Ecological drivers of avian haemosporidian richness, host specificity, and network centrality in South America

Published: 2 November 2026| Version 1 | DOI: 10.17632/myjk6wvgww.1
Contributors:
,
,
,
,
,

Description

We tested three hypotheses about avian haemosporidians (Plasmodium, Haemoproteus subgenera Haemoproteus and Parahaemoproteus, and Leucocytozoon) in resident South American birds: (1) lineage richness increases with host beta-diversity; (2) richness and host specificity are shaped by climate, human footprint and geographic gradients, with genus-specific responses reflecting vector ecology; and (3) generalist lineages and widely distributed hosts are more central in host–parasite networks. The dataset contains 6,024 curated records, each the detection of a cytochrome b (cytb) lineage in one host individual at a georeferenced locality, compiled from MalAvi (accessed 7 June 2024; N = 4,300) and three Colombian institutional databases (GERPH–Universidad Nacional, N = 237; Universidad del Valle, N = 1,397; EcoEvo–Universidad de Antioquia, N = 90). They cover 1,400 lineages (73 H. (Haemoproteus), 431 H. (Parahaemoproteus), 641 Plasmodium, 255 Leucocytozoon) in 833 bird species. Files: Registros_suramerica.txt (all records: lineage, host species, genus, detection, coordinates in WGS84) and the same records split by genus (*_Norm.txt); specificity_per_cell_per_lineage.txt (one row per lineage × grid cell, with network size, d′, H′₂ and their null-model means, SD and P values, and betweenness centrality, BC); Suppl_BC_parasites.csv and Suppl_BC_hosts.csv (connector nodes with Z-scores); predictor rasters on a 0.18° (~20 × 20 km) grid (WorldClim BIO1, BIO4, BIO12, BIO15; human footprint; host richness; host beta-diversity) plus shapefiles; and three R scripts reproducing richness/accumulation analyses, per-cell networks and GAMs. Main findings: richness peaks in the Amazon basin and Tropical Andes, but only 42–50% of the Jackknife2-estimated diversity is known. H′₂ and d′ exceeded null expectations, with high spatial variation. No lineage was a consistent generalist across cells, so generalism depends on the local host pool. Most lineages were peripheral (BC = 0 in 89–95% of records), and centrality mostly tracked the number of hosts infected. Sampling effort explained most variation in richness; after controlling for it, environmental and host-community variables explained little, and no association survived Benjamini–Hochberg correction. Interpretation: data are presence-only (MalAvi does not report uninfected hosts), so they represent detections, not prevalence, and analyses should account for sampling effort. Network metrics apply only to cells with ≥ 5 lineages and ≥ 5 host species; H. (Haemoproteus) had no qualifying cells, and Leucocytozoon results (17 cells) are exploratory. Lineages are cytb barcodes used as species proxies. The data can support macroecological meta-analyses, tests of new predictors (e.g., vectors) and biogeographic comparisons of host–parasite associations.

Files

Steps to reproduce

Data compilation: South American records were compiled from MalAvi (accessed 7 June 2024), the GERPH collection (Universidad Nacional de Colombia), the Animal Ecology Group database (Universidad del Valle, largely from GenBank) and the EcoEvo collection (Universidad de Antioquia), including published and unpublished cytb sequences. Curation: we kept sequences ≥ 300 bp and/or with ≤ 5 synonymies confirmable in MalAvi (otherwise names were assigned by publication priority). Coordinates and host taxonomy were verified against SACC and Birds of the World. Only resident bird species were retained, and records were split by genus and subgenus according to vector group. Spatial framework: records were assigned to a 0.18° (~20 × 20 km) WGS84 grid. Sampling completeness (Number of lineages.R): lineage accumulation curves were computed with vegan::specaccum (rarefaction, 100 permutations), total richness was estimated with Jackknife2 and representativeness as observed/estimated richness, overall and per WWF biome (biomes with < 10 records excluded). Predictors: host richness from BirdLife ranges (letsR); host Sørensen beta-diversity (divraster; retained over PhyloSørensen, computed with 1,000 BirdTree trees, because r > 0.99); WorldClim 2.1 BIO1, BIO4, BIO12 and BIO15 at 10 arc-min (|r| < 0.7 per genus); and the human footprint index. All layers were resampled to the analysis grid. Local networks (Redesxcelda.R; bipartite): for each cell and genus, a lineage × host matrix weighted by infected individuals was built, keeping cells with ≥ 5 lineages and ≥ 5 hosts. d′ and H′₂ were compared with 1,000 Vázquez null matrices (vaznull) and recalculated with binary matrices (Spearman’s ρ). Betweenness centrality (BC) was computed for lineages and hosts, with Z-scores and P values from 199 null networks per cell, and the BC–number-of-hosts correlation was tested with 999 within-cell permutations. Lineages in ≥ 3 cells were classified by mean d′ and its coefficient of variation. Statistical models (GAMs_final.R; mgcv): GAMs fitted by maximum likelihood with shrinkage splines (bs = “ts”, k = 3–5) and standardised predictors. Richness: negative binomial (Poisson for H. (Haemoproteus)) with log(records per cell) as effort covariate. d′ (cell random effect) and H′₂ (log network size covariate): beta regression after Smithson–Verkuilen transformation. BC > 0: binomial, one predictor at a time. A spatial smooth s(lon, lat) was kept if it lowered AIC by ≥ 2; P values were adjusted with Benjamini–Hochberg. Diagnostics: residual Moran’s I, concurvity, k.check, deviance partitioning; sensitivity: quasibinomial d′ models and richness models without effort. Software: R (raster, sp, sf, letsR, divraster, vegan, bipartite, mgcv, dplyr, tidyr, ggplot2, patchwork, ggrepel). Seeds: 2026 (GAMs), 123 (BC). Update file paths and run Number of lineages.R → Redesxcelda.R → GAMs_final.R.

Institutions

Categories

Ecology, Parasitology, Ornithology, Biological Interaction

Funders

Licence