Macroevolutionary trends in leaf venation network architecture

dor ABSTRACT A diversity of leaf venation architectures can be observed in both living and fossil plants. However, many aspects of venation evolution remain elusive, limiting our understanding of when, why, and in which clades the architectural innovations arose; and how they impacted leaf functioning. Here, we used a unique image dataset of 1,000 extant and extinct plant taxa to phylogenetically reconstruct ca. 400 Myr of evolutionary dynamics of venation architecture across clades and vein sizes. Overall, venation networks evolved from having fewer veins with convoluted loops, to having more veins, straighter loops, and more internal free-ending veins. These changes only occurred in smaller and medium (minor) veins, while larger (major) veins remained unaltered. The architectural diversity (disparity) followed a non-linear trajectory, with ferns and early-diverging seed plants first filling the limits of the observed morphospace, and then with recent angiosperms lineages initiating a second phase of morphospace exploration likely still on course. Vein evolution mostly occurred independent of fluctuations in climate conditions, but may have been driven by insect herbivore evolution. We discuss the implications of those novel findings for inferring past plant ecophysiology, reconstructing paleoclimate from fossil leaves, and understanding internal and external drivers of venation network diversification. METHODS Obtaining images of leaf venation networks. We compiled 1,000 leaf images of extant (N = 880) and fossil (N = 120) species, distributed in 784 genera, 325 families, 90 orders, and 7 plant clades. To create this unique dataset, we brought together samples from multiple paleobotany and cleared leaf collections. Images of extant species were selected from four collections: (1) Smithsonian National Cleared Leaf Collection: comprises 7,000 cleared leaf images available at https://collections.peabody.yale.edu/pb/nclc/; (2) University of California Museum of Paleontology (‘UCMP’) Cleared Leaf Collection: comprises over 2,000 cleared leaf samples of extant species, including the Daniel I. Axelrod collection. Low-resolution images of some samples are available at (https://ucmp.berkeley.edu/collections/paleobotany-collection/) and high-resolution images are available here at (UCMP_samples folder); (3) Wilf collection: an open-access database of 26,176 leaf images of extant species available at (https://phytokeys.pensoft.net/article/72350/); and (4) Macrosystem Ecology Laboratory at Berkeley (‘MEL’) leaf collection: a database of 326 cleared leaf images of extant species prepared by our team from collections in various localities (Costa Rica, Ghana, Ecuador, United States). The original cleared leaf images for the 122 leaf samples collected at the University of California Botanical Garden at Berkeley (UCBG) are available at doi: 10.5061/dryad.1g1jwsv36, while Ghana leaf samples are available at (ADD DOI). Cleared leaves from Ecuador and Costa Rica used in this study are available here at Costa Rica_samples and Ecuador_samples folders. Matos et al. 2024 describe how the UCBG-MEL leaf samples were collected, chemically cleared, and imaged. Ecuador samples: Leaf samples for the Ecuador dataset were collected in June and July, 2019 at the Smithsonian CTFS long-term forest dynamics plot at Yasuni Biological Research Station (also known as Estacion Biologica Yasuni). The tree species were chosen based on the species list of Cardenas et al. (2014). Therefore, the trees chosen for leaf sampling may or may not have been within the CTFS survey plot as these trees were chosen based on their proximity to the research station. The samples were collected with the assistance of field botanist Pablo Alvia, and researchers Dr. Luiza Aparecido (then Arizona State University, ASU) and Emily Guevara Heredia (PUCE), where they were pressed, dried and stored at the research station and later transported to the lab at ASU for chemical processing and imaging of the leaf venation architecture (conducted by Miguel Duarte and Martha Ryan, both ASU, in addition to Emily G. Heredia). Costa Rica samples: Leaf samples for the Costa Rica dataset were collected incidental to a plot-based survey of woody plant community composition and traits along altitudinal gradients. Species were selected for this study on the basis of their biomass dominance, or when diameter data were not available, based on abundance. Four 0.1 ha plots were established at elevations ranging from 75 m to 3250 m within the Río Savegre watershed on the Central Pacific slope of Costa Rica. Plot elevations were determined in advance of field work, with final locations determined by feasibility of access from field stations to the nearest area of apparently undisturbed forest at the pre-selected elevation. Plots were established and inventoried in April 2011, using the Gentry 0.1 ha method, which includes stems of all woody plants ≥2.5cm diameter at breast height encountered in ten 2 m x 50 m subplots, for a total plots area of 1000 m 2 (Phillips et al., 2003; Phillips and Miller, 2002) . In each plot, at least one specimen of every species encountered was collected, preserved in the field in newspaper saturated with 70% ethanol, and dried three to seven days later in standard plant presses at the Herbarium of the National Biodiversity Institute of Costa Rica (Index Herbariorum acronym INB, now part of CR). Each specimen was assigned a unique collection number with Brad Boyle as senior collector, followed by an integer identifier. Specimens identifications were completed at INB by Brad Boyle with the assistance of local and international taxonomic experts. After drying, a portion of each collection was deposited as a permanent specimen voucher with the Herbarium of the National Museum of Costa Rica (acronym CR). Duplicate specimens bearing the same collection numbers were transported to the Brian Enquist Lab at the University of Arizona and stored in standard herbarium cabinets. Leaf samples used for the present analysis were obtained from the Enquist Lab duplicate specimens (see Costa_rica_specimens.csv file for more information about collection sites). We carefully evaluated the leaf images available from all those collections above to determine which samples were suitable for our work. We selected only sharp images showing the whole leaf or with at least 75% of the leaf area present, and with no major damage (e.g. tears, holes, folds, bubbles, herbivory, or material deterioration). For many monocot species, leaves were too long (>30 cm) to be wholly processed and/or imaged, so only a leaf segment representing ca. 20-50% of the total leaf area was analyzed. Notably, due to the variable resolution of the leaf images of extant species, our study was able to accurately resolve any vein with width ≥10 𝜇m, but occasionally missed small veins in species with minor venation smaller than this threshold. Therefore, our dataset is not directly comparable to previous studies using microscopy of minor veins. Images of extinct species were compiled from fossil leaves from six museum paleobotany collections: (1) Yale Peabody Museum (https://peabody.yale.edu/): 4,300 typed and imaged specimens, including Triassic and late Cretaceous floras from Arizona, New York, New Jersey, and southern New England, which are freely available at GBIF (www.gbif.org); (2) Natural History Museum of Berlin (https://www.museumfuernaturkunde.berlin/): 29,826 fossil records from the Paleozoic to Cenozoic eras; high-resolution images for most of this collection are available at GBIF too; (3) University of California Museum of Paleontology: ca. 12,000 typed plant fossil specimens available at https://ucmpdb.berkeley.edu/. (4) Illinois State Museum (https://www.illinoisstatemuseum.org/): 600 plant fossils, including many Carboniferous fossils from the Mazon Creek formation; (5) Florida Museum (https://www.floridamuseum.ufl.edu/): ca. 250,000 specimens, ranging from the Proterozoic to the Pleistocene; and (6) University of Alberta Paleobotanical Collection: ca. 127,000 plant fossil specimens, available at https://search.museums.ualberta.ca/. In addition to those museum collections, we also sampled images of fossil leaves from the Wilf collection41, which compiles 4,076 fossil samples dated from Late Cretaceous to Eocene, including specimens from the Florissant Fossil Beds National Monument (late Eocene, Colorado, US). Although leaf fossils are abundant in many depositional settings, the preservation of whole-leaf venation networks in fossils is rare. Thus, instead of whole leaves, we sampled leaf sections (ranging from 19 to 32,000 mm2) from any leaf fossil image where tertiary or smaller veins were distinguishable. We also only retained fossil samples whose taxonomic identity and age could be retrieved, since that information was required for grafting fossil taxa to the phylogenetic tree of extant species. We recognize that our survey of leaf fossil images is not exhaustive, and it is potentially biased towards taxa more likely to fossilize well (e.g. species with bundle sheath extensions). Despite those limitations, our fossil dataset yielded key coverage of important internal nodes of the vascular plants phylogenetic tree, ranging in age from 11.6 to 358.9 Mya, the time period of vascular land plant evolution. When compiling leaf images of both extant and fossil species, we also aimed to maximize phylogenetic coverage, by sampling species from all major vascular plant clades: (1) ferns, (2) gymnosperms (including Bennettitales, Caytoniales, Cordaitanthales, Medullosales), (3) basal angiosperms (including Amborellales, Nymphaeales, Austrobaileyales, Chloranthales, and Magnoliids), (4) monocots (including Commelinids), (5) basal eudicots (including Ranunculales, Proteales, Trochodendrales, Buxales, Gunnerales, Dilleniales, Saxifragales), (6) rosids, and (7) asterids. We limited our analysis to one sample (one leaf) per species, so our study did not address intraspecific variation in venation traits. This approach is justified since our study focuses on interspecific variation in those traits across the whole plant phylogeny. Obtaining the phylogenetic tree of extant and fossil taxa. We assembled a time-calibrated phylogenetic supertree for all 1,000 vascular plant species evaluated in this study (see Figure 1 in figures folder). First, we used the R-package ‘U.Taxostand’ (Zhang and Qian, 2023) to standardize spellings and nomenclature for all extant species names, following the World Plants (worldplants.de) database and the APG IV classification (APG IV, 2016). Similarly, fossil species names were standardized following the Paleobiology (paleobiodb.org) and Fossilworks (fossilworks.org) databases. After the nomenclature standardization, we built a phylogenetic tree for all extant species using the ‘V.PhyloMaker2’ R-package (Jin and Qian, 2022). To graft our fossil species to the extant phylogenetic tree, we used the function ‘tree.merger’ from the ‘RRphylo’ R-package (Castiglione et al., 2022). See '1_create_phylo_tree.R' file in scripts folder. Extracting key venation architectural traits. We used GIMP version 2.10.32 (https://www.gimp.org/downloads/) and ImageJ version 1.53t (https://imagej.nih.gov/) to pre-process all leaf images. Image pre-processing involved cropping the leaf image, preparing masks to delineate leaf boundaries, and determining image resolution, which ranged from 0.003 to 0.061 pixels mm-1 in extant species and from 0.006 to 0.179 pixels mm-1 in fossil species. For some species with a large midrib (i.e. midvein), a mask of the midrib was also prepared to prevent it from splitting into multiple segments during the network extraction (see Matos et al., 2024 for more details about the image pre-processing step). After pre-processing, leaf image segmentation (i.e. conversion of a coloured leaf cleared image into a binary image where veins pixels are shown in white and non-veins pixels are shown in black) was done automatically using LeafVeinCNN program versions 1.0.7 (software and manual available at https://doi.org/10.5281/zenodo.4007731) and 2.12 (available at Matos et al., 2024). We configured LeafVeinCNN to use an ensemble average of three convolutional neural network predictions(CNNs) to segment veins. Since our CNNs were not trained to segment fossil images, which have different color, contrast, and textures than modern cleared leaf images, all our leaf fossil samples were carefully hand-traced to identify the position, size, and connectivity of all veins. The segmentation process resulted in over 19 million veins segmented and resolved veins with width ≥10 𝜇m, and occasionally missed some very small veins. All segmented venation networks were then processed in the LeafVeinCNN program to (1) produce a spatial graph representation of the entire venation network, (2) generate hierarchical loop decompositions (Katifori and Magnasco, 2012), and (3) calculate a series of multiscale venation statistics that describes how the venation network topology and geometry change across vein sizes. Here, we focused on four of those multiscale traits: vein density (VD), minimum spanning tree ratio (MST), loop elongation ratio (ER), and circularity ratio (CR) (Blonder et al., 2020). Identifying evolutionary trends in single venation traits. We investigated how each of those four key venation traits (VD, MST, ER, CR) varied over time at each vein spatial scale (i.e. at small, medium, large sizes). To make an interpretable assessment of trait evolution at different vein spatial scales we binned VD, MST, ER, and CR values for each species at three classes to represent small, medium, and large vein widths. First standardized (i.e. z-transformed) vein width values for each species, by dividing each vein width by the maximum width value of that species. Therefore, scaled vein sizes varied between 0 and 1 across all species. Next, for each species we classified vein sizes into small (0 < scaled vein width ≤ 0.3), medium (0.3 < scaled vein width ≤ 0.6), and large (scaled vein width > 0.6 mm) size classes. After classifying veins into size classes, we calculated the median VD, MST , CR and ER for each class, see '2_prepare_data_for_ace.R' file in scripts folder. We used the ‘RRphylo’ R-package (Castiglione et al, 2018) to investigate the rates and trajectories of diversification over time in each of our four key architecture traits at each vein size class. ‘RRphylo’ performs a phylogenetic ridge regression (a maximum-likelihood method) on a phylogenetic tree and associated trait data to return both the ancestral estimates at internal nodes and the branch-wise rates of trait evolution. To test for the existence of evolutionary trends (i.e. directional patterns of evolutionary trait mean change over time) in leaf venation architecture, we applied the function ‘search.trend’ from the ‘RRphylo’ package for each trait at each vein size class, see '3_run_ace.R' and '4_plot_ace.R' files in scricpts folder. To assess whether the evolutionary trends identified in this study were robust to sampling and phylogenetic uncertainty, we applied the ‘overfitRR’ function (Serio et al, 2019) from the “RRphylo” R-package, see '5_overfitRR.R' file in scripts folder. Identifying evolutionary trends in multi-trait venation network architecture. To investigate in which clades the evolution of novel architectural trait-combinations have originated and how the diversity of architectures varied over time we performed a disparity (aka multidimensional space occupancy) analysis using the ‘dispRity’ R-package (Guillerme et al, 2018). First, we carried out a principal component analysis (PCA) across the four venation traits (VD, ER, CR and MST) at each vein size class (small, medium, and large) to define our multidimensional venation architectural space (i.e. morphospace). Using this 12-dimensional architectural space, we calculated two metrics of disparity that describe two complementary aspects of the multidimensional space occupancy: (1) the median centroids (i.e. the median distance between each element and the centroid of the ordinated space), which describe the position of a element in the morphospace compared to a fixed point in this space (in our case the space centroid); and (2) the sum variances (i.e. the sum of variance of each dimension of the ordinated space), which characterizes the size of morphospace occupied by each element. To test whether different plant clades occupy different regions of the architectural space (i.e. whether they differ in their median centroids and/or sum variances), we used the function ‘dispRity.per.group’, which measured the disparity between each clade, see '6_run_disparity.R' file in scripts folder. To test how the space occupancy varied across time (e.g. test whether the disparity metrics increased or decreased over time), we used the ‘dispRity.through.time’ function to conduct a disparity-through-time (DTT) analyses using the time-slicing method (Guillerme and Cooper, 2018), using two alternative models: a punctuated model (i.e. “proximity”) and a gradual model (i.e. “gradual.splits”) of evolution. We also quantified the evolutionary mode of leaf venation architecture diversification of vascular plants by using the ‘model.test’ function to fit five alternative modes of disparity changes through time, see '6_run_disparity.R' file in scripts folder. Identifying main environmental drivers of venation architecture evolution. To explore abiotic and biotic factors that may have influenced leaf venation evolution in vascular plants, we compared the reconstructed evolutionary trajectories of single venation traits and the patterns of venation architectural disparity with global average temperature (Foster et al, 2017), atmospheric CO2 concentration (Scotese et al, 2021), and insect diversification rates (Condamine et al, 2016) for the Phanerozoic eon. Time-binned median values of each of these time-series data sets were calculated and used in the generalized least squares (GLS) regression analyses as implemented in the R-package “nlme”, see '7_time_series.R' file in scripts folder. All analyses were carried out using R version 4.3.1 (R Core Team, 2023). DATA DESCRIPTION Data and R-scripts to reproduce the analysis and figures in the paper entitled: Macroevolutionary trends in leaf venation network architecture. 1. data folder: 1.1 ace_scaled_change_10myr.csv: rate of change of median trait value per 10 Myrs of four key venation traits (VD - vein density, MST - minimum spanning tree ratio, CR - loop circularity ratio, and ER - loop elongation ratio) across three vein sizes (small, medium, large) and seven major plant clades (ferns, gymnosperms, basal angiosperms, monocots, basal eudicots, rosids, and asterids); 1.2 ace_scaled_FINAL.csv: median values of four key venation traits (VD - vein density, MST - minimum spanning tree ratio, CR - loop circularity ratio, and ER - loop elongation ratio) across three vein sizes (small, medium, large) for 1,000 extant (N = 880) and extinct (N = 120) plant taxa, and ancestral state estimations for 999 internal nodes of the phylogenetic tree obtained via phylogenetic ridge regression analysis. age2 = species or node age in Mya; time-slice = binned age into 40 time bins; class = categorical variable indicating wether the values in a row represent a tip extant species, tip extinct species, or internal node; clade1-3 = species plant clade. 1.3 CR_large_rate_box.csv: evolutionary rates for loop circularity ratio at large vein sizes (CR large) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.4 CR_medium_rate_box.csv: evolutionary rates for loop circularity ratio at medium vein sizes (CR medium) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.5 CR_small_rate_box.csv: evolutionary rates for loop circularity ratio at small vein sizes (CR small) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.6 data_for_ace_rmin_scaled_final.csv: median values of four key venation traits (VD - vein density, MST - minimum spanning tree ratio, CR - loop circularity ratio, and ER - loop elongation ratio) across three vein sizes (small, medium, large) for 1,000 extant (N = 880) and extinct (N = 120) plant taxa, and ancestral state estimations for 999 internal nodes of the phylogenetic tree obtained via phylogenetic ridge regression analysis. 1.7 ER_large_rate_box.csv: evolutionary rates for loop elongation ratio at large vein sizes (ER large) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.8 ER_medium_rate_box.csv: evolutionary rates for loop elongation ratio at medium vein sizes (ER medium) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.9 ER_small_rate_box.csv: evolutionary rates for loop elongation ratio at small vein sizes (ER small) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.10 evol_trends_v5_spp_list.csv: list of 1,000 plant taxa investigated in this study. Species name = following World Plant database; Genus = species genus; Family = species family; Order = species order; Clade = species clade; Status = categorical variable describing species status extant or not extant; Database = collection from which the leaf image was obtained; Image resolution = in mm per pixel; Old file name = sample name in the original database; Threshold = threshold value used during the leaf image segmentation in LeafVeinCNN; TotA: total leaf area; file = file name in this database. 1.11 FADLAD.csv: fossil species ages. FAD = first age of occurrence, LAD = last age of occurrence. 1.12 fadlad_ast.csv: first and last age of occurrence of asterids fossils. 1.13 fadlad_bangio.csv: first and last age of occurrence of basal angiosperms fossils. 1.14 fadlad_beu.csv: first and last age of occurrence of basal eudictos fossils. 1.15 fadlad_fern.csv: first and last age of occurrence of ferns fossils. 1.16 fadlad_gym.csv: first and last age of occurrence of gymnosperms fossils. 1.17 fadlad_mono.csv: first and last age of occurrence of monocots fossils. 1.18 fossil_info_v5.csv: information used to graft fossil species to the phylogenetic tree of extant species. sample_name = sample names; species = species name; genus = species genus; family = species family; order = species order; clade = species clade; bind = fossil species name to be binded to the tree; reference and reference 2 = position in the extant tree where the fossil species is to be grafted; tipage_low = first age of species fossil occurrence; tipage_up = last age of species fossil occurrence; poly = whether the group is dichotmomic or polytomic; link = to the resource used for fossil age and taxonomic position; resolution = fossil leaf image resolution in mm per pixel; leaf-section = whether the segmentation was done in the whole leaf or a leaf section; sister_taxa = sister taxa of the fossil species; database = database from where the fossil leaf images were obtained. 1.19 HLD_final_1000.csv: results of hierarchical loop decomposition (HLD) for all 1,000 plant taxa used in this study. file = sample file name; spp_name = species name; Genus = species genus; Family = species family; Order = species order; Clade = species clade; Database = database collection where the leaf image file was obtained from; rmin = vein width in mm; ER = loop elongation ratio; CR = loop circularity ratio; VD = vein density in mm mm-2; MST = minimum spanning tree ratio. 1.20 MST_large_rate_box.csv: evolutionary rates for minimum spanning tree ratio at large vein sizes (MST large) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.21 MST_medium_rate_box.csv: evolutionary rates for minimum spanning tree ratio at medium vein sizes (MST medium) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.22 MST_small_rate_box.csv: evolutionary rates for minimum spanning tree ratio at small vein sizes (MST small) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.23 node_clades_info.csv: information about the internal nodes in the phylogenetic tree for the 1,000 plant taxa investigated in this study. tree = tree file name; node_number = number of the internal node in each tree; obs = node label; median time (Mya) = median time for the node in million of years ago; Confidence interval = confindence interval for the node age in Mya; number of studies = supporting the median node age or resource used to obtain the node median age. 1.24 node_names.csv: names of the internal nodels in the phylogenetic tree for the 1,000 plant taxa investigated in this study. 1.25 nodes_ast.csv: names of the internal nodels in the phylogenetic tree for the asterids clade. 1.26 nodes_bangio.csv: names of the internal nodels in the phylogenetic tree for the basal angiosperms clade. 1.27 nodes_beu.csv: names of the internal nodels in the phylogenetic tree for the basal eudicots clade. 1.28 nodes_ferns.csv: names of the internal nodels in the phylogenetic tree for the ferns clade. 1.29 nodes_gym.csv: names of the internal nodels in the phylogenetic tree for the gymnosperms clade. 1.30 nodes_mono.csv: names of the internal nodels in the phylogenetic tree for the monocots clade. 1.31 nodes_rosids.csv: names of the internal nodels in the phylogenetic tree for the rosids clade. 1.32 time_series_final.csv: time-series for abiotic and biotic factors used in this study. time_id: time identification; time_slice = time slice; N = number of samples in each time slice; median_var_pun = median value of the disparity metric (sum variances) under the punctuated model of evolution; median_var_pun_25 = 25% confidence interval for median_var_pun; median_var_pun97 = 97% confidence interval for median_var_pun; median_cent_pun = median value of the disparity metric (median centroids) under the punctuated model of evolution; median_cent_pun25 = 25% confidence interval for median_cent_pun; median_cent_pun97 = 97% confidence interval for median_cent_pun; median_var_grad = median value of the disparity metric (sum variances) under the graduate model of evolution; median_var_grad_25 = 25% confidence interval for median_var_grad; median_var_grad97 = 97% confidence interval for median_var_grad; median_cent_grad = median value of the disparity metric (median centroids) under the graduate model of evolution; median_cent_grad25 = 25% confidence interval for median_cent_grad; median_cent_grad97 = 97% confidence interval for median_cent_grad; co2_median = reconstruction of median global atmospheric CO2 concentration during the Phanerozoic eon from https://doi.org/10.1016/j.earscirev.2021.103503; co2_lw95 = 95% lower confidence interval for co2 concentration; co2_up95 = 95% upper confidence interval for co2 concentration; co2_up68 = 68% upper confidence interval for co2 concentration; co2_lw68 = 68% lower confidence interval for co2; temp = reconstruction of mean global air temperature for the Phanerozoic eon fromhttps://doi.org/10.1038/ncomms14845; insect_rate_all = insect diversification rate from https://doi.org/10.1038/srep19208. 1.33 tree_final_1000.tre: phylognetic tree for all 1,000 plant taxa investigated in this study. 1.34 VD_large_rate_box.csv: evolutionary rates for vein density at large vein sizes (VD large) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.35 VD_medium_rate_box.csv: evolutionary rates for vein density at medium vein sizes (VD medium) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 1.236 VD_small_rate_box.csv: evolutionary rates for vein density at small vein sizes (VD small) for 1,000 extant and extinct plant taxa and 999 internal nodes of the phylogenetic tree. 2. scripts folder: 2.1 1_create_phylo_tree: R-script to produce the phylogenetic tree for the 880 extant species and to graft the 120 fossil species to it. 2.2 2_prepare_data_for_ace: R-script to prepare the dataset for the ancestral state reconstruction analysis (ace) of key leaf venation architecture traits across vein sizes. 2.3 3_run_ace: R-script to run the ancestral state reconstruction analysis using phylogenetic ridge regressions. 2.4 4_plot_ace: R-script to plot the results from the ancestral state reconstruction analysis. 2.5 5_overfitRR: R-script to run the bootstrapping analysis on the ancestral state reconstruction results. 2.6 6_run_disparity: R-script to run the disparity through time analysis. 2.7 7_time_series: R-script to run the generalized least square analysis to test for potential drivers of leaf venation network evolution. 3. figures folder: figures produced using the R-scripts above. 4. results folder: results produced using the R-scripts above. 5. CNNs folder (part1, part2, and part3): leaf segmentations produced using the LeafVeinCNN, where vein pixels are shown in white and non-vein pixels are shown in black. 6. masks folder (part1 and part2): leaf masks used during the LeafVeinCNN analysis to delimit the leaf boundaries. 7. R project: to run the R-scripts above. 8. Costa Rica samples: cleared leaf images collected from Costa Rica. 9. Ecuador samples: cleared leaf images collected from Ecuador. 10. UCMP samples: high resolution images from the University of California Museum of Paleontology.