Available online at www.sciencedirect.com
ScienceDirect Phenotyping and beyond: modelling the relationships between traits Christine Granier and Denis Vile Plant phenotyping technology has become more advanced with the capacity to measure many morphological and physiological traits on a given individual. With increasing automation, getting access to various traits on a high number of genotypes over time raises the need to develop systems for data storage and analyses, all congregating into plant phenotyping pipelines. In this review, we highlight several studies that illustrate the latest advances in plant multi-trait phenotyping and discuss future needs to ensure the best use of all these quantitative data. We assert that the next challenge is to disentangle how plant traits are embedded in networks of dependencies (and independencies) by modelling the relationships between them and how these are affected by genetics and environment. Addresses Laboratoire d’Ecophysiologie des Plantes sous Stress Environnementaux, INRA-Supagro 2 Place Viala, 34060 Montpellier, France Corresponding authors: Granier, Christine ([email protected]) and Vile, Denis ([email protected])
Current Opinion in Plant Biology 2014, 18:96–102 This review comes from a themed issue on Genome studies and molecular genetics Edited by Kirsten Bomblies and Olivier Loudet For a complete overview see the Issue and the Editorial Available online 15th March 2014 1369-5266/$ – see front matter, # 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.pbi.2014.02.009
Introduction With the rapid development of sequencing technologies over recent decades, whole genomes of many plant species are now available and the race towards the functional characterisation of thousands of genes has started [1,2]. Two big concerns often discussed in the literature rapidly arose: the need for automated high-throughput systems to record traits on large numbers of samples (numerous organelles, organs or individuals of large populations of genotypes), and the availability of sets of information about experimental protocols and growing conditions to ensure data reuse and meta-analyses [3,4]. Plant phenotyping, i.e. the process of recording quantitative and qualitative plant traits, is not a new research activity, but it has been the backbone of most studies in Current Opinion in Plant Biology 2014, 18:96–102
ecology, agronomy and ecophysiology to explore plant functional diversity, compare the performance of species/ varieties or study plant responses to the environment. However, the word phenotype, and then phenotyping, has not been often used in functional ecology [5], mainly because average trait values were used to represent each species until now. However, the renewed interest in the role of intraspecific variability in the ecology of plant communities [6], as well as the unprecedented access to genomic data, helped to reincorporate this term in ecological studies. Even if there is nothing really new behind the definition of ‘phenotyping’, nor behind the use of terms such as ‘phenomics’; i.e. the full set of phenotypic features of an individual, their use (and abuse) need a careful inspection in order to avoid them boiling down to ‘high-throughput measurements’ because ‘high throughput measurements’ are instrumental means but are not a goal per se. Here, we review some advances in plant phenotyping, highlighting that having an automaton with a camera for plant imaging is not a sine qua non condition for finding promising associations between a phenotype and its underlying genotype. In addition, measuring one trait at one date on a very large number of genotypes does not necessarily give insights into plant functioning or into the genetic control of this trait. As illustrated by a few recent studies, we show that taking into account environmental and temporal variation of the phenotype and considering the phenotype at different levels of integration, i.e. from subcellular, cellular, tissue, organ to whole-plant level, might enhance our understanding of genotype-phenotype relationships. However, when such datasets are properly acquired, their analyses require a conceptual and statistical corpus, that is not always in the plant biologist’s know-how.
Alleviating the bottleneck caused by the lack of high-throughput tools to measure traits associated with gene function: from phenotyping platforms to phenotyping pipelines of analyses Plant phenotyping relies on skills and technologies that are used to characterise qualitative or quantitative traits regardless of the throughput of the analyses. To match the rapid increase in genetic resources, the development of plant phenotyping platforms has been initiated since the 2000s and they are now common tools either commercially available or developed by scientific groups (Figure 1a). www.sciencedirect.com
Multi-trait phenotyping Granier and Vile 97
Figure 1
Fv/Fm
(e) Raw data
0.8 0.5
Area (cm2)
0
Leaves Bolted
(c) Image analysis (b) Imaging station
Frequency (%)
32.5
16
No
400 300 200 100 0 0
5000
10000 15000
Epidermal cell area (μm) 2,9 2,8 2,7
Nuclei
Photosynthesis (μmol s-1 cm-2)
(d) Other phenotyping technologies
2,6 2,5 2,4 2,3 2,2
(a) Phenotyping platform with environmental sensors
Coloration intensity
Time after set-up in the gas exchange chamber
0 1 2 3 4 5 6 7 8 9 10 11
Days
0.5
0
-2 12 17 19 21 25 28 Days
80 60 40 20 0
0 10 20 30
40 50
Soil humidity (%)
Epidermal cell number in leaf 6 (103)
10
1
Rosette area (cm2)
20
Whole-rosette Fv /Fm
Rosette area (cm2)
(f) Integrating time, environment and genetics in data analyses 30
80
Ler
60 40
An-1
20 0
0
10 10 30 40
50
Soil humidity (%) Current Opinion in Plant Biology
An example of multi-trait phenotype pipeline. Arabidopsis thaliana plants are grown in controlled environmental conditions with automatic watering, imaging and recording of micrometeorological conditions (a) [9]. The platform is equipped with an imaging station that includes different types of cameras as illustrated in (b) with RGB vertical and horizontal cameras, an infra-red camera and a fluorescence camera. After automatic image acquisition and storage, images are processed to extract useful traits as shown for whole rosette area, rosette fluorescence (Fv/Fm) and temperature (b,c). Other phenotypic traits are measured manually with more or less invasive technologies such as plant gas exchange, microscopic observations or cellular analyses by flow cytometry (d). Tissue microscopic observations also need image processing as shown for the measurements of epidermal cell area (c,e). Raw data are extracted and processed to include temporal variation of the trait (as shown for dynamic changes in growth and whole rosette Fv/Fm (f)), trait response to environmental conditions [as illustrated by the response curve of rosette area to soil water content (f)] or genetic variation of the phenotype [as illustrated by the response curves of epidermal cell number to soil water content for two A. thaliana accessions, Ler and An-1 (f)].
In most phenotyping platforms, plants or plant parts are automatically imaged by different types of cameras (Figure 1b) enabling the non-destructive measurements of many plant traits [7]. Depending on the systems, images can be captured at high speeds, thereby offering the possibility to acquire images of many plants over time and therefore allowing the consideration of dynamic aspects. Depending on the experimental facilities, plants do not move but are imaged by the moving imaging station set up on a robotic arm [8,9]; whereas in other facilities conveyor belts drive the plants below a motionless imaging www.sciencedirect.com
station [10,11,12,13]. Different orthogonal views of specific plant parts such as whole shoot or root systems are automatically acquired either by using cameras positioned around the plant or by rotation of the plant in front of a camera. The different orthogonal images are combined to extract sets of morphological traits including angles, lengths, widths, diameters and areas [14–16]. Fluorescence imaging allows measuring the photosystem II status in planta [8,17,18] whereas thermal infra-red imaging gives access to leaf surface temperature [19]. Hyperspectral imaging systems are also used to capture different plant Current Opinion in Plant Biology 2014, 18:96–102
98 Genome studies and molecular genetics
phenotypes [20,21]. Partially thanks to the part of automation implemented by those platforms, phenotyping is carried out more accurately, with higher efficiency and in most case on larger number of plants than was previously possible [11]. Systems for automatic image storage have been developed to accompany high-throughput image acquisition [22]. When acquisition and storage are not limiting, the throughput of the analyses is conditioned by the computational image processing (Figure 1c). In this context, many groups have developed automatic or semi-automatic tools for image analysis and batch processing of high number of images and have made them available to the scientific community for seed, hypocotyl, root and leaf growth phenotyping [23–29]. Whole analysis pipelines of plant phenotypes now include the system for plant cultivation, the system for automatic image acquisition, the image analysis tools for automatic extraction of phenotype data and the information system for data storage and sharing [12,22]. The throughput of phenotypic data acquisition is now often compatible with the phenotyping of genetically structured populations for quantitative genetics analyses [11,26] or large collections of accessions for Genome Wide Association studies [30]. However, whole automated pipelines of analyses, i.e., when plants are imaged automatically, images are stored automatically and processed automatically to extract pre-selected measurements; can be dangerous without supervision that requires manual processes. A first reason is that one can miss interesting phenotypes that could have been detected if plants and/or images would have been processed manually [12]. Another one is that automated measurements and manual ones can be complementary to ensure data quality control. A compromise need to be maintained between the efficiency of performing largescale experiments and the flexibility of trait measurements that is a tribute to the technology. In functional genetics analyses, phenotype datasets provided automatically by the platforms are sometimes completed by other traits at other scales [12,31], measured manually with minimally invasive or even invasive technologies (Figure 1d).
From individual datasets to meta-analyses: a need to integrate environment and time Whereas most phenotyping studies have been performed and analysed separately to answer specific questions, there is ample, yet unexploited information in databases that needs to be integrated in meta-analyses. However, many studies have highlighted the inconsistency in trying to characterise one genotype by a unique phenotype, and the difficulty in comparing genotype descriptors among datasets. The dynamic variation in plant phenotype and its environmental plasticity have both to be considered in the analyses of raw data (Figure 1e,f). Current Opinion in Plant Biology 2014, 18:96–102
First, plant phenotypes strongly depend on the environment, leading to differential ranking of genotypes in various environments, the well-known but sometimes forgotten process of genotype-by-environment interaction. The phenotype plasticity that can be encountered when the considered genotype is grown under various environmental conditions can be an obstacle in joint analyses of different datasets as phenotypic plasticity in itself is truly a complex and quantitative trait [32,33]. The thorough characterisation of plant traits absolutely requires the recording and storing of environmental conditions in which the plant was grown and the trait was evaluated [22]. This has led many groups to quantify environmental conditions sensed by the plants with appropriate sensors. A few platforms are devoted to the analysis of plant phenotypic traits in response to environmental variations such as soil water content [9,11,13]. Soil watering is then performed automatically by a system of pot weighing and irrigation at a target weight. In the platforms, traits are measured at a given soil humidity. In some cases, genotypes are compared on the basis of response curves of the trait to the environment to take into account this environmental induced-plasticity (Figure 1f) [34,35]. Another difficulty is the dynamic variation of the phenotype (Figure 1f). Plants expanding rapidly early in development do not necessarily have a larger final leaf area; highlighting that growth phenotype measured at one date does not necessarily reflect the final phenotype [36]. Automated image acquisition together with temporally resolved imaging assays are now combined to give new insights on trait ontogeny rather than the mean trait value at a given date or stage [11,30,37,38,39,40]. The ontogenic changes of quantitative trait loci (QTL) profiles becomes common in quantitative genetics as shown for the analysis of leaf production [41], root gravitropism [10] or whole projected leaf area [11]. In these studies, two methods are used to introduce the temporal resolution: either the authors analyse how the QTL profiles of a given trait change over time or they calculate integrated dynamic variables from the temporal evolution of this trait to detect QTLs of these integrated variables. There are increasing efforts to ensure that phenotypic measurements are now stored together with protocols and environmental conditions within appropriate data management systems to enable their sharing and reuse [22,42].
From single-trait approaches at one date towards multi-trait approaches over time: the next challenge The phenotype cannot be restricted to one trait characterising a genotype. It is the whole set of morphological, structural, physiological and biochemical traits that characterise a genotype at a given stage or date, in a www.sciencedirect.com
Multi-trait phenotyping Granier and Vile 99
Figure 2
PHENOTYPIC DATA
Modeling traits-traits and traits-environment relationships Bivariate relationships
Networks
(a) Genetically driven spectra Ler/Ler Cvi/Cvi Ler/Cvi Cvi/Ler
rate (nmol g Photosynthetic
(g) Trait space Endoreduplication
Whole-plant leaf area
Leaf number
100
(b) Relationship QTL 3000
10
Epidermal cell size
)
Photosynthetic rate
Leaf number Leaf area Whole-plant leaf area
Transpiration rate
(e) Correlative networks
Cell density Stomatal density
B Whole-plant leaf area
2000 1000 0
Fv/Fm
Flowering time
Ag e
LMA (g m -2 )
A
Epidermal cell number
at f
10
4000
Days to flowering
low er ing
100 10
Rosette temperature
Leaf area
100
(d
-1
-1 s )
(d) Causal networks
1000
Epidermal cell 2 area (μm )
Dimensionality reduction
(h) Projection of individuals 0
2
4
6
8
10
Epidermal cell number (104)
10 C 8 6 4 2 0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 2
Leaf number
Leaf area
Water availability Epidermal cell size
Days to flowering
Leaf 6 area (mm )
Epidermal cell number
(c) Environmental effect on allometric relationships
HTxWW CTxWW
Temperature HTxWD
CTxWW HTxWW CTxWD HTxWD
CTxWD
Mixed-models
1
Transpiration (mg d )
10000
(f) yi = Xiβi + Ziui + ei
1000
100 1
10
100
1000
Size at reproduction (mg)
Current Opinion in Plant Biology
Modelling traits-traits and traits-environment relationships. Bivariate relationships between traits: (a) genetically driven spectrum of leaf and wholeplant functioning [50]; (b) allelic effects on the relationships between leaf size-related traits (relationship QTL; [46]); (c) effects control (CT) and high (HT) air temperature and well-watered (WW) and water deficit (WD) soil conditions on the allometric relationship of transpiration (Vasseur et al. unpublished). Trait networks as modelled by (d) causal relationships (e.g. [46]; structural equation modelling) and (e) correlative networks (e.g. Bayesian networks). (f) General form of mixed models of the genotype-by-environment effects on single and multiple traits. The responses yi are modelled by a set of fixed (bi; means of trait i within environment across genotypes) and random (ui, random genotypic effects ui N(0, G) with G the genetic covariance matrix; and ei, the non-genetic residuals) effects. Xi and Zi are design matrices. The structure of mixed models is expandable in many ways. (g) Example of dimensionality reduction of the phenotypic space of plant growth, morphology and physiology as depicted in the first plane of a principal component analysis performed on mean trait values per genotype. (h) Projection of individuals from the Ler x Cvi RIL population (n = 120) of A. thaliana within a trait space similar to (a) under CT and HT and in WW and WD (Vasseur et al. unpublished); the first two axes show the additive and interactive effects of WD and HT (see [55]).
precise environmental context (Figures 1 and 2). The high dimensionality of the phenotype both fascinates and terrifies biologists. This has led many of them to consider a restricted number of biologically relevant traits (i.e. relevant for testing biological hypotheses), www.sciencedirect.com
with successful advances in some cases [43]. For instance, plant breeding has been successful in selecting for a very limited number of traits (e.g. seed mass or yield), one-by-one, although there has been co-selection for a correlated suite of traits. However, resolving the Current Opinion in Plant Biology 2014, 18:96–102
100 Genome studies and molecular genetics
‘many-to-many’ relationships involved in the genotypeto-phenotype map requires advanced mathematical and statistical methods [44]. Indeed, univariate statistical methods remain useful for hypothesis testing on single traits (see Figure 1), but many studies emphasize the importance of considering the covariations between traits (Figure 2a–c), such as trade-offs and allometries, in order to avoid confounding effects and misinterpretation especially when these covariations change with ontogeny or with the environment [10,45]. Moreover, inter-relationships between variables impair any causal mechanism to be ascribed; this is especially the case with correlated environmental variables. Multivariate methods, such as partial regression, are often used to circumvent this problem. Path analysis is another method used to control for covariations between variables and test hypothetical causal graphs for an interpretative approach (Figure 2d) [46–48]. Genetic effects have successfully been included in such methods to search for changes in trait-trait or trait-environment relationships (Figure 2a– c). For instance, genome-wide screens allow identifying ‘relationship QTLs’ that cause variation of trade-offs and allometric relationships [46,49,50]. QTLs identified by classical interval mapping have been successfully included in path models to account for genetic effects on causal networks of traits [45,51]. Bayesian procedures are classically used to study biochemical and molecular regulatory pathways. Systems biology will need to integrate all these procedures together in order to link complex regulatory pathways and correlation networks between traits and the environment (Figure 2e) [52,53]. Other multivariate approaches (principal components analyses, partial least square analysis and clustering and random forest algorithms), aim at reducing the dimensionality of the phenotypic space while preserving the major axes of functional variation (Figure 2g,h) [54–56]. These methods provide an efficient way to extract specific traits relevant for the discrimination of groups that can be further investigated using conventional statistical procedures [57]. In a recent meta-analysis, Laughlin showed that, although authors tend to underestimate the intrinsic dimensionality of plant traits at the interspecific level, a limited number of independent axes of variation remain sufficient to explain the patterns of biodiversity along environmental gradients [54]. Mixed-effect models are among the most efficient tools that can take both the hierarchical (e.g. tissue type within individuals within genotypes within species within communities; treated as random effects) and the multifactorial nature (time, age, environment, experimental blocks, year; often treated as fixed effects) of high-dimensional phenotype data into account (e.g. [58,59]; Figure 2f). To take the multifactorial nature of traits into account in genetic and evolutionary studies, some authors advocate Current Opinion in Plant Biology 2014, 18:96–102
for the systematic use of ‘function-valued approaches’ [38], i.e. statistical analyses of continuous functions of the standard model for quantitative traits (Pij = mj + Gi + GEij; [60]). The availability of non-linear modelling procedures, as well as advances in multi-trait modelling [61], will represent an important step to orchestrate the genotype-to-phenotype map. We advocate for the use of a combination of methods, embedded in a pipeline of analysis, in order to increase the discrimination efficiency of biologically meaningful variables. Plant traits are embedded in complex networks of dependencies (and independencies) that form the phenotypic space. Statistical and modelling tools are available to help disentangle those networks but should be planned before setting experiments.
Conclusions and future directions During the last years, research in biology has benefited from the development of methods allowing the highthroughput measurements of plant traits at different levels, including whole plant and plant parts growth-related traits, cells and tissues developmental, biochemical and physiological states (including metabolome, proteome and transcriptome). Developments in imaging and image processing enable capturing how these traits vary over time [62]. In addition, the use of micrometeorological sensors enables placing these traits in an environmental context. These have generated an exponential number of highly multidimensional datasets that will benefit to the integration of the multi-scale functioning of organisms. A main challenge is now to analyse and model the circuitry that links the different levels of whole-plant organization in response to environmental factors: phenology, leaf, root and whole-plant growth, reproductive development, cell production and expansion, in pace with physiological and molecular modules. One limitation can arise from mastering the statistical and mathematical techniques. But once the acquisition of phenotypic data is no longer limiting for biologists, efficient support by statisticians to help design experiments and analyse the data should be easily graspable, and will pave the way for the development of efficient multidisciplinary simulation models.
Conflicts of interest The authors declare that there are no conflicts of interest related to this publication.
Acknowledgements We would like to thank Franc¸ois Vasseur, Justine Bresson and Maryline Lie`vre for providing unpublished data and images for Figures 1 and 2. We also thank Boris Parent, Cyrille Violle and Michael Mielewczik for critical reading and comments. This review was made possible by a series of grants between 2002 up to now, supporting the development of phenotyping platforms and activities in our group: the DAGOLIGN Research Training Network (an European Community Human Potential Program, HPRN-CT-2002-00267), grants from the Institut National de Recherches Agronomiques (INRA, ‘Ecogene’), the GABI-GENOPLANTE project (AF 2001 094), GENOPLANTE (GPLA-06014G), ARABRAS (ERAPG-003-03), the www.sciencedirect.com
Multi-trait phenotyping Granier and Vile 101
AGRON-OMICS Integrated Project funded by the sixth European Framework Programme (LSHG-CT-2006-037704), the European Plant Phenotyping Network funded by the FP7 Research Infrastructures Programme of the European Union (EPPN, grant agreement no. 284443) and the EIT Climate-KIC project AgWaterBreed.
different conclusions can be drawn with or without taking into consideration environment and/or ontogeny in their analyses.
References and recommended reading
14. Iyer-Pascuzzi AS, Symonova O, Mileyko Y, Hao YL, Belcher H, Harer J, Weitz JS, Benfey PN: Imaging and analysis platform for automatic phenotyping and trait ranking of plant root systems. Plant Physiol 2010, 152:1148-1157.
Papers of particular interest, published within the period of review, have been highlighted as: of special interest of outstanding interest 1.
Bolger ME, Weisshaar B, Scholz U, Stein N, Usadel B, Mayer KFX: Plant genome sequencing applications for crop improvement. Curr Opin Plant Biotechnol 2014, 26:31-37.
2.
Yang W, Duan L, Chen G, Xiong L, Liu Q: Plant phenomics and high-throughput phenotyping: accelerating rice functional genomics using multidisciplinary technologies. Curr Opin Plant Biol 2013, 16:180-187.
3.
Fiorani F, Schurr U: Future scenarios for plant phenotyping. Annu Rev Plant Biol 2013, 64:267-291.
4.
Furbank RT, Tester M: Phenomics – technologies to relieve the phenotyping bottleneck. Trends Plant Sci 2011, 16:635-644.
5.
Craine JM, Engelbrecht BMJ, Lusk CH, McDowell N, Poorter H: Resource limitation, tolerance, and the future of ecological plant classification. Front Plant Sci 2012, 3:246.
6.
Violle C, Enquist BJ, McGill BJ, Jiang L, Albert CH, Hulshof C, Jung V, Messier J: The return of the variance: intraspecific variability in community ecology. Trends Ecol Evol 2012, 27: 244-252.
7. Dhondt S, Wuyts N, Inze´ D: Cell to whole-plant phenotyping: the best is yet to come. Trends Plant Sci 2013, 18:428-439. This review gives a recent overview of different imaging tools and imaging processes that are revolutionizing plant phenotyping from low-to highthroughput, considering different temporal and spatial resolutions. 8.
9.
Jansen M, Gilmer F, Biskup B, Nagel KA, Rascher U, Fischbach A, Briem S, Dreissen G, Tittmann S, Braun S et al.: Simultaneous phenotyping of leaf growth and chlorophyll fluorescence via GROWSCREEN FLUORO allows detection of stress tolerance in Arabidopsis thaliana and other rosette plants. Funct Plant Biol 2009, 36:902-914. Granier C, Aguirrezabal L, Chenu K, Cookson SJ, Dauzat M, Hamard P, Thioux JJ, Rolland G, Bouchier-Combaud S, Lebaudy A et al.: PHENOPSIS, an automated platform for reproducible phenotyping of plant responses to soil water deficit in Arabidopsis thaliana permitted the identification of an accession with low sensitivity to soil water deficit. New Phytol 2006, 169:623-635.
10. Moore CR, Johnson LS, Kwak IY, Livny M, Broman KW, Spalding EP: High-throughput computer vision introduces the time axis to a quantitative trait map of a plant growth response. Genetics 2013, 195:1077-1086. This paper highlights the necessity to combine automated image acquisition with custom analysis algorithms and high-throughput computing resources to capture the time course of gravitropism in genetically structured populations of A. thaliana. The whole pipeline of analyses developed in this project is used to add time as a third dimension to a quantitative trait locus (QTL) map for plant root gravitropism. 11. Tisne S, Serrand Y, Bach L, Gilbault E, Ben Ameur R, Balasse H, Voisin R, Bouchez D, Durand-Tardif M, Guerche P et al.: Phenoscope: an automated large-scale phenotyping platform offering high spatial homogeneity. Plant J 2013, 74:534-544. 12. Arvidsson S, Pe´rez-Rodrı´guez, Mueller-Roeber B: A growth phenotyping pipeline for Arabidopsis thaliana integrating image analysis and rosette area modeling for robust quantification of genotype effects. New Phytol 2011, 191: 895-907. This paper first introduces the notion of the phenotyping pipeline. The authors combine data automatically extracted and processed from a high-throughput phenotyping platform with manually annotated information to analyse shoot growth in A. thaliana. They illustrate that very www.sciencedirect.com
13. Skirycz A, Vandenbroucke K, Clauw P, Maleux K, De Meyer B, Dhondt S, Pucci A, Gonzalez N, Hoeberichts F, Tognetti VB et al.: Survival and growth of Arabidopsis plants given limited water are not equal. Nat Biotechnol 2011, 29:212-214.
15. Clark RT, MacCurdy RB, Jung JK, Shaff JE, McCouch SR, Aneshansley DJ, Kochian LV: Three-dimensional root phenotyping with a novel imaging and software platform. Plant Physiol 2011, 156:455-465. 16. Vasseur F, Pantin F, Vile D: Changes in light intensity reveal a major role for carbon balance in Arabidopsis responses to high temperature. Plant Cell Environ 2011, 34:1563-1576. 17. Rousseau C, Belin E, Bove E, Rousseau D, Fabre F, Berruyer R, Guillaumes J, Manceau C, Jacques MA, Boureau T: High throughput quantitative phenotyping of plant resistance using chlorophyll fluorescence image analysis. Plant Methods 2013, 9:17. 18. Woo NS, Badger MR, Pogson BJ: A rapid, non-invasive procedure for quantitative assessment of drought survival using chlorophyll fluorescence. Plant Methods 2008, 4:27. 19. Costa JM, Grant OM, Chaves MM: Thermography to explore plant–environment interactions. J Exp Bot 2013, 64:3937-3949. 20. Matsuda O, Tanaka A, Fujita T, Iba K: Hyperspectral imaging techniques for rapid identification of Arabidopsis mutants with altered leaf pigment status. Plant Cell Physiol 2012, 53:1154-1170. 21. Mahlein AK, Steiner U, Hillnhutter C, Dehne HW, Oerke EC: Hyperspectral imaging for small-scale analysis of symptoms caused by different sugar beet diseases. Plant Methods 2012, 8:3. 22. Fabre J, Dauzat M, Negre V, Wuyts N, Tireau A, Gennari E, Neveu P, Tisne S, Massonnet C, Hummel I, Granier C: PHENOPSIS DB: an information system for Arabidopsis thaliana phenotypic data in an environmental context. BMC Plant Biol 2011, 11:77. 23. Pound MP, French AP, Atkinson J, Wells DM, Bennet JM, Pridmore T: RootNav: navigating images of complex root architectures. Plant Physiol 2013, 162:1802-1814. 24. Lobet G, Draye X: Novel scanning procedure enabling the vectorization of entire rhizotron-grown root systems. Plant Methods 2013, 9:2-10. 25. De Vylder J, Vandenbussche F, Hu Y, Philips W, Van Der Straeten D: Rosette Tracker: an open source image analysis tool for automatic quantification of genotype effects. Plant Physiol 2012, 160:1149-1159. 26. Herridge RP, Day RC, Baldwin S, Macknight RC: Rapid analysis of seed size in Arabidopsis for mutant and QTL discovery. Plant Methods 2011, 7:3. 27. Hartmann A, Czauderna T, Hoffmann R, Stein N, Schreiber F: HTPheno: an image analysis pipeline for high-throughput plant phenotyping. BMC Bioinformatics 2011, 12:148. 28. Weight C, Parnham DRW: LeafAnalyser: a computational method for rapid and large-scale analyses of leaf shape variation. Plant J 2008, 53:578-586. 29. Bylesjo¨ M, Segura V, Soolanayakanahally RY, Rae AM, Trygg J, Gustafsson P, Jansson S, Street NR: LAMINA: a tool for rapid quantification of leaf size and shape parameters. BMC Plant Biol 2008, 8:82. 30. Zhang X, Hause RJ, Borevitz JO: Natural genetic variation for growth and development revealed by high-throughput phenotyping in Arabidopsis thaliana. G3 (Bethesda) 2012, 2: 29-34. 31. Massonnet C, Tisne S, Radziejwoski A, Vile D, De Veylder L, Dauzat M, Granier C: New insights into the control of endoreduplication: endoreduplication could be driven by Current Opinion in Plant Biology 2014, 18:96–102
102 Genome studies and molecular genetics
organ growth in Arabidopsis leaves. Plant Physiol 2011, 157: 2044-2055. 32. Perez-Harguindeguy N, Diaz S, Garnier E, Lavorel S, Poorter H, Jaureguiberry P, Bret-Harte MS, Cornwell WK, Craine JM, Gurvich DE, Urcelay C et al.: New handbook for standardised measurement of plant functional traits worldwide. Aust J Bot 2013, 61:167-234. 33. Massonnet C, Vile D, Fabre J, Hannah MA, Caldana C, Lisec J, Beemster GTS, Meyer RC, Messerli G, Gronlund JT et al.: Probing the reproducibility of leaf growth and molecular phenotypes: a comparison of three Arabidopsis accessions cultivated in ten laboratories. Plant Physiol 2010, 152:2142-2157. 34. Malosetti M, Ribaut JM, van Eeuwijk FA: The statistical analysis of multi-environment data: modeling genotype-byenvironment interaction and its genetic basis. Front Physiol 2013, 4:44. 35. Reymond M, Muller B, Leonardi A, Charcosset A, Tardieu F: Combining quantitative trait loci analysis and an ecophysiological model to analyze the genetic variability of the responses of maize leaf growth to temperature and water deficit. Plant Physiol 2003, 131:664-675. 36. Lievre M, Wuyts N, Cookson SJ, Bresson J, Dapp M, Vasseur F, Massonnet C, Tisne S, Bettembourg M, Balsera C et al.: Phenotyping the kinematics of leaf development in flowering plants: recommendations and pitfalls. WIREs Dev Biol 2013, 2:809-821. 37. Andrade-Sanchez P, Gore MA, Heun JT, Thorp KR, Carmo-Silva AE, French AN, Salvucci ME, White JW: Development and evaluation of a field-based high-throughput phenotyping platform. Funct Plant Biol 2013 http://dx.doi.org/10.1071/FP13126. 38. Stinchcombe JR, Kirkpatrick M: Genetics and evolution of function-valued traits: understanding environmentally responsive phenotypes. Trends Ecol Evol 2012, 27:637-647. The Function-valued Traits Working Group advocate for the systematic analysis of trait variation as a function of time and/or environment. While challenging this would considerably improve our ability to disentangle the genotype-to-phenotype map. 39. Cole B, Kay SA, Chory J: Automated analysis of hypocotyl growth dynamics during shade avoidance in Arabidopsis. Plant J 2010, 65:991-1000. 40. Walter A, Silk WK, Schurr U: Environmental effects on spatial and temporal patterns of leaf and root growth. Annu Rev Plant Biol 2009, 60:279-304. 41. Mendez-Vigo B, de Andres MT, Ramiro M, Martinez-Zapater JM, Alonso-Blanco C: Temporal analysis of natural variation for the rate of leaf production and its relationship with flowering initiation in Arabidopsis thaliana. J Exp Bot 2010, 61: 1611-1623. 42. Billiau K, Sprenger H, Schudoma C, Walther D, Kohl KI: Data management pipeline for plant phenotyping in a multisite project. Funct Plant Biol 2012, 39:948-957. 43. Chapuis R, Delluc C, Debeuf R, Tardieu F, Welcker C: Resiliences to water deficit in a phenotyping platform and in the field: how related are they in maize? Eur J Agron 2012, 42:59-67. The authors show that sensitivity of maize leaf growth to water deficit measured in a phenotyping platform set up in a greenhouse is highly correlated with that of seed number in a network of field experiments. 44. Houle D, Govindaraju DR, Omholt S: Phenomics: the next challenge. Nat Rev Genet 2010, 11:855-866. 45. Fournier-Level A, Wilczek AM, Cooper MD, Roe JL, Anderson J, Eaton D, Moyers BT, Petipas RH, Schaeffer RN, Pieper B et al.: Paths to selection on life history loci in different natural environments across the native range of Arabidopsis thaliana. Mol Ecol 2013, 22:3552-3566. The authors used common garden experiments to demonstrate the importance of considering environmental variation in QTL detection, and how this genetic information can be included in path models aiming at understanding the determinants of plant fitness.
Current Opinion in Plant Biology 2014, 18:96–102
46. Tisne´ S, Reymond M, Vile D, Fabre J, Dauzat M, Koornneef M, Granier C: Combined genetic and modeling approaches reveal that epidermal cell area and number in leaves are controlled by leaf and plant developmental processes in Arabidopsis. Plant Physiol 2008, 148:1117-1127. 47. Vile D, Shipley B, Garnier E: A structural equation model to integrate changes in functional strategies during old-field succession. Ecology 2006, 87:504-517. 48. Shipley B: Cause and Correlation in Biology: A User’s Guide to Path Analysis, Structural Equations, and Causal Inference. Cambridge, UK: Cambridge University Press; 2000, . 49. Pavlicev M, Wagner GP: A model of developmental evolution: selection, pleiotropy and compensation. Trends Ecol Evol 2012, 27:316-322. This paper highlights the potential evolutionary role of rQTL, i.e. QTL that affect the relationships between traits such as allometric relationships or trade-offs. Examples of rQTL in plants can be found in [45] and [49]. 50. Vasseur F, Violle C, Enquist BJ, Granier C, Vile D: A common genetic basis to the origin of the leaf economics spectrum and metabolic scaling allometry. Ecol Lett 2012, 15:1149-1157. 51. Li R, Tsaih SW, Shockley K, Stylianou IM, Wergedal J, Paigen B, Churchill GA: Structural model analysis of multiple quantitative traits. PLoS Genet 2006, 2:e114. 52. Poorter H, Anten NP, Marcelis LFM: Trait correlation networks: a whole-plant perspective on the recently criticized leaf economic spectrum. New Phytol 2013:1-5. 53. Poorter H, Lambers H, Evans JH: Physiological mechanisms in plant growth models: do we need a supracellular systems biology approach? Plant Cell Environ 2013, 36:1-18. In this paper, the authors present trait correlation networks based on various large-scale screenings and they show how a correlation network changes when the environment changes: a first step towards dose– response curve of trait correlation networks across environmental gradient. 54. Laughlin D: The intrinsic dimensionality of plant traits and its relevance to community assembly. J Ecol 2013 http://dx.doi.org/ 10.1111/1365-2745.12187. 55. Vile D, Pervent M, Belluau M, Vasseur F, Bresson J, Muller B, Granier C, Simonneau T: Arabidopsis growth under prolonged high temperature and water deficit: independent or interactive effects? Plant Cell Environ 2012, 35:702-718. 56. Westoby M, Falster DS, Moles AT, Vesk PA, Wright IJ: Plant ecological strategies: some leading dimensions of variation between species. Ann Rev Ecol Evol Syst 2002, 33:125-159. 57. Topp CN, Iyer-Pascuzzi AS, Anderson JT, Lee CR, Zurek PR, Symonova O, Zheng Y, Bucksch A, Mileyko Y, Galkovskyi T et al.: 3D phenotyping and quantitative trait locus mapping identify core regions of the rice genome controlling root architecture. Proc Natl Acad Sci USA 2013, 110:E1695-E1704. 58. van Eeuwijk FA, Bink MCAM, Chenu K, Chapman S: Detection and use of QTL for complex traits in multiple environments. Curr Opin Plant Biol 2010, 13:193-205. 59. Boer MP, Wright D, Feng LZ, Podlich DW, Luo L, Cooper M, van Eeuwijk FA: A mixed-model quantitative trait loci (QTL) analysis for multiple-environment trial data using environmental covariables for QTL-by-environment interactions, with an example in maize. Genetics 2007, 177:1801-1813. 60. Kirkpatrick M, Lofsvold D, Bulmer M: Analysis of the inheritance, selection and evolution of growth trajectories. Genetics 1990, 124:979-993. 61. Blonder B, Lamanna C, Violle C, Enquist B: The n-dimensional hypervolume. Global Ecol Biogeogr 2014. (in press). 62. Sozzani R, Busch W, Spalding EP, Benfey PN: Advanced imaging techniques for the study of plant growth and development. Trends Plant Sci 2014. (in press).
www.sciencedirect.com
ScienceDirect Phenotyping and beyond: modelling the relationships between traits Christine Granier and Denis Vile Plant phenotyping technology has become more advanced with the capacity to measure many morphological and physiological traits on a given individual. With increasing automation, getting access to various traits on a high number of genotypes over time raises the need to develop systems for data storage and analyses, all congregating into plant phenotyping pipelines. In this review, we highlight several studies that illustrate the latest advances in plant multi-trait phenotyping and discuss future needs to ensure the best use of all these quantitative data. We assert that the next challenge is to disentangle how plant traits are embedded in networks of dependencies (and independencies) by modelling the relationships between them and how these are affected by genetics and environment. Addresses Laboratoire d’Ecophysiologie des Plantes sous Stress Environnementaux, INRA-Supagro 2 Place Viala, 34060 Montpellier, France Corresponding authors: Granier, Christine ([email protected]) and Vile, Denis ([email protected])
Current Opinion in Plant Biology 2014, 18:96–102 This review comes from a themed issue on Genome studies and molecular genetics Edited by Kirsten Bomblies and Olivier Loudet For a complete overview see the Issue and the Editorial Available online 15th March 2014 1369-5266/$ – see front matter, # 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.pbi.2014.02.009
Introduction With the rapid development of sequencing technologies over recent decades, whole genomes of many plant species are now available and the race towards the functional characterisation of thousands of genes has started [1,2]. Two big concerns often discussed in the literature rapidly arose: the need for automated high-throughput systems to record traits on large numbers of samples (numerous organelles, organs or individuals of large populations of genotypes), and the availability of sets of information about experimental protocols and growing conditions to ensure data reuse and meta-analyses [3,4]. Plant phenotyping, i.e. the process of recording quantitative and qualitative plant traits, is not a new research activity, but it has been the backbone of most studies in Current Opinion in Plant Biology 2014, 18:96–102
ecology, agronomy and ecophysiology to explore plant functional diversity, compare the performance of species/ varieties or study plant responses to the environment. However, the word phenotype, and then phenotyping, has not been often used in functional ecology [5], mainly because average trait values were used to represent each species until now. However, the renewed interest in the role of intraspecific variability in the ecology of plant communities [6], as well as the unprecedented access to genomic data, helped to reincorporate this term in ecological studies. Even if there is nothing really new behind the definition of ‘phenotyping’, nor behind the use of terms such as ‘phenomics’; i.e. the full set of phenotypic features of an individual, their use (and abuse) need a careful inspection in order to avoid them boiling down to ‘high-throughput measurements’ because ‘high throughput measurements’ are instrumental means but are not a goal per se. Here, we review some advances in plant phenotyping, highlighting that having an automaton with a camera for plant imaging is not a sine qua non condition for finding promising associations between a phenotype and its underlying genotype. In addition, measuring one trait at one date on a very large number of genotypes does not necessarily give insights into plant functioning or into the genetic control of this trait. As illustrated by a few recent studies, we show that taking into account environmental and temporal variation of the phenotype and considering the phenotype at different levels of integration, i.e. from subcellular, cellular, tissue, organ to whole-plant level, might enhance our understanding of genotype-phenotype relationships. However, when such datasets are properly acquired, their analyses require a conceptual and statistical corpus, that is not always in the plant biologist’s know-how.
Alleviating the bottleneck caused by the lack of high-throughput tools to measure traits associated with gene function: from phenotyping platforms to phenotyping pipelines of analyses Plant phenotyping relies on skills and technologies that are used to characterise qualitative or quantitative traits regardless of the throughput of the analyses. To match the rapid increase in genetic resources, the development of plant phenotyping platforms has been initiated since the 2000s and they are now common tools either commercially available or developed by scientific groups (Figure 1a). www.sciencedirect.com
Multi-trait phenotyping Granier and Vile 97
Figure 1
Fv/Fm
(e) Raw data
0.8 0.5
Area (cm2)
0
Leaves Bolted
(c) Image analysis (b) Imaging station
Frequency (%)
32.5
16
No
400 300 200 100 0 0
5000
10000 15000
Epidermal cell area (μm) 2,9 2,8 2,7
Nuclei
Photosynthesis (μmol s-1 cm-2)
(d) Other phenotyping technologies
2,6 2,5 2,4 2,3 2,2
(a) Phenotyping platform with environmental sensors
Coloration intensity
Time after set-up in the gas exchange chamber
0 1 2 3 4 5 6 7 8 9 10 11
Days
0.5
0
-2 12 17 19 21 25 28 Days
80 60 40 20 0
0 10 20 30
40 50
Soil humidity (%)
Epidermal cell number in leaf 6 (103)
10
1
Rosette area (cm2)
20
Whole-rosette Fv /Fm
Rosette area (cm2)
(f) Integrating time, environment and genetics in data analyses 30
80
Ler
60 40
An-1
20 0
0
10 10 30 40
50
Soil humidity (%) Current Opinion in Plant Biology
An example of multi-trait phenotype pipeline. Arabidopsis thaliana plants are grown in controlled environmental conditions with automatic watering, imaging and recording of micrometeorological conditions (a) [9]. The platform is equipped with an imaging station that includes different types of cameras as illustrated in (b) with RGB vertical and horizontal cameras, an infra-red camera and a fluorescence camera. After automatic image acquisition and storage, images are processed to extract useful traits as shown for whole rosette area, rosette fluorescence (Fv/Fm) and temperature (b,c). Other phenotypic traits are measured manually with more or less invasive technologies such as plant gas exchange, microscopic observations or cellular analyses by flow cytometry (d). Tissue microscopic observations also need image processing as shown for the measurements of epidermal cell area (c,e). Raw data are extracted and processed to include temporal variation of the trait (as shown for dynamic changes in growth and whole rosette Fv/Fm (f)), trait response to environmental conditions [as illustrated by the response curve of rosette area to soil water content (f)] or genetic variation of the phenotype [as illustrated by the response curves of epidermal cell number to soil water content for two A. thaliana accessions, Ler and An-1 (f)].
In most phenotyping platforms, plants or plant parts are automatically imaged by different types of cameras (Figure 1b) enabling the non-destructive measurements of many plant traits [7]. Depending on the systems, images can be captured at high speeds, thereby offering the possibility to acquire images of many plants over time and therefore allowing the consideration of dynamic aspects. Depending on the experimental facilities, plants do not move but are imaged by the moving imaging station set up on a robotic arm [8,9]; whereas in other facilities conveyor belts drive the plants below a motionless imaging www.sciencedirect.com
station [10,11,12,13]. Different orthogonal views of specific plant parts such as whole shoot or root systems are automatically acquired either by using cameras positioned around the plant or by rotation of the plant in front of a camera. The different orthogonal images are combined to extract sets of morphological traits including angles, lengths, widths, diameters and areas [14–16]. Fluorescence imaging allows measuring the photosystem II status in planta [8,17,18] whereas thermal infra-red imaging gives access to leaf surface temperature [19]. Hyperspectral imaging systems are also used to capture different plant Current Opinion in Plant Biology 2014, 18:96–102
98 Genome studies and molecular genetics
phenotypes [20,21]. Partially thanks to the part of automation implemented by those platforms, phenotyping is carried out more accurately, with higher efficiency and in most case on larger number of plants than was previously possible [11]. Systems for automatic image storage have been developed to accompany high-throughput image acquisition [22]. When acquisition and storage are not limiting, the throughput of the analyses is conditioned by the computational image processing (Figure 1c). In this context, many groups have developed automatic or semi-automatic tools for image analysis and batch processing of high number of images and have made them available to the scientific community for seed, hypocotyl, root and leaf growth phenotyping [23–29]. Whole analysis pipelines of plant phenotypes now include the system for plant cultivation, the system for automatic image acquisition, the image analysis tools for automatic extraction of phenotype data and the information system for data storage and sharing [12,22]. The throughput of phenotypic data acquisition is now often compatible with the phenotyping of genetically structured populations for quantitative genetics analyses [11,26] or large collections of accessions for Genome Wide Association studies [30]. However, whole automated pipelines of analyses, i.e., when plants are imaged automatically, images are stored automatically and processed automatically to extract pre-selected measurements; can be dangerous without supervision that requires manual processes. A first reason is that one can miss interesting phenotypes that could have been detected if plants and/or images would have been processed manually [12]. Another one is that automated measurements and manual ones can be complementary to ensure data quality control. A compromise need to be maintained between the efficiency of performing largescale experiments and the flexibility of trait measurements that is a tribute to the technology. In functional genetics analyses, phenotype datasets provided automatically by the platforms are sometimes completed by other traits at other scales [12,31], measured manually with minimally invasive or even invasive technologies (Figure 1d).
From individual datasets to meta-analyses: a need to integrate environment and time Whereas most phenotyping studies have been performed and analysed separately to answer specific questions, there is ample, yet unexploited information in databases that needs to be integrated in meta-analyses. However, many studies have highlighted the inconsistency in trying to characterise one genotype by a unique phenotype, and the difficulty in comparing genotype descriptors among datasets. The dynamic variation in plant phenotype and its environmental plasticity have both to be considered in the analyses of raw data (Figure 1e,f). Current Opinion in Plant Biology 2014, 18:96–102
First, plant phenotypes strongly depend on the environment, leading to differential ranking of genotypes in various environments, the well-known but sometimes forgotten process of genotype-by-environment interaction. The phenotype plasticity that can be encountered when the considered genotype is grown under various environmental conditions can be an obstacle in joint analyses of different datasets as phenotypic plasticity in itself is truly a complex and quantitative trait [32,33]. The thorough characterisation of plant traits absolutely requires the recording and storing of environmental conditions in which the plant was grown and the trait was evaluated [22]. This has led many groups to quantify environmental conditions sensed by the plants with appropriate sensors. A few platforms are devoted to the analysis of plant phenotypic traits in response to environmental variations such as soil water content [9,11,13]. Soil watering is then performed automatically by a system of pot weighing and irrigation at a target weight. In the platforms, traits are measured at a given soil humidity. In some cases, genotypes are compared on the basis of response curves of the trait to the environment to take into account this environmental induced-plasticity (Figure 1f) [34,35]. Another difficulty is the dynamic variation of the phenotype (Figure 1f). Plants expanding rapidly early in development do not necessarily have a larger final leaf area; highlighting that growth phenotype measured at one date does not necessarily reflect the final phenotype [36]. Automated image acquisition together with temporally resolved imaging assays are now combined to give new insights on trait ontogeny rather than the mean trait value at a given date or stage [11,30,37,38,39,40]. The ontogenic changes of quantitative trait loci (QTL) profiles becomes common in quantitative genetics as shown for the analysis of leaf production [41], root gravitropism [10] or whole projected leaf area [11]. In these studies, two methods are used to introduce the temporal resolution: either the authors analyse how the QTL profiles of a given trait change over time or they calculate integrated dynamic variables from the temporal evolution of this trait to detect QTLs of these integrated variables. There are increasing efforts to ensure that phenotypic measurements are now stored together with protocols and environmental conditions within appropriate data management systems to enable their sharing and reuse [22,42].
From single-trait approaches at one date towards multi-trait approaches over time: the next challenge The phenotype cannot be restricted to one trait characterising a genotype. It is the whole set of morphological, structural, physiological and biochemical traits that characterise a genotype at a given stage or date, in a www.sciencedirect.com
Multi-trait phenotyping Granier and Vile 99
Figure 2
PHENOTYPIC DATA
Modeling traits-traits and traits-environment relationships Bivariate relationships
Networks
(a) Genetically driven spectra Ler/Ler Cvi/Cvi Ler/Cvi Cvi/Ler
rate (nmol g Photosynthetic
(g) Trait space Endoreduplication
Whole-plant leaf area
Leaf number
100
(b) Relationship QTL 3000
10
Epidermal cell size
)
Photosynthetic rate
Leaf number Leaf area Whole-plant leaf area
Transpiration rate
(e) Correlative networks
Cell density Stomatal density
B Whole-plant leaf area
2000 1000 0
Fv/Fm
Flowering time
Ag e
LMA (g m -2 )
A
Epidermal cell number
at f
10
4000
Days to flowering
low er ing
100 10
Rosette temperature
Leaf area
100
(d
-1
-1 s )
(d) Causal networks
1000
Epidermal cell 2 area (μm )
Dimensionality reduction
(h) Projection of individuals 0
2
4
6
8
10
Epidermal cell number (104)
10 C 8 6 4 2 0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 2
Leaf number
Leaf area
Water availability Epidermal cell size
Days to flowering
Leaf 6 area (mm )
Epidermal cell number
(c) Environmental effect on allometric relationships
HTxWW CTxWW
Temperature HTxWD
CTxWW HTxWW CTxWD HTxWD
CTxWD
Mixed-models
1
Transpiration (mg d )
10000
(f) yi = Xiβi + Ziui + ei
1000
100 1
10
100
1000
Size at reproduction (mg)
Current Opinion in Plant Biology
Modelling traits-traits and traits-environment relationships. Bivariate relationships between traits: (a) genetically driven spectrum of leaf and wholeplant functioning [50]; (b) allelic effects on the relationships between leaf size-related traits (relationship QTL; [46]); (c) effects control (CT) and high (HT) air temperature and well-watered (WW) and water deficit (WD) soil conditions on the allometric relationship of transpiration (Vasseur et al. unpublished). Trait networks as modelled by (d) causal relationships (e.g. [46]; structural equation modelling) and (e) correlative networks (e.g. Bayesian networks). (f) General form of mixed models of the genotype-by-environment effects on single and multiple traits. The responses yi are modelled by a set of fixed (bi; means of trait i within environment across genotypes) and random (ui, random genotypic effects ui N(0, G) with G the genetic covariance matrix; and ei, the non-genetic residuals) effects. Xi and Zi are design matrices. The structure of mixed models is expandable in many ways. (g) Example of dimensionality reduction of the phenotypic space of plant growth, morphology and physiology as depicted in the first plane of a principal component analysis performed on mean trait values per genotype. (h) Projection of individuals from the Ler x Cvi RIL population (n = 120) of A. thaliana within a trait space similar to (a) under CT and HT and in WW and WD (Vasseur et al. unpublished); the first two axes show the additive and interactive effects of WD and HT (see [55]).
precise environmental context (Figures 1 and 2). The high dimensionality of the phenotype both fascinates and terrifies biologists. This has led many of them to consider a restricted number of biologically relevant traits (i.e. relevant for testing biological hypotheses), www.sciencedirect.com
with successful advances in some cases [43]. For instance, plant breeding has been successful in selecting for a very limited number of traits (e.g. seed mass or yield), one-by-one, although there has been co-selection for a correlated suite of traits. However, resolving the Current Opinion in Plant Biology 2014, 18:96–102
100 Genome studies and molecular genetics
‘many-to-many’ relationships involved in the genotypeto-phenotype map requires advanced mathematical and statistical methods [44]. Indeed, univariate statistical methods remain useful for hypothesis testing on single traits (see Figure 1), but many studies emphasize the importance of considering the covariations between traits (Figure 2a–c), such as trade-offs and allometries, in order to avoid confounding effects and misinterpretation especially when these covariations change with ontogeny or with the environment [10,45]. Moreover, inter-relationships between variables impair any causal mechanism to be ascribed; this is especially the case with correlated environmental variables. Multivariate methods, such as partial regression, are often used to circumvent this problem. Path analysis is another method used to control for covariations between variables and test hypothetical causal graphs for an interpretative approach (Figure 2d) [46–48]. Genetic effects have successfully been included in such methods to search for changes in trait-trait or trait-environment relationships (Figure 2a– c). For instance, genome-wide screens allow identifying ‘relationship QTLs’ that cause variation of trade-offs and allometric relationships [46,49,50]. QTLs identified by classical interval mapping have been successfully included in path models to account for genetic effects on causal networks of traits [45,51]. Bayesian procedures are classically used to study biochemical and molecular regulatory pathways. Systems biology will need to integrate all these procedures together in order to link complex regulatory pathways and correlation networks between traits and the environment (Figure 2e) [52,53]. Other multivariate approaches (principal components analyses, partial least square analysis and clustering and random forest algorithms), aim at reducing the dimensionality of the phenotypic space while preserving the major axes of functional variation (Figure 2g,h) [54–56]. These methods provide an efficient way to extract specific traits relevant for the discrimination of groups that can be further investigated using conventional statistical procedures [57]. In a recent meta-analysis, Laughlin showed that, although authors tend to underestimate the intrinsic dimensionality of plant traits at the interspecific level, a limited number of independent axes of variation remain sufficient to explain the patterns of biodiversity along environmental gradients [54]. Mixed-effect models are among the most efficient tools that can take both the hierarchical (e.g. tissue type within individuals within genotypes within species within communities; treated as random effects) and the multifactorial nature (time, age, environment, experimental blocks, year; often treated as fixed effects) of high-dimensional phenotype data into account (e.g. [58,59]; Figure 2f). To take the multifactorial nature of traits into account in genetic and evolutionary studies, some authors advocate Current Opinion in Plant Biology 2014, 18:96–102
for the systematic use of ‘function-valued approaches’ [38], i.e. statistical analyses of continuous functions of the standard model for quantitative traits (Pij = mj + Gi + GEij; [60]). The availability of non-linear modelling procedures, as well as advances in multi-trait modelling [61], will represent an important step to orchestrate the genotype-to-phenotype map. We advocate for the use of a combination of methods, embedded in a pipeline of analysis, in order to increase the discrimination efficiency of biologically meaningful variables. Plant traits are embedded in complex networks of dependencies (and independencies) that form the phenotypic space. Statistical and modelling tools are available to help disentangle those networks but should be planned before setting experiments.
Conclusions and future directions During the last years, research in biology has benefited from the development of methods allowing the highthroughput measurements of plant traits at different levels, including whole plant and plant parts growth-related traits, cells and tissues developmental, biochemical and physiological states (including metabolome, proteome and transcriptome). Developments in imaging and image processing enable capturing how these traits vary over time [62]. In addition, the use of micrometeorological sensors enables placing these traits in an environmental context. These have generated an exponential number of highly multidimensional datasets that will benefit to the integration of the multi-scale functioning of organisms. A main challenge is now to analyse and model the circuitry that links the different levels of whole-plant organization in response to environmental factors: phenology, leaf, root and whole-plant growth, reproductive development, cell production and expansion, in pace with physiological and molecular modules. One limitation can arise from mastering the statistical and mathematical techniques. But once the acquisition of phenotypic data is no longer limiting for biologists, efficient support by statisticians to help design experiments and analyse the data should be easily graspable, and will pave the way for the development of efficient multidisciplinary simulation models.
Conflicts of interest The authors declare that there are no conflicts of interest related to this publication.
Acknowledgements We would like to thank Franc¸ois Vasseur, Justine Bresson and Maryline Lie`vre for providing unpublished data and images for Figures 1 and 2. We also thank Boris Parent, Cyrille Violle and Michael Mielewczik for critical reading and comments. This review was made possible by a series of grants between 2002 up to now, supporting the development of phenotyping platforms and activities in our group: the DAGOLIGN Research Training Network (an European Community Human Potential Program, HPRN-CT-2002-00267), grants from the Institut National de Recherches Agronomiques (INRA, ‘Ecogene’), the GABI-GENOPLANTE project (AF 2001 094), GENOPLANTE (GPLA-06014G), ARABRAS (ERAPG-003-03), the www.sciencedirect.com
Multi-trait phenotyping Granier and Vile 101
AGRON-OMICS Integrated Project funded by the sixth European Framework Programme (LSHG-CT-2006-037704), the European Plant Phenotyping Network funded by the FP7 Research Infrastructures Programme of the European Union (EPPN, grant agreement no. 284443) and the EIT Climate-KIC project AgWaterBreed.
different conclusions can be drawn with or without taking into consideration environment and/or ontogeny in their analyses.
References and recommended reading
14. Iyer-Pascuzzi AS, Symonova O, Mileyko Y, Hao YL, Belcher H, Harer J, Weitz JS, Benfey PN: Imaging and analysis platform for automatic phenotyping and trait ranking of plant root systems. Plant Physiol 2010, 152:1148-1157.
Papers of particular interest, published within the period of review, have been highlighted as: of special interest of outstanding interest 1.
Bolger ME, Weisshaar B, Scholz U, Stein N, Usadel B, Mayer KFX: Plant genome sequencing applications for crop improvement. Curr Opin Plant Biotechnol 2014, 26:31-37.
2.
Yang W, Duan L, Chen G, Xiong L, Liu Q: Plant phenomics and high-throughput phenotyping: accelerating rice functional genomics using multidisciplinary technologies. Curr Opin Plant Biol 2013, 16:180-187.
3.
Fiorani F, Schurr U: Future scenarios for plant phenotyping. Annu Rev Plant Biol 2013, 64:267-291.
4.
Furbank RT, Tester M: Phenomics – technologies to relieve the phenotyping bottleneck. Trends Plant Sci 2011, 16:635-644.
5.
Craine JM, Engelbrecht BMJ, Lusk CH, McDowell N, Poorter H: Resource limitation, tolerance, and the future of ecological plant classification. Front Plant Sci 2012, 3:246.
6.
Violle C, Enquist BJ, McGill BJ, Jiang L, Albert CH, Hulshof C, Jung V, Messier J: The return of the variance: intraspecific variability in community ecology. Trends Ecol Evol 2012, 27: 244-252.
7. Dhondt S, Wuyts N, Inze´ D: Cell to whole-plant phenotyping: the best is yet to come. Trends Plant Sci 2013, 18:428-439. This review gives a recent overview of different imaging tools and imaging processes that are revolutionizing plant phenotyping from low-to highthroughput, considering different temporal and spatial resolutions. 8.
9.
Jansen M, Gilmer F, Biskup B, Nagel KA, Rascher U, Fischbach A, Briem S, Dreissen G, Tittmann S, Braun S et al.: Simultaneous phenotyping of leaf growth and chlorophyll fluorescence via GROWSCREEN FLUORO allows detection of stress tolerance in Arabidopsis thaliana and other rosette plants. Funct Plant Biol 2009, 36:902-914. Granier C, Aguirrezabal L, Chenu K, Cookson SJ, Dauzat M, Hamard P, Thioux JJ, Rolland G, Bouchier-Combaud S, Lebaudy A et al.: PHENOPSIS, an automated platform for reproducible phenotyping of plant responses to soil water deficit in Arabidopsis thaliana permitted the identification of an accession with low sensitivity to soil water deficit. New Phytol 2006, 169:623-635.
10. Moore CR, Johnson LS, Kwak IY, Livny M, Broman KW, Spalding EP: High-throughput computer vision introduces the time axis to a quantitative trait map of a plant growth response. Genetics 2013, 195:1077-1086. This paper highlights the necessity to combine automated image acquisition with custom analysis algorithms and high-throughput computing resources to capture the time course of gravitropism in genetically structured populations of A. thaliana. The whole pipeline of analyses developed in this project is used to add time as a third dimension to a quantitative trait locus (QTL) map for plant root gravitropism. 11. Tisne S, Serrand Y, Bach L, Gilbault E, Ben Ameur R, Balasse H, Voisin R, Bouchez D, Durand-Tardif M, Guerche P et al.: Phenoscope: an automated large-scale phenotyping platform offering high spatial homogeneity. Plant J 2013, 74:534-544. 12. Arvidsson S, Pe´rez-Rodrı´guez, Mueller-Roeber B: A growth phenotyping pipeline for Arabidopsis thaliana integrating image analysis and rosette area modeling for robust quantification of genotype effects. New Phytol 2011, 191: 895-907. This paper first introduces the notion of the phenotyping pipeline. The authors combine data automatically extracted and processed from a high-throughput phenotyping platform with manually annotated information to analyse shoot growth in A. thaliana. They illustrate that very www.sciencedirect.com
13. Skirycz A, Vandenbroucke K, Clauw P, Maleux K, De Meyer B, Dhondt S, Pucci A, Gonzalez N, Hoeberichts F, Tognetti VB et al.: Survival and growth of Arabidopsis plants given limited water are not equal. Nat Biotechnol 2011, 29:212-214.
15. Clark RT, MacCurdy RB, Jung JK, Shaff JE, McCouch SR, Aneshansley DJ, Kochian LV: Three-dimensional root phenotyping with a novel imaging and software platform. Plant Physiol 2011, 156:455-465. 16. Vasseur F, Pantin F, Vile D: Changes in light intensity reveal a major role for carbon balance in Arabidopsis responses to high temperature. Plant Cell Environ 2011, 34:1563-1576. 17. Rousseau C, Belin E, Bove E, Rousseau D, Fabre F, Berruyer R, Guillaumes J, Manceau C, Jacques MA, Boureau T: High throughput quantitative phenotyping of plant resistance using chlorophyll fluorescence image analysis. Plant Methods 2013, 9:17. 18. Woo NS, Badger MR, Pogson BJ: A rapid, non-invasive procedure for quantitative assessment of drought survival using chlorophyll fluorescence. Plant Methods 2008, 4:27. 19. Costa JM, Grant OM, Chaves MM: Thermography to explore plant–environment interactions. J Exp Bot 2013, 64:3937-3949. 20. Matsuda O, Tanaka A, Fujita T, Iba K: Hyperspectral imaging techniques for rapid identification of Arabidopsis mutants with altered leaf pigment status. Plant Cell Physiol 2012, 53:1154-1170. 21. Mahlein AK, Steiner U, Hillnhutter C, Dehne HW, Oerke EC: Hyperspectral imaging for small-scale analysis of symptoms caused by different sugar beet diseases. Plant Methods 2012, 8:3. 22. Fabre J, Dauzat M, Negre V, Wuyts N, Tireau A, Gennari E, Neveu P, Tisne S, Massonnet C, Hummel I, Granier C: PHENOPSIS DB: an information system for Arabidopsis thaliana phenotypic data in an environmental context. BMC Plant Biol 2011, 11:77. 23. Pound MP, French AP, Atkinson J, Wells DM, Bennet JM, Pridmore T: RootNav: navigating images of complex root architectures. Plant Physiol 2013, 162:1802-1814. 24. Lobet G, Draye X: Novel scanning procedure enabling the vectorization of entire rhizotron-grown root systems. Plant Methods 2013, 9:2-10. 25. De Vylder J, Vandenbussche F, Hu Y, Philips W, Van Der Straeten D: Rosette Tracker: an open source image analysis tool for automatic quantification of genotype effects. Plant Physiol 2012, 160:1149-1159. 26. Herridge RP, Day RC, Baldwin S, Macknight RC: Rapid analysis of seed size in Arabidopsis for mutant and QTL discovery. Plant Methods 2011, 7:3. 27. Hartmann A, Czauderna T, Hoffmann R, Stein N, Schreiber F: HTPheno: an image analysis pipeline for high-throughput plant phenotyping. BMC Bioinformatics 2011, 12:148. 28. Weight C, Parnham DRW: LeafAnalyser: a computational method for rapid and large-scale analyses of leaf shape variation. Plant J 2008, 53:578-586. 29. Bylesjo¨ M, Segura V, Soolanayakanahally RY, Rae AM, Trygg J, Gustafsson P, Jansson S, Street NR: LAMINA: a tool for rapid quantification of leaf size and shape parameters. BMC Plant Biol 2008, 8:82. 30. Zhang X, Hause RJ, Borevitz JO: Natural genetic variation for growth and development revealed by high-throughput phenotyping in Arabidopsis thaliana. G3 (Bethesda) 2012, 2: 29-34. 31. Massonnet C, Tisne S, Radziejwoski A, Vile D, De Veylder L, Dauzat M, Granier C: New insights into the control of endoreduplication: endoreduplication could be driven by Current Opinion in Plant Biology 2014, 18:96–102
102 Genome studies and molecular genetics
organ growth in Arabidopsis leaves. Plant Physiol 2011, 157: 2044-2055. 32. Perez-Harguindeguy N, Diaz S, Garnier E, Lavorel S, Poorter H, Jaureguiberry P, Bret-Harte MS, Cornwell WK, Craine JM, Gurvich DE, Urcelay C et al.: New handbook for standardised measurement of plant functional traits worldwide. Aust J Bot 2013, 61:167-234. 33. Massonnet C, Vile D, Fabre J, Hannah MA, Caldana C, Lisec J, Beemster GTS, Meyer RC, Messerli G, Gronlund JT et al.: Probing the reproducibility of leaf growth and molecular phenotypes: a comparison of three Arabidopsis accessions cultivated in ten laboratories. Plant Physiol 2010, 152:2142-2157. 34. Malosetti M, Ribaut JM, van Eeuwijk FA: The statistical analysis of multi-environment data: modeling genotype-byenvironment interaction and its genetic basis. Front Physiol 2013, 4:44. 35. Reymond M, Muller B, Leonardi A, Charcosset A, Tardieu F: Combining quantitative trait loci analysis and an ecophysiological model to analyze the genetic variability of the responses of maize leaf growth to temperature and water deficit. Plant Physiol 2003, 131:664-675. 36. Lievre M, Wuyts N, Cookson SJ, Bresson J, Dapp M, Vasseur F, Massonnet C, Tisne S, Bettembourg M, Balsera C et al.: Phenotyping the kinematics of leaf development in flowering plants: recommendations and pitfalls. WIREs Dev Biol 2013, 2:809-821. 37. Andrade-Sanchez P, Gore MA, Heun JT, Thorp KR, Carmo-Silva AE, French AN, Salvucci ME, White JW: Development and evaluation of a field-based high-throughput phenotyping platform. Funct Plant Biol 2013 http://dx.doi.org/10.1071/FP13126. 38. Stinchcombe JR, Kirkpatrick M: Genetics and evolution of function-valued traits: understanding environmentally responsive phenotypes. Trends Ecol Evol 2012, 27:637-647. The Function-valued Traits Working Group advocate for the systematic analysis of trait variation as a function of time and/or environment. While challenging this would considerably improve our ability to disentangle the genotype-to-phenotype map. 39. Cole B, Kay SA, Chory J: Automated analysis of hypocotyl growth dynamics during shade avoidance in Arabidopsis. Plant J 2010, 65:991-1000. 40. Walter A, Silk WK, Schurr U: Environmental effects on spatial and temporal patterns of leaf and root growth. Annu Rev Plant Biol 2009, 60:279-304. 41. Mendez-Vigo B, de Andres MT, Ramiro M, Martinez-Zapater JM, Alonso-Blanco C: Temporal analysis of natural variation for the rate of leaf production and its relationship with flowering initiation in Arabidopsis thaliana. J Exp Bot 2010, 61: 1611-1623. 42. Billiau K, Sprenger H, Schudoma C, Walther D, Kohl KI: Data management pipeline for plant phenotyping in a multisite project. Funct Plant Biol 2012, 39:948-957. 43. Chapuis R, Delluc C, Debeuf R, Tardieu F, Welcker C: Resiliences to water deficit in a phenotyping platform and in the field: how related are they in maize? Eur J Agron 2012, 42:59-67. The authors show that sensitivity of maize leaf growth to water deficit measured in a phenotyping platform set up in a greenhouse is highly correlated with that of seed number in a network of field experiments. 44. Houle D, Govindaraju DR, Omholt S: Phenomics: the next challenge. Nat Rev Genet 2010, 11:855-866. 45. Fournier-Level A, Wilczek AM, Cooper MD, Roe JL, Anderson J, Eaton D, Moyers BT, Petipas RH, Schaeffer RN, Pieper B et al.: Paths to selection on life history loci in different natural environments across the native range of Arabidopsis thaliana. Mol Ecol 2013, 22:3552-3566. The authors used common garden experiments to demonstrate the importance of considering environmental variation in QTL detection, and how this genetic information can be included in path models aiming at understanding the determinants of plant fitness.
Current Opinion in Plant Biology 2014, 18:96–102
46. Tisne´ S, Reymond M, Vile D, Fabre J, Dauzat M, Koornneef M, Granier C: Combined genetic and modeling approaches reveal that epidermal cell area and number in leaves are controlled by leaf and plant developmental processes in Arabidopsis. Plant Physiol 2008, 148:1117-1127. 47. Vile D, Shipley B, Garnier E: A structural equation model to integrate changes in functional strategies during old-field succession. Ecology 2006, 87:504-517. 48. Shipley B: Cause and Correlation in Biology: A User’s Guide to Path Analysis, Structural Equations, and Causal Inference. Cambridge, UK: Cambridge University Press; 2000, . 49. Pavlicev M, Wagner GP: A model of developmental evolution: selection, pleiotropy and compensation. Trends Ecol Evol 2012, 27:316-322. This paper highlights the potential evolutionary role of rQTL, i.e. QTL that affect the relationships between traits such as allometric relationships or trade-offs. Examples of rQTL in plants can be found in [45] and [49]. 50. Vasseur F, Violle C, Enquist BJ, Granier C, Vile D: A common genetic basis to the origin of the leaf economics spectrum and metabolic scaling allometry. Ecol Lett 2012, 15:1149-1157. 51. Li R, Tsaih SW, Shockley K, Stylianou IM, Wergedal J, Paigen B, Churchill GA: Structural model analysis of multiple quantitative traits. PLoS Genet 2006, 2:e114. 52. Poorter H, Anten NP, Marcelis LFM: Trait correlation networks: a whole-plant perspective on the recently criticized leaf economic spectrum. New Phytol 2013:1-5. 53. Poorter H, Lambers H, Evans JH: Physiological mechanisms in plant growth models: do we need a supracellular systems biology approach? Plant Cell Environ 2013, 36:1-18. In this paper, the authors present trait correlation networks based on various large-scale screenings and they show how a correlation network changes when the environment changes: a first step towards dose– response curve of trait correlation networks across environmental gradient. 54. Laughlin D: The intrinsic dimensionality of plant traits and its relevance to community assembly. J Ecol 2013 http://dx.doi.org/ 10.1111/1365-2745.12187. 55. Vile D, Pervent M, Belluau M, Vasseur F, Bresson J, Muller B, Granier C, Simonneau T: Arabidopsis growth under prolonged high temperature and water deficit: independent or interactive effects? Plant Cell Environ 2012, 35:702-718. 56. Westoby M, Falster DS, Moles AT, Vesk PA, Wright IJ: Plant ecological strategies: some leading dimensions of variation between species. Ann Rev Ecol Evol Syst 2002, 33:125-159. 57. Topp CN, Iyer-Pascuzzi AS, Anderson JT, Lee CR, Zurek PR, Symonova O, Zheng Y, Bucksch A, Mileyko Y, Galkovskyi T et al.: 3D phenotyping and quantitative trait locus mapping identify core regions of the rice genome controlling root architecture. Proc Natl Acad Sci USA 2013, 110:E1695-E1704. 58. van Eeuwijk FA, Bink MCAM, Chenu K, Chapman S: Detection and use of QTL for complex traits in multiple environments. Curr Opin Plant Biol 2010, 13:193-205. 59. Boer MP, Wright D, Feng LZ, Podlich DW, Luo L, Cooper M, van Eeuwijk FA: A mixed-model quantitative trait loci (QTL) analysis for multiple-environment trial data using environmental covariables for QTL-by-environment interactions, with an example in maize. Genetics 2007, 177:1801-1813. 60. Kirkpatrick M, Lofsvold D, Bulmer M: Analysis of the inheritance, selection and evolution of growth trajectories. Genetics 1990, 124:979-993. 61. Blonder B, Lamanna C, Violle C, Enquist B: The n-dimensional hypervolume. Global Ecol Biogeogr 2014. (in press). 62. Sozzani R, Busch W, Spalding EP, Benfey PN: Advanced imaging techniques for the study of plant growth and development. Trends Plant Sci 2014. (in press).
www.sciencedirect.com
