis [6,7], explains species diversity as a function of habitat diversity; thus, habitat heterogeneity allows coexistence by reducing competition. Additionally, a hypothesis based on biotic interactions [8,9] suggests that the strength of ecological interactions is positively correlated with diversity. Yet, gradients of parasite diversity can be highly complex and contrasting between taxa, implying that such hypotheses may or may not apply to different parasite taxa [10–12].

Although high diversity and abundance of parasites are usually associated with tropical areas, some parasite groups such as avian haemosporidians and mammal fleas do not exhibit the classic latitudinal gradient of diversity [13,14]. The diversity of these parasite groups may be better explained by uneven macroevolutionary processes across space; for example, the differential rates of host switching, host co-speciation and diversification events that have produced evolutionarily unique and diverse parasite assemblages in non-tropical regions [15,16]. Also, factors related to habitat heterogeneity (e.g. climatic diversity, vegetation complexity, landscape attributes, land use change, anthropogenic activities and host richness) can generate diversity gradients of parasites, but these relationships have been poorly explored in parasite assemblages at continental scales. In addition, parasites can indirectly interact through host immune systems, where such interactions can determine the success of other parasites in the same hosts [17]. Therefore, the way parasites exploit resources, and the direction of their interactions can shape parasite assemblages and generate diversity gradients that do not necessarily follow those exhibited by free-living organisms [18,19]. Parasite specialization is thought to be a positive driver of parasite richness, allowing coexistence by niche partitioning [13,20]. However, there remains a paucity of studies delving into the relationship between the resource use of parasites within assemblages and its diversity.

Avian haemosporidians, also known as avian malaria and related parasites (genera Plasmodium, Haemoproteus and Leucocytozoon), infect most bird clades of almost every ecosystem of the world [13,21,22]. Each haemosporidian genus is transmitted by different families of dipteran vectors that have different environmental restrictions [23], resulting in exclusive responses of abundance and distribution of each genus to environmental gradients [24,25]. Heterogeneous patterns of abundance across haemosporidian lineages of the same genus are known to emerge with variation in host identity, vegetation structure [26] and human disturbance [27]. Previous research on spatial patterns of haemosporidians has suggested that the abundance and the geographical distribution of the different haemosporidian groups are influenced by a combination of climatic conditions [25,28], landscape attributes [24,26] and within-host dynamics (host-related natural history) [28–30]. Furthermore, the distribution of haemosporidians is also shaped by their host specificity, where sizes of geographical ranges of parasites are linked to the degree of parasite specificity [31], and Pinheiro et al. [32] noted that lineage richness is correlated to specialization in South American assemblages. However, among the diverse set of reported predictors that constrain haemosporidian distributions, there is a lack of synthesis of the relative importance and direct relationships between landscape attributes, host specificity of parasite assemblages, and different dimensions of parasite diversity such as taxonomic diversity, phylogenetic diversity and phylogenetic structure, which accounts for the number of parasite lineages, variety of evolutionary histories within assemblages and the phylogenetic relatedness of species within assemblages, respectively.

Here, we used the MalAvi database to describe and evaluate taxonomic diversity (i.e. SR), phylogenetic diversity and phylogenetic structure of haemosporidians across the globe. We then explored the relationships between these haemosporidian diversity metrics and variables related to energy input (temperature, precipitation, net radiation and actual evapotranspiration) and to habitat heterogeneity measured by climatic diversity, landscape and ecosystem heterogeneity, human disturbance level and host richness. We also evaluated how metrics related to ecological interactions linked to parasite resource use (i.e. host species) can explain parasite diversity by using a novel assemblage-level metric (degree of generalism) obtained from measuring the host specificity of co-occurring parasites (see table 1 for a description of the evaluated hypotheses on how the predictors may affect parasite diversity metrics).

Table 1.

Predictions for taxonomic and phylogenetic diversity and phylogenetic structure of haemosporidians in relation to energy input variables, habitat heterogeneity and ecological interactions.

2. Methods

(a) . Parasite occurrence data

To obtain the occurrence of parasite lineages, we acquired records from MalAvi [22], a global database with at least 2400 haemosporidian lineages of mtDNA sequences of a cytochrome b (cyt-b) gene fragment (479 bp) that has been used as a barcode [33]. We considered each haplotype as a unique parasite lineage [33] and in the case of duplicated sequences with different names, we randomly assigned duplicated sequences to the same lineage name. We then downloaded the Host and Sites Table (accessed in March 2023), we validated geographical coordinates at country level, and the final table had 10 302 records based on 286 references published between 2006 and 2023.

To reduce sampling bias, we identified Nearctic, Neotropical and Palearctic zoogeographic realm levels (as defined by Holt et al. [34]) as suitable for this study because they are currently the best-sampled regions for avian haemosporidian parasites (figure 1a). Since most of the records in the Palearctic region were in the west, we just considered the West Palearctic region as defined by Snow et al. [35]. We used a Behrmann equal-area grid with a cell size of 27 × 27 km² (approx. 0.25° at the equator) and intersected it with data of each haemosporidian genus. We obtained three datasets, one for each haemosporidian genus, containing cells from the three zoogeographic realms with at least two lineages. We considered each of these cells as a single assemblage of parasites.

Figure 1.

(a) Studied haemosporidian assemblages across Nearctic, Neotropical and West Palearctic regions. (b) Boxplots for diversity metrics of parasite assemblages by genus. SR, species richness; RPD, residual phylogenetic diversity; PSV, phylogenetic species variability.

(b) . Parasite phylogenetic trees and diversity metrics

We created three independent cyt-b alignments for each genus: Plasmodium alignment had 692 lineages, Haemoproteus alignment had 668 lineages and Leucocytozoon alignment had 563 lineages. For each genus, we applied Bayesian methods using MrBayes [36] at CIPRES portal (phylo.org). We constructed a general time-reversible model with invariant gamma-distributed substitution model (GTR + I + G), as this was the best model using sump function in MrBayes [37]. We ran two MCMC simulations simultaneously with four chains each, sampling every 5000 generations until convergence (maximum of 300 million). We settled a prior substitution rate with a mean of 0.004 and standard deviation of 0.00033 as these values were obtained from the substitution rates per million years calculated in different models by Pacheco et al. [38] for the whole cyt-b gene. The lineage GALLUS06 (Leucocytozoon) was used as an outgroup for Plasmodium and Haemoproteus, whereas PADOM09 (Plasmodium) was used as an outgroup for Leucocytozoon. Relative time was calculated using the mean branch length method [39].

The convergence of the two chains was assumed once the average standard deviation of the posterior probability was less than 0.01. The values of the potential scale reduction factor (PSRF) were between 0.9999 and 1.0004 and values of average effective sample size (ESS) were above 300. Then, a ‘burn-in' of 25% of the trees was discarded. Given that the phylogenetic uncertainty in trees based on the cyt-b fragment was high, we accounted for this variation to ensure the robustness of the diversity metrics. To do so, we employed the workflow of the treespace R package [40]. We used a sample of 10 000 trees to build a matrix based on the concordance of the phylogenetic hypotheses according to the Kendall Colijn metric. We then conducted a two-dimensional hierarchical clustering analysis to identify 100 distinct groups of trees and constructed a maximum clade credibility (MCC) tree for each of these groups.

We analysed variations in taxonomic diversity, phylogenetic diversity and phylogenetic structure. We measured taxonomic diversity as the number of different lineages in each cell, or SR. This metric was squared root transformed to facilitate the modelling framework. To measure phylogenetic diversity for each cell, we used their corresponding residuals from a generalized additive model (GAM) regression between Faith's phylogenetic diversity [41] and SR across cells. The resulting residual phylogenetic diversity (RPD) decouples the quantity of evolutionary history measured as the sum of phylogenetic branch lengths in each assemblage from SR, with positive values indicating greater evolutionary histories relative to the number of species in an assemblage whereas negative values indicate that the parasite assemblage has lower evolutionary history than expected by the number of species [42–44]. To measure the phylogenetic structure or the degree of phylogenetic relatedness of co-occurring parasite species in each cell, we calculated the phylogenetic species variability (PSV) as implemented in the R package Picante, version 1.8.2 [45]. This metric measures the expected among-species variance in a hypothetical trait that evolved neutrally; when species are closely related, this variance is reduced, and PSV can take values closer to 0, indicating phylogenetic clustering. PSV can take values closer to 1 when species are fully dispersed across the phylogeny [46,47].

(c) . Predictors of haemosporidian diversity

We assessed the effects of 11 variables as predictors of each diversity metric variation for the three haemosporidian genera. All predictors were calculated at 0.25°×0.25° resolution and overlaid with the cells containing haemosporidian assemblages. Due to the different units and magnitudes of the predictors, these were standardized to 0 mean and standard deviation of 1 to compare the predictors' effects.

(i) . Energy input predictors

To measure energy availability in the environment, we used both variables that proxy solar radiation such as temperature (annual mean temperature-Bio1 from WorldClim v.2) [48] and net radiation from CERES product of Nasa Earth Observations (https://neo.gsfc.nasa.gov/). We also included variables that proxy energy–water dynamics such as precipitation (annual precipitation-Bio 12) and actual evapotranspiration (AET) from USGS [49].

(ii) . Habitat heterogeneity predictors

We used the temperature annual range (Bio 7) and precipitation seasonality (Bio 15) to account for climatic heterogeneity (WorldClim v.2) [48]. To measure landscape heterogeneity, for each locality, we obtained mean proportions of land cover types from MODIS (MOD12Q1) from 2000 to 2020, then we calculated Shannon index using proportions by locality and used it as a measure of landscape heterogeneity. We also obtained monthly averages of the Enhanced Vegetation Index (EVI) from the last 20 years from MODIS (MOD12A3) (approx. 0.05° resolution). We used rasterDIV R package [50] to calculate Rao's Q as a measure of the diversity of EVI data and used it as a proxy for ecosystem heterogeneity based on vegetation structure. We also included measures of the level of anthropogenic disturbance in each parasite assemblage as a factor of biotic homogenization that shapes habitat heterogeneity for parasites. For this, we first used the ‘Human footprint' dataset published by [51], which has an approximately 0.01° resolution and scores anthropogenic disturbance from 0 to 50 based on anthropogenic pressures such as croplands, built-up environments and electrical power infrastructure, among others. Second, we used global human population density (GPW version 4, from CIESIN, https://sedac.ciesin.columbia.edu/) with data on human density by km² at approximately 0.01° resolution as a proxy of urbanization. Finally, we calculated host (bird species) richness as the number of species in each assemblage as a proxy of niche availability for parasites. For that, we used the letsR package for R [52] and geographical distribution range maps from Birdlife International (www.birdlife.org).

(iii) . Ecological predictor

To calculate the degree of generalism of parasite assemblages, we first constructed a maximum clade credibility tree with host species reported in the MalAvi database by downloading host phylogenetic hypotheses from birdtree.org [53]. Subsequently, Faith's phylogenetic diversity was calculated for the pool of host species reported for each parasite lineage in our dataset. This metric, akin to the phylogenetic interaction niche of Dehling et al. [54], illustrates the evolutionary host's range of each lineage across its geographical distribution. The phylogenetic interaction niche of lineages with only one host reported was calculated as the branch length of that unique host.

Following a methodology similar to Vimal & Devictor's [55] averaged specialization measure, we computed the average phylogenetic interaction niches of coexisting lineages within each parasite assemblage (i.e. grid cell). This calculation provides insights into the degree of generalism exhibited by each parasite assemblage. Haemosporidian assemblages composed of generalist lineages are expected to have higher degree of generalism.

(d) . Variable importance and model building

To evaluate the influence of predictors on parasite diversities, we used GAMs. GAMs represent an extension of generalized linear models introducing flexibility through the incorporation of smooth functions, typically modelled using splines, which are piecewise polynomials [56]. The utilization of splines enables GAMs to adapt to nonlinear structures in the data and makes them a powerful tool for modelling complex nonlinear relationships. We used machine learning algorithms to train GAMs and to detect the most important variables explaining the three measures of diversity for each haemosporidian genus. The diversity and distribution of haemosporidians are known to be geographically structured at the zoogeographic region level [15]; thus we included the centroid latitude and longitude of each assemblage to compare the importance of studied predictors with geography. We used the train and varimp functions in the caret R package [57] for these purposes. The importance scores of each predictor are based on their contribution in terms of R² to models produced by random permutations of the predictors fitted in each model. Importance values were scaled to show that scores closer to 100 indicate that a specific variable is the most important predictor of a model, and values closer to 0 suggest that a variable has little effect on dependent variables. Furthermore, we performed variable importance analysis for the 100 diversity measurements derived from the MCC trees and calculated standard deviation error bars for each predictor. We subsequently implemented a cut-off at the 25th percentile to exclude predictors with extremely low importance scores and to refine our model selection process, predictors with scores or lower error bars equal to or below this threshold were excluded from the model-building process.

To detect robust predictors and explore their relationships with haemosporidian diversity, we used the SR as well as the PSV and RPD calculated from the 100 MCC trees. We searched for the most suitable geographically explicit models to explain different dimensions of the diversity of the three haemosporidian genera, while looking for the most parsimonious models of each diversity metric for the three datasets. For this, we implemented the procedure of Nussbaum et al. [58] in the geoGAM package for R, which searches for the best models based on a component-wise gradient boosting algorithm [59]. This procedure fits GAMs iteratively, adding predictors sequentially, accounting for residual autocorrelation, and for spatially changing dependence (nonstationary effects) between predictors and dependent variables. For the models with phylogenetic diversity metrics, we only report predictors that were significant (mean p-value below 0.05) and were present in at least 80% of the models with diversity metrics taken from the 100 different phylogenetic hypotheses.

We examined the Pearson correlations among predictors in each of the three final datasets. Most of the correlations have coefficients below 0.6, except for temperature which showed a positive correlation with AET (up to 0.82) and a negative correlation with the diurnal temperature range (up to 0.78; electronic supplementary material, figure S1). To measure overfitting in the final models, particularly when the model had more than one predictor, we calculated concurvity estimates using concurvity function of the mgcv R package. This metric can be viewed as a generalization of co-linearity in GAMs, where 1 is the maximum value of concurvity. Notably, in all cases, our models exhibited concurvity values below 0.4 (electronic supplementary material, tables S1–S3).

3. Results

In total, we analysed 764 assemblages of three haemosporidian genera (Plasmodium, Haemoproteus and Leucocytozoon) distributed in three zoogeographic realms (table 2 and figure 1a). Diversity values among assemblages were similar among genera (figure 1b).

Table 2.

Details of the datasets for the three genera, with the final number of records, cells with data, parasite lineages and host species.

dataset	no. points	cells with data	no. parasite lineages	no. hosts
Plasmodium	1156	331	719	1215
Haemoproteus	889	243	696	1202
Leucocytozoon	675	187	604	587

The diversity of each parasite genus responded to different sets of predictors, and relevant predictors also varied between different parasite diversity metrics even within the same genus (figure 2). Among energy input variables, net radiation was important for explaining SR of the three haemosporidian genera as well as for the phylogenetic structure of Haemoproteus. Temperature and evapotranspiration were important, albeit to a lesser degree, in explaining SR for Plasmodium and Leucocytozoon. Additionally, precipitation was found to be an important factor in explaining variation in RPD of Plasmodium assemblages. Turning to habitat heterogeneity predictors, landscape heterogeneity was an important factor for SR of Leucocytozoon and Haemoproteus, and ecosystem heterogeneity was one of the most important predictors of Leucocytozoon SR, and bird richness was a significant variable for SR across all three haemosporidian genera. Regarding human-related predictors, human population density and human footprint were found to influence the phylogenetic structure (PSV) of Haemoproteus. Furthermore, the degree of generalism was an important predictor of Plasmodium SR, and consistently played a key role in explaining phylogenetic diversity metrics (RPD and PSV) across all three genera.

Figure 2.

Importance scores of variables to explain variation in three metrics of diversity across assemblages of Plasmodium, Haemoproteus and Leucocytozoon: (a) taxonomic diversity (SR); (b) residual phylogenetic diversity (RPD); and (c) phylogenetic species variability (PSV). For phylogenetic metrics, error bars represent the standard deviation from the mean importance scores (based on the measurements of the 100 MCC phylogenetic trees). AET, actual evapotranspiration.

Variable importance analysis confirmed that geography, particularly latitude, is an important factor for predicting SR of Haemoproteus and Leucocytozoon. When considering geography in the geo-GAMs, we fitted the best models with a reduced set of statistically significant predictors that explained between 18% and 49% of the variation in SR of single-genus haemosporidian assemblages. Our analysis did not reaveal a strong association between the studied predictors and Plasmodium RPD. Models for RPD of Haemoproteus and Leucocytozoon had low performance (R² between 7% and 16%) and they were not statistically significant (p < 0.05; table 3). Final geo-GAMs also explained between 20% and 32% of the variation in PSV of the parasite assemblages (table 3; details of the models in electronic supplementary material, table S4).

Table 3.

Final geoGAMs for explaining taxonomic diversity and phylogenetic structure variation across haemosporidian assemblages from three zoogeographic realms. Model fitness metrics include deviance explained (dev. explained) and R². SR, species richness; RPD, residual phylogenetic diversity; PSV, phylogenetic species variability. Bold values indicate overall statistical significance of models at p < 0.05. For phylogenetic metrics, the displayed values represent the mean and standard deviation calculated from 100 models.  

genus	metric	significant predictors (p < 0.05)a	model p-value	dev. explained	R2
Plasmodium	SR	s(AET)	<0.005	0,49	0.44
		s(precipitation)
		s(Rad. Balance)
		s(temperature)
		s(degree of generalism)
	PSV	s(degree of generalism)	<0.001 ± 0	0.22 ± 0.033	0.207 ± 0.033
Haemoproteus	SR	s(precipitation)	<0.001	0.26	0.21
		s(rad)
	RPD	s(temperature)	0.988 ± 0.007	0.09 ± 0.001	0.076 ± 0.01
	PSV	s(human_footprint)	<0.001 ± 0	0.324 ± 0.065	0.286 ± 0.059
Leucocytozoon	SR	s(temperature)	<0.001	0,18	0.13
		s(degree_of_generalism)
	RPD	s(degree of generalism)	1 ± 0	0.09 ± 0.001	0.076 ± 0.01
	PSV	s(degree of generalism)	<0.001 ± 0	0.31 ± 0.058	0.276 ± 0.051

^aIn phylogenetic models, only predictors that were significant at least in 80% of the models were included.

(a) . Energy input

Final models of SR of each haemosporidian genus showed that energy predictors (i.e. AET, temperature, precipitation and net radiation) were significant for explaining Plasmodium SR (R² = 0.44, dev. explained = 49%), temperature and precipitation exhibited nonlinear and positive associations with Plasmodium SR (figure 3a,b). Conversely, AET and net radiation exhibited a pattern with peaks of Plasmodium SR in regions with high radiation and evapotranspiration (figure 3c,d). Haemoproteus SR (R² = 0.21, dev. explained = 26%) was explained by precipitation with a positive nonlinear relationship (figure 3g) and by radiation balance with a nonlinear pattern with two peaks of SR at regions with moderate and high net radiation (figure 3h). In addition, temperature was the only predictor in the final model for Leucocytozoon SR (R² = 0.13, dev. explained = 18%) suggesting a slight negative nonlinear relationship with peaks of richness in regions with moderate temperatures (figure 3j).

Figure 3.

Relationships between the best predictors obtained from geoGAMs for species richness (SR), residual phylogenetic diversity (RPD) and phylogenetic species variability (PSV) of Plasmodium (a–f), Haemoproteus (g–i) and Leucocytozoon (j–l). AET, actual evapotranspiration; D. generalism, degree of generalism; H. footprint, human footprint.

(b) . Habitat heterogeneity

Among habitat heterogeneity predictors, geographical models suggested that landscape heterogeneity variables such as human footprint explain phylogenetic structure (PSV) of Haemoproteus assemblages (R² = 0.28, dev. explained = 32%), with more phylogenetically clustered assemblages as human disturbance increases, but with dispersed assemblages at more extreme values of human footprint (figure 3i).

(c) . Ecological interactions

Finally, the degree of generalism was a significant predictor of SR and PSV for Plasmodium and Leucocytozoon assemblages. Plasmodium SR exhibited a peak at a low degree of generalism, then decreased as the degree of generalism increased (figure 3e).

PSV of Plasmodium assemblages was explained only by the degree of generalism (R² = 0.2, dev. explained = 22%), where phylogenetic dispersion of parasite assemblages decreased as their degree of generalism increased, and there was a strong increase in phylogenetic dispersion at extremely generalist assemblages (figure 3f). The degree of generalism had a nonlinear relationship with Leucocytozoon SR, with an N-shaped pattern, showing high SR at a moderate and high degree of generalism (figure 3k). The PSV of Leucocytozoon (R² = 0.27, dev. explained = 32%) assemblages was explained by a quadratic relationship with the degree of generalism, where higher PSV increased up to a moderate degree of generalism, and then PSV decreased at extreme values of generalism (figure 3l).

4. Discussion

To foster our understanding about the heterogeneous patterns of parasite diversity at continental scales, we explored the relationships of three dimensions of haemosporidian diversity (i.e. SR, phylogenetic diversity and phylogenetic structure) with a set of variables related to energy input, environmental heterogeneity and ecological interactions, all of which are hypothesized as drivers of parasite distribution. When measuring the relative importance of these divers for explaining each haemosporidian diversity metric, we found that host richness—a measure of habitat heterogeneity for parasites—was important for the taxonomic diversity of Plasmodium, Haemoproteus and Leucocytozoon, and the degree of generalism was consistently the most important predictor for phylogenetic diversity and phylogenetic structure of the three haemosporidian genera. Yet, when accounting for geography, we were unable to fit statistically significant models where degree of generalism would explain spatial variation in phylogenetic diversity (RPD). We did find statistically significant effects, however, of degree of generalism on parasite taxonomic diversity (SR) and phylogenetic structure (PSV) of Plasmodium and Leucocytzoon assemblages. Additionally, we found that different subsets of energy input variables explained SR of the three genera, and that anthropogenic disturbance explained phylogenetic structure (PSV) of Haemoproteus.

(a) . Energy input

Our study revealed distinct impacts of Earth's energy budget components on taxonomic diversity of the three haemosporidian genera. All energy predictors were significant in the model for Plasmodium SR, showing positive relationships and suggesting a higher SR in regions with higher energy input. This observation aligns with existing literature that has previously associated temperature as a crucial predictor for the distribution of Plasmodium [60,61]. Moreover, we noted a decline in SR of Plasmodium assemblages in regions characterized by elevated values of AET and radiation. This decline can be attributed to the conditions found in neotropical mountain ecosystems, which can feature high radiation and evapotranspiration levels but relatively low temperatures. Indeed, low temperatures pose a limit to the sporogonic development of Plasmodium, limiting its distribution in mountain ecosystems such as the Andes [61,62].

We found no evidence to support the energy input hypothesis for Haemoproteus assemblages. Instead, our analysis revealed two distinct peaks of SR in relation to precipitation and net radiation, which are proxies for an Earth's energy component related to structure and type of land covers as well as to primary production and water availability [63]. Our model suggested that richer Haemoproteus assemblages are associated with different regimes of energy, such as those in tropical regions with high radiation and precipitation as well as the non-tropical regions with low to moderate radiation and precipitation (i.e. Nearctic savanna and Iberian mountains in the West Palearctic) [13,21,64].

Similarly, we found that Leucocytozoon assemblages exhibited higher SR in regions characterized by lower to moderate temperatures. Leucocytozoon is known to adapt well to both tropical and temperate environments [13,65,66]. However, abundant and diverse assemblages of Leucocytozoon in the Neotropical region are mainly restricted to highlands or non-equatorial systems [24,65,66]. This is intriguing because Leucocytozoon is distributed in the Afrotropical region and has been detected in Amazonian lowlands in migratory birds [65,66]. Therefore, factors other than temperature may act to limit the distribution of Leucocytozoon to montane systems in the Neotropical region. For example, it has been suggested that the presence of specific vector groups exclusively distributed in temperate or mountainous systems (i.e. Diptera: Simuliidae, genus Gigantodax spp.) may play a significant role in constraining the Neotropical distribution of these parasites [67].

(b) . Habitat heterogeneity

Host richness, often used as a metric for habitat heterogeneity for the environment of parasites [11,20], was an important predictor of the SR of the three haemosporidian genera. However, the importance of host richness as a predictor diminished when we considered spatial autocorrelation in our final models. This outcome contrasts with the hypothesis of higher host diversity positively correlating with parasite diversity [11,60]. Instead, our models suggested that lineage richness of haemosporidian assemblages is better explained by predictors linked to Earth's energy budget components, which are also known to be influential factors in explaining bird richness at continental scales [68].

Habitat heterogeneity provided by landscape attributes related to vegetation structure and density are known to explain variation in haemosporidian transmission rates at regional [69–71] and global scales [25]. In this study, we did not find compelling evidence to support the idea that the diversity of land covers or the diversity of ecosystems based on vegetation types predict parasite diversity. These results suggest that landscape attributes may have a stronger connection with other aspects of haemosporidian diversity, such as abundance, rather than being prominent drivers of taxonomic and phylogenetic diversity at the global scale.

We included metrics of human disturbance as a proxy for biotic homogenization processes and expected that such disturbance would reduce haemosporidian diversity across local assemblages at continental scales. We found that the taxonomic diversity (SR) and phylogenetic diversity (RPD) of haemosporidians were not significantly affected by human disturbances, suggesting that most disturbed areas can harbour as much diversity as less disturbed areas, as observed in some regional systems [72,73]. This pattern can be explained due to the highly heterogeneous nature of urban environments, which may contain contrasting habitats (well-preserved urban forests versus built-up areas with impervious surfaces and low percentages of green spaces) in short spatial distances, thus producing strong environmental gradients that will affect diversity in myriad ways [74]. We also hypothesized that human disturbance could modify parasite assemblages based on their evolutionary relatedness, producing either phylogenetically dispersed or clustered assemblages. We found support for this hypothesis in Haemoproteus assemblages because human footprint was one of the most important predictors for their PSV. We observed that Haemoproteus PSV presented a nonlinear relationship with the human footprint, with an initial negative relationship suggesting that assemblages at disturbed places may be composed of evolutionarily related (i.e. phylogenetically clustered) lineages. This pattern may be produced if increasing levels of human disturbance (as measured by human footprint scores) non-randomly remove lineages along the local phylogeny, where only some Haemoproteus-related lineages (i.e. specific clades) are resistant to human transformations (e.g. [27]). Subsequently, we observed that the phylogenetic dispersion of Haemoproteus assemblages increases at high levels of human disturbance. Phylogenetically dispersed assemblages in urban areas can be observed in highly heterogeneous cities capable of supporting different evolutionary histories of hosts (i.e. cities with varying degrees of greenery and blue spaces that add habitat heterogeneity to urban environments promoting the presence of native and non-native species [75]).

(c) . Ecological interactions

We expected to observe a negative relationship between the degree of generalism and parasite diversity, following the principle of niche partitioning [20]. Although these relationships were not straightforward due to complex nonlinear relationships and genus-specific patterns for each diversity metric, we found partial support for this hypothesis for Plasmodium and Leucocytozoon SR. Although we observed that the highest SR in both genera were found at moderate to low levels of generalism, generalist assemblages had the lowest number of parasite lineages. This is in line with related research on bird haemosporidians where parasite lineage richness was linked to specificity of parasites in interaction networks in South America [32] as well as in African lowland rainforests [76].

We also expected that if host specificity is fixed within co-occurring and closely related haemosporidian lineages, we would observe phylogenetically clustered assemblages (low PSV) in both assemblages composed by generalist lineages and assemblages composed by specialist lineages. We observed that the degree of generalism was the only predictor explaining variation in the phylogenetic structure of Plasmodium and Leucocytozoon assemblages. In the case of Plasmodium, PSV decreased as the degree of generalism increased, showing that generalist assemblages may be composed of closely related lineages. However, phylogenetic dispersion increased in extremely generalist assemblages, where unrelated generalist lineages of Plasmodium may coexist. By contrast, PSV of Leucocytozoon assemblages and the degree of generalism displayed a parabolic-like relationship, showing that phylogenetic relatedness of Leucocytozoon lineages within assemblages is higher when the parasites are mainly specialists. However, as the degree of generalism increases at high values, PSV decreased. Despite the complex patterns observed between the degree of generalism and diversity, our results suggest that lineage-specific eco-evolutionary traits (e.g. host specificity) within a host assemblage can shape parasite interactions and coexistence at the global scale. Therefore, it is valuable to analyse the degree of generalism as a property of parasite assemblages that can explain variation in different dimensions of diversity.

(d) . Study limitations and considerations

The patterns exposed here must be taken with caution because our models explained only a fraction of the variation in haemosporidian diversity, and the calculation of some metrics was limited by data availability. First, the parasite's host specificity calculated from the MalAvi database had a high proportion of lineages that interacted with only one host species. Rather than a real specialization of all these lineages, some of them may show subsampling. In addition, we acknowledge that host specificity is a labile trait that can change in response to environmental conditions and is scale-dependent [77,78]; also, the availability of weighted data and spatio-temporal variation on haemosporidian–host interactions could bring more resolution to the patterns discussed here. Yet, to overcome some of these caveats, we used an assemblage level metric that reflects the mean phylogenetic diversity of the hosts being parasitized within a parasite assemblage and found that this metric could explain a significant proportion of the variation in lineage richness and phylogenetic structure of Plasmodium and Leucocytozoon assemblages.

Second, diversity metrics based on phylogenies from a short fragment of mtDNA cyt-b presented uncertainty. Nonetheless, we addressed this issue by testing whether predictors were robust by evaluating 100 models fitted with diversity metrics obtained from a representative set of possible phylogenetic hypotheses. Thus, predictors such as the degree of generalism for Plasmodium and Leucocytozoon, as well as the human footprint for Haemoproteus, may be robust since they were consistently important in explaining variation in taxonomic diversity and phylogenetic structure across different phylogenetic hypotheses.

Finally, transmission dynamics of haemosporidian parasites are associated with their vector abundance and distribution [23], which is in turn linked to energy input variables such as precipitation and temperature [61], and landscape modification [79]. Therefore, vector assemblages might be an important factor shaping the assemblages of avian haemosporidian parasites. Unfortunately, current knowledge of interactions between haemosporidians and their vectors is insufficient. Diptera families known to interact with haemosporidian genera are widespread, ubiquitous and respond differently to the same environmental conditions [21,23,67]. Data at lower taxonomic scales are required to think about specific groups of competent vectors driving the diversity of haemosporidian parasites across environmental gradients.

5. Conclusion

We synthesized the relationships between a diverse set of predictors related to energy input, habitat heterogeneity, ecological interactions and the local diversity of avian haemosporidians across three zoogeographic realms. We identified key predictors and described patterns that help explain variations in both richness and phylogenetic structure of the three haemosporidian genera. We observed that distinct components of energy budget affect differentially haemosporidian genera. Plasmodium taxonomic richness showed a pattern consistent with the energy input hypothesis, while Haemoproteus and Leucocytozoon richness displayed nonlinear relationships with energy and exhibited peaks of richness in both temperate and tropical regions with low and high energy input, respectively. Processes related to landscape homogenization, such as human disturbance, can shape the phylogenetic structure of Haemoproteus assemblages but not that of the other haemosporidian genera. Furthermore, our results supported the partition niche principle for Plasmodium and Leucocytozoon assemblages as taxonomic diversity was lower in assemblages with higher degree of generalism. We also identified nonlinear relationships between the degree of generalism of Plasmodium and Leucocytozoon and their phylogenetic relatedness within assemblages, suggesting that the degree of generalism reflects assembly processes related to resource (host) use within haemosporidian assemblages that can modify their SR and phylogenetic structure.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

Code scripts and datasets associated with this paper are archived in the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.hx3ffbgkg [80] and also can be cloned at the GitHub repository of O.D.H.C. at https://github.com/OscarHCECO/GlobalHaemosporidianDiversity.

Supplementary material is available online [81].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors' contributions

O.D.H.C.: conceptualization, data curation, formal analysis, investigation, methodology, visualization, writing—original draft, writing—review and editing; E.J.T.-R.: conceptualization, data curation, formal analysis, methodology, writing—review and editing; F.V.: conceptualization, data curation, formal analysis, writing—review and editing; L.C.-V.: supervision, writing—review and editing; D.S.-A.: conceptualization, investigation, methodology, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This study was partially supported by the CONAHCYT programme Problemas Nacionales 2015-01-1628, CONACYT Ciencia Básica 2011-01-168524, and by startup funds provided to D.S.-A. by USF. We also thank the National Council of Science and Technology (CONAHCYT-Mexico) for the doctoral scholarship granted to O.D.H.C. (CONAHCYT, 2020-000026-02NACF-27722) and the graduate programme of the Institute of Ecology AC (INECOL). E.J.T.-R. was supported by a postdoctoral fellowship from CONAHCYT.