Drivers of bromeliad leaf and floral bract variation across a latitudinal gradient in the Atlantic Forest

Abstract Aim Understanding the complex interaction and relative contributions of factors involved in species and trait diversification is crucial to gain insights into the evolution of Neotropical biodiversity. Here, we investigated the drivers of morphological variation in bromeliads along a latitudinal gradient in a biodiversity hotspot. Location Atlantic Forest, Brazil. Taxon A species complex in the genus Vriesea (Bromeliaceae). Methods We measured shape and size variation for 208 floral bracts and 176 leaves in individuals from 14 localities using geometric morphometrics. We compiled data for two chloroplast regions (matK and trnL‐F) from 89 individuals to assess genetic diversity, population structure and phylogenetic relationships. We tested the influence of climate, altitude and genetic distance on morphological traits using linear statistical models. Results Temperature seasonality is a main driver of floral bract shape. Together with precipitation, it also explains changes in leaf size across the latitudinal gradient. Shifts in morphological traits are correlated with genetic structure and partly support the recent taxonomic delimitation proposed for the species complex. The species started to diversify in the Pliocene ca. 5 Mya. We detected a phylogeographical break in species distribution into northern and southern clades between the Bocaina region and the southern portion of the Atlantic Forest. Main Conclusions We identify how geography and environmental changes through time shape floral bracts and leaves in similar ways. At highly seasonal sites with lower annual precipitation (in the southern subtropical portion of the Atlantic Forest), leaves are larger and floral bracts are wide‐elliptic, making them better suited for increased water accumulation. In contrast, at less seasonal sites (in the tropical north, where rainfall is more abundant and temperatures are higher), leaves are narrower and floral bracts are lanceolate‐shaped, facilitating water drainage. The biogeographical break we identified suggests a role of tectonic activity and climatic oscillations in promoting species divergence and diversification.


| INTRODUC TI ON
The high biodiversity found in the Neotropical region is driven by multiple abiotic and biotic factors occurring simultaneously over different spatial and temporal scales (Antonelli et al., 2018;Antonelli & Sanmartín, 2011). Variation through space and environment can promote adaptive responses in species (Morales, De-la-Mora, & Piñero, 2018;Nosil, 2012). When phenotypes are correlated with environmental gradients, these responses often arise by selection (local adaptation), phenotypic plasticity, or some combination, and can influence the patterns of gene flow and distribution (Wanderley et al., 2017). Understanding the complex interaction and relative contributions of factors involved in the diversification of species and traits is crucial to gain insights into the evolution of the Neotropical biota.
In plants, the large variation of leaf shape and size reflects physiological demands imposed by environment (Givnish, 1979;Nicotra et al., 2011), and is observed both among and within species (Morello, Sassone, & López, 2018). Shape and size variability may have distinct patterns in distinct organisms, with size being considered more labile with variation mostly influenced by environment (phenotypic plasticity), while shape can be more conserved and variation usually genetically structured (Cardini & Elton, 2009;Chitwood et al., 2014;Maestri et al., 2016).
Vegetative and reproductive traits can undergo different selective pressures. The lability of vegetative traits often prevails over that found for reproductive ones (Chalcoff, Ezcurra, & Aizen, 2008).
Reproductive accessory organs such as showy floral bracts have an important role in pollination, and hence reproductive isolation (Bergamo et al., 2019). Floral bracts can have a variety of colours, shapes and scents decisive to reproductive success, besides their function to protect the flowers and therefore are often more conserved than leaves (Pélabon, Armbruster, & Hansen, 2011). Unlike the poorly studied floral bracts, leaves present a well-known pattern, where small-sized leaves are often found in dry and hot places, as well as at high latitudes and elevations to avoid transpirational water loss, while large sized leaves are typical of high humidity and warm conditions (Wright et al., 2017).
Within the Neotropical region, the Atlantic Forest is exceptionally diverse (Myers, Mittermeier, Mittermeier, da Fonseca, & Kent, 2000). The past and present environmental dynamics have a role in biome diversification and current species distribution. Both the Pliocene geological processes and subsequent Pleistocene climatic oscillations and the strong gradient across its latitudinal range may lead to genetic and morphological divergence not only within species, but also within complexes of closely related taxa (Alvares, Stape, Sentelhas, de Moraes, & Sparovek, 2013;Turchetto-Zolet, Pinheiro, Salgueiro, & Palma-Silva, 2013). Vriesea (Bromeliaceae) is one of the most representative genera of Atlantic Forest epiphytes (Ramos et al., 2019), where more than 90% of its 225 species occur (BFG, 2018). Recent diversification of the genus has caused low infrageneric resolution and unclear species boundaries (Costa, Gomesda-Silva, & Wanderley, 2014;Gomes-da-Silva & Souza-Chies, 2017).
One of the many species complexes within Vriesea includes V. incurvata complex'). The Vriesea incurvata complex is widely and continuously distributed across the Atlantic Forest. A recent taxonomic revision using morphological multivariate analysis showed a northsouth gradient in the width and length of both leaves and floral bracts and identified five taxa, including a new variety V. taritubensis var. patens (Neves, Uribbe, Jacques, Zanella, & Costa, 2018). Here, we aim to assess the drivers of morphological trait variation in the V.
incurvata complex along a latitudinal gradient in the Atlantic Forest.
We build upon previous work (Neves et al., 2018) using geometric morphometrics and the estimation of genetic structure and diversity. We evaluate shape and size of floral bracts and leaves of taxa in the complex. We use climatic, altitudinal and plastid genetic data to address the roles of both historical gene flow and environment in shaping morphological traits and species limits.
We test for trait variation in accordance with two hypotheses: (a) if shape is less labile than size (Cardini & Elton, 2009;Maestri et al., 2016) and is diagnostic for taxa (Neves et al., 2018), then we expect genetic structure to be a main predictor; (b) if vegetative characters are more labile than reproductive ones (Pélabon et al., 2011), then we expect variation in leaf traits to be better predicted by environmental shifts, while variation in floral bracts should be better predicted by genetic divergence.
We interpret our results in the light of the traditional taxonomic delimitation of the complex and the historical biogeographical patterns that can predict current distribution of taxa. By identifying the factors driving trait variation along the Atlantic Forest, we identify the importance of seasonality and rainfall regime as predictors of plant morphology and the influence of abiotic processes on the Atlantic Forest, including the effect of Pleistocene climatic oscillations on current species distribution and genetic diversity.

| Species and locality description
Taxa within the V. incurvata complex are abundant and widely distributed throughout the south-eastern and southern part of the Atlantic  Figure 1a).
All individuals are epiphytes generally found near rivers and waterfalls, in altitudes varying from 0 to 1,000 m a.s.l., from Rio de Janeiro (22°31′ S) to Rio Grande do Sul (29°44′ S) states (Neves et al., 2018), Brazil. The most notable attribute of these species is their showy red floral bracts covering the inflorescences, filled with a transparent and odourless mucilage. Leaves are highly polymorphic within and among taxa (Neves et al., 2018). Their spiral disposition forms a rosette, and together with leaf sheath size and shape these factors are important for determining the capacity of leaves for water and nutrient storage (Benzing, 2000). Besides the species V. incurvata and V. taritubensis (with three varieties), Neves et al. (2018) distinguished another species, V. sucrei L.B.Sm. & Read, the most divergent in the complex. Its inclusion was needed to determine the position of another binomial involved in the taxonomic revision but we did not include V. sucrei in the present work as it is clearly morphologically distinct.
F I G U R E 1 (a) Distribution of taxa based on herbarium records: Vriesea taritubensis var. brevisepala in blue, V. taritubensis var. taritubensis in green, V. taritubensis var. patens in yellow and V. incurvata in red. Sampling localities marked with a circle for morphological data and a cross for genetic data (see codes in Table S1, Appendix S2). The Atlantic Forest (grey), Serra do Mar (light green) and Serra dos Órgãos and Serra da Bocaina (dark green) are shown. The black line shows a major break in species distribution between the Serra da Bocaina region and southern Atlantic Forest. Map of Brazil with the Atlantic Forest area in upper left. (b) A median-joining network shows two main haplogroups, only V. taritubensis var. taritubensis and V. taritubensis var. patens shared haplotypes; each circle represents one haplotype, with its diameter proportional to its total frequency; more than one mutational step are indicated by numbers. (c) The Neighbour-joining tree shows the distance in bract shape among localities. The consensus shapes for each taxon (coloured line) and for the whole sampling (black dotted line) display changes from lanceolate to wide-elliptic [Colour figure can be viewed at wileyonlinelibrary.com] The Atlantic Forest presents high latitudinal (ca. 27°S) and altitudinal (0-2,892 m a.s.l.) variation. Along the study area, the climate ranges from tropical with dry winter at Serra dos Órgãos to humid subtropical with hot summer or temperate summer towards the south, with no distinct dry season. Warmer temperatures are registered in the southeast ranges of both varieties of V. taritubensis, and a cooler and marked seasonal climate is found further south where V.
incurvata occurs (Alvares et al., 2013;Ribeiro, Metzger, Martensen, Ponzoni, & Hirota, 2009). Precipitation ranges from 4,000 mm/year in the coast to 1,000 mm/year in inland areas. Higher precipitation levels are registered especially for Serra da Bocaina region (23°S), explained by the orographic effect caused by the mountains, within the range of V. taritubensis var. taritubensis and V. taritubensis var.

| Geometric morphometrics
To explore shape and size variation we sampled and georeferenced individuals from 14 localities, 208 individuals for floral bract and 176 individuals for leaf datasets ( Figure 1; Table S1, Appendix S2). We selected localities based on a previous study of herbarium collections (Neves et al., 2018), considering accessibility and availability of individuals. We sampled localities spread out along the whole geographical distribution of species, including the type localities of V. incurvata and V. taritubensis. We randomly sampled 8-19 individuals per locality.
We scanned one floral bract from the middle of the inflorescence and one leaf from the middle of the rosette of each specimen. We then characterized shape by a set of seven landmarks for bracts profile (bracts in side view) and nine for leaves. We discriminate landmarks into anatomical, when located at homologous points with biological meaning; mathematical, when located at extreme points with some geometrical property; and pseudo-landmarks, when located between other landmarks (Dryden & Mardia, 2016). For bract profile, we placed three anatomical landmarks (1, 2, 5) at the base and apex; two mathematical landmarks (6, 7) and two pseudo-landmarks (3, 4), at the widest point of the margin and its dorsal opposite and at the midpoint of margin contraction and its dorsal opposite. Such landmarks at the greatest and least width of bract margins reflect greater or lesser cell activity, and their dorsal opposites are needed to minimize the distance among base and tip describing the protuberance they can present ( Figure S1a, Appendix S2). The same for leaf, we placed three anatomical landmarks (1, 2, 6) at the sheath base and apex; four mathematical landmarks (3,4,8,9) at sheath widest points and at the transition sheath/blade; and two pseudo-landmarks (5, 7) at the level 3/4 of blade ( Figure S1b, Appendix S2). We used these set of landmarks to better describe the shape of structures without making use of numerous equidistant pseudo-landmarks along the contour itself (Bensmihen et al., 2008) or the Elliptical Fourier Descriptors (Chitwood & Otoni, 2017), alternatives when homologous points are lacking. To be consistent, a single person (B. Neves) manually digitized the landmarks for each image using TpsDig 2 (Rohlf, 2010). We imported the file containing raw data to MorphoJ 1.06b (Klingenberg, 2011). We then extracted shape information performing a Generalized Procrustes Analysis which removes size, rotation and translation effects (Rohlf & Slice, 1990).
We examined potential bias due to variation within a single individual by testing multiple leaves spirally arranged at the middle of the rosette through a Multivariate Analysis of Variance and Principal Component Analysis (PCA) plot. We also tested for digitization error using Procrustes ANOVA in two sets of data (original and replicated measures of the same leaf) following Klingenberg and McIntrye (1998). We analysed leaves and floral bracts separately, as well as shape and size. We used the centroid size (the square root of the sum of all squared distances from each landmark to the centroid) as a measure of size. We tested allometry by performing a regression analysis between shape and size of each bract and leaf. Because size predicted 47% of leaf shape variation (and 2% for bract), we ran further analyses with the size-corrected shape matrix resulting from the residuals of the regression. We examined shape and size data performing Procrustes ANOVA, PCA and Canonical Variate Analysis (CVA) (Campbell & Atchley, 1981;Neff & Marcus, 1980), testing for differences among localities and taxa. Due to the marked differentiation in bract shape reflecting the studied taxa, we made a Neighbour-joining (Saitou & Nei, 1987) tree using Mahalanobis distances, appropriate for morphometric data (Mahalanobis, 1936) and reconstructed the consensus shape for each taxon. All analyses were performed in MorphoJ except for ANOVA and Neighbour-joining using package 'vegan' (Oksanen et al., 2017) in R 3.5.0 (R Core Team, 2018).

| Genetic analyses
To assess genetic diversity and population structure, and to infer the relationships among individuals, we concatenated sequences from two chloroplast regions: trnL-trnF intron and intergenic spacer and matK gene for 89 individuals in 15 localities. Among the 15 localities, we have morphological data for nine of them. We collected genetic data independently from the morphological data, individuals were randomly sampled according to availability in the field. We used plastidial data (trnL-trnF and matK) from V. incurvata sequenced by Zanella, Palma-Silva, Goetze, and Bered (2016) and from 40 new samples of V. taritubensis (Table S1, Appendix S2). Data were deposited in GenBank with accession numbers MN593031 -MN593118.
For details about DNA processing and PCR conditions see Zanella et al. (2016). We performed multiple sequence alignment in MUSCLE (Edgar, 2004), implemented in MEGA 7.0 (Kumar, Stecher, & Tamura, 2016). Mononucleotide repeat length variations were excluded and indels of more than 1 bp were coded as single mutational event.
We ran a Bayesian analysis in BEAST 2.5.2 (Bouckaert et al., 2014) to estimate the temporal divergence of species and varieties using the CIPRES Science Gateway 3.3 (Miller, Pfeiffer, & Schwartz, 2010 (Gernhard, 2008) was used to date the tree. Markov chains were run for 150,000,000 steps, sampling every 15,000 steps, totalling 10,000 trees. We used secondary calibration to date the tree

| Climate and topography
To identify the most important environmental variables in explaining trait variation, we extracted a total of 19 bioclimatic variables from 20 localities (Table S1, Appendix S2). Data were obtained from the Worldclim database (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005) with a spatial resolution of 30 arc-seconds. In addition, data on altitude were extracted from GPS coordinates in the field. We first performed a PCA with the correlation matrix of all environmental variables to avoid colinearity and overfitting the full model. We used the packages 'raster' (Hijmans et al., 2015) to extract climate variables and 'stats' to perform a PCA in R. The results of the PCA are described in Appendix S1. Because the first three principal components accounted for 96% of variance and were highly correlated with temperature seasonality (BIO4), annual precipitation (BIO12) and altitude (Table S2, Appendix S2), we proceeded the further analyses using these three variables.
To investigate the correlation of the genetic structure, bioclimatic variables and altitude with morphometric traits variation, we ran multiple linear models and to identify the best model in explaining traits variation, we performed model selection based on AIC (Burnham and Anderson 2002). We ran analyses using the R package 'stats' and 'MuMIn'. We included data only for localities where both genetic and morphological data were collected (9 of the 20 localities Table   S1, Appendix S2). We used floral bract shape, floral bract size, leaf shape and leaf size, separately, as response variables. The models were fit with genetic distance across localities, temperature seasonality, annual precipitation and altitude as explanatory factors (response variable ~ pco1gen + pco2gen + temp season + annual prec + alt). To describe shape, we used the regression scores from the size versus shape interaction exported from MorphoJ that captured size-free information, then we performed a PCA with those regression scores and summarized shape in one value using the first PC (which explained 61% of bract shape and 71% of leaf shape variance) as the descriptor. For size, we used the log-transformed centroid to control for large variation of size in the sampling (Klingenberg, 2016). To describe genetic distance, we first calculated the pairwise F ST between localities using DnaSP. F ST is frequently used as a measure of distance for depicting genetic variation (Wright, 1951). Then we performed a Principal Coordinate Analysis (PCoA) on the pairwise F ST matrix using the R package 'ape' (Paradis, Claude, & Strimmer, 2004) and selected the first two PCo following the broken stick criterion (Jackson, 1993;  (Table S2, Appendix S2). All explanatory variables were standardized to zero for the mean and unit variance. Additionally, we tested for spatial autocorrelation in size and shape of bract and leaf with Moran's correlograms using the R package 'ncf'. Because significant values were found for bract shape, we corrected for spatial autocorrelation effects by performing Principal Coordinates of Neighbourhood Matrices (PCNM; Borcard & Legendre, 2002) with the R package 'vegan'. The Euclidean distance matrix of the geographical coordinates from the localities was converted into a truncated distance matrix, focusing on neighbouring sites which is then subject to a PCoA. The PCNM axes were used as explanatory variables for bract shape and the one that showed significant explanatory power (PCNM 1) was then incorporated to the model.

| Floral bract
The shape of floral bract significantly discriminated three of the four taxa as currently defined morphologically (df = 3, F = 140.8, p < 2e-16). We identified a shift from lanceolate to wide-elliptic along the

| Leaf
A significant difference in leaf shape between taxa (df = 3, F = 4.919, p = .00265) and localities (df = 13, F = 2.87, p = .000945) was indicated. The PCA showed shape changes from narrow-obovate to linear along the first PC, which accounted for 75% of the total variance ( Figure 4). Despite the overlap due to the high variation within and between localities, we detected a gradual change pointing to the separation of V. taritubensis var. brevisepala that represents an enlargement of leaf sheaths, from large-obovate to elliptic (see first CV of CVA in Figure 4b).
taritubensis (with the largest leaves found in locality III and the smallest in locality VII) and within V. incurvata (locality XIII presents the smallest, and localities XVIII and XX the largest leaves) (Figure 3b

| Genetic structure and diversity
We identified 24 polymorphic sites (8 transitions, 4 transversions and 12 indels) resulting in 27 haplotypes from the cpDNA dataset (Table S1, Appendix S2). Vriesea incurvata, the most widespread taxon, presented the highest number of haplotypes, followed by Vriesea taritubensis var. taritubensis. The four taxa showed high haplotype and nucleotide diversity ( Table 1). The main source of genetic variation was found among taxa (79%), followed by 17% within localities and 2% among localities within taxa (Table S3, Appendix S2). We found high genetic differentiation among taxa through pairwise F ST estimates, excepted between V. taritubensis var. taritubensis and V. taritubensis var. patens that showed low genetic differentiation (Table 2). On the other hand, low genetic distance across localities within each taxon was observed ( Figure S2, Appendix S2). Two main haplogroups were found: one corresponding to V. incurvata, another to V. taritubensis.
A subgroup can also be identified dividing V. taritubensis var. brevisepala from V. taritubensis var. taritubensis and V. taritubensis var. patens

| Predictors of morphological variation
Temperature seasonality and genetics were the main drivers of floral bract shape and together explained a large proportion of the trait variance (66%). Annual precipitation, temperature seasonality and genetics were recovered as drivers of leaf size changes and our best fit model for leaf size explained 28% of overall variation. Only a small portion of the variation was predicted by bract size and leaf shape models (6% and 9%, respectively) and both traits correlated with genetics and altitude (Table 3). For complete information on best models selected see Table S4, Appendix S2.

| D ISCUSS I ON
We used an interdisciplinary approach to investigate the factors underlying morphological trait variation in bromeliads across a latitudinal gradient in the Atlantic Forest. Temperature seasonality is the main driver of floral bract shape and together with annual precipitation explains leaf size variation. The genetic structure (Figure 1b and Figure   S2a) correlates with morphological trait variation and all models indicate an influence of genetic inheritance (Table 3). This finding rejects our two hypotheses as follows (a) vegetative and reproductive characters are both labile and (b) shape is not more conserved than size, and they both vary with environmental heterogeneity. Our results also reveal a north-south phylogeographical break between the Bocaina region and southern portion of Atlantic Forest (Figures 1 and 5), from where V. taritubensis and V. incurvata likely diverged around 5 Mya.
Divergence was likely a consequence of tectonic events that continued shaping the complex topography of the Serra do Mar at that time, resulting in marked habitat heterogeneity (Brunes, Sequeira, Haddad, & Alexandrino, 2010;Ribeiro, 2006).

F I G U R E 3
Boxplots of centroid size variation of (a) bracts and (b) leaves in a latitudinal gradient across 14 localities. Vriesea taritubensis var. brevisepala in blue, V. taritubensis var. taritubensis in green, Vriesea taritubensis var. patens in yellow and V. incurvata in red. For localities number see Table S1 in Appendix S2 [Colour figure can be viewed at wileyonlinelibrary.com]

| Climatic drivers of morphological traits
We found clear trends in shape and size that are congruent with climatic variation along the Atlantic Forest latitudinal gradient ( Figure 1). Floral bracts are wide-elliptic and leaf sizes larger (wider blades and sheaths) in localities at highly seasonal sites with lower precipitation levels (V. incurvata distribution) in the southern, subtropical portion of the Atlantic Forest (Figures 1, 2, 3 and 4). This phenotype relates to bracts and rosettes with greater capacity to intercept and accumulate water, ensuring survival in dryer periods. In contrast, our data show lanceolate bracts and narrower leaf blades and sheaths in localities at the Bocaina region (V. taritubensis var.

F I G U R E 4 (a) Principal Component
Analysis and (b) Canonical Variate Analysis computed on leaf landmarks. Variation of shape from narrow-obovate to linear is shown along the first PC. The first CV relate to a change of leaf sheaths from large-obovate to elliptic which separate blue and red groups. Percentage of total variance for each axis is indicated. Vriesea taritubensis var. taritubensis in green, Vriesea taritubensis var. patens in yellow, V. taritubensis var. brevisepala in blue and V. incurvata in red [Colour figure can be viewed at wileyonlinelibrary.com] taritubensis var. brevisepala distribution), where annual precipitation is moderate (ranging from 1,336 to 1,740 mm) and there is little seasonal temperature variation (Figures 1 and 2).
The size and shape of leaves have been widely associated with temperature and moisture variables (Nicotra et al., 2011;Royer, Wilf, Janesko, Kowalski, & Dilcher, 2005;Wright et al., 2017). A TA B L E 2 Pairwise genetic divergence (F ST values) for Vriesea incurvata complex species based on plastid sequence data (trnL-trnF + matK) F I G U R E 5 Maximum Clade Credibility tree from BEAST based on trnL-trnF and matK sequences with posterior probabilities above 0.8 shown for the main clades. Two major clades splitting V. taritubensis (green, yellow and blue) and V. incurvata (red) specimens are recovered. Stars represent calibration points. For localities number see Table S1 in Appendix S2 [Colour figure can be viewed at wileyonlinelibrary.com] global assessment of climatic drivers of leaf size demonstrated no effective thermal constraint acting in warm and ever-wet tropical climates, as sufficient water is commonly available for transpirational cooling and plants are not exposed to frost damage (Wright et al., 2017). At a small scale and focusing on a clade of understorey epiphytes in a humid mountain tropical forest, we still detect climate influencing leaf size changes (Table 3). Besides considering leaf area itself, bromeliad tanks store water that is then available for a longer time period than a single rainfall event. Being large leaved at seasonal sites with lower mean annual precipitation is advantageous if this size translates into wider blades and sheaths to form a robust tank to retain water. Our results contrast to the pattern of a decrease in size at sites with lower rainfall and hotter temperatures with increased irradiance (Peppe et al., 2011;Ribeiro et al., 2016;Wright et al., 2017), but is consistent if interpreted in a more comprehensive way regarding plant architecture.
Furthermore, the rainwater drainage capacity of the narrow-leaved plants generally found in humid regions can prevent leaf surface water retention, which can affect gas exchange and inhibit photosynthesis (Benzing, 2000;Martin & Siedow, 1981). Bromeliads show a large variety of leaf shape and size and tank forms (Benzing, 2000). We hypothesize the pattern we found here to occur in another epiphytic tank-forming bromeliads which are widely distributed across Atlantic Forest.
We show bract shape changes follow similar trends as leaf size in being strongly correlated with temperature seasonality (Table 3; Figures 1 and 2). Contrary to our findings, Pélabon et al. (2011) found leaf size to be more sensitive to environment than floral bract size.
The bracts of V. incurvata complex are leaf-like, boat-shaped and distichally arranged along the inflorescence axis. Such display is similar to the water impound tank of most bromeliads and likely controls flower humidity, reflecting the similar pattern of variation found for the leaves.
Floral bract size and leaf shape variation were poorly explained by our models (Table 3). Both traits show high local and intraspecific variation that could be influenced by other abiotic local conditions and/or biotic interaction, or still by stochastic factors not accounted for in this study. Leaf shape is shown to be driven by both evolutionary and environmental forces (Chitwood & Sinha, 2016). We identified some influence of both genetic structure and climate driving leaf shape, despite the diffuse pattern recovered among taxa and along the latitudinal gradient in the forest (Table 3, Figure 4). Finally, most morphometric studies assessing leaf shape variation have been focused on individual taxa and eudicotyledons net-veined leaves with distinct petioles (Royer, Meyerson, Robertson, & Adams, 2009). Our study is one of the first to present an integrative and ecological assessment of shape of leaves and floral bracts in bromeliads and, as far as we know, in monocotyledons.  Figures 1 and 5). Vriesea incurvata and V. taritubensis started to diverge in the Pliocene (5 Mya; Figure 5) and probably experienced a long period of genetic isolation in consequence of geographical and climatic barriers, and/or influenced by low or absent gene flow.
On the other hand, the V. incurvata and V. taritubensis varieties showed moderate to high intraspecific genetic diversity and intraspecific gene flow (Table 1; Tables S1 and S3, Figure S2, Appendix S2), as has been observed for other bromeliad species (Goetze, Zanella, Palma-Silva, Büttow, & Bered, 2017;Palma-Silva et al., 2009;Zanella et al., 2016). The low genetic variation among localities within taxa suggests that intra-taxon gene flow has been maintained ( Figure S2, Appendix S2). The V. incurvata complex is pollinated by hummingbirds such as Phaethornis eurynome, generalist and non-territorial long-distance dispersers (Machado & Semir, 2006). Seeds are wind dispersed, which may also facilitate the maintenance of gene flow and localities connectivity.
We show floral bract shape and chloroplast divergence support different taxa, except for V. taritubensis var. taritubensis and V. taritubensis var. patens that share haplotypes and the same floral bract shape (Figure 1). The two species V. incurvata and V. taritubensis are clearly distinct from each other and the varieties of V. taritubensis are not monophyletic ( Figure 5). Our results partly corroborate the recent taxonomic circumscription proposed by Neves et al. (2018) using traditional morphometrics.
The new variety V. taritubensis var. patens proposed by these authors was described based on size and orientation of the inflorescence. We found three exclusive haplotypes for V. taritubensis var.
patens and a difference in size separating it from V. incurvata (Figures   1b and 3a). Our PCoA results based on pairwise F ST distance across all localities also showed a separation of V. taritubensis var. patens (locality VIII) along PCo 2 ( Figure S2a, Appendix 1). Despite this lack of evidence we find here for V. taritubensis var. patens, we suggest taxonomic decisions be made using more robust molecular data.

| Phylogenetic and biogeographical patterns in the Vriesea incurvata complex
We identify a north-south phylogeographical break between the Bocaina region and southern Atlantic Forest (Figures 1 and 5). The divergence between V. incurvata and V. taritubensis started around 5 Mya, followed by the split of the V. taritubensis varieties around 2.5 Mya, during the Pliocene-Early Pleistocene ( Figure 5). This temporal framework is in accordance with that estimated by Kessous et al. (2019) for Vriesea species diversification (starting ca. 3 Mya and increasing during Pleistocene) and also coincides with the divergence of other widely distributed Atlantic Forest lineages such as reptiles and amphibians (Brunes et al., 2010;Fitzpatrick, Brasileiro, Haddad, & Zamudio, 2009;Grazziotin, Monzel, Echeverrigaray, & Bonatto, 2006).  Zanella, 2013). We suggest the break to be the result of increased topographic discontinuity of the Serra do Mar mountains, which led to geographical isolation. Furthermore, the central Atlantic Forest is a convergence zone of numerous distinct bromeliad lineages and is highly species rich (Fontoura, Scudeller, & Costa, 2012), consistent with the higher biodiversity levels found in the central portion (Carnaval et al., 2009).
Pollen data from Holocene vegetation support the existence of historical refugia in the Serra da Bocaina and Serra dos Órgãos

| CON CLUS IONS
Our findings identify the role of climate, genetics and topography in shaping morphological traits in the Atlantic Forest. We provide a broad-scale quantification of trait variability at both inter-and intraspecific levels. Importantly, we show that reproductive (floral bracts) and vegetative (leaves) traits respond in similar ways to abiotic factors, contradicting previous suggestions. Temperature seasonality and rainfall regime are associated with the morphological diversification of these bromeliads as they drive traits variation that are diagnostic for taxa and have an ecological role. Morphological variation and genetic structure patterns coincided with recurrent biogeographical breaks in space (central region) and time (Late Pliocene) with other widely distributed Atlantic Forest organisms. We shed new light on the species delimitation of one of many species complexes of morphologically similar taxa within Vriesea. Because the Atlantic Forest is highly threatened and suffers continued habitat destruction, studies like this help to unveil combined evolutionary and ecological dynamics generating species diversification, before much biodiversity is lost.

ACK N OWLED G EM ENTS
We thank Simon Mayo, Renan Maestri, and members of the

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study will be openly avail-