Morphometric analysis of Passiflora leaves: the relationship between landmarks of the vasculature and elliptical Fourier descriptors of the blade

  1. Daniel H. Chitwood
  2. Wagner C. Otoni

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

No abstract available

Article activity feed

  1. GigaScience

    Leaf shape among

    Reviewer 2. Tim Dickinson.

    Review of GIGA-D-16-00069, "Morphometric analysis of Passiflora leaves I: the relationship between landmarks of the vasculature and elliptical Fourier descriptors of the blade" by D.H. Chitwood and W.C. Otoni.

    This is a valuable contribution, both in terms of (a) the data collected (leaf outlines and landmarks from a large sample of Passiflora species exhibiting a wide range of leaf shapes), and (b) the well thought out analyses undertaken with care and, for the most part, attention to detail. The paper also showcases, at least for me, the range of tools now available for comparative biologists and others who seek to analyze patterns of morphological variation and ontogenetic change. Until relatively recently it appeared that there was no real replacement for MorphoSys (Meacham, 1993) and the ability of this MS-DOS based system for outline capture and measurement. In many ways there still isn't, leaving workers to pick through an assortment of programs with more limited scopes or (better), to learn to use the imageJ toolbox effectively (Schneider, Rasband, and Eliceiri, 2012). Or SHAPE (Iwata and Ukai, 2002), if only outline data are needed. And in the meantime, adepts from the morphometrics world have embraced R (R Core Team, 2016) as a data analysis and graphics environment, resulting in packages like shapes (Dryden, 2016) and Momocs (Bonhomme et al., 2014). Others (e.g. Jensen, Ciofani, and Miramontes, 2002), myself included, have compared results from different analytical methods based, variously, on landmark and outline data from a common sample of study objects.

    The authors' correlation analysis (Fig. 7) is especially valuable (and as far as I know entirely original) for the way it enables the authors to infer connections between particular landmarks (their x or y coordinates) and particular elliptic Fourier coefficients (A, B, C, or D) for a given harmonic. Although they tend to contrast their landmark data as reflecting leaf vasculature features (because these provide the locations of several of their landmarks), as against overall shape features provided by their outline data, the authors also emphasize the complementarity of these two aspects of their study objects (p. 5). In their subsequent ms (GIGA-D-16-00070) the authors are more adroit in emphasizing this complementarity. Here, they occasionally sound as if they think these associations are somehow intrinsic in these data sources, as if the would obtain even in objects other than leaves (pp. 2, 3, 6). It may be that what's at issue here is just unnecessarily bringing forward interpretations of their results (as if they described the results) before those results have been fully presented. I have three concerns about this ms. First, the authors have been variously careless about details of spelling ("fratal," "heteroblatsy") and the agreement between verb and noun ("data" is a plural noun) and somewhat cavalier about explaining terms when first introduced ("eigenleaf"; "amplification factor," in the caption to Fig. 1).

    Second, basic details of the authors' sampling are unclear, probably because they are so abundantly obvious to them. What is the actual sample size for their study? They refer (pp. 2, 4, 5, 10) to having analyzed more than 3,300 leaves, and (on p. 5) to making available a dataset comprising "...555 scans of leaves from 40 different species of Passiflora..." Presumably, the data released represent a subset of the total sample, but it would help to make this, and their overall sampling strategy, explicit by tabulating how many vines (with what numbers of leaves) were studied for each of the 40 species. These details are important in order to dispel any idea that their discriminant analyses are overfitted because of the large numbers of descriptors (30, 80, or 110) and binary-valued dummy variables for species and leaf position (39, 10); Gittins 'monograph on canonical analysis (Gittins, 1985) references simulation studies suggesting that upwards of 20 times as many study objects as descriptors (measured, plus those designating groups) are needed in order for an analysis to be anything other than a deterministic description of a particular dataset. One of the exciting aspects of this study is the refreshingly large sample size that appears to have been used.

    The third, related, concern is that the way in which leaf position on individual vines was recorded is unclear, given that nothing seems to be said about whether the same number of leaves was produced on all vines of all species during the period during which the study material was grown. Numbering leaves from the youngest leaf at the tip of the shoot, to the base, suggests that the youngest leaf would be numbered 1, the next 2, and so on to N, the most basal leaf on the vine. If the total number of leaves varies from vine to vine then there won't be an exact homology between leaf positions. Figures 5 and 6 suggest that leaf position 1 (basal-most, left-most, respectively) is in fact the most basal leaf. Would this position in fact correspond to the cotyledons, or to the first post-cotyledonary leaf? Clarification would be helpful. I also wonder about the continued reference to "node position" and "heteroblastic node position" as well as, in one case, "shoot position." I suggest that a more transparent usage would be to refer throughout to leaf position (numbered from the most basal leaf), or to the position of a leaf on a shoot. Reference to "heteroblastic node position" seems completely unnecessary, since heteroblasty is an emergent property of some or all of the shoots studied, as seen from the way in which leaves vary over the course of shoot development in shape (or some other property) from the most basal to the most apical.

    BONHOMME, V., S. PICQ, C. GAUCHEREL & J. CLAUDE. 2014. Momocs: Outline Analysis Using R. Journal of Statistical Software 56: 1-24. DRYDEN, I. L. 2016. shapes package, version 1.1-13. website: https://www.maths.nottingham.ac.uk/personal/ild/shapes/index.html. GITTINS, R. 1985. Canonical Analysis. Springer-Verlag, Berlin. IWATA, H. & Y. UKAI. 2002. SHAPE: A computer program package for quantitative evaluation of biological shapes based on elliptic Fourier descriptors. Journal of Heredity 93: 384-385. JENSEN, R. J., K. M. CIOFANI & L. C. MIRAMONTES. 2002. Lines, outlines, and landmarks: morphometric analyses of leaves of Acer rubrum, Acer saccharinum (Aceraceae) and their hybrid. Taxon 51: 475-492. MEACHAM, C. A. 1993. MorphoSys An interactive machine vision program for acquisition of morphometric data. In R. Fortuner [ed.], Advances in computer methods for systematic biology: Artificial intelligence, databases, computer vision, 393-402. R CORE TEAM. 2016. R: A language and environment for statistical computing. website: http://www.R-project.org. SCHNEIDER, C. A., W. S. RASBAND & K. W. ELICEIRI. 2012. NIH Image to ImageJ: 25 years of image analysis. Nature Methods 9: 671-675.

    Re-review; The authors have responded to my comments, and for the most part have satisfied my concerns. In this regard, the copy of the original ms marked up to show the changes the authors have made is extremely helpful. The material on github consists of the data files and R code used to produce the figures, and does not appear provide direct access to the original leaf spectra that would show the variation observed in outline and landmark position. For the publication, I trust that the journal will require that the raw data referenced as "44. Giga DB reference" will consist of a series of files, e.g. one for each species, comprising complete leaf spectra, with the leaves labeled acropetally, i.e. in the heteroblastic sequence from base to tip, for each plant referenced in Figure S1A. If these are provided as vector files (outlines plus landmarks) this data repository should not be excessively large. The vector format will also make it possible for the interested reader to zoom in and out without loss of resolution in order to examine the patterns of variation at different scales.

  2. GigaScience

    BACKGROUND

    **Reviewer 1. Christopher Jiggins. ** This paper assesses leaf shape variation in Passiflora using two different morphometric methods. I will say from the start that I am not an expert in these techniques and am not in a position to review the methods or statistics in this paper, so my comments are somewhat superficial and the paper should also be reviewed by a morphometrics/statistical expert.

    Background: I'd like to see a bit more justification for the study- Is this aimed at developing automated methods for species identification, or quantifying ecological communities and their diversity of leaf shape? Or is it purely an assessment of the methods for shape quantification (Passiflora have such dramatic shape variation it should be an easy test case?) The analysis considers 'species' and developmental stage 'heteroblasty' separately and as independent factors. However I think this is the wrong approach because the plasticity of leaves is highly species dependent - so in analysing plasticity, species should be included in the model, such that the interaction of stage and species is considered. It is not really surprising that when considered independently, leaf position is not a very good predictor of leaf shape because the developmental process is different in each species. Or alternatively the leaf position analysis could be conducted on each species separately, if the data are sufficient.

    It is interesting that more juvenile leaves show less species-specificity as detected in the dataset - anecdotally I have noticed the same thing.

    Discussion: Are there any ecological implications of this work? Are the species sympatric, or where do they come from. I'd be interested to know about disparity in leaf shape among sympatric species, and whether this might be greater than expected by chance, as would perhaps be predicted by the Heliconius coevolution hypothesis. It would be interesting to think further about which ecological questions could be addressed with these methods

    Page 3 Line 21 Refer to recent paper on shape learning Dell'aglio, D. D., Losada, M. E. & Jiggins, C. D. Butterfly Learning and the Diversification of Plant Leaf Shape. Front. Ecol. Evol. 4, 81 (2016).

  4. Version published to 10.1093/gigascience/giw008
  5. Version published to 10.1101/067512v2 on bioRxiv
  6. Version published to 10.1101/067512v1 on bioRxiv