Keywords: Elliptical Fourier Descriptors, heteroblasty, landmarks, leaf morphology, leaf shape, leaves, morphometrics,Passiflora, Procrustes analysis

2.1. Data description

Portions of these Materials and Methods are recapitulated from Chitwood and Otoni (2016a). The two manuscripts analyze different biological problems arising from the same dataset (Chitwood & Otoni,2016b). Recapitulating the methods here is meant to aid the reader and is in line with Committee on Publication Ethics (COPE) guidelines.

The 555 original scans used for analysis (Chitwood & Otoni,2016b) represent 40 different species ofPassiflora in which the order of leaves arising from the vine is recorded (starting with “1” for the youngest leaf scanned from the growing tip of each vine). We importantly note: the numbering of nodes in the raw scans described above (Chitwood & Otoni,2016b), starting at the tip of the shoot, is opposite from the numbering of nodes presented in the manuscript, in which numbering (starting with “1”) begins with the oldest leaf at the base of the shoot. The reason for this opposite numbering in the manuscript is that by beginning the counting of nodes with “1” at the shoot base the numbering aligns with the heteroblastic series (which begins with the first emerged leaf).

Data from more than 3,300 leaves, and the code to analyze the data and reproduce the figures in this manuscript, are provided, including both landmark data (found in the GitHub repository under PassifloraLeaves/Paper2/Figure2/1.all_landmarks/0.refromated_Proc_coord.txt), measuring the vasculature, lobes and sinuses and Elliptical Fourier Descriptor (EFD) data (found in the GitHub repository under PassifloraLeaves/Paper2/Figure2/0.harmonics/0.passiflora_nef.txt), which quantifies leaf outlines (Chitwood,2016).

2.2. Plant materials and growth conditions

Passiflora germplasm was kindly provided by R. Silva (Viveiros Flora Brasil, Araguari, MG, Brazil), Dr. F.G. Faleiro (EMBRAPA Cerrados, Planaltina, DF, Brazil), Prof. M.M. Souza (Universidade Estadual de Santa Cruz—UESC, Ilhéus, BA, Brazil), M. Peixoto (Mogi das Cruzes, SP, Brazil), Prof. M.L. Silva (Universidade do Estado de Mato Grosso, Tangará da Serra, MT, Brazil), and Prof. C.H. Bruckner (Universidade Federal de Viçosa, Viçosa, MG, Brazil).

The plants were germinated from seed, planted between late October 2015 and early March 2016, in Viçosa, at the Federal University of Viçosa, MG, Brazil. The populations were raised and maintained under polycarbonate‐covered glasshouse conditions, equipped with automatic environmental control using exhaust fans and evaporative cooling panels (with expanded clay wettable pads). Seeds for eachPassiflora species were sown in 128 cell propagation plastic trays (GPlan Comércio de Produtos Agrícola s EIRELI—ME, São Paulo, SP, Brazil) filled with horticultural organic Tropstrato HT Hortaliças substrate (Vida Verde Indústria e Comércio de Insumos Orgânicos Ltda, Mogi Mirim, SP, Brazil). After germination (30–40 days), plantlets were individually transplanted to 5‐L capacity plastic pots (EME‐A‐EME Ind. Com. Ltda., Petrópolis, RJ, Brazil) filled with horticultural substrate. Each pot received 5 g of Osmocote^® Plus NPK 15‐09‐12 3‐4 month controlled release fertilizer (Scotts, USA). Plants were irrigated on a daily basis with tap water, and no phytosanitary control was applied. Vines exhibited variable germination and growth rates. Leaves were harvested at a time point after which the first ten leaves had fully expanded so that analyses across these nodes to measure heteroblasty were comparable. For each species, the following number of vines was harvested:P. actinia 5,P. alata 7,P. amethystina 6,P. biflora 5,P. caerulea 3,P. capsularis 6,P. cincinnata 2,P. coccinea 6,P. coriacea 12,P. cristalina 7,P. edmundoi 5,P. edulis 3,P. foetida 6,P. galbana 7,P. gibertii 5,P. gracilis 2,P. hatschbachii 7,P. kermesina 3,P. ligularis 7,P. malacophylla 7,P. maliformis 5,P. micropetala 6,P. miersii 7,P. miniata 8,P. misera 8,P. mollissima 6,P. morifolia 2,P. mucronata 5,P. nitida 4,P. organensis 4,P. pohlii 1,P. racemosa 7,P. rubra 6,P. setacea 6,P. sidifolia 6,P. suberosa 13,P. tenuifila 4,P. tricuspis 8,P. triloba 6,P. villosa 4.

For scanning, a multifunction printer (Canon PIXMA MX340 Wireless Office All‐in‐One Printer, model 4204B019, USA) was used. A 20‐cm metallic ruler was positioned at the bottom of each scanned sheet as a size marker. Leaves were carefully detached, from the base to the tip of the shoot, and affixed to an A4 paper sheet, adaxial face down, using 12‐mm‐double‐sided tape (Scotch Model 9400, 3M do Brasil, SP, Brazil). The numbers written near each leaf indicate position in the shoot, in a tip‐to‐base direction, starting with the youngest leaf at the tip of the shoot.

2.3. Morphometric and statistical analyses

All morphometric data and code used for statistical analysis are available on GitHub (Chitwood,2016). Landmarks, as described in the text, were placed on leaves in ImageJ (Abràmoff, Magalhães, & Ram,2004). Procrustes superimposition was performed using the shapes package (Dryden,2015) in R (R Development Core Team,2016) with the procGPA function using reflect=TRUE.

To isolate outlines for Elliptical Fourier Descriptor (EFD) analysis, the “Make Binary” function in ImageJ (Abràmoff et al.,2004) was found to be sufficient to segment leaves. The wand tool was used to select individual binary leaf outlines, which were pasted into a new canvas, which was subsequently saved as an individual image, which was named by vine and node position from which the leaf was derived. The binary images were batch converted into RGB.bmp files and read into SHAPE, which was used to perform chain‐code analysis (Iwata, Niikura, Matsuura, Takano, & Ukai,1998; Iwata & Ukai,2002). The resulting chain‐code.chc file was then used to calculate normalized EFDs. The resulting normalized EFD.nef file was then read into Momocs (version 0.2‐6) (Bonhomme, Picq, Gaucherel, & Claude,2014) in R. The harmonic contributions to shape were visualized using the hcontrib function. Averaged leaf outlines were calculated using the meanShapes function.

Unless otherwise noted, all visualization was performed using ggplot2 in R (Wickham,2009). Analysis of variance (ANOVA) was performed using the aov function fitting the model trait ~ species*heteroblasty. Linear discriminant analysis (LDA) was performed using the lda function and subsequent prediction of species identity or heteroblastic node position performed using the predict function with MASS (Venables & Ripley,2002). For prediction, LDA was performed with CV=“TRUE,” which is a “leave‐one‐out” cross‐validation approach in which for each leaf an LDA is performed excluding the leaf after which it is assigned to the resulting LDA space it was excluded from. Hierarchical clustering was performed using the hclust function. t‐distributed Stochastic Neighbor Embedding (t‐SNE) was performed using the Rtsne package (Krijthe,2015) in R with perplexity=40.

3.1. Heteroblastic changes in Passiflora leaf shape

The >3,300 leaves analyzed in this study come from 40 different species ofPassiflora and were collected in order from the base of the vine onwards to the growing tip, keeping track of the node from which each leaf originated. Among the 40Passiflora species sampled, the degree of heteroblastic shape change is variable (Figure 1). Previously (Chitwood & Otoni,2016a), we defined seven species groups (classes) based on their clustering in a principal component analysis (PCA) and qualitative shape differences. Members of Class G, with lance‐shaped leaves, exhibit little to no changes in leaf shape across the leaf series, and similarly Class E and Class F leaves generally remain round or lance‐shaped regardless of shoot position. Members of Class C and Class D, which are characterized by different numbers of lobes later in the series, begin with a rounder, less lobed juvenile leaf shape at the base of the vine. Members of Class A and Class B, with wide “bat‐like” leaves, begin with a rounder leaf shape that transitions to a stereotypical wide leaf (Figure 1).

Generally, if heteroblastic changes in leaf shape are observed, the juvenile leaf form (at the base of the shoot) tends to be rounder and exhibit less lobing compared to adult leaves at the shoot tip (Figure 1b). To substantiate this statement, we used 15 landmarks measuring vascular patterning and the position of the lobes and sinuses, and Elliptical Fourier Descriptors (EFDs), quantifying the contours of leaves, to visualize shape changes through the heteroblastic series (Figure 2a,b). Landmarks arex,y coordinates that define the position of homologous points found on every leaf, such that leaf shape is represented as a set ofx,y points shared between samples (Bookstein,1997). EFDs convert shapes into chain code, a string of numbers that records pixel movements to perfectly recapitulate the shape (Freeman,1974). The chain code is treated as a wave function and decomposed into a harmonic series using a Fourier transform (Kuhl & Giardina,1982). The coefficients of the harmonic series represent shape and are used in subsequent analyses.

Averaged landmarked leaves (Figure 2C,D) and EFD‐derived outlines (Figure 2e) across the leaf series compared to the first leaf show that leaves become progressively more dissected. Qualitatively, most of these heteroblastic changes occur earlier in the leaf series, and the remainder of leaves later in the shoot show little further changes in shape. This result is consistent with the previous observation (Chitwood & Otoni,2016a) that shape features in juvenile leaves in the first nodes allow them to be correctly assigned to the predicted node at higher rates than leaves later in the series. We return to this idea later and test the hypothesis that across species juvenile leaves are more like each other than leaves later in the series that exhibit divergent shapes.

3.2. Allometric changes in leaf shape across the heteroblastic series

To explore the relative contributions of leaf subareas to differences in leaf shape across the heteroblastic leaf series in different species, we performed an allometric analysis. Changes in leaf shape, whether between species or within the heteroblastic series, often correlate linearly with size (Chitwood, Rundell et al.,2016); whether such relationships exist in this dataset remains an open question. Procrustes‐aligned leaves were divided into subregions (Figure 2f), including the areas of the midvein and proximal vein, and distal and proximal leaf blade areas. The rationale of using Procrustes‐aligned leaf shapes (which have been scaled, translated, and rotated to superimpose leaves for analysis) is to analyze the relative contributions of vein and blade area to total leaf area. When plotting the square root of each subregion area against the square root of the overall Procrustes‐aligned leaf area, clear linear relationships are observed (Figure 2g). Notably, blade areas expand at a higher rate compared to vein areas as the total leaf area increases (i.e., the slope of the blade subregions is greater than the vein subregions). That total area of the leaf occupied by blade expands at a faster rate than the area occupied by veins is consistent with the previous results observed in grapevines (Chitwood, Klein et al.2016; Chitwood, Rundell et al.,2016).

When the results are plotted for each species, the general trend of blade area expanding at the expense of vein area is observed, but there is a large amount of variation between species (Figure 3). For example,P. misera exhibits linear relationships in which both blade regions have similar slopes that are greater than the vein regions. Contrastingly, the distribution of total leaf area forP. caerulea is bimodal, and differences in each population of leaves contribute to widely different slopes between the distal and proximal blade subregions. The distinctness of eachP. caerulea subpopulation reflects the discrete transformation of leaf shape from entire to highly dissected and palmate (Figure 1a). This is reflected when the heteroblastic node number is projected onto the plots (Figure 4), revealing distinct populations of juvenile and adult leaves with different ratios of blade and vein areas that contribute to each leaf type. Other species vary in the extent that the heteroblastic series is defined by the linear allometric relationships contributing to differences in leaf shape across the leaf series.

Although overall blade subregions expand at faster rates compared to vein subregions, and this relationship is mostly linear across all species (Figure 2g), species vary widely in the relative ratios of these regions across allometric lines (Figure 3) and the heteroblastic series (Figure 4).

3.3. Statistical effects of species, heteroblasty, and interaction

Previously, we demonstrated using a linear discriminant analysis (LDA) that not only can the leaves ofPassiflora species be predicted regardless of the node from which they arise, but surprisingly too that the leaf node can be predicted regardless of the species (Chitwood & Otoni,2016a). This result is consistent with work in grapevine (Chitwood, Klein et al.2016; Chitwood, Rundell et al.,2016) and demonstrates that within the complex shapes of leaves, some attributes vary independently by species, whereas others by node position.

To more rigorously quantify these effects inPassiflora, we performed an analysis of variance (ANOVA) for eachx andy landmark coordinate as well as coefficients of the harmonic series resulting from an Elliptical Fourier Descriptor (EFD) analysis (Figure 2a,b). Each trait was modeled by species*heteroblasty, which takes into consideration the additive species and heteroblastic effects as well as their interaction (Figures 5 and6); that is, there are species effects (Figure 5, shape effects due to differences between species regardless of node), heteroblasty effects (topmost row of Figure 6, shape effects due to differences between nodes regardless of species), and species × heteroblasty interaction effects (all other rows besides the topmost of Figure 6, shape effects across nodes that vary between species). Each shape feature, whether landmarks (Figure 2a) or EFDs (Figure 2b), is modeled independently (see column headings for Figures 5 and6).P. actinia was set as the intercept, primarily because its species epithet is alphabetically first, but conveniently, the round shape ofP. actinia leaves means that as effect sizes are compared against this species (by definition,P. actinia effects sizes are 0), the magnitude of leaves of other species deviating from a round shape is being measured.

The largest effect on leaf shape is species (species effects are about a magnitude greater than other statistical effects; compare the legend for species effect size in Figure 5 to the legend for other effect sizes in Figure 6). The strongest effect sizes are seen inx landmark coordinates of the petiolar junction inP. micropetala andP. gracilis, likely owing to the unique intersection of veins in these species. But more general trends are also observable. Thex andy landmarks of the distal sinus and harmonic coefficients from the EFD analysis are widely affected in Class C (lobed species) and Class B (wide, wing‐shaped species), reflecting the deviation of these leaf shape classes from the round shape ofP. actinia leaves.

Heteroblasty and interaction effects (Figure 6) were about a magnitude less than species effects (compare legend between Figures 5 and6). The heteroblasty effect is negligible compared to the interaction effect (Figure 6, see top row). This does not mean there are no shape attributes modulated by node position independent from species effects, as we previously showed node position can be predicted independently from species identity (albeit to a much less degree and most strongly for the first two leaves in the heteroblastic series; Chitwood & Otoni,2016a). Rather, it indicates that heteroblastic‐independent effects on leaf shape are slight compared to species and interaction effects. The strongest species × heteroblasty interaction effects are found in the EFD harmonic coefficients ofP. cincinnata, the leaves of which are highly lobed. Generally, like species effects (Figure 5), the interaction effects are strongest for the lobed (Class C) and winged (Class A and Class B) leaf shapes (Figure 6).

Given that (i) we previously showed that node position can be predicted independently from species identity, although strongest for the first two leaves (Chitwood & Otoni,2016a) and (ii) that the interaction effect sizes between species and heteroblasty are much stronger than heteroblasty alone (Figures 5,6), we hypothesized that evolutionary differences in leaf shape between species ofPassiflora arise within a heteroblastic context, which we explore more thoroughly, below.

3.4. Divergent heteroblastic trajectories and similar juvenile leaf shapes

Many pieces of evidence suggest the earliest leaf shapes in the heteroblastic series are similar acrossPassiflora species and that leaves later in the series differ between species. When the average shape of leaves across the heteroblastic series is compared using landmarks (Figure 2d) and contours derived from Elliptical Fourier Descriptors (EFDs) (Figure 2e), it is evident that the more lobed leaf shape characteristic of later leaves is achieved within the first two nodes.

To test the idea that juvenile leaves at the base of the shoot are more similar between species than those later in the series, we performed a linear discriminant analysis (LDA) using both landmark and EFDs to discriminate leaves from each node by species identity (Figure 7a). Although there is wide variability of the ability to discriminate the leaves of each species, the LDAs using leaves from nodes 1 and 2 performed poorly in their ability to discriminate leaves by species compared to subsequent nodes. From the overall average correct reassignment rate for the LDA performed for each node (see bottom of Figure 7a), it is evident that leaves from nodes 1 and 2 have less distinctive features differentiating leaves from species compared to later nodes. This suggests that leaves from nodes 1 and 2 are more similar in shape between species than leaves from later nodes.

A more direct test of variability in leaf shape is to measure the standard deviation of the raw traits used to measure leaf shape. Shape features of the leaf were hierarchically clustered and the standard deviation for each across species for each node calculated (Figure 7b). Most shape features either have reduced or unchanged standard deviation values in the first 1–3 leaves of the series compared to later leaves. A minority of shape features, especially thex coordinate values for landmarks defining the petiolar junction and bases of the major veins (landmarks 1–6), have increased standard deviation values in the earlier nodes. The results suggest that generally leaf shape is less variable in leaves from the first nodes (Figure 7b) consistent with the observation that juvenile leaves discriminate species less than leaves later in the series (Figure 7a). Exceptionally, the landmarks defining the petiolar junction in thex coordinate direction are more variable in juvenile than adult leaves (Figure 7b).

To visualize the divergent heteroblastic trajectories leading to disparate leaf shapes betweenPassiflora species, we used a t‐distributed Stochastic Neighbor Embedding (t‐SNE) approach to reduce the dimensionality of the data (Figure 8a) (Krijthe,2015). t‐SNE separates species classes and benefits from no assumptions of linearity and reducing the data to strictly two dimensions. By doing so, each sampled vine can be visualized as a vector in two‐dimensional space, with the base and tip of the vector corresponding to the first sampled node at the base of the shoot and the furthest sampled node at the tip of the shoot, respectively (Figure 8b). Each vector, therefore, is a representation of the shape space traversed over nodes across the heteroblastic series.

The bases of the vectors representing each vine tend to cluster together, with similar Dimension 1 values but varying across Dimension 2. From this common region representing a shared juvenile leaf shape, the directions of each vector for different species classes vary, representing differing heteroblastic trajectories leading to disparate adult leaf shapes at nodes toward the shoot tips. To better visualize the divergent heteroblastic trajectories of each species class, each vector base was centered to the origin (Figure 8c). After centering, it is apparent that different species classes vary drastically—sometimes diametrically opposed—in the direction of their heteroblastic shape changes.

Collectively, the inability of leaves from the first nodes to successfully discriminate different species (Figure 7a), the reduced variability in shape features of leaves from the first nodes (Figure 7b) and the divergent heteroblastic trajectories between species classes (Figure 8) demonstrate that juvenile leaves betweenPassiflora species are similar and that divergent heteroblastic trajectories are responsible for the distinctive leaf shapes between species.

1. Abràmoff, M. D., Magalhães, P. J., & Ram, S. J. (2004). Image processing with ImageJ. Biophotonics International, 11(7), 36–42. [Google Scholar]
2. Allsopp, A. (1953a). Experimental and analytical studies of pteridophytes XIX. Investigations on Marsilea 2. Induced reversion to juvenile stages. Annals of Botany, 17, 37–55. 10.1093/oxfordjournals.aob.a083342 [DOI] [Google Scholar]
3. Allsopp, A. (1953b). Experimental and analytical studies of pteridophytes XXI. Investigations on Marsilea 3. The effect of various sugars on development and morphology. Annals of Botany, 17, 447–463. 10.1093/oxfordjournals.aob.a083362 [DOI] [Google Scholar]
4. Allsopp, A. (1954). Juvenile stages of plants and the nutritional status of the shoot apex. Nature, 173, 1032–1035. 10.1038/1731032a0 [DOI] [Google Scholar]
5. Allsopp, A. (1967). Heteroblastic development in vascular plants. Advances in Morphogenesis, 6, 127–171. 10.1016/B978-1-4831-9953-5.50008-1 [DOI] [PubMed] [Google Scholar]
6. Bonhomme, V., Picq, S., Gaucherel, C., & Claude, J. (2014). Momocs: Outline analysis using R. Journal of Statistical Software, 56, 1–24. 10.18637/jss.v056.i13 [DOI] [Google Scholar]
7. Bookstein, F. L. (1997). Morphometric tools for landmark data: Geometry and biology. Cambridge, UK: Cambridge University Press. [Google Scholar]
8. Chitwood, D. H. (2016). PassifloraLeaves. GitHub. Retrieved fromhttps://github.com/DanChitwood/PassifloraLeaves
9. Chitwood, D. H., Headland, L. R., Kumar, R., Peng, J., Maloof, J. N., & Sinha, N. R. (2012). The developmental trajectory of leaflet morphology in wild tomato species. Plant Physiology, 158, 1230–1240. 10.1104/pp.111.192518 [DOI] [PMC free article] [PubMed] [Google Scholar]
10. Chitwood, D. H., Klein, L. L., O'Hanlon, R., Chacko, S., Greg, M., Kitchen, C., … Londo, J. P. (2016). Latent developmental and evolutionary shapes embedded within the grapevine leaf. New Phytologist, 210, 343–355. 10.1111/nph.13754 [DOI] [PMC free article] [PubMed] [Google Scholar]
11. Chitwood, D. H., Kumar, R., Headland, L. R., Ranjan, A., Covington, M. F., Ichihashi, Y., … Sinha, N. R. (2013). A quantitative genetic basis for leaf morphology in a set of precisely defined tomato introgression lines. Plant Cell, 25, 2465–2481. 10.1105/tpc.113.112391 [DOI] [PMC free article] [PubMed] [Google Scholar]
12. Chitwood, D. H., & Otoni, W. C. (2016a). Morphometric analysis ofPassiflora leaves: The relationship between landmarks of the vasculature and elliptical Fourier descriptors of the blade. GigaScience, 6(1), 1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
13. Chitwood, D. H., & Otoni, W. C. (2016b). Supporting data for “Morphometric analysis of Passiflora leaves: The relationship between landmarks of the vasculature and elliptical Fourier descriptors of the blade”. GigaScience Database. 10.5524/100251 [DOI] [PMC free article] [PubMed] [Google Scholar]
14. Chitwood, D. H., Ranjan, A., Kumar, R., Ichihashi, Y., Zumstein, K., Headland, L. R., … Sinha, N. R. (2014). Resolving distinct genetic regulators of tomato leaf shape within a heteroblastic and ontogenetic context. Plant Cell, 26, 3616–3629. 10.1105/tpc.114.130112 [DOI] [PMC free article] [PubMed] [Google Scholar]
15. Chitwood, D. H., Ranjan, A., Martinez, C. C., Headland, L. R., Thiem, T., Kumar, R., … Sinha, N. R. (2014). A modern ampelography: A genetic basis for leaf shape and venation patterning in grape. Plant Physiology, 164, 259–272. 10.1104/pp.113.229708 [DOI] [PMC free article] [PubMed] [Google Scholar]
16. Chitwood, D. H., Rundell, S. M., Li, D. Y., Woodford, Q. L., Yu, T. T., Lopez, L. R., … Londo, J. P. (2016). Climate and developmental plasticity: Interannual variability in grapevine leaf morphology. Plant Physiology, 170, 1480–1491. 10.1104/pp.15.01825 [DOI] [PMC free article] [PubMed] [Google Scholar]
17. Chitwood, D. H., & Sinha, N. R. (2014). Plant development: Small RNAs and the metamorphosis of leaves. Current Biology, 24, R1087–R1089. 10.1016/j.cub.2014.10.013 [DOI] [PubMed] [Google Scholar]
18. Cushman, J. A. (1902). Studies of localized stages of growth in some common New England plants. American Naturalist, 36, 865–885. 10.1086/278228 [DOI] [Google Scholar]
19. Cushman, J. A. (1903). Studies of localized stages in some plants of the botanic gardens of Harvard University. American Naturalist, 37, 243–259. 10.1086/278284 [DOI] [Google Scholar]
20. Dell'aglio, D. D., Losada, M. E., & Jiggins, C. D. (2016). Butterfly learning and the diversification of plant leaf shape. Frontiers in Ecology and Evolution, 4, 81. [Google Scholar]
21. Dryden, I. L. (2015). shapes: Statistical Shape Analysis. R package version 1.1‐11. Retrieved fromhttps://CRAN.R-project.org/package=shapes
22. Feldman, L. J., & Cutter, E. G. (1970). Regulation of leaf form inCentaurea solstitalis L. I. Leaf development on whole plants in sterile culture. Botanical Gazette, 131, 31–39. 10.1086/336509 [DOI] [Google Scholar]
23. Freeman, H. (1974). Computer processing of line‐drawing images. ACM Computing Surveys (CSUR), 6(1), 57–97. 10.1145/356625.356627 [DOI] [Google Scholar]
24. Friedman, W. E., & Diggle, P. K. (2011). Charles Darwin and the origins of plant evolutionary developmental biology. Plant Cell, 23, 1194–1207. 10.1105/tpc.111.084244 [DOI] [PMC free article] [PubMed] [Google Scholar]
25. Gilbert, L. E. (1975). Ecological consequences of a coevolved mutualism between butterflies and plants. In Gilbert L. E., & Raven P. H. (Eds.), Coevolution of animals and plants (pp. 210–240). Austin, TX: University of Texas Press. [Google Scholar]
26. Gilbert, L. E. (1982). The coevolution of a butterfly and a vine. Scientific American, 247, 110–121. 10.1038/scientificamerican0882-1107134958 [DOI] [Google Scholar]
27. Goebel, K. (1908). Einleitung in die experimentelle Morphologie der Pflanzen. Leipzig, Germany: B.G. Teubner. [Google Scholar]
28. Goethe, J. W. (1952). Botanical writings. Trans. B. Mueller, introduction C. J. Engard. Honolulu, HI: University of Hawaii Press. [Google Scholar]
29. Iwata, H., Niikura, S., Matsuura, S., Takano, Y., & Ukai, Y. (1998). Evaluation of variation of root shape of Japanese radish (Raphanus sativus L.) based on image analysis using elliptic Fourier descriptors. Euphytica, 102, 143–149. 10.1023/A:1018392531226 [DOI] [Google Scholar]
30. Iwata, H., & Ukai, Y. (2002). SHAPE: A computer program package for quantitative evaluation of biological shapes based on elliptic Fourier descriptors. Journal of Heredity., 93, 384–385. 10.1093/jhered/93.5.384 [DOI] [PubMed] [Google Scholar]
31. Jones, C. S. (1992). Comparative ontogeny of a wild cucurbit and its derived cultivar. Evolution, 46, 1827–1847. 10.1111/evo.1992.46.issue-6 [DOI] [PubMed] [Google Scholar]
32. Jones, C. S. (1995). Does shade prolong juvenile development? A morphological analysis of leaf shape changes in Cucurbita argyrosperma subsp. sororia (Cucurbitaceae). American Journal of Botany., 82, 346–359. 10.2307/2445580 [DOI] [Google Scholar]
33. Klucking, E. P. (1992). Leaf venation patterns, Vol. 6 (pp. 222–262). Berlin, Germany: J. Cramer. Passifloraceae. [Google Scholar]
34. Krijthe, J. (2015). Rtsne: T‐distributed stochastic neighbor embedding using Barnes‐Hut implementation. R package version 0.10. Retrieved fromhttps://CRAN.R-project.org/package=Rtsne
35. Kuhl, F. P., & Giardina, C. R. (1982). Elliptic Fourier features of a closed contour. Computer Graphics and Image Processing, 18(3), 236–258. 10.1016/0146-664X(82)90034-X [DOI] [Google Scholar]
36. MacDougal, J. M. (1994). Revision of Passiflora subgenus Decaloba section Pseudodysosmia (Passifloraceae). Systematic Botany Monographs, 41, 1–146. 10.2307/25027834 [DOI] [Google Scholar]
37. Njoku, E. (1956). Studies in the morphogenesis of leaves XI. The effect of light intensity on leaf shape inIpomea caerulea. New Phytologist, 55, 91–110. 10.1111/nph.1956.55.issue-1 [DOI] [Google Scholar]
38. Njoku, E. (1971). Effect of sugars and applied chemicals on the heteroblastic development inIpomoea purpurea grown in aseptic culture. American Journal of Botany, 58, 61–64. 10.2307/2441307 [DOI] [Google Scholar]
39. R Development Core Team(2016). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Retrieved fromhttp://www.R-project.org [Google Scholar]
40. Roebbelen, G. (1957). Uber Heterophyllie beiArabidopsis thaliana (L.) Heynh. Berichte Der Deutschen Botanischen Gesellschaft, 70, 39–44. [Google Scholar]
41. Venables, W. N., & Ripley, B. D. (2002). Modern applied statistics with S, 4th ed.. New York, NY: Springer. 10.1007/978-0-387-21706-2 [DOI] [Google Scholar]
42. Wickham, H. (2009). ggplot2: Elegant graphics for data analysis. New York, NY: Springer‐Verlag. 10.1007/978-0-387-98141-3 [DOI] [Google Scholar]
43. Yang, L., Xu, M., Koo, Y., He, J., & Poethig, R. S. (2013). Sugar promotes vegetative phase change in Arabidopsis thaliana by repressing the expression of MIR156A and MIR156C. eLife, 2, e00260. [DOI] [PMC free article] [PubMed] [Google Scholar]
44. Yu, S., Cao, L., Zhou, C. M., Zhang, T. Q., Lian, H., Sun, Y., … Wang, J. W. (2013). Sugar is an endogenous cue for juvenile‐to‐adult phase transition in plants. eLife, 2, e00269. [DOI] [PMC free article] [PubMed] [Google Scholar]