Geometric Morphometrics of Leaf Blade Shape in Montrichardia Linifera (Araceae) Populations from the Rio Parnaba Delta, Northeas

bs_bs_banner

Botanical Journal of the Linnean Society, 2012, 170, 554–572. With 7 ﬁgures

Geometric morphometrics of leaf blade shape in Montrichardia linifera (Araceae) populations from the Rio Parnaíba Delta, north-east Brazil Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 MARIA FRANCILENE SOUZA SILVA1, IVANILZA MOREIRA DE ANDRADE1 and SIMON JOSEPH MAYO2*

1Departamento de Biologia, Universidade Federal do Piauí, Campus Ministro Reis Velloso, Avenida São Sebastião 2819, Parnaíba CEP 64202-020, Brazil 2Herbarium, Royal Botanic Gardens, Kew, Richmond TW9 3AE, UK

Received 13 February 2012; revised 25 June 2012; accepted for publication 28 August 2012

Leaf characters of populations of the aquatic macrophyte Montrichardia linifera were studied using geometric morphometrics to compare variation with traditional circumscriptions of the two recognized species. Two hundred and ten individuals were sampled from seven populations in the delta region of the Rio Parnaíba, north-east Brazil. Six landmarks of the leaf blade were digitized from images and analysed with MorphoJ software. Procrustes- aligned configurations were studied using principal component analysis and canonical variates analysis in the pooled data and individual populations. Sinus shape variation was studied using landmark configurations of the posterior lobe basiscopic lamina. Covariation of leaf blade shape, basiscopic lamina shape, secondary vein number and petiole ligule length was investigated with partial least squares analysis. Allometry of these variables with leaf blade centroid size was investigated using multivariate regression, linear modelling and analysis of covariance. Measured variables varied continuously over the ranges previously reported for the two species. The characters of the two species morphotypes covaried and were only partly influenced by allometric effects. Symmetric shape variables predominated, but a distinctive left- and right-handed asymmetry occurred in all populations. Genetic and ecological studies are needed to investigate the significant inter-population differences further. The study offers a methodology for a broader combined morphometric/molecular investigation. © 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572.

ADDITIONAL KEYWORDS: allometry – Maranhão – MorphoJ – multivariate analysis – Piauí – taxonomy.

INTRODUCTION Carlsen, 2007; Nadruz Coelho, 2010; D. Barabé, pers. comm.), but some (e.g. Howard, 1979a, b; Bunting, The genus Montrichardia Crueg. consists of large, 1980; Lins, 1994; Bunting, 1995; Lins & Oliveira, freshwater aquatic plants (helophytes) widely dis- 1995) have noted that in natural populations the tributed in the Neotropics from Mexico (Croat, two main species morphotypes are sympatric with Fernández-Concha & González, 2005) to the state of many intermediate forms. Because of this variation, Rio de Janeiro in Brazil (Mayo, Bogner & Boyce, Howard (1979b: 286) and Bunting (1980: 209) con- 1997; CATE-Araceae in Haigh et al., 2009). Two sidered that only one taxon was recognizable at species, M. arborescens (L.) Schott and M. linifera species rank. Lins (1994) drew attention to different (Arruda) Schott, are recognized by most authors ecological preferences, M. linifera colonizing banks of (Engler, 1911; Jonker-Verhoef & Jonker, 1953; Croat newly deposited river silt and M. arborescens occur- & Lambert, 1986; Croat, 1994; Bunting, 1995; Boggan ring at a later stage of vegetational succession under et al., 1997; Croat & Grayum, 1999; Aymard, Croat & greater competition from other plant species, leaving open the possibility that the two species are merely *Corresponding author. ecotypes. Simmonds (1950a: 400, 1950b: 288) recog- E-mail: [email protected] nized M. arborescens var. aculeata (Meyer) Engl.

554 © 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 MONTRICHARDIA LEAF MORPHOMETRICS 555 as a plant of greater height, more spiny stems, rough- crucial for understanding the ecology of this aquatic ened petioles and basal ribs with a longer denuded plant. zone (‘veins more exposed in the sinus’). He suggested Reproductive morphology and ecology of Montri- that var. aculeata preferred more saline (brackish) chardia have been studied intensively in French water, a view also expressed by Lins (1994), who also Guiana by Gibernau et al. (2003), Chouteau, Barabé observed that the most aculeate forms occurred in & Gibernau (2006), Boubes & Barabé (1997), Barabé coarse sand substrates. & Lacroix (2001) and Barabé, Lavallée & Gibernau We sought to answer the following questions. (1) (2008). Weber & Halbritter (2007) reported an Are distinct morphotypes recognizable that corre- unusual exploding pollen behaviour in Montrichar-

spond to the diagnoses of the two species? (2) Does the dia. The chemical toxicity of Montrichardia has Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 range of variation of the selected characters include been studied by Amarante et al. (2009) and the chem- the diagnostic values of the two species? (3) Are istry of leaf decomposition and mineralization there significant differences between populations? (4) by Bianchini, Pacobahyba & Cunha-Santino (2002) Can any of the character trends be explained by and Cunha-Santino, Pacobahyba & Bianchini allometry? (2003, 2004). Herrera et al. (2008) reported a well- documented leaf fossil of Montrichardia from the TAXONOMIC CHARACTERS OF THE TWO SPECIES Palaeocene of Colombia. Landmark geometric morphometrics has been Taxonomic treatments with descriptions and keys to applied to only a limited extent in comparative both species have been published by Engler (1911), studies of leaf shape (e.g. McLellan & Endler, 1998; Jonker-Verhoef & Jonker (1953), Croat & Lambert Jensen, Ciofani & Miramontes, 2002; Volkova & (1986) and Bunting (1995). The distinctions used by Shipunov, 2007; Viscosi et al., 2009; Magrini & Scop- these authors to separate the two species are given in pola, 2010; Viscosi, Loy & Fortini, 2010; Volkova Table S1. The most frequently used characters are et al., 2011), despite its considerable potential as leaf shape, number of secondary veins (= primary presented recently by Viscosi & Cardini (2011) and lateral veins) on each side of the leaf anterior lobe and Klingenberg et al. (2012). This is the first such study the shape of the sinus between the posterior lobes; on a taxon of Araceae, although previous studies have these are the characters we have studied most inten- been made using the morphometric techniques of sively. Other less emphasized characters are the eigenshape analysis (Ray, 1992) and elliptic Fourier length of the petiole sheath ligule (cusp) (Jonker- analysis (Andrade et al., 2008, 2010; Andrade & Verhoef & Jonker, 1953; Croat & Lambert, 1986), Mayo, 2010; Soares et al., 2011). presence of prickles on the stem (Engler, 1911; Croat Montrichardia plants are adapted to colonize the & Lambert, 1986), plant stature (Bunting, 1995) and silt that accumulates along river margins (Lins, stem thickness (Croat & Lambert, 1986). Of these, 1994; Lins & Oliveira, 1995). The stem consists of a only ligule length has been studied in detail here. subterranean branching rhizome that produces numerous vertical branches, the visible stems of the PREVIOUS STUDIES colony. Mature vertical stems can reach up to 6 m in This is the first comparative study of quantitative leaf height or more and typically bear a terminal crown morphology in Montrichardia. Conventional taxo- of three to six leaves; the leaves produced during its nomic descriptions of one or both species are given by main extension growth are soon deciduous leaving Engler (1911), Standley (1928, 1944), Jonker-Verhoef a bare stem with conspicuous annular scars. After & Jonker (1953), Standley & Steyermark (1958), the first flowering, a new article (module) is formed Croat (1978), Howard (1979a), Bunting (1980), Croat from the penultimate foliage leaf before the spathe, & Lambert (1986), Mayo (1986), Bunting (1995), so that the mature plant may create a relatively Martínez & Croat (1997), Croat & Stiebel (2001), short apical sympodium (Blanc, 1978), but these sec- Fournet (2002) and Grayum (2003). Other previous ondary flowering articles have much shorter inter- studies relevant to our work are those of Blanc (1978) nodes. Even when there is a sympodium present, on shoot architecture, Keating (2003) and Macedo the leaves are always found as a crown on the most et al. (2005) on vegetative anatomy and Lins (1994) distal article. and Lins & Oliveira (1995), who studied root We carried out a quantitative study of variation in morphology and anatomy of populations of Montri- morphological characters used by taxonomists to dis- chardia near Belém, Pará state, in Brazilian Amazo- tinguish the two species in a series of seven Montri- nia. The study of Lins & Oliveira is especially chardia populations in one area, the delta region of important, not only because ecological details of both the Rio Parnaíba, a major river that forms the border taxa are compared, but also because of the emphasis between the states of Piauí and Maranhão in north- on seed development and root anatomy, which are east Brazil. Our approach was to focus on diagnostic

Table 1. Sampled populations of Montrichardia linifera from the states of Piauí and Maranhão, Brazil. Thirty plants were sampled from each population, one mature leaf from each plant. Voucher specimens deposited at the Herbário MRV (‘Profa. Ana Maria Giulietti’), Universidade Federal do Piauí, Campus Ministro Reis Velloso, Parnaíba, Piauí, Brazil

Municipality and Abbreviated state Full locality name locality name Voucher specimens Coordinates

Parnaíba, Piauí Igarapé do Alto do Batista AltoBatista M.F.S. Silva 183 2°54.8′S, 41°47.7′W Ilha Grande, Piauí Ilha das Batatas IlhaBatatas M.F.S. Silva 336 2°49.9′S, 41°49.9′W Parnaíba, Piauí Igarapé das Parelhas Parelhas M.F.S. Silva 129 2°54.6′S, 41°47.8′W Parnaíba, Piauí Antigo Braço do Rio Igaraçu RioIgaraçu M.F.S. Silva 230 2°55.7′S, 41°46.4′W Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 Água Doce, Maranhão São Raimundo SãoRaimundo M.F.S. Silva 327 3°4.7′S, 42°15.0′W Ilha Grande, Piauí Tatus Vala TatusVala M.F.S. Silva 234 2°49.8′S, 41°49.8′W Tutóia, Maranhão Tutóia Araticum TutóiaAraticum M.F.S. Silva 299 2°46.8′S, 42°20.9′W

morphological characters, gather data from multiple populations, quantify the selected variables and study their range and covariance in the pooled and individual populations. A secondary aim of the study was to provide a detailed quantitative description of the Montrichardia populations in one area, which can be used for future comparison with populations in other regions of tropical America.

MATERIAL AND METHODS STUDY AREA AND POPULATIONS Seven populations were sampled in the delta region of the Rio Parnaíba in the Brazilian states of Piauí and Maranhão (Table 1, Fig. 1). Stands of Montrichardia are abundant throughout the delta, both along the main river channels and along the banks of the quieter water in the narrow and tortuous igarapés that criss-cross the region. The sampled populations were either close to the urban environment of the city of Parnaíba (AltoBatista, Parelhas, RioIgaraçu) or in the neighbourhood of a small river port (IlhaBatatas, TatusVala) or more distant from population centres (SãoRaimundo, TutóiaAraticum). The Rio Parnaíba forms the border between the states of Piauí and Maranhão and its delta is a major geographical feature of the Brazilian coastline (Fig. 1), with the main river dividing there into ﬁve Figure 1. Location of the study populations. A, map of major channels that debouche directly into the open Brazil showing the state limits of Maranhão and Piauí, sea [Igaraçu, Canárias, Caju, Carrapato (Melancieira) with the study area in the rectangle. B, delta region of the and Tutóia]. A labyrinthine network of smaller chan- Rio Parnaíba showing the location of the populations as nels (igarapés) forms c. 80 alluvial islands on an squares, and the location of the city of Parnaíba. See extensive ﬂuvial–marine plain, with large mobile Table 2 for further details of populations. sand dunes in some areas and extensive stands of restinga (Santos Filho, 2009), mangrove and car- naúba palm (Copernicia prunifera (Mill.) H.E.Moore) tropical semi-arid, hot and humid in the rainy season with a rich fauna (Silva, 2004; Lustosa, 2005); it is a (January to May) and with a 6-month dry season. sanctuary for many species of migratory birds (MMA/ Mean annual rainfall is > 1200 mm, mean maximum SDS, 2002) and an important commercial source for temperature is 32 °C and mean minimum 22 °C (Pre- crab meat (Crepani & Medeiros, 2005). The climate is feitura Municipal de Parnaíba, 2012). The region is

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 MONTRICHARDIA LEAF MORPHOMETRICS 557 considered important for the development of ecotour- six landmarks in all 210 leaves of the sample was ism (Silva, 2004). carried out using the software tpsDig ver. 2.16 (Rohlf, 2010a). Landmarks 1, 2, 4 and 5 represent the petiole–leaf blade junction, the apex of the anterior LEAF SAMPLING lobe and the apices of the two posterior lobes, respec- Leaves were sampled from the mature crowns of tively. These landmarks mapped the relative positions individual aerial stems. In this report, we present of the most obvious and stable structural features of results based on a subsample of 30 leaves from each the leaf. The three apical landmarks correspond to population, in which one leaf was chosen from the the endpoints of the major vascular structures of the crown of a different plant, avoiding leaves with

leaf blade. Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 damaged or incomplete blades. In less extensive popu- Although the secondary veins are conspicuous lations, leaves were taken from plants 1–2 m apart, structures, their homology relations in comparing one and in larger populations at c. 10 m apart, to reduce leaf to another are not easy to establish in the the possibility of re-sampling the same clone. Each absence of ontogenetic studies because of their vari- leaf was pinned to a horizontal board on a white ation in number. We therefore selected the vein on paper background using drawing pins alongside a each side that arose nearest to the petiole–leaf blade ruler scale and images were captured with a centrally junction (Fig. 2A) and used their apical positions as positioned Sony Cybershot (14.1 megapixel) digital landmarks 3 and 6. These two landmarks were posi- camera. tioned on the leaf margin closest to the junction between the secondary vein and the submarginal collective vein that connects the distal ends of the MORPHOLOGICAL CHARACTERS secondary veins. The point of origin of these two Leaf blade shape central veins varied from the midrib to the basal ribs Leaf shape was quantiﬁed by using the geometric or exactly at the angle between midrib and basal rib morphometric (GMM) analysis of landmarks, a well- (Fig. 2A). The position of landmarks 3 and 6 reﬂected based mathematical and statistical methodology (e.g. both the relative width of the leaf blade in its central Dryden & Mardia, 1998; Zelditch et al., 2004; Klin- region and the magnitude of the angle between the genberg, 2011; Viscosi & Cardini, 2011; Klingenberg veins and midrib. et al., 2012) with a range of free software available for carrying out computations and visualizing results (e.g. Rohlf, 2010a, b, c; Klingenberg, 2011). Basiscopic lamina of posterior lobes We selected six landmarks to capture the main The shape of the sinus is conditioned by the width of features of the leaf shape (Fig. 2A). Digitization of the lamina expansion growth on the basiscopic side of

Figure 2. Leaf blade of Montrichardia as seen in abaxial view. A, six leaf blade landmarks are numbered in boxes and shown as circles. CA, distance from leaf apex to tip of right-hand posterior lobe; CB, length of midrib; CLD, length of right-hand basal rib from petiole junction (Landmark 1) to lobe tip (Landmark 5); CLE, length of left-hand basal rib from petiole junction (Landmark 1) to lobe tip (Landmark 4); LJ, width of leaf blade at petiole junction; LPONT, distance between posterior lobe tips (Landmarks 4 and 5). B, twelve landmarks of the basiscopic lamina. Leaves not to scale.

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 558 M. F. S. SILVA ET AL. Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019

Figure 3. Variation in leaf blade shape in Montrichardia linifera populations. A, RioIgaraçu (leaf 218-A). B, TatusVala (leaf 239-A). C, TutoiaAraticum (leaf 285-A). D, SãoRaimundo (leaf 312-A). E, IlhaBatatas (leaf 350-A). F, AltoBatista (leaf 438-A). G, RioIgaraçu (leaf 210-A). H, Parelhas (leaf 421-A). J, TatusVala (leaf 250-A). Leaves not to scale. each posterior lobe (the side on the basal side of the the apex of the basal rib (= apex of the posterior lobe), basal ribs) and by the angle between the two basal respectively. The basiscopic margin was then digitized ribs (Fig. 3). In leaves with broad basiscopic laminae with tpsDig ver. 2.16 (Rohlf, 2010a) using the ‘curve’ the posterior lobes overlap making it impossible to tool and normalized to ten equally spaced points capture the shape of the sinus in its entirety from using the ‘Resample curve’ feature. The TPS file thus images. In order to obtain a quantitative measure of generated of all 210 leaves was passed through the the sinus shape we therefore used the basiscopic tpsUtil software (Rohlf, 2010b) using the feature lamina of only the left-hand posterior lobe. When the ‘Append tps Curve to landmarks’, which converted right-hand lobe was uppermost and overlapped the the curve points to pseudo-landmarks 3 to 12. left-hand lobe, the image was flipped prior to digitization and hence the reflection of the right-hand lobe Secondary vein number was used instead. Landmarks 1 and 2 (Fig. 2B) cor- Secondary vein number was counted for each leaf responded to the junction of basal rib and midrib, and from the captured images. Vein number was recorded

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 MONTRICHARDIA LEAF MORPHOMETRICS 559 on both sides of each leaf as follows: (1) all veins minant function analysis (DFA), respectively, of the arising from the midrib and basal rib combined Procrustes-aligned conﬁgurations. This allowed sepa- (ToAcroL and TotAcroR); (2) only veins arising from rate analysis of the symmetric and asymmetric com- midrib, including veins arising exactly at the angle ponents and all pairwise comparisons of populations. between midrib and basal rib (AntL and AntR); (3) DFA also provided a ‘leave-one-out’ cross-validation only veins arising on the basiscopic side of the basal test for each population pair comparison, free from rib (PostBasiL and PostBasiR). Veins in category (3) the assumption that the covariance structure within were almost always much thinner than those in the all the groups is the same (Klingenberg, 2011). other categories. Thinner veins lying between the EXCEL (Microsoft Office Excel 2003) with STAT-

main secondaries (‘interprimaries’) were often seen in PLUS add-in (Berk & Carey, 2000) was used for Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 categories (1) to (2) but were not counted. calculating univariate statistics of linear measurements. R statistical software (R Development Core Ligule length and linear measurements of the Team, 2011, ver. 2.12.2) was used to carry out regres- leaf blade sion analysis of linear measurements. ANCOVA and The ligule is a projecting free lobe at the apex of the ANOVA were carried out using BIOMstat software petiole sheath. Linear measurements (in cm) of ligule (Rohlf & Slice, 2003) to investigate allometric effects (Clig) and leaf blade were made using a ruler in in the variables describing basiscopic lamina shape, freshly collected leaves; all measurements were made secondary vein number and ligule length. PAST by the first author. On the leaf blade, the distance (Hammer, Harper & Ryan, 2001, ver. 2.02) was used from the apex of the anterior lobe to the apex of the for Mantel tests of matrix correlation, ANOVA and right-hand (seen abaxially) posterior lobe (CA) was Kruskal–Wallis tests of population differences. measured and used to scale the images digitized for Two-block partial least-squares analysis (PLS, Zeld- landmarks (Fig. 2A). The precise definition of the itch et al., 2004), as implemented in MorphoJ software measurement of leaf length and width in the pub- (Klingenberg, 2011), was used to investigate the cov- lished taxonomic descriptions is ambiguous as the ariation between the symmetric component of shape endpoints of the measured segments are not indi- variables derived from the six-landmark leaf shape cated. At least two possibilities for length are CA or configurations as block 1, with block 2 composed of by summing the length of the midrib and one (or the various combinations of shape variables derived from mean of both) basal ribs. Width is also not straight- the 12-landmark basiscopic lamina configurations, forward. The point of widest width is often different secondary vein characters and ligule length. on each side of the blade and has no consistent landmark associated with it. We used the width at the petiole junction (LJ), roughly perpendicular to the RESULTS midrib as the latter not always straight. We also LEAF BLADE SHAPE recorded the distance between the posterior lobe apices (LPONT), midrib length (CB) and lengths of Symmetric variation left-hand (CLE) and right-hand (CLD) basal ribs, as Visual inspection showed that leaf blade shape varied viewed abaxially. The measurements made are pre- considerably across the pooled sample between sented in Table S3. cordate, sagittate and intermediate forms (Fig. 3). The Procrustes fit of the full data set of pooled populations separated the shape variation (as expressed DATA ANALYSIS by the tangent sums of squares) into a symmetric The six-landmark configurations representing the leaf component, which explained 69.2% of the total, and blade shape were analysed using MorphoJ software an asymmetric component, which explained 30.8%. (Klingenberg, 2011). This performs a full Procrustes The main trends of shape variation in the pooled fit and separates the variation into symmetric and population sample of 210 leaves were shown by the asymmetric components of aligned configurations principal components (PCs) of the symmetric covari- that can then be studied separately. Covariance ance matrix; the first three PCs explained 96.2% of matrices of the two components were computed and the total symmetric variation (Table 2). The trends of subjected to principal component analysis (PCA), shape variation expressed by the four PCs are shown which produced four principal component (PC) shape in Figure 4A–D and the plot of the first two PCs in variables. Separate Procrustes fits and PCAs of each Figure 5A, illustrating the distributions of population population were also carried out. means. The symmetric shape variables consist of a The relative similarity and discrimination of decomposition of the leaf shape variation, with the the seven populations was analysed with MorphoJ following trends from low to high values on each axis: using canonical variates analysis (CVA) and discri- (1) angle of divergence of the basal ribs increases

(Fig. 4A; PC1) accompanied by a slight foreshortening divergence of the basal ribs and slight lengthening of the leaf and a more acute angle of the lateral of anterior lobe and shortening of posterior lobes secondary veins; (2) diverging angle between the (Fig. 4D; PC4). These four shape variables, as princi- lateral secondary veins and midrib (Fig. 4B; PC2) pal components, are independent by deﬁnition. accompanied by diverging basal ribs and slight Box plots of the individual populations on the ﬁrst lengthening of the anterior lobe and shortening of three PCs (Fig. 6A–C) show that the Parelhas popula- posterior lobes; (3) narrowing of the leaf in the tion stands out in having on average the least diver- central region and slight overall leaf lengthening gent basal ribs (PC1, low values), most widely angled (Fig. 4C; PC3); (4) vertical relative movement of the central secondary veins (PC2, high values) and rela-

petiole–blade junction, accompanied by a slight tively narrowest leaves (PC3, high values). The Tatus- Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 Vala population is noteworthy in having on average the most divergent basal ribs (PC1, high values). Table 2. Leaf blade shape variables in Montrichardia Canonical variate analysis (in MorphoJ) of the sym- linifera. Principal component analysis (PCA) of the symmetric component showed that the most important metric and asymmetric components of variation in 210 shape trends separating the populations were in a Procrustes-aligned six-landmark configurations of leaf different order of importance. CV1 (Fig. 5B, 57.0% of blades. Computed using MorphoJ (Klingenberg, 2011) symmetric variation) expressed a somewhat similar shape trend to PC2, also reflected by the São Principal Cumulative Raimundo and Parelhas populations having the components Eigenvalues % variance % variance lowest and highest average values on these two axes, respectively (Fig. 6B, D). CV2 (22.3% of variation) Symmetric component (69.2% of total tangent sums of squares) PC1 0.00310590 46.043 46.043 was similar to PC3, and CV3 (17.2% of variation) was PC2 0.00238874 35.411 81.454 similar to PC1, the first three CV axes thus account- PC3 0.00099411 14.737 96.191 ing for 96.6% of the among-group variation. PC4 0.00025697 3.809 100.000 The inter-population morphological distances Asymmetric component (30.8% of total tangent sums of squares) (Mahalanobis and Procrustes, Table 3) showed that PC1 0.00222079 74.446 74.446 Parelhas was the most divergent population overall PC2 0.00042273 14.171 88.617 and IlhaBatatas the least, as also shown by the PC3 0.00027592 9.249 97.866 PC4 0.00006365 2.134 100.000 positions of these populations on the CV1 axis alone (Fig. 6D). Procrustes distances between populations

Table 3. Difference in leaf blade shape among seven populations of Montrichardia linifera. Mahalanobis and Procrustes distances computed from a canonical variates analysis of the symmetric component of a matrix of 210 Procrustes-aligned six-landmark conﬁgurations of leaf shape. Each population sample consisted of 30 individuals. P-values for the signiﬁcance of the interpopulation distances were computed using permutation tests (10 000 replications). Computed with MorphoJ (Klingenberg, 2011)

AltoBatista IlhaBatatas Parelhas RioIgaraçu SãoRaimundo TatusVala TutoiaAraticum

Mahalanobis distances: upper triangle are P-values; lower triangle are distances between populations AltoBatista 0.1472 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 IlhaBatatas 0.5897 < 0.0001 0.0002 < 0.0001 < 0.0001 0.0195 Parelhas 2.013 2.3318 < 0.0001 < 0.0001 < 0.0001 < 0.0001 RioIgaraçu 1.6001 1.1452 2.9036 < 0.0001 < 0.0001 0.1758 SãoRaimundo 1.6906 1.5092 3.2691 2.0557 < 0.0001 < 0.0001 TatusVala 1.6855 1.6786 2.4382 1.6515 1.863 < 0.0001 TutoiaAraticum 1.4039 0.8291 2.8271 0.6378 1.7171 1.8378 Procrustes distances: upper triangle are P-values; lower triangle are distances between populations AltoBatista 0.27 < 0.0001 < 0.0001 < 0.0001 < 0.0001 0.0009 IlhaBatatas 0.0183 < 0.0001 0.001 0.0001 < 0.0001 0.1383 Parelhas 0.0517 0.0669 < 0.0001 < 0.0001 < 0.0001 < 0.0001 RioIgaraçu 0.0525 0.0413 0.085 < 0.0001 < 0.0001 0.1252 SãoRaimundo 0.0723 0.0609 0.1143 0.0695 < 0.0001 0.0012 TatusVala 0.0803 0.077 0.0946 0.0555 0.0763 < 0.0001 TutoiaAraticum 0.0427 0.0253 0.0862 0.0267 0.0546 0.0714

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 MONTRICHARDIA LEAF MORPHOMETRICS 561 Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019

Figure 4. Leaf blade. A–D, symmetric principal components (PCs), showing shapes for scores of -0.1 (left) and +0.1 (right). A, PC1 (46.0% of total symmetric variation). B, PC2 (35.4% variation). C, PC3 (14.7% variation). D, PC4 (3.8% variation) showing shapes for scores of -0.05 (left) and +0.05 (right). E, consensus configuration (centroid). F, asymmetric PC1 (74.5% of total asymmetric variation) showing shapes for scores of -0.08 (left) and +0.08 (right). G, basiscopic lamina. PC1 showing shapes for scores of -0.15 (left) and +0.15. H–J, first pair of singular axes of partial least squares analysis of leaf blade and basiscopic lamina. H, block 1, leaf blade showing shapes for scores of -0.1 (left) and +0.1 (right). J, block 2, basiscopic lamina showing shapes for scores of -0.1 (left) and +0.1 (right). Pooled data from 210 individuals. Computed with MorphoJ (Klingenberg, 2011). were highly significant in all pairwise compari- There was no correlation between geographical and sons except AltoBatista–IlhaBatatas, IlhaBatatas– morphological distance among the populations Tutoia-Araticum and RioIgaracu–TutoiaAraticum. (Mantel test in PAST: correlation R =-0.017, P Mahalanobis distances were similar, but the (uncorrelated) = 0.540). IlhaBatatas–Tutoia-Araticum pair was also significantly different. Asymmetry The cross-validation results of population pairwise In a PCA of the asymmetry covariance matrix, the discriminant function analysis (see also Support- first PC accounted for 74.5% of total asymmetric ing Information, Table S2) showed misclassifica- variation (Table 2) and expressed a left-handed–right- tion rates ranging from a high value of 46.7% handed asymmetry (Fig. 4F), which had a distinct (RioIgaraçu–TutoiaAraticum) to a low value of 3.3% two-cluster structure in the data and was also (TutoiaAraticum–Parelhas, Parelhas–SãoRaimundo). expressed in each population (Fig. 7). A separate

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 562 M. F. S. SILVA ET AL. Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019

Figure 5. See caption on next page.

Figure 5. Leaf shape variables in Montrichardia linifera. A, principal component (PC) analysis of symmetric matrix of Procrustes-aligned six-landmark leaf blade configurations. Plot of PC1 46.0% of total symmetric variation) and PC2 (35.4% variation) showing 90% confidence ellipses of population means. B, canonical variate analysis of the same matrix. Plot of CV1 (57.0% of symmetric variation) and CV2 (22.3% variation) showing 90% confidence ellipses of population means. Shapes corresponding to low and high values on the axes are shown in their corresponding positions. Al: AltoBatista, IB: IlhaBatatas, Pa: Parelhas, Ri: RioIgaraçu, Sã: SãoRaimundo, Ta: TatusVala, Tu: TutoiaAraticum. Pooled data from 210 individuals. Computed with MorphoJ (Klingenberg, 2011). ᭣ Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019

Figure 6. Comparison of leaf characters among seven populations of Montrichardia linifera. A–C, principal component (PC) analysis of symmetric Procrustes-aligned six-landmark leaf blade configurations: A, PC1 (see Fig. 4A). B, PC2 (see Fig. 4B). C, PC3 (see Fig. 4C). D, canonical variate analysis of symmetric Procrustes-aligned six-landmark leaf blade configurations: CV1 (see Fig. 5B). E, principal component analysis of Procrustes-aligned twelve-landmark basiscopic lamina configurations: PC1 (see Fig. 4G). F, number of secondary veins in the basiscopic lamina (see Table 6). Thirty individuals sampled per population. Population names on x-axis (see Table 2). Small circles represent outliers. Thick horizontal lines represent median values. Box plots computed and plotted in R (R Development Core Team, 2011).

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 564 M. F. S. SILVA ET AL. Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019

Figure 7. Leaf shape in Montrichardia linifera, showing separation of left- and right-handed groups on the PC1 (74.5% of total asymmetric variation) axis. Shapes are shown for scores of -0.08 (left), +0.08 (right) and the mean shape in the centre (score 0.00). The 30 individuals of the RioIgaraçu population are shown as larger (black) dots superimposed on smaller (light blue) dots representing the remaining 190 individuals of the study sample. Data are six-landmark configurations of leaf shape in seven populations each of 30 individuals. Computed in MorphoJ (Klingenberg, 2011). analysis of the RioIgaraçu population (details not and high values narrow ones. The TatusVala and given here), which included all mature leaves of the Parelhas populations, with the narrowest basiscopic plants sampled (82 leaves from 31 plants), showed laminas, showed the highest PC1 values (Fig. 6E). that the handedness of each plant is constant. Each Visual inspection (Fig. 3) showed that sinus shape population consisted of a mixture of left- and right- varied widely across the populations, from completely handed plants, as indicated in Figure 7, in which the closed (posterior lobes overlapping) to very open. individuals of the RioIgaraçu population are shown Partial least-squares analysis showed significant divided between the two clusters. All the other popu- covariation between the symmetric component of leaf lations showed a comparable division. blade shape and basiscopic lamina shape; 96.7% of CVA of the asymmetry component showed that the total covariation was expressed by the first pair of asymmetric variation significantly separated only a singular axes (PLS 1, RV coefficient = 0.16), in a few population pairs. Comparison of individual popu- highly significant correlation (P < 0.0001 permutation lations showed that asymmetric variation varied from test 10 000 replicates). These two correlated shape 26.9% (SãoRaimundo) to 46.0% (Parelhas) of the total variables expressed a relationship such that, as the tangent sum of squares. angle between the central secondary veins and the midrib increases (Fig. 4H), the basiscopic lamina becomes narrower (Fig. 4J). The leaf shape variable BASISCOPIC LAMINA SHAPE – SINUS BETWEEN expressed in the PLS analysis (block 1, Fig. 4H) is POSTERIOR LOBES most similar to the second symmetric shape variable Principal component analysis of the 12-landmark of the PCA analysis (PC2, Fig. 4B), differing by model of the basiscopic lamina shape captured 94.3% slightly greater lateral compression of the blade. of variance in the first four PCs, with the first being the most important (72.9% total variance). This shape SECONDARY VEIN NUMBER variable (Fig. 4G) consists essentially in a lateral Secondary vein number in the anterior lobe (Table 4, contraction, low values representing broad laminas Anterior lobe) varied from four to eight on each side;

Table 4. Secondary vein (= primary lateral vein) number per side in leaf blades of seven populations of Montrichardia linifera. Counts of veins on left and right sides are pooled (30 leaves sampled per population, 60 pooled counts for each population). Calculated in EXCEL, bootstrappping in R.

2.5% and 97.5% 95% conﬁdence bootstrapped Minimum and intervals of quantiles of Sample Populations maximum no. Mean the mean Median the median Mode no.

Anterior lobe only Pooled 4–8 5.58 5.50–5.65 6 5.5–6.0 6 420 Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 RioIgaraçu 4–7 5.05 4.86–5.24 5 5.0–5.0 5 60 SãoRaimundo 4–6 5.18 5.02–5.35 5 5.0–5.0 5 60 TatusVala 4–7 5.38 5.15–5.61 5 5.0–6.0 5 60 TutoiaAraticum 4–7 5.53 5.37–5.69 6 5.0–6.0 6 60 Parelhas 4–7 5.75 5.54–5.96 6 6.0–6.0 6 60 AltoBatista 5–8 6.00 5.84–6.16 6 6.0–6.0 6 60 IlhaBatatas 5–8 6.13 5.95–6.31 6 6.0–6.0 6 60 Acroscopic veins (anterior + posterior lobes) Pooled 6–11 9.07 8.97–9.16 9 9.0–9.0 9 420 RioIgaraçu 7–11 8.73 8.47–8.99 9 8.0–9.0 8 60 Parelhas 6–11 8.77 8.44–9.09 9 8.0–9.0 9 60 SãoRaimundo 7–11 8.98 8.73–9.24 9 9.0–9.0 9 60 TatusVala 6–11 9.00 8.69–9.31 9 8.5–9.0 8–9 60 AltoBatista 7–11 9.10 8.88–9.32 9 9.0–9.0 9 60 TutoiaAraticum 8–11 9.27 9.08–9.46 9 9.0–9.0 9 60 IlhaBatatas 8–11 9.62 9.41–9.82 10 9.5–10.0 10 60 Basiscopic veins (posterior lobes) Pooled 0–6 2.46 2.34–2.58 3 2.0–3.0 3 420 Parelhas 0–3 1.10 0.86–1.34 1 1.0–1.0 1 60 TatusVala 0–4 1.12 0.84–1.39 1 1.0–1.0 0 60 AltoBatista 0–3 2.25 2.07–2.43 2 2.0–2.0 2 60 IlhaBatatas 0–5 2.78 2.54–3.03 3 3.0–3.0 3 60 RioIgaraçu 2–6 3.17 2.95–3.38 3 3.0–3.0 3 60 TutoiaAraticum 1–5 3.22 2.99–3.45 3 3.0–4.0 3 60 SãoRaimundo 2–5 3.58 3.37–3.80 4 3.0–4.0 4 60 six was the commonest number. In M. arborescens acroscopic (external) side of the basal rib, plus those there are typically only three or four such secondary of the corresponding side of the anterior lobe (Table 4, veins per side (Table S1), but in this study no mature Anterior + posterior lobes). The commonest value was crown leaves were observed with three veins. The nine veins per side, varying between six and 11. The vein number on each side often differed because of the proportion of leaves with unequal numbers of veins left-/right-handed asymmetry of the blade. on each side of the blade was higher than in the Comparisons between populations were made by previous case, which considered only those of the pooling the counts for the left and right sides of the anterior lobe (36.7% compared with 24.8% of leaves), anterior lobe. The non-parametric Kruskal–Wallis but the populations differed less among themselves; test showed signiﬁcant differences in the median Kruskal–Wallis tests showed fewer and weaker sig- number of veins between many population pairs niﬁcant differences between population pairs. (Table 5) and gave similar results for pairwise com- The most distinct inter-population differences parison to ANOVA, despite non-normality of the data. were found in basiscopic secondary vein number There was no correlation between mean secondary (Table 4), which varied from 0 to six (mode three). vein number and geographical location of the popu- TatusVala and Parelhas, the two populations with lations [Mantel test in PAST: correlation R =-0.200, the narrowest basiscopic laminas, showed distinctly P (uncorrelated) = 0.800]. lower vein numbers (Fig. 6F). Multivariate regres- We also investigated the overall vein numbers on sion in MorphoJ of basiscopic lamina shape each side of the leaf, i.e. the veins arising on the against basiscopic secondary vein number as the

Table 5. Secondary vein number in anterior lobe of leaves of Montrichardia linifera. Mann–Whitney pairwise comparisons using Kruskal–Wallis test as implemented in PAST (Hammer et al., 2001); H = 81.1, Hc = 93.91, probability (P) that medians of all sample populations are the same < 0.001. Upper triangle Bonferroni uncorrected; lower triangle Bonferroni corrected. 30 individuals in each of seven populations

RioIgaraçu TatusVala TutoiaAraticum SãoRaimundo IlhaBatatas Parelhas AltoBatista

RioIgaraçu 0 0.034 0.000 0.217 0.000 0.000 0.000 TatusVala 0.723 0 0.325 0.241 0.000 0.029 0.000 TutoiaAraticum 0.007† 1.000 0 0.012 0.000 0.121 0.001 SãoRaimundo 1.000 1.000 0.249 0 0.000 0.000 0.000 Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 IlhaBatatas 0.000‡ 0.000‡ 0.001‡ 0.000‡ 0 0.026 0.354 Parelhas 0.000‡ 0.605 1.000 0.005† 0.540 0 0.142 AltoBatista 0.000‡ 0.004† 0.015* 0.000‡ 1.000 1.000 0

*P Յ 0.05. †P Յ 0.01. ‡P Յ 0.001.

Table 6. Length of petiole sheath apical ligule in seven length and basiscopic sinus shape) was studied using populations of Montrichardia linifera. Calculated in two-block PLS analysis in MorphoJ. A series of analy- EXCEL and STATPLUS (Berk & Carey, 2000) ses with different combinations of characters was carried out, with the first block always consisting of Minimum 95% confidence the symmetric aligned coordinates of the six leaf Ligule length and intervals of Sample blade landmarks. The most significant covariation (cm) maximum Mean the mean no. was found when the second block consisted of anterior Pooled populations 0.1–6.7 3.03 2.84–3.22 210 lobe secondary vein number, basiscopic secondary SãoRaimundo 0.5–5.1 2.38 2.01–2.75 30 vein number and the first principal component of TatusVala 0.1–4.5 2.41 1.94–2.87 30 basiscopic lamina shape (Table 7). When ligule length Parelhas 0.6–5.3 2.45 1.97–2.93 30 (as natural logarithm) was included in the second TutoiaAraticum 0.5–6.7 2.89 2.40–3.38 30 block, covariation was slightly reduced (RV coeffi- AltoBatista 1.3–5.6 3.33 2.92–3.75 30 < RioIgaraçu 0.1–6.0 3.44 2.86–4.02 30 cient: 0.2274) but still significant (P 0.0001). IlhaBatatas 0.7–6.4 4.32 3.83–4.80 30 The leaf shape variable on the x-axis of the first pair of singular axes (PLS1) resembled the second principal component of the PCA (Fig. 4B), i.e. low values corresponded to more acutely angled central independent variable (pooled data from all popu- secondary veins. On the y-axis, high values correlations) showed significant covariation, broader sponded with narrow basiscopic laminas and few basiscopic laminas having higher secondary vein basiscopic secondary veins and low values with broad number [predicted sum of squares 35.26% of total, basiscopic laminas and many veins. In the second < P (independence) 0.0001]. pair of singular axes (PLS2) (also highly significant; Table 7), the x-axis shape trend resembled that of the first PC of the PCA (Fig. 4A), i.e. increasing diver- LIGULE LENGTH gence of the basal ribs, whereas the y-axis repre- Ligule length varied overall from 0.1–6.7 cm sented a decreasing trend in anterior lobe secondary (Table 6), with the range including those of both vein number. Ligule length, when included, decreased species morphotypes (see also Supporting Informa- with decreasing numbers of anterior lobe secondary tion, Table S1) in all populations. The shortest mean veins and was negatively correlated with increasingly lengths were found in SãoRaimundo, TatusVala and divergent basal ribs. Parelhas, and the longest in AltoBatista, RioIgaraçu and IlhaBatatas.

ALLOMETRY COVARIATION OF THE TAXONOMIC Highly signiﬁcant differences between populations in CHARACTERS USING PLS leaf size were demonstrated using leaf centroid size Covariation of leaf blade shape and the other taxo- based on the six-landmark conﬁguration data. The nomic characters (secondary vein number, ligule populations with the smallest leaves were Parelhas

Table 7. Covariation between leaf blade characters in Montrichardia linifera. Two-block partial least squares (PLS) analysis of pooled data from 210 leaves. Block 1: symmetric component of the matrix of aligned six-landmark configurations of the leaf blade. Block 2: standardized values of (1) secondary vein numbers on each side of leaf blade (anterior lobe veins, basiscopic lamina veins); (2) first principal component of the matrix of aligned 12-landmark configurations of the basiscopic lamina. RV coefficient (overall strength of association between blocks) = 0.236. Permutation test against null hypothesis of independence: 10 000 replicates, P < 0.0001. Computed in MorphoJ (Klingenberg, 2011)

Singular values and pairwise correlations of PLS scores between blocks

P-value P-value Pairs of Singular (permutation % total (permutation Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 singular axes value test) covariation Correlation test)

PLS1 0.0439 < 0.0001 65.18 0.5926 < 0.0001 PLS2 0.0320 < 0.0001 34.65 0.4502 < 0.0001 PLS3 0.0020 0.5744 0.1320 0.1780 0.5744 PLS4 0.0011 0.1807 0.0420 0.0645 0.1807

and TutoiaAraticum and those with the largest were tend to be associated with larger leaves. ANCOVA in IlhaBatatas and RioIgaraçu. ANOVA carried out in BIOMstat showed highly signiﬁcant differences BIOMstat showed that 24.68% of the variance com- between populations in the PC1 basiscopic lamina ponent was expressed among populations and 75.32% shape variable when adjusted to take account of inter- within (ANOVA: Fs = 10.8323, P < 0.0001). Conse- population differences in centroid size (differences quently, each of the characters studied was further among adjusted means: F = 28.11, P < 0.0001; differ- investigated for covariation with leaf centroid size. ences among slopes: F = 0.42, P = 0.8660).

Leaf blade shape Secondary vein number Multivariate regression of the four symmetric princi- Multivariate regression (MorphoJ) of four secondary pal components of leaf shape (dependent variables) on vein number variables (Table 8) on centroid size of the centroid size of the six-landmark configurations (inde- leaf blade showed significant covariation [11.49% of pendent variable) showed significant covariation total sums of squares, P (independent) < 0.0001]. (P-value: < 0.0001), but this accounted for only 4.4% Regressions of each of these vein number variables of the total variation (sums of squares). Regressing against centroid size showed that the basiscopic veins the matrix of symmetric aligned coordinates on cen- were the most important (Table 8), i.e. larger leaves troid size produced an identical result. Separate tended to have more basiscopic veins. regressions for each principal component on centroid ANCOVAs showed that the secondary vein number size showed that the most important shape variable in the leaf anterior lobe differed significantly between (PC1, Fig. 4A) was not involved in the allometric populations after the effect of leaf centroid size was relationship (P-value: 0.0679), whereas the three taken into account [differences among adjusted other PCs each showed significant covariation means: F = 9.295,11.841, P < 0.0001 (both); differ- (Table 8). As the slopes of the regressions were sig- ences among slopes: F = 1.484, 1.3770, P = 0.1854, nificantly different, the influence of size on inter- 0.2284]. ANCOVAs of basiscopic vein number also population differences in these leaf shape variables showed highly significant differences between popu- could not be further tested using ANCOVA. lations after removing the effect of size [differences among adjusted means: F = 38.226,44.943, P < 0.0001 Basiscopic lamina shape (both), differences among slopes: F = 0.608, 0.822, Multivariate regression of the aligned coordinate P = 0.7239, 0.5537]. matrix of the 12 landmarks of the basiscopic lamina on centroid size of the leaf blade showed no significant Ligule length relationship (1.11% of total predicted sum of squares, Linear regression using R software showed that ligule P = 0.0915). However, regression of the first principal length was significantly predicted by leaf blade cen- component of this matrix on leaf blade centroid size troid size [intercept: -1.372, SE 0.420, t =-3.264, covaried negatively (24.34% of total predicted sum of P < 0.01; slope: 0.121, SE 0.011, t = 10.754, P < 0.0001 squares, P < 0.0001), the other PCs having much adjusted R2 = 0.358; F = 115.7 (d.f. 1 and 205), weaker signals. Thus, broader basiscopic laminas P < 0.0001]. Logged values for both axes increased

Table 8. Allometric relationships in leaves of Montrichardia linifera studied with univariate and multivariate regression. Characters regressed on centroid size of 210 Procrustes-aligned six-landmark conﬁgurations of the leaf blade. Permuta- tion test against the null hypothesis of independence used 10 000 randomization rounds. Computed in MorphoJ (Klingenberg, 2011)

Regression % of total sums of Structure Variable coefficient squares predicted P-value

Six-landmark configurations of Configurations matrix 4.40 < 0.0001 the leaf blade Pooled (PCs 1 to 4) 4.40 < 0.0001 PC1 0.00100061 1.57 0.0679 Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 PC2 -0.00156394 4.99 0.0011 PC3 -0.00157547 12.16 < 0.0001 PC4 -0.00040929 3.18 0.0103 Twelve-landmark configurations Configurations matrix 1.11 0.0915 of basiscopic lamina PC1 -0.00577306 24.34 < 0.0001 Secondary vein number Pooled (AntL, 11.49 < 0.0001 PostBasiL, AntR, PostBasiR) AntL 0.01480211 1.66 0.0590 PostBasiL 0.07185951 16.20 < 0.0001 AntR 0.01455670 1.52 0.0687 PostBasiR 0.07336329 14.74 < 0.0001 Ligule of petiole sheath Length 0.10520742 26.97 < 0.0001

non-normality and heteroscedasticity and so were not tive shape variables that would classify the popula- used. The same regression in MorphoJ gave a value tions into groups corresponding to the arborescens for predicted sums of squares of 26.97% of total sums and linifera morphotypes (Fig. 6). of squares. ANCOVA in BIOMSTAT showed highly Do the quantitative shape variables adequately significant differences between populations in ligule express the qualitative distinction between cor- length when adjusted for differences among popula- date and sagittate shapes? The descriptive terms tions in centroid size of the leaf blade (differences ‘sagittate’ and ‘cordate’ mean, ‘arrow-shaped’ and among adjusted means: F = 8.81, P < 0.0001; differ- ‘heart-shaped’, respectively (Stearn, 1973; Ellis et al., ences among slopes: F = 1.65, P = 0.1345). 2009) but, having no precise definition, can only be interpreted intuitively. An objection could be that, as this qualitative distinction involves the outline shape DISCUSSION (contour) of the leaves, our analysis has failed to SYMMETRIC LEAF BLADE SHAPE VARIATION model sufficiently the traditional character distinc- Visual inspection (Fig. 3) of the range of leaf blade tion. However, the geometry of the four landmarks at shape found across all populations showed cordate, the three lobe tips and the petiole junction clearly sagittate and intermediate forms and therefore con- plays an influential role in the recognition of the two founds this character as a discriminator of the two forms, and the two lateral landmarks add information Montrichardia spp. if all populations are regarded as from the leaf contour, making the six-landmark model belonging to a single species. The principal compo- the simplest that combines information from the most nent and canonical variate analyses showed a con- obvious positional homologies and the contour, all of tinuous range of variation along all axes, except the which are involved in the intuitive perception of first PC of the asymmetric component, which shape difference. Most of the shape variables of our expresses a distinct asymmetry discussed separately quantitative model expressed shapes that resemble below. The populations have significantly different cordate–sagittate trends. Principal components 1–3 ranges of variation along these shape variable (Fig. 4A–C) all involve divergence of the basal ribs, axes and the DFA cross-validation tests show that lengthening of the anterior lobe or narrower leaf certain pairwise distinctions are especially strong; for width at the centre compared with that at the poste- example, TutoiaAraticum–Parelhas (see also Support- rior lobe tips. These variables are independent ele- ing Information, Table S2). Nevertheless, no clear ments of a mathematical decomposition of the shape, overall distinction could be found in these quantita- as PCA is designed to produce uncorrelated variables.

Although evidence from morphogenetic studies is CONCLUSIONS needed to give them greater biological meaning, they Morphological variation in the populations in the Rio demonstrate that the intuitive notions of cordate and Parnaíba delta shows that leaf shape, ligule length sagittate shape each include many different possible and sinus shape varied continuously over the ranges forms. reported for the two currently recognized Montrichar- dia spp. However, as anterior lobe secondary vein number was never as low as three per side (a diag- ASYMMETRIC LEAF SHAPE VARIATION nostic value for M. arborescens), the populations This appears to be the first published report of left- studied are nearer the M. linifera morphotype. The and right-handedness in leaves of Montrichardia, covariation study indicates that the association of Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 despite this asymmetry being obvious on visual morphological characters implicit in the two species inspection (Fig. 3). Although of intrinsic interest, it morphotypes does occur within the populations of the was largely a ‘nuisance’ factor as far as the aims of study area, but as a continuous trend rather than in the present study were concerned. Nevertheless, the two distinct groups. It seems likely, given the previ- population in which symmetrical leaf shape differed ous reports of Howard (1979b), Lins (1994) and most from the others (Parelhas) was also that which Bunting (1980), that such a trend also occurs in exhibited the largest proportion of asymmetric shape Montrichardia populations elsewhere. Despite this variation (46% of total). The probable explanation for continuous rather than discrete trend of variation, we left- or right-handedness of the leaf blade is that it is found significant inter-population morphological dif- established in the seedling, when the first leaf is ferences and these need to be further investigated by initiated and is maintained during the development comparing quantitative shape variables and genetic of the plant as a consequence of the genetic spiral and ecological data. The methodology used here could (D. Barabé, pers. comm.). also be further employed to compare populations in other regions, particularly those in which M. arborescens is recognized as predominant, such as in north- ALLOMETRY ern South America and the Caribbean. The most important symmetric shape variable of the Principal components of shape data are dependent leaf blade (PC1, Fig. 4A) did not covary with size, on the shape variation embodied in the data matrix although overall there was a significant but minor from which they are generated. This means that the covariation of leaf blade shape with size when all the shape variables illustrated here will not correspond data were pooled. Allometric effects also explained exactly to those that might emerge in a more global only a minor part of the character differences between analysis that, for example, compared the Piauí popu- populations. Basiscopic lamina shape, secondary vein lations of M. linifera with populations of the same or numbers and ligule length showed stronger allometric other species from elsewhere. Indeed, it is to be effects, but among-population differences in these expected that, as the global range of shape variation variables were still significant when overall size in the data set increases, so the predominating effects were removed. PC shape variables will change more noticeably. Although we cannot conclude that variables established by one study will be applicable in another, we COVARIATION can postulate that, as the scope of a study expands, the set of predominating shape variables will gradu- The results of the PLS analyses of leaf blade shape ally converge on that which best describes the global against the other three taxonomic characters sup- morphology of the species under study, provided the ported the correlation implicit in the distinction of the latter can be defined a priori on other criteria. This two species morphotypes. The leaf blade shape vari- emphasizes how crucial the accumulation of data ables on the first two (highly significant) pairs of from the broadest possible range of populations is to singular axes (resembling Fig. 4B and A, respectively) achieve an accurate global picture of species mor- can both be regarded as contributing to a trend from phologies. The landmark data and the images on cordate to sagittate leaf forms. Given this interpreta- which they are based remain unchanged and are tion, it is thus possible to conclude that relatively reusable as the scope of analysis widens. cordate leaf blades are correlated with higher numbers of anterior lobe secondary veins, longer ligules and broader basiscopic laminas, while rela- ACKNOWLEDGEMENTS tively sagittate leaves are correlated with fewer anterior lobe secondary veins, shorter ligules and The authors are grateful to the following orga- narrower basiscopic laminas. nizations for financial and infrastructural support:

© 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 170, 554–572 570 M. F. S. SILVA ET AL. the Brazilian Government’s Conselho Nacional de C. 1997. Checklist of the plants of the Guianas (Guyana, Desenvolvimento Científico e Tecnológico (CNPq) for Surinam, French Guiana), 2nd edn. Washington, DC: financial support for the project Biodiversidade de Smithsonian Institution. Macrófitas Aquáticas do Delta do Parnaíba, which Boubes C, Barabé D. 1997. Flower and inflorescence devel- includes a productivity grant and grant for young opment in Montrichardia arborescens (L.) Schott (Araceae). scientists (PIBIC); the Universidade Federal do Piauí; International Journal of Plant Sciences 158: 408–417. the Sistema de Autorização e Informação em Biodi- Bunting GS. 1980 (‘1979’). Sinopsis de las Araceae de Ven- versidade (ICMBio-SISBIO), and the Royal Botanic ezuela. Revista de la Facultad de Agronomia (Maracay) 10: 208–209. (Montrichardia). Gardens, Kew. We are also grateful to Professor Dr Bunting GS. 1995. Araceae. In: Berry PE, Holst BK,

Leandro Rabello Monteiro (Universidade Estadual do Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 Yatskievych K, eds. Flora of the Venezuelan Guayana, Norte Fluminense Darcy Ribeiro) for valuable advice Vol. 2. Portland, OR: Timber Press, 629–631. Fig. 562 on morphometric analysis. (Montrichardia). Chouteau M, Barabé D, Gibernau M. 2006. Pollen–ovule REFERENCES ratios in some Neotropical Araceae and their putative significance. Plant Systematics and Evolution 257: 147– Amarante CB, Silva JCF, Solano FAR, Nascimento LD, 157. Moraes LG, Silva GF, Uno WS. 2009. Estudo espec- Crepani E, Medeiros JS. 2005. Carcinicultura em api- trométrico das folhas da Aninga (Montrichardia linifera) cum no litoral do Piauí: uma análise com sensoriamento coletadas à margem do Rio Guamá no campus da UFPA, remoto e geoprocessamento. Available at: http://www.dsr. Belém – PA. Uma contribuição ao estudo químico da família inpe.br/dsr/simeao/Publicacoes/Carcinicultura.pdf (accessed Araceae. Revista Científica da Universidade Federal do 13 October 2012). Pará 7: 1–19. Croat T. 1978. Flora of Barro Colorado Island. Stanford, CA: Andrade IM, Mayo SJ. 2010. Molecular and morphometric Stanford University Press, 207 (Montrichardia). patterns in Araceae from fragmented northeast Brazilian Croat T. 1994. Araceae. In: Cremers G, Hoff M, eds. forests. In: Seberg O, Petersen G, Barfod AS, Davis JR, eds. Inventaire taxonomique des plantes de la Guyane Française Diversity, phylogeny and evolution in the monocotyledons. IV – Les Monocotyledones (Orchidacées, Cyperacées et Århus: Aarhus University Press, 115–128. Poacées exclues). Paris: Muséum National d’Histoire Andrade IM, Mayo SJ, Kirkup D, Van den Berg C. 2008. Naturelle, 29. (Montrichardia). Comparative morphology of populations of Monstera Schott Croat TB, Fernández-Concha GC, González LI. 2005. (Araceae) from natural forest fragments in Northeast Brazil Montrichardia arborescens (L.) Schott (Araceae) newly using elliptic Fourier Analysis of leaf outlines. Kew Bulletin reported for Mexico. Aroideana 28: 86–87. 63: 193–211. Croat TB, Grayum MH. 1999. Araceae. In: Jørgensen PM, Andrade IM, Mayo SJ, Kirkup D, Van den Berg C. 2010. León-Yánez S, eds. Catalogue of the vascular plants of Elliptic Fourier analysis of leaf shape in Anthurium sinua- Ecuador. St Louis, MO: Missouri Botanical Garden Press, tum Benth. ex Schott and A. pentaphyllum (Aubl.) G.Don 240. (Montrichardia). (Araceae) in forest fragment populations from Northeast Croat TB, Lambert N. 1986. The Araceae of Venezuela. Brazil. Kew Bulletin 65: 3–20. Aroideana 9: 75. (Montrichardia). Aymard G, Croat TB, Carlsen M. 2007. Araceae. In: Duno Croat TB, Stiebel T. 2001. Araceae. In: Stevens WD, Ulloa de Stefano R, Aymard G, Huber O, eds. Flora vascular de los Ulloa C, Pool A, Montiel O, eds. Flora de Nicaragua: Intro- Llanos de Venezuela. Caracas: FUDENA – Fundación ducción Gimnospermas y Angiospermas, Monographs in Empresas Polar – FIBV, 221–222. Systematic Botany from the Missouri Botanical Garden, Barabé D, Lacroix C. 2001. The developmental floral mor- Vol. 85,(Montrichardia). St. Louis: Missouri Botanical phology of Montrichardia arborescens (Araceae) revisited. Garden Press, 162–163. Botanical Journal of the Linnean Society 135: 413–420. Cunha-Santino MB, Pacobahyba LD, Bianchini I Jr. Barabé D, Lavallée K, Gibernau M. 2008. Pollen viability 2003. Changes in the amount of soluble carbohydrates and and germination in some Neotropical aroids. Botany 86: polyphenols contents during decomposition of Montrichar- 98–102. dia arborescens (L.) Schott. Acta Amazonica 33: 469–476. Berk KN, Carey P. 2000. Data analysis with Microsoft Excel. Cunha-Santino MB, Pacobahyba LD, Bianchini I Jr. Pacific Grove, CA: Duxbury. 2004. O/C stoichiometry from mineralization of Montrichar- Bianchini I Jr, Pacobahyba LD, Cunha-Santino MB. dia arborescens (L.) Schott. Acta Limnologica Brasiliensia 2002. Aerobic and anaerobic decomposition of Montrichar- 16: 351–357. dia arborescens (L.) Schott. Acta Limnologica Brasiliensia Dryden IL, Mardia KV. 1998. Statistical shape analysis. 14: 27–34. Chichester: John Wiley and Sons. Blanc P. 1978. Aspects de la ramification chez des Aracées Ellis B, Daly DC, Hickey L, Johnson KR, Mitchell JD, tropicales. Unpublished Ph.D. Thesis, Université Pierre et Wilf P, Wing SL. 2009. Manual of leaf architecture. Ithaca, Marie Curie, Paris. NY: Cornell University Press. Boggan J, Funk V, Kelloff C, Hoff M, Cremers G, Feuillet Engler A. 1911. Araceae–Lasioideae. In: Engler A, ed. Das

pflanzenreich. IV.23C (Heft 48). Leipzig: W. Engelmann, Lustosa AHM. 2005. Sustentabilidade: os catadores de 121–125. caranguejo do Delta do Parnaíba. Unpublished M.Sc. Fournet J. 2002. Flore illustrée des phanérogames de Guad- Thesis, Universidade Federal do Piauí, Teresina. eloupe et de Martinique, 2nd edn. Paris: CIRAD, 1780–1781. Macedo EG, Filho BGS, Potiguara RCV, Santos DSB. Fig. 165.10 (Montrichardia). 2005. Anatomia e arquitetura foliar de Montrichardia linif- Gibernau M, Barabé D, Labat D, Cerdan P, Dejean A. era (Arruda) Schott (Araceae) espécie da várzea amazônica. 2003. Reproductive biology of Montrichardia arborescens Boletim do Museu Paraense Emílio Goeldi, Ciências Natu- (Araceae) in French Guiana. Journal of Tropical Ecology 19: rais 1: 19–43. 103–107. Magrini S, Scoppola A. 2010. Geometric morphometrics as Grayum MH. 2003. Araceae. In: Hammel BE, Grayum MH, a tool to resolve taxonomic problems: the case of Ophioglos- Herrera C, Zamora N, eds. Manual de plantas de Costa sum species (ferns). In: Nimis PL, Vignes Lebbe R, eds. Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 Rica, Vol. II. Gimnospermas y Monocotiledóneas (Agavaceae Tools for identifying biodiversity: progress and problems. – Musaceae). St Louis, MO: Missouri Botanical Garden Paris: EUT, 251–256. Press, 135. (Montrichardia). Martínez RV, Croat TB. 1997. Araceae. In: Martínez RV, ed. Haigh A, Mayo SJ, Croat T, Reynolds L, Mora Pinto M, Flórula de las Reservas Biológicas de Iquitos, Perú. St Louis, Boyce PC, Lay L, Bogner J, Clark B, Kostelac C, Hay MO: Missouri Botanical Garden Press, 747–748. Lamina A. 2009. Interactive web-taxonomy for the Araceae. 94C (Montrichardia). www.cate-araceae.org. Blumea 54: 13–15. Mayo SJ. 1986. Araceae. In: Cope F, Philcox D, eds. Flora of Hammer Ø, Harper DAT, Ryan PD. 2001. PAST: paleon- Trinidad and Tobago, Vol. III, Part IV. Trinidad: Ministry of tological statistics software package for education and data Agriculture, Lands and Food Production, 326–328. (Montri- analysis. Palaeontologia Eletronica 4: 1–9. chardia). Herrera FA, Jaramillo CA, Dilcher DL, Wing SL, Gómez Mayo SJ, Bogner J, Boyce PC. 1997. The genera of Araceae. NC. 2008. Fossil Araceae from a Paleocene Neotropical Kew: Royal Botanic Gardens. rainforest in Colombia. American Journal of Botany 95: McLellan T, Endler JA. 1998. The relative success of some 1569–1583. methods for measuring and describing the shape of complex Howard RA. 1979a. Flora of the Lesser Antilles, Vol. 3. objects. Systematic Biology 47: 264–281. Jamaica Plain, MA: Arnold Arboretum, Harvard University: MMA/SDS. 2002. Zoneamento ecológico-econômico do baixo 389, Fig. 84 (Montrichardia). Rio Parnaíba: subsídios técnicos, Relatório Final. Brasília: Howard RA. 1979b. Nomenclatural notes on the Araceae of Ministério do Meio Ambiente. the Lesser Antilles. Journal of the Arnold Arboretum Nadruz Coelho MA. 2010. Montrichardia. In: Forzza RC, ed. Harvard University 60: 272–289. Catálogo de plantas e fungos do Brasil. Rio de Janeiro: Jensen RJ, Ciofani KM, Miramontes LC. 2002. Lines, Andrea Jakobsson Estudio, Instituto de Pesquisas Jardim outlines, and landmarks: morphometric analyses of leaves Botânico do Rio de Janeiro, 654. of Acer rubrum, Acer saccharinum (Aceraceae) and their Prefeitura Municipal de Parnaíba. 2012. Dados Gerais. hybrid. Taxon 51: 475–492. Available at: http://www.parnaiba.pi.cnm.org.br/ (accessed Jonker-Verhoef AME, Jonker FP. 1953. Araceae. In: Pulle 13 October 2012). A, ed. Flora of Suriname. Amsterdam: The Royal Tropical R Development Core Team. 2011. R: a language and envi- Institute, 75–79. (Montrichardia). ronment for statistical computing. Vienna: R Foundation for Keating RC. 2003 (‘2002’). IX. Acoraceae and Ara- Statistical Computing. ceae. In: Gregory M, Cutler DF, eds. Anatomy of the Ray TS. 1992. Landmark eigenshape analysis: homologous monocotyledons. Oxford: Clarendon Press, 124–126. contours: leaf shape in Syngonium (Araceae). American (Montrichardia). Journal of Botany 79: 69–76. Klingenberg CP. 2011. MorphoJ: an integrated software Rohlf FJ. 2010a. tpsDig, digitize landmarks and outlines, package for geometric morphometrics. Molecular Ecology ver. 2.16. Stony Brook, NY: Department of Ecology and Resources 11: 353–357. Evolution, State University of New York. Klingenberg CP, Duttke S, Whelan S, Kim M. 2012. Rohlf FJ. 2010b. tpsUtil, file utility program, ver. 1.46. Stony Developmental plasticity, morphological variation and Brook, NY: Department of Ecology and Evolution, State evolvability: a multilevel analysis of morphometric integra- University of New York. tion in the shape of compound leaves. Journal of Evolution- Rohlf FJ. 2010c. tpsRelw, Relative warps analysis, ver. 1.49. ary Biology 25: 115–129. Stony Brook, NY: Department of Ecology and Evolution, Lins ALFA. 1994. Aspectos morfológicos e anatômicos de State University of New York. raízes do gênero Montrichardia Crüger. (Araceae). Unpub- Rohlf FJ, Slice DE. 2003. BIOMstat, statistical software lished M.Sc. Thesis, Universidade Federal do Rio Grande do for biologists, ver. 3.30. Setauket, NY: Exeter Publishing, Sul, Porto Alegre. Ltd. Lins ALFA, Oliveira PL. 1995 (‘1994’). Origem, aspectos Santos Filho FS. 2009. Composição florística e estrutural da morfologicos e anatomicos das raizes embrionarias de Mon- vegetação de restinga do Estado do Piauí. Unpublished trichardia linifera (Arruda) Schott (Araceae). Boletim do Doctoral Thesis, Programa de Pós-Graduação em Botânica, Museu Paraense Emílio Goeldi, Série Botânica 10: 221–236. Universidade Federal Rural de Pernambuco, Recife.

Silva MFL. 2004. Ecoturismo no Delta do Parnaiba-PI e geometric morphometrics: a simpliﬁed protocol for begin- entorno: turismo e sustentabilidade. Undergraduate Disser- ners. PLoS ONE 6: 1–20. tation. Universidade de Brasília. Brasília. Viscosi V, Fortini P, Slice DE, Loy A, Blasi C. 2009. Simmonds NW. 1950a. The Araceae of Trinidad and Tobago, Geometric morphometric analyses of leaf variation in B.W.I. Kew Bulletin 5: 391–406. four oak species of subgenus Quercus (Fagaceae). Plant Simmonds NW. 1950b. Notes on the biology of the Araceae of Biosystems 143: 575–587. Trinidad. Journal of Ecology 38: 277–291. Viscosi V, Loy A, Fortini P. 2010. Geometric morphometric Soares MLC, Mayo SJ, Gribel R, Kirkup D. 2011. Elliptic analysis as a tool to explore covariation between shape and Fourier analysis of leaf outlines in ﬁve species of Heteropsis other quantitative leaf traits in European white oaks. In: (Araceae) from the Reserva Florestal Adolpho Ducke, Nimis PL, Vignes Lebbe R, eds. Tools for identifying biodi- Manaus, Amazonas, Brazil. Kew Bulletin 66: 463–470. versity: progress and problems. Paris: EUT, 257–261. Downloaded from https://academic.oup.com/botlinnean/article-abstract/170/4/554/2416148 by guest on 15 July 2019 (published online 15 Oct 2011). Volkova PA, Kasatskaya SA, Boiko AA, Shipunov AB. Standley PC. 1928. Flora of the Panama Canal Zone. Con- 2011. Stability of leaf form and size during specimen prepa- tributions from the United States National Herbarium, Vol. ration of herbarium specimens. Feddes Repertorium 121: 27. Washington, DC: U.S. Government Printing Office, 105, 219–225. Plate 9 (Montrichardia). Volkova PA, Shipunov AB. 2007. Morphological variation of Standley PC. 1944. Flora of Panama, Part II. Fascicle 3. Nymphaea (Nymphaeaceae) in European Russia. Nordic Araceae. Annals of the Missouri Botanical Garden 31: Journal of Botany 25: 329–338. 59–60. (Montrichardia). Weber M, Halbritter H. 2007. Exploding pollen in Mon- Standley PC, Steyermark JA. 1958. Flora of Guatemala. trichardia arborescens (Araceae). Plant Systematics and Fieldiana: Botany 24: 335–337. Evolution 263: 51–57. Stearn WT. 1973. Botanical Latin. Newton Abbot: David and Zelditch ML, Swiderski DL, Sheets HD, Fink WL. 2004. Charles. Geometric morphometrics for biologists: a primer. Amster- Viscosi V, Cardini A. 2011. Leaf morphology, taxonomy and dam: Elsevier Academic Press.

SUPPORTING INFORMATION Additional Supporting Information may be found in the online version of this article: Table S1. Recognized taxonomic distinctions of Montrichardia arborescens and M. linifera. Table S2. Cross-validation tests carried out as part of a discriminant function analysis of all population pairs using the symmetric component of a matrix of 210 Procrustes-aligned six-landmark conﬁgurations of leaf shape. Table S3. Summary of leaf measurements.