The drivers of avian‐haemosporidian prevalence in tropical lowland forests of New Guinea in three dimensions

Abstract Haemosporidians are among the most common parasites of birds and often negatively impact host fitness. A multitude of biotic and abiotic factors influence these associations, but the magnitude of these factors can differ by spatial scales (i.e., local, regional and global). Consequently, to better understand global and regional drivers of avian‐haemosporidian associations, it is key to investigate these associations at smaller (local) spatial scales. Thus, here, we explore the effect of abiotic variables (e.g., temperature, forest structure, and anthropogenic disturbances) on haemosporidian prevalence and host–parasite networks on a horizontal spatial scale, comparing four fragmented forests and five localities within a continuous forest in Papua New Guinea. Additionally, we investigate if prevalence and host–parasite networks differ between the canopy and the understory (vertical stratification) in one forest patch. We found that the majority of Haemosporidian infections were caused by the genus Haemoproteus and that avian‐haemosporidian networks were more specialized in continuous forests. At the community level, only forest greenness was negatively associated with Haemoproteus infections, while the effects of abiotic variables on parasite prevalence differed between bird species. Haemoproteus prevalence levels were significantly higher in the canopy, and an opposite trend was observed for Plasmodium. This implies that birds experience distinct parasite pressures depending on the stratum they inhabit, likely driven by vector community differences. These three‐dimensional spatial analyses of avian‐haemosporidians at horizontal and vertical scales suggest that the effect of abiotic variables on haemosporidian infections are species specific, so that factors influencing community‐level infections are primarily driven by host community composition.


| INTRODUC TI ON
Parasites are ubiquitous, diverse, and play major ecological roles in terrestrial and aquatic ecosystems (García del Río et al., 2020;Poulin, 1999), where they are a prominent selective force that influences fitness, distribution, and evolution of hosts (Poulin, 1998).
Haemosporidians (Phylum Apicomplexa) are blood parasites transmitted by dipteran vectors and are among the most common parasites in vertebrates (Soares et al., 2017), including in birds Valkiūnas, 2005). Infections in birds generally impact host fitness negatively (Atkinson, 2009;LaPointe et al., 2012;Rivero & Gandon, 2018) and the introduction of haemosporidians to naïve bird communities (e.g., on previously isolated islands) can have dramatic consequences and even lead to population collapses or species extinctions (Ewen et al., 2012;Freed et al., 2005).
A first step towards understanding what determines local, regional, and ultimately global avian-haemosporidian prevalence patterns is to decipher the factors that govern prevalence in individual host species at local spatial scales. We address this by investigating haemosporidian prevalence, host specificity, and host-parasite networks of lowland bird species in multiple forest localities in close geographic proximity in Papua New Guinea. We sampled 10 abundant bird species along an east-west axis (spanning ~70 km) in 4 fragmented forest patches (~48 km apart) as well as 5 localities within a single continuous primary forest (~14 km apart) (Figure 1). Additionally, in one locality within the continuous primary forest (Figure 1) (Table S1).
To explore the effect of vertical stratification on avianhaemosporidian associations, we captured birds from the forest floor to 27 m above ground in 2013 at the Swire Station locality within WCA (Figure 1) using stacked mist nets (for details see Chmel et al., 2016 Figure   S1, Table S2).
Body mass and tarsus length was measured for all individuals sampled, and 10-20 µl of blood was obtained from the brachial artery and stored in 70% ethanol until DNA extractions.
To test for sex-specific differences, we sexed individuals using PCRs with the primers 2550F and 2718R for Passeriformes and Columbiformes, and p2 and p8 for Coraciiformes (Fridolfsson & Ellegren, 1999). Heterogametic females and homogametic males were distinguished through visualization of PCR products on a 2% agarose gel.

| Molecular identification of haemosporidians
DNA was extracted using the Qiagen DNeasy ® blood and tissue kit (Hilden, Germany), following the manufacturer's guidelines, with a prolonged incubation period (approximately 12 h at 56°C).
Haemosporidians were identified through nested PCRs with slight modifications to a well-established protocol (Bensch et al., 2000;Hellgren et al., 2004). The initial PCRs were conducted in triplicates using HaemNF1 (5′-CATATATTAAGAGAAITATGGAG-3′) and HaemNR3 (5′-ATAGAAAGATAAGAAATACCATTC-3′) primers and the PCR master mix contained a total volume of 25 μl per sample (12.5 μl of VWR RedTaq polymerase ® , 1 μl of 10 mM concentration of each primer, 8.5 μl of autoclaved MilliQ water, and 2 μl of the DNA template). These PCRs were conducted under an initial step of 3 min at 94°C and 20 cycles of 30 s at 94°C, 30 s at 50°C, 45 s at 72°C, and 10 min at 72°C. We then proceeded with the second PCRs targeting specific haemosporidian genera (Haemoproteus and Plasmodium), using HaemR2 (5′-GCATTATCTGGATGTGATAATGGT-3′) and HaemF (5′-ATGGTGCTTTCGATATGCATG-3′) primers (Hellgren et al., 2004). We did not investigate Leucocytozoon parasites due to their low prevalence in New Guinea (Bodawatta et al., 2020). The second PCR was set up using 10 μl of Qiagen multiplex master mix (Hilden, Germany), 1 μl of 10 mM concentration of each primer, and 8 μl of 10× diluted product from the first PCR. The second PCR was conducted with an initial step of 3 min at 94°C and 35 cycles of 30 s at 94°C, 30 s at 50°C, 45 s at 72°C, and 10 min at 72°C. Every PCR round contained a positive control and a negative control for every 16 samples. Final PCR products were visualized on a 2% agarose gel containing GelGreen ® stain at 90 V for approximately 1 h.

| Host-parasite networks, lineage specificity, and host phylogeny
To explore host-parasite network structures in different sampling localities we calculated the network-level specificity index (H 2 ′) for bird-haemosporidian communities using the R package bipartite v2.15 (Dormann et al., 2008). An H 2 ′ index close to 1 indicates specialized host-parasite communities with more one-to-one interactions between host species and parasite lineages, while indices closer to 0 indicate more generalized networks (Blüthgen et al., 2006). We then compared the observed network specificity values with specificities expected by chance through generating 1,000 random networks, to investigate whether observed values deviate significantly from network specificities expected by chance. We also investigated haemosporidian lineage-level specificity on the most common lineages (infecting >2 individuals) between the continuous forest and the fragmented forest patches (combining all localities within each category). We used the threshold of >2 individuals as the majority of our haemosporidian lineages only infected one bird species. We calculated specificity for each lineage using Rao's quadratic entropy, while incorporating phylogenetic distances among host species (accounting for the importance of host evolutionary histories on haemosporidian specificity levels) using the raoD function in the R package picante v1.8.2 (Kembel et al., 2010). Higher Rao values indicate more generalists while lower values indicate more specialist lineages (Ellis et al., 2020).

| Environmental data
Environmental variables for individual sampling localities (e.g., maximum and minimum temperature, elevation, and distance to large water bodies [rivers and the sea]) were gathered from online databases (see below). We used the distance to rivers as a proxy for habitat availability for vectors, but we do acknowledge that this parameter is suboptimal to fully understand the habitat availability for vectors, as vectors can breed in small water pools, such as water retained in tree stumps and bromeliads. Nevertheless, this index still provides an indication of water availability in the area.
We used NDVI to estimate forest greenness as a proxy for forest structure. NDVI has been used extensively to evaluate forest structure (Grace & Gates, 1982), yet we acknowledge the inherent limitations (e.g., not capturing the changes in forest interior) of this measure. Nonetheless, NDVI provides a normalized value for forest greenness that is comparable across study sites and even between studies. NDVI was calculated using the following equation where NIR is the near-infrared and RED the visible band (Myneni et al., 1995). It measures the degree of absorption by chlorophyll in red wavelengths (Myneni et al., 1995), the index values fall between −1 and 1, with values around −1 representing clouds and water, values around 0 representing bare soil, and values close to 1 representing forested areas with maximum greenness (i.e., forest cover) (Atoyan et al., 2018). For environmental variables that had a low resolution for the exact GPS coordinate, we used the value of the adjacent pixel (<800 m from the original point) to that locality. We used the NNJoin Plugin v3.1.3 (Tveite, 2019), to calculate nearest neighbor relationships (Eppstein et al., 1997) from each locality to rivers, roads, and the sea (Tables S1 and S4; Figures S2 and S3).
Furthermore, because M. nigra and P. kirhocephalus had high parasite prevalence in all localities, with little to no variation (99%-100%), they were excluded from the linear models (Table 1).
First, we examined the collinearity of abiotic variables using Pearson's correlation tests with the function ggpairs from the R package GGally v2.0.0 (Schloerke et al., 2019), and found that multiple variables that were significantly correlated with each other ( Figure   S4). Thus, for the final analyses, we only included variables that were not collinear (NDVI, Minimum temperature, Distance to roads).
Although, NDVI was positively, yet nonsignificantly, correlated with the vegetation type (Pearson correlation: r = .6270, p = .1001), we chose to include NDVI rather than vegetation type due to NDVI being more accurate. We further checked spatial autocorrelation of environmental variables considering latitude and longitude of the sampling localities using Monte Carlo tests with the function mantel.rtest from the R package ade4 v1.7-18 (Thioulouse et al., 2018) and package to account for host phylogenetic relationship (Li et al., 2020). Here we included host species, site, sex, NDVI, distance to the roads, and minimum temperature as the independent variables and the distance to sea as a random effect to control for the spatial arrangement of the sampling sites (our sampling sites are located in an east-west spatial scale from the sea: Figure 1). Following the guidelines in Crawley (2013) Figure S5) and not between locations (binomial GLM: LR χ 2 = 12.18, df = 8, p = .1431; Figure S6) suggesting that some bird species are more susceptible to infections than others.
However, prevalence did not differ between the sexes (binomial GLM: LR χ 2 = 2.032, df = 3, p = .5658). The strong host species effect further supported conducting statistical analyses on both host community and species levels.

| More specialized host-parasite networks in localities within continuous forest
Host-parasite network structure was more specialized than expected by chance throughout continuous forest localities (H 2 ′ = 0.7645, null mean 1,000 random iterations : 0.6063, p < .0001;

| Species-specific effects of abiotic factors on Haemoproteus prevalence between localities
Despite the significant effect of locality on Haemoproteus prevalence, there was no significant difference between continuous and fragmented forests (binomial GLM: LR χ 2 = 0.0646, df = 1, p = .7993; Figure S9). We found 27 unique haemosporidian haplotypes in the forest fragments and 26 in the continuous forest ( Figure S10). At the bird community level, NDVI was the only significant predictor of Haemoproteus prevalence, which decreased with increasing NDVI (binomial PGLMM: Std.error = 3.381, Zscore = −2.480, p = .0131, Figure S11). However, species-level analyses (  Figure 3).
Interestingly, there were more specialist lineages (when considering lineages in >two individuals) in the canopy than the understory ( Figure S13C), suggesting that the high prevalence observed in the canopy may be a result of the presence of more specialized lineages ( Figure 4).

| DISCUSS ION
We investigated the influence of environmental and anthropogenic factors on avian-haemosporidian (Haemoproteus and Plasmodium) parasite prevalence, distribution, specificity, and host-parasite network structures in tropical lowland birds at horizontal and vertical spatial scales. Haemoproteus was the most common parasite genus, aligning with previous work in Papua New Guinea (Bodawatta et al., F I G U R E 2 Bird-haemosporidian networks and host specificity in different forest categories (continuous vs. fragmented). Networks indicate combined host-parasite communities of the five localities within Wanang Conservation Area (a) and four fragmented forest patches (b). An H 2 ′ index closer to 1 indicates that host-parasite communities are more specialized (many one-to-one associations), while highly generalized networks have H 2 ′ indices closer to 0. Observed H 2 ′ and the average H 2 ′ acquired from 1,000 null models are given within parentheses. Asterisks indicate significantly different observed H 2 ′ values compared to the null expectation. Haemoproteus lineage names are given in gray and Plasmodium lineage names are in black. c. Association between host specificity of the most abundant haemosporidian lineages (Rao's quadratic entropy) and the lineage abundance in the continuous forest and fragmented forests. Parasite lineages found in both forest categories are given in gray. Large Rao values indicate generalist lineages while small values indicate specialist lineages. Host specificity was only calculated for parasite lineages that infected more than two bird individuals 2020), who sampled at a site less than 25 km from our study sites.
However, Haemoproteus prevalence was overall markedly lower in the previous study (~15%). The lowland study site sampled by Bodawatta et al. (2020) represents the lowest part of the Mount Wilhelm elevational gradient and includes the total bird community, while our localities are part of an extensive lowland area and only include 10 abundant bird species. Topographical differences and sampled avian communities of the two localities may thus at least in part explain the observed prevalence-level differences. Plasmodium prevalences were low (5%) in the longitudinal study, consistent with findings by Bodawatta et al. (2020). This could be explained by hosts being less susceptible to this genus (Lima & Pérez-Tris, 2020) or geographic variation in the distribution and density of Plasmodium vectors (Ferreira et al., 2020).
At the bird community level, NDVI (greenness) was the only variable that significantly influenced Haemoproteus prevalence.
However, at the bird species level, the picture is less clear, with species-specific effects of minimum temperature, distance to roads, and NDVI. Furthermore, we found vertical segregation in host-parasite interactions with higher prevalence in the canopy than the understory. Haemoproteus accounted for the majority of infections in both strata, but with a higher relative proportion in the canopy. Collectively, this not only suggests that specific vector communities may influence the transmission of particular malarial lineages but also that adaptation to particular ecological niches of a host species makes them differentially susceptible to pathogens.

| More specialized host-parasite networks in undisturbed forests
We did not find significant differences in Haemoproteus prevalence between the localities within the continuous forest and the fragmented forests, which aligns with results from regional spatial scale studies in the Neotropics and the Afrotropics (Belo et al., 2011;Chasar et al., 2009;Loiseau et al., 2010;Rivero de Aguilar et al., 2018;Sebaio et al., 2012). However, our findings contrast a study from tropical Australia which found higher Haemoproteus prevalence in continuous than fragmented forest (Laurance et al., 2013). Higher prevalence levels in continuous forest have been speculated to be a result of higher vector abundances (Mangudo et al., 2017;Zhou et al., 2007). Thus, forest structure could indirectly affect parasite infection risk through influencing the vector abundances (Mangudo et al., 2017;Zhou et al., 2007), implying that vector sampling across forest types is needed to decipher their potential effects on prevalence levels between continuous and fragmented forests.
While prevalences did not differ between the continuous and fragmented forests, host-parasite network structures were notably different, where continuous forests harbored significantly more specialized networks than fragmented forests. The greater specialization in continuous forests could imply that undisturbed forests may provide more stable environments with higher host species richness (Bregman et al., 2014;Sam et al., 2014;Van Hoesel et al., 2020) that could lead to more specialized associations. Highly specialized avian-haemosporidian networks has been observed before in an undisturbed tropical lowland bird community in Ecuador (Svensson-Coelho et al., 2014). Fragmented forests, on the other hand, tend to favor generalist parasite lineages (driving observed random host-parasite network structures), which is evident by the observed association between host specificity of lineages and their abundances in the fragmented but not in the continuous forest.
This aligns with the niche-breadth hypothesis (Ellis et al., 2020;Pinheiro et al., 2016), predicting that generalist parasite lineages with broader host niches perform better in small forest patches than specialist lineages. The differences in lineage specificity and abundances in fragmented versus continuous forests may thus result from (i) changes in bird communities (abundances and densities) (Bodawatta et al., 2020;Fecchio, Wells, et al., 2019), (ii) changes in the potential for competition between haemosporidian lineages (Bodawatta et al., 2020), and/or (iii) changes in environmental variables associated with forest fragmentation (Afrane et al., 2006).

| Haemosporidian prevalence levels depict species-specific responses to environmental and anthropogenic factors
Of the environmental variables, only increased NDVI (greenness) led to significantly reduced Haemoproteus prevalence, suggesting that minimum temperature and distance to roads do not significantly affect community-level haemosporidian prevalence at local spatial scales. NDVI appears to be a good predictor for vector abundance and distribution (Roiz et al., 2015) and has been shown to be-in contrast to our findings-positively associated with Haemoproteus prevalence in seasonal temperate regions . However, our tropical localities had very high (0.8-1.0) NDVI with minor differences between sites, compared to studies in temperate regions (Fecchio et al., 2020;Ferraguti et al., 2018;Roiz et al., 2015). The species-specific effect of different environmental variables on Haemoproteus prevalence ( Figure 2) aligns with results from other studies on bird species from both temperate and tropical regions (Isaksson et al., 2013;Samuel et al., 2015;Santiago-Alarcon et al., 2019;Van Hoesel et al., 2020). Despite the overall nonsignificant effect of distance to roads (a proxy for anthropogenic influence), it did significantly affect prevalence in three bird species, suggesting F I G U R E 3 Associations of Haemoproteus prevalence with minimum temperature (°C), distance to roads (m) and NDVI across all sampling sites for all species sampled in 2015. Pitohui kirhocephalus and Melanocharis nigra were not included in these analyses as all individuals were infected. Dots represent infected (100%) or noninfected (0%) individuals and lines ( (Chasar et al., 2009;Gonzalez-Quevedo et al., 2014), or no (Sehgal, 2015) effects. This is consistent with our findings, as the magnitude of anthropogenic activity effects on prevalence differs from bird species to species.
Temperature tends to positively impact haemosporidian prevalence at a regional scale in both tropical and temperate regions (Chapa-Vargas et al., 2020;Padilla et al., 2017;Sehgal, 2015;Zamora-Vilchis et al., 2012). However, we found that increased minimum temperature (even minor differences, ~1.5°C) had a negative effect on Haemoproteus prevalence across multiple species (albeit only significantly for A. insularis) (Figure 2). Consensus on the effect of temperature on parasite prevalence in birds in the Australo-Papuan region is lacking, as studies have shown positive (Zamora-Vilchis et al., 2012) or no (Bodawatta et al., 2020) effects. Areas with lower temperatures experience more rainfall in our study region, indicating potentially more vector breeding habitats (Lapointe et al., 2012;Sehgal, 2015) that could lead to higher vector abundances and increased prevalence. In summary, our findings imply that the sum of species-specific responses to different environmental variables dictate community-level effects of abiotic factors in tropical bird communities.

| Higher prevalence and reduced diversity of haemosporidians in the canopy
Higher Haemoproteus prevalence in the canopy than the understory aligns with previous findings from the Afrotropics (Lutz et al., 2015).
However, Plasmodium prevalence was higher in the understory than canopy, which may reflect higher mosquito abundances (Plasmodium vectors) at the forest floor. Our finding suggests that the pattern might be opposite for biting midge vectors of Haemoproteus that are conceivably higher in the canopy as they tend to prefer these sites to ground strata (Černý et al., 2011;Garvin & Greiner, 2003;Swanson & Adler, 2010;Swanson et al., 2012). Only 5 of the 32 haemosporidian lineages were shared between the strata, likely due to vertical segregation of vector species (Henry & Adkins, 1975), implying that investigations of canopy and understory bird communities in a locality is needed to fully capture host-vector-parasite diversity and associations.

| CON CLUS IONS
Our results demonstrate that interactions between haemosporidian parasites and tropical avian hosts are influenced by a multitude of factors at different taxonomic levels and spatial scales.
Forest structure influences associations between particular host species and parasite lineages, while parasite prevalence of a set of host species (the community) is driven by a combination of species-specific environmental effects. Vertical separation within a single locality appears to expose avian hosts to markedly different parasite pressures, which is likely driven by vector communities. Taken together, these results emphasize the importance of investigating avian-haemosporidian associations in space, for both individual host species and at the host community level. Finally, the species-specific effects of environmental variables and vertical stratification on parasite prevalence accentuate that the factors driving these interactions can differ between global, regional, and local spatial scales.

ACK N OWLED G M ENTS
We thank the Carlsberg Foundation for a Distinguished Associate Research Centre for enabling us to conduct field work and Nick Bos for providing advice on statistics.

CO N FLI C T O F I NTE R E S T
All the authors declare there is no competing interest related to the material of this manuscript.

O PEN R E S E A RCH BA D G E S
This article has been awarded Open Data, Open Materials Badges.
All materials and data are publicly accessible via the Open Science Framework at https://doi.org/10.5281/zenodo.5776763.