Co-occurrence or dependence? Using spatial analyses to explore the interaction between palms and triatomines (Chagas disease insect vectors)

Background Triatomine kissing bugs are responsible for the vectorial transmission of the parasite Trypanosoma cruzi, etiological agent of Chagas disease, a zoonosis affecting 10 million people and with 25 million at risk of infection. Triatomines are associated with particular habitats that offer shelter and food. Several triatomine species of the Rhodnius genus have close association with palm crowns, where bugs can obtain blood from the associated fauna. The Rhodnius - palm interaction has been reported in several places of Central and South America. However, the association in the distributions of Rhodnius species and palms has not been quantitatively determined. Methodology/Principal Findings Broad distributions of eight Rhodnius species and 16 palm species with Rhodnius-infestation reports were estimated using Ecological Niche Models. Rhodnius species distributions in their total range were compared to their distributions in areas with palms. Rhodnius species presence was found to be higher in areas with palms. However, that tendency notoriously depended on palm species. Rhodnius species presence increased several times in areas with particular palm species. Moreover, a possible relationship was found between Rhodnius and palm species richness, indicating the Amazon region as the convergent region where several Rhodnius and palm species intersected. Finally, palm distribution was evaluated as predictor of Rhodnius species distributions, but their inclusion in the distributions models did not improve their performance. Conclusions/Significance The distributions of some Rhodnius and palm species showed a high spatial association, which can be based on species interaction or niche similarity. Based on distribution convergence, the Amazon region appear to be the origin of the Rhodnius-palm association. The direct relationship between palms and Rhodnius species richness could be based on the habitat heterogeneity offered by different palm species. Despite spatial association, palm presence would not be a relevant predictor of Rhodnius species distributions in comparison to other environmental variables. Inclusion of other input data as hosts’ distribution could help to increase model predictability. Author summary The infestation of palms with Rhodnius genus kissing bugs (Chagas disease vectors) is important from the public health perspective, since insects living in palms can infest nearby houses. The migration of these bugs to households could threaten vector control programs since reinfestation of treated dwellings can occur. Association between Rhodnius and palms species distributions has been previously suggested but never quantitatively determined. The strong association between one palm species and one Rhodnius species can be used as a factor to predict the presence of Rhodnius bugs in definite areas. In this study, we estimated by models the distributions of eight Rhodnius species and 18 Rhodnius-infested palm species. Rhodnius distributions models showed a biased presence toward areas with certain palm species. That specific association was very strong in some cases; however, the presence of associated palm species was used in Rhodnius distributions models, but that did not improve the predictability of the models. Palm presence appear to be not essential for the Rhodnius current distribution because they could inhabit other habitats; but that association could be relevant to the Rhodnius evolutionary and biogeographic history.


Introduction
in previous reported areas as a possible consequence of vector control initiatives [74], and 178 the species presence has not been associated with palm trees [14,17,75]. 179 To reduce the effect of sampling bias in the occurrence data set, spatial thinning was 180 performed with the "spThin" R package [76] using a minimum nearest neighbor distance 181 greater than or equal to 10 km.   Most of the Rhodnius models predicted an area of distribution adjusted to the 298 occurrence points (Fig 1). However, R. prolixus, R. pallescens, and R. colombiensis, models 299 showed over-prediction areas outside occurrences (Fig 1) Considering performance, all the palm models showed pAUC ratios significantly 320 higher than the null model line except in A. speciosa (Table 3). Like Rhodnius models, 10 321 percentile omission rates were higher than the expected value, but the zero percentile 322 omission rate were very close to the expected one. Three species showed very high omission 323 rates values with both thresholds: P. aequatorialis, Sa. mauritiiformis and Sy. oleracea.

324
Attalea speciosa, showed a very high 10 percentile omission rate but low zero percentile.

325
Every palm model predicted an area of distribution adjusted to the occurrence points (Fig 2), species. In contrast, topographic variables showed low influence in the models.

331
As an alternative to decrease the ten-percentile omission rates, ENMs were repeated 332 using as layers, the first 16 PCAs obtained from the original 42 variables (which covered 333 90% of the environmental variation). However, omission rates did not improve (S1 Table) 334 and the initial ENM were used for the further analysis.

336
Association between Rhodnius species and palms distributions.

337
Rhodnius species prevalence was higher in areas with palm presence compared to the 338 entire area, except for two species, R. prolixus and R. colombiensis (Table 4). However, 339 differences between prevalence values were small; all the odds ratios were close to 1. Palm 340 prevalence (presence of at least one palm species) was very high in all the Rhodnius species 341 distribution areas (Fig 3). In some cases, as in R. robustus and R. pallescens, presence of palm trees covered almost the entire area (  To compare if Rhodnius-palm spatial association could be explained by niche similarity, 380 niche overlap was compared between Rhodnius-palm pairs with and without spatial 381 association (S2 Table). To this, n-dimensional hypervolumes overlapping was calculated by 382 the function "dynRB_VPa" in the "dynRB" R package [86]. In all the Rhodnius species but 383 R. colombiensis, mean niche overlap was higher in Rhodnius-palm pairs with spatial 384 association than in pairs without the association. ). More than 60% of the area predicted for Rhodnius (i.e. area with at least one Rhodnius 389 species), was predicted to be occupied by two or more Rhodnius species (Fig 4 up). In the 390 limits of this region, only one Rhodnius species is predicted as present. The Amazon region was also the area with the highest predicted richness of palm species (species considered in 392 this study), and 87% of the area predicted as present for palms (i.e. area with at least one 393 palm species) is predicted to be occupied by two or more palm species. Almost all the 394 considered area, from Guatemala to northern Argentina, had a continuous presence of palms 395 (species with infestation reports).   Rhodnius models with palm trees distributions as predictors. 413 When Rhodnius models were run with palm distributions as predictors, performance 414 behavior was similar to the previous models. Partial AUC ratios were significantly higher than the null model line except for R. neglectus and R. colombiensis (Table 2B); 10 percentile 416 omission rates were higher than expected (and sometimes much higher), but zero percentile 417 omission rates were closer to the expected values (Table 2B). The three species with the 418 lowest occurrence number (R. nasutus, R. colombiensis and R. ecuadoriensis) had both 419 omission rates very far from the expected values. As predictor, palm distributions showed to 420 be not very relevant for Rhodnius models (Table 2B). Palm importance was low in all 421 Rhodnius species but R. pallescens and R. colombiensis. In those species, however, the 422 models did not show any increase in performance using palm distributions. Spatial 423 differences in the predictions of models with and without palm distributions were scarce and 424 disperse, and they are mainly located in the edges of the presence areas (Fig. 6).  Considering the association between Rhodnius species and palms, the prevalence of 435 Rhodnius species was not much higher in palm tree areas than in the total modeled areas.

436
That could be a consequence of the palm presence area, which was very big and, in some 437 cases, encompassed all the extension. If palm area and total modeled area are similar,

Rhodnius prevalence in both areas cannot be very different. That is demonstrated in the odd
439 ratios values that were all very close to one. Therefore, based on Rhodnius prevalence 440 comparison, the association between Rhodnius species and palm trees presence cannot be 441 clearly determined.

442
In contrast, when palm species were considered, prevalence comparisons showed 443 greater differences. Each Rhodnius species' prevalence increased in specific palm areas 444 compared to the entire area. Comparisons were several times higher in some cases (Table 5).

445
That showed a clear spatial association between the presence of Rhodnius species and certain 446 palm species. Rhodnius prevalence difference can be enormous between palm species. For 447 instance, R. robustus presence was 150 times higher in As. aculeatum areas than in Ac. 448 aculeata areas, and R. prolixus presence was 17 times higher in A. aculeata areas than in O. 449 bataua areas. Hence, the palm species appears to be key for Rhodnius-palm association.

488
Despite infesting several palm species (Table 1) Venezuela and Guiana could be considered as overprediction, since in that region, there are 496 no occurrences nor predicted palm presence (Fig 1). However, those highlands are similar to 497 the Andean zones where R. prolixus has been intensively reported.

498
Like R. prolixus, R. pallescens was also reported in several palm species (Table 1) 499 but it only showed high association with E. oleifera. This palm has a broad distribution very 500 similar to that of R. pallescens [19,36]. No association was found with A. butyracea even  Rhodnius nasutus was also reported in several palm species (Table 1), but spatial 506 association was found only with Cp. prunifera and At. speciosa (the latter with low odd ratio 507 but close to 2). Copernicia prunifera was the most distributed palm species inside R. nasutus 508 area, and their association have been frequently reported [5,49,[91][92][93][94]. Considering the other palms with infestation reports, Ac. aculeata and M. flexuosa showed no association and Sy. 510 oleracea prevalence was very small as to be considered in the analysis (lower than 0.10).

511
In contrast to the previous Rhodnius species, all the palm species associated with R.  In R. ecuadoriensis, a high spatial association was seen with P. aequatorialis; this 516 relationship has been deeply studied [47,48]. In the north of R. ecuadoriensis distribution, 517 presence is related to palm trees presence (Fig. 3); while in the south, presence is related to 518 domiciliation process with no palm trees [29,95]. In R. colombiensis, clear association was 519 found with As. aculeatum and not with At. butyracea, the only species with infestation 520 reports. Nevertheless, R. colombiensis distribution appears to be underestimated by the 521 models, which produced very high omission rates (Table 2A).   Rhodnius genus distribution (Fig 4), while several species occur in the Amazon region [14].

531
High Rhodnius species richness shown in the Amazon appears to be related with high 532 palm richness. Palms are considered to be suitable habitats for Rhodnius since they offer food and shelter. Habitat quality offered by palms could be heterogeneous among palm species 534 [6], and ecological heterogeneity has been proposed as a driver of species richness for several 535 reasons [97]: 1) Different habitat types could increase the available niche space and allow 536 more species to coexist; 2) There would be more diversity in shelter and refuges from adverse

555
As distribution models, ENM are severely dependable on available information. The 556 low occurrence number in some species and the biased distribution of information (e.g. some areas intensively sampled in comparison to others) could limit the validity of the conclusions.

558
The presence of Rhodnius triatomines in more palm species than those considered in this 559 study cannot be excluded, and the conclusions here are limited to one small subgroup of palm 560 species inside the huge diversity of palm trees found in the Rhodnius presence area [19].