Biogeography and conservation status of the pineapple family (Bromeliaceae)

To provide distribution information and preliminary conservation assessments for all species of the pineapple family (Bromeliaceae), one of the most diverse and ecologically important plant groups of the American tropics—a global biodiversity hotspot. Furthermore, we aim to analyse patterns of diversity, endemism and the conservation status of the Bromeliaceae on the continental level in the light of their evolutionary history.

While recently some progress has been made in providing largescale distribution information of plant species in the Neotropics (e.g. www.biend ata.org; Antonelli, Ariza, et al., 2018), most of the current macroecological and macroevolutionary understanding of the Neotropics at the continental scale is based either on relatively well-studied animal groups, for which standardized distribution information is available ("range maps", www.iucn.org; Guedes et al., 2018;Quintero & Jetz, 2018) or on trees (ter Steege et al., 2013).
Despite the crucial importance of non-tree plants for understanding biodiversity and ecosystem functioning, there are still large gaps in the knowledge of their distributions in the Neotropics (Engemann et al., 2015;Feeley, 2015), and scarce and spatially biased knowledge of their distribution is a major obstacle to understanding macroevolutionary and macroecological processes.
The pineapple family (Bromeliaceae) is one of the most speciesrich and ecologically important plant families of the Neotropics with 3,503 known species (Butcher & Gouda, 2017). Bromeliads are an abundant and diverse element of many habitats, from the evergreen rain forests of Amazonia to the Atacama Desert. The ecological and evolutionary success of the Bromeliaceae is likely related to the repeated evolution of physiological (e.g. CAM photosynthesis) and morphological (e.g. a tank-like growth and trichomes for water and nutrient uptake via the leaves) key innovations (Crayn, Winter, & Smith, 2004;Silvestro, Zizka, & Schulte, 2014). Approximately 1,552 bromeliad species are epiphytes (WCSP, 2017), mostly in wet tropical forests, and they are often important ecosystem engineers, providing habitat for numerous animal species (Benzing, 2008;Givnish et al., 2011;Versieux et al., 2012).
The Bromeliaceae is virtually endemic to the Americas (one species occurs in West Africa). The family likely originated on the Guiana shield and radiated in the last 20 million years with a subsequent dispersal across the Neotropics (Givnish et al., 2011). Currently, no comprehensive species-level phylogeny of the Bromeliaceae exists, but the taxonomy broadly reflects the evolutionary history of the family with eight subfamilies (Brocchinioideae, Lindmanioideae, Tillandsioideae, Hechtioideae, Navioideae, Pitcairnioideae, Puyoideae and Bromelioideae) forming clades of different ages. The in-situ radiation of Bromeliaceae in the Neotropics, its high diversity, and the adaptation to a wide range of environmental conditions make the family a model to understand the evolutionary history of the Neotropics, and have triggered research interest in its morphology, physiology and diversification (Barfuss et al., 2016;Cáceres, Schulte, Schmidt, & Zizka, 2011;Crayn, Winter, Schulte, & Smith, 2015;Givnish et al., 2011Givnish et al., , 2014Males & Griffiths, 2018;Schuetz, Krapp, Wagner, & Weising, 2016;Silvestro et al., 2014). Yet, no upto-date treatment of the biogeography of Bromeliaceae exists, and the geographic distribution of many species is poorly known.
This lack of knowledge is especially problematic, since large parts of the Neotropics are under human land use pressure (Soares-Filho et al., 2013). The on-going habitat loss has raised concern that many plant species in the region are threatened with extinction and many of them might go extinct before they are known to science (Lees & Pimm, 2015;ter Steege et al., 2015;Wearn, Reuman, & Ewers, 2012 We expect continental-scale centres of bromeliad diversity and endemism in three regions (Smith, 1934;Smith & Downs, 1974, 1977, 1979: the Andes, a major centre of diversification for many bromeliad genera (Jabaily & Sytsma, 2013;Wagner et al., 2013), the Atlantic Forest in eastern Brazil where especially the subfamily Bromelioideae radiated (Martinelli et al., 2008) and  (Givnish et al., 2011). The diversity of bromeliads in the Amazonian lowlands is supposed to be comparatively low, but previous estimates might be biased by a lack of sampling in Amazonia.

How do distinct evolutionary lineages within the Bromeliaceae differ in distribution?
The subfamilies of the Bromeliaceae represent evolutionary coherent groups of different ages and differ in species richness as well as morphological and physiological traits. We expect a larger geographic distribution for the more species-rich subfamilies, related to the evolution of tank habit and CAM photosynthesis in the family.

How many species of Bromeliaceae are threatened with extinction?
Based on the high number of local endemics in the Bromeliaceae (Martinelli et al., 2013(Martinelli et al., , 2008Wagner et al., 2013) and results from a regional assessments of the Bromeliaceae of Chile  and Brazil Martinelli et al., 2013), we expect a relative high number of threatened species compared with other plant families.

4.
Where are hotspots of bromeliad conservation? Due to the decrease of tropical forest area we expect the epiphytic species to be especially endangered. In contrast, we expect the species of the Andes and the Guiana highlands to be generally less endangered due to lower human land use pressure.

| ME THODS
We compiled a database of geographic occurrence records for Bromeliaceae from publicly available sources (GBIF.org, 2017, www. idigB io.org, http://splink.cria.org.br, www.tropi cos.org) and own fieldwork and databases (data from BN in the Atlantic Forest, from DC in Panama and Costa Rica and from GZ in Chile). For the public databases, we downloaded data on the family level ("Bromeliaceae") and then resolved names on the species level using an up-to-date taxonomic list (Butcher & Gouda, 2017). For those species where we could not obtain occurrences with this procedure, we used Gouda, Butcher & Gouda (cont. updated) to obtain the locality of the type specimen and georeferenced them manually using Google Earth (https ://www.Google.com/earth/ ), if necessary.
Since occurrence records from public databases are error prone (Maldonado et al., 2015), we removed spatial errors following Zizka cates, respectively (a total of 12 models per species). For two species (Aechmea bracteata and Hohenbergia stellata), the models with 16 times the number of pseudo-absences did not converge and we restricted these species to models with two times and eight times the number of pseudo-absences. We then selected the model with the best TSS value for the projection of species' distribution in space (Liu, Newell, & White, 2019). In cases of equal TSS values, we picked the model using less pseudo-absence points. We fitted the models to the first three principal components of 19 bioclim variables from the CHELSA project (Karger et al., 2017), downscaled to 25 × 25 km, using the "sdm v1.0-41" package (Naimi & Araújo, 2016). We restricted the projections to the same buffered convex hull used for sampling the pseudo-absences. We then converted the projected distributions into presence/absence using a threshold of equal specificity and sensitivity (Liu, Newell, & White, 2016) and converted the raster distributions into range polygons. (b) For species with 15 > n ≥ 10 records after filtering and thinning, we followed the same procedure, except that we used only the first two principal components of the climate data. (c) For species with 10 > n ≥ 3 records after filtering, we used a pseudo-spherical convex hull generated with the "speciesgeocodeR v. 2.0-10" package (Töpel et al., 2016) as a representation of the geographic range; and (d) for species with n < 3, we used a spatial buffer with 50 km radius (the grain of the diversity analyses) to represent the species range.
We overlaid the estimated ranges to visualize species richness patterns for the Bromeliaceae and its subfamilies using the get_range and map_richness function of the novel "bromeliad" package, based on a 100 × 100 km grid. We then used the same grid to estimate the weighted endemism (Crisp & Laffan, 2001) as implemented in r (Guerin, Ruokolainen, & Lowe, 2015) and identified areas of high and low endemism using a significance test based on deviance from the expected endemism, given the observed species richness (Guerin et al., 2015).
To address question 2-distribution of evolutionary distinct lineages-we visualized the genus richness and the distribution of the eight subfamilies within the Bromeliaceae. Taxonomic ranks are an imperfect approximation for evolutionary history, but since a species-level phylogenetic tree for the family is missing and the subfamilies likely represent evolutionary clades (Givnish et al., 2011), we used them as proxy for evolutionary history.

To address question 3-number of threatened species in
Bromeliaceae-we used our database of occurrence records to generate automated conservation assessments (AA) using the "ConR v 1.2.2" package in r (Dauby et al., 2017). ConR calculates the extent of occurrence (EOO), the area of occupancy (AOO) and the number of locations (the latter following a slightly different approach than suggested by the IUCN) for each species based on occurrence records and uses this data to assign each species a threat status follow- To address question 4-distribution of threatened Bromeliaceae species-we first visualized the distribution of all Possibly Threatened species in a 100 × 100 km grid. Furthermore, we classified each species into 12 major biomes (Olson et al., 2001) to identify the number and fraction of possibly threatened species in each biome. We classified species as present in a biome if at least 5% of its occurrence records were in this biome, since this threshold replicated independent distribution data best . Tillandsia recurvata (L.) L. with 2,433 records, the median number of records per species was 3. A total of 370 species had more than 14 records, 212 species had between 9 and 15 records (hence distributions for 582 species were estimated using niche models); 1,061 species had between 4 and 9 records; and 1,629 species had less than three records. We could not obtain occurrence records for 231 accepted names. Figure S1.1 in Appendix S1 shows the density of occurrence records across the study area.  3 and 4). Members of the three species-rich subfamilies occurred across the entire range of the family but differed in their diversity centres. The Bromelioideae was most species-rich in eastern Brazil, whereas the Pitcairnioideae and Tillansioideae were most diverse in the northern Andes (Figures 3 and 4).

| Conservation assessment
The automated conservation assessment (AA) identified 2,638 species (81% of the evaluated species) as Possibly Threatened (

| D ISCUSS I ON
Here, we provide modelled distribution ranges for 3,272 species The results presented here are the first comprehensive treatment of the Bromeliaceae biogeography since Smith and Downs (1974), Smith and Downs (1977) and Smith and Downs (1979), who gave distribution maps for subfamilies and genera in their old circumscription for the then much smaller number of bromeliad species based on a much smaller dataset. Other previous biogeographic studies in the family had limited taxon and locality sampling (Benzing, 2000;Canela, Paz, & Wendt, 2003;Givnish et al., 2011Givnish et al., , 2014Males & Griffiths, 2018;Smith, 1934;Smith & Downs, 1974, 1977, 1979 and focused on individual taxa (Canela et al., 2003;Leme, Heller, Zizka, & Halbritter, 2017;Peters, 2009;Zizka, Horres, Nelson, & Weising, 1999;Zizka, Trumpler, & Zöllner, 2002) or geographic regions (Cáceres, 2012;Judith et al., 2013;Zizka et al., 2009; www.flora dobra sil.jbrj.gov.br). Our distribution maps are available in Appendix S2, and we supply all species ranges under a CC-BY license via the bromeliad r package, which also includes functions for publication-level species richness maps for individual genera, traits or conservation categories (Appendix S4).
We compiled our dataset of geographic occurrence records from publicly available sources subjected to automatic cleaning and manually curated datasets. The dataset is not complete, and we included data based on a compromise between data precision and data availability. To overcome the generally scarce and biased sampling in the Neotropics, we combined multiple range modelling algorithms. While each of these algorithms as well as their combination has limitations, we are confident that our range maps are an adequate representation of Bromeliaceae distribution given the grain of our analyses and enable a comprehensive assessment of the Bromeliaceae biogeography. The openly accessible distribution ranges will serve as a resource to the bromeliad research community and will hopefully enable future studies to relate species distribution to physiological and morphological adaptations in a more detailed manner.

| Diversity and endemism
The major diversity hotspots we identified-the Atlantic Forest, the northern Andes and Central America (including Southern Mexico)confirm the centres of diversity identified in previous studies (e.g. Smith, 1934). Novel and noteworthy are two species-rich "corridors", the first extending from the northern Andes over north-western Amazonia (The Napo and Imeri province sensu Morrone, 2014) and    Figure   S1.4). The Puyoideae, with the single genus Puya, is ecologically well characterized by its Andean distribution, and only few species in this group extend to the Guiana Highlands (P. floccosa, P. grafii, P. harrylutheri, P. sanctaecrucis) or Northern and Central Chile west of the Andes (P. chilensis, P. alpestris, P. boliviensis, P. coerulea, P. gilmartiniae, P. venusta) (Jabaily & Sytsma, 2010, 2013Zizka et al., 2009).
The distribution of individual genera within the Bromeliaceae reflects changes in environmental niche, related to the evolution of key physiological and morphological traits (Males & Griffiths, 2018). For instance, several of the early branching lineages within the subfamily Bromelioideae exclusively comprise species doing C3 photosynthesis (e.g. Greigia, Ochagavia, Fascicularia and Fernseea; Crayn et al., 2015), which fits with their distribution in cold to temperate and moist, mostly Andean areas. Contrary to our expectations, the Pitcairnioideae, which are rarely epiphytes (Zotz, 2013) and generally lack key innovations such as tank habit and highly ab-

| Conservation status
Our automated conservation assessment (AA) provides information for 3,032 species for which no full assessment was available from IUCN before. In contradiction to our expectation, the proportion of endangered species was higher in terrestrial and lithophytic species than in epiphytes, suggesting that the conservation of habitats where terrestrial bromeliads are diverse might especially benefit bromeliad conservation (See Figure S1.5 for species richness patterns of Neotropics (question 2). We provide distribution maps and shapefiles of species ranges via the "bromeliad" r package upon publication of this study.
We found 81% of the evaluated species as Possibly Threatened with extinction (question 3) in many cases in agreement with independent reference assessments. The proportion of Possibly Threatened species was particularly high for terrestrial species and in the subfamilies Lindmanioideae, Navioideae and Puyoideae. Most Possibly Threatened species occurred in the Atlantic Forest and the Central Andes, especially in Tropical rain forests (question 4). This high number is worrying, especially because of the ecological keystone role of many bromeliad species. The automated assessment presented here can act as a data-driven baseline to direct more detailed conservation assessment, which might include data on population dynamics and specific threats.

ACK N OWLED G EM ENTS
We thank Michelin Middeke, Frank Lappe and Tanja Jungcurt for help with finding and georeferencing specimen locations. We thank Derek Butcher and Eric Gouda for continuously updating the bromeliad taxon list and thereby providing an invaluable tool for biogeographic analysis. AZ is thankful for funding by iDiv via the German and https ://github.com/idiv-biodi versi ty/brome liad; see Appendix S4 for a tutorial on how to use the package and extract data available upon publication).