Current Biology

Review

Plant Phenomics, From Sensors to Knowledge

François Tardieu1,*, Llorenç Cabrera-Bosquet1, Tony Pridmore2, and Malcolm Bennett3,*
1INRA, Laboratoire d’Ecophysiologie des Plantes sous Stress Environnementaux, F34060, Montpellier, France
2School of Computer Science, University of Nottingham, NG8 1BB, UK
3Plant & Crop Sciences, School of Biosciences, University of Nottingham, LE12 3RD, UK

*Correspondence: francois.tardieu@inra.fr (F.T.), malcolm.bennett@nottingham.ac.uk (M.B.)

http://dx.doi.org/10.1016/j.cub.2017.05.055

Major improvements in crop yield are needed to keep pace with population growth and climate change. While
plant breeding efforts have greatly benefited from advances in genomics, profiling the crop phenome (i.e., the
structure and function of plants) associated with allelic variants and environments remains a major technical
bottleneck. Here, we review the conceptual and technical challenges facing plant phenomics. We first
discuss how, given plants’ high levels of morphological plasticity, crop phenomics presents distinct chal-
lenges compared with studies in animals. Next, we present strategies for multi-scale phenomics, and
describe how major improvements in imaging, sensor technologies and data analysis are now making
high-throughput root, shoot, whole-plant and canopy phenomic studies possible. We then suggest that
research in this area is entering a new stage of development, in which phenomic pipelines can help re-
searchers transform large numbers of images and sensor data into knowledge, necessitating novel methods
of data handling and modelling. Collectively, these innovations are helping accelerate the selection of the
next generation of crops more sustainable and resilient to climate change, and whose benefits promise to
scale from physiology to breeding and to deliver real world impact for ongoing global food security efforts.

Introduction then discuss the challenges of studying these processes in

Genetic improvement of crop resilience to abiotic stresses and the naturally fluctuating conditions in the field versus controlled
new pests arising from climate change is imperative to ensure environment conditions [11–14]. Finally, we suggest that phe-
future food security [1,2]. The increasing use of gene editing nomics is entering a new stage of development, necessitating
[3,4] and continued exploitation of natural genetic variability novel methods for data handling, statistical approaches and
[5,6] provide invaluable opportunities for generating novel alleles modelling to connect and interpret the knowledge generated
and selecting natural sources of genetic variation for crop at different scales (see Box 1).
improvement [2]. This requires analysis of hundreds of lines
grown under diverse environmental scenarios. While genotyping One Genotype, Many Phenotypes in Plants
has reached this throughput at a relatively low cost through ad- Plant phenomics does not consist of solely associating a geno-
vances in DNA marker assays and sequencing technologies [7], type to one phenotype in a given condition (e.g., in a controlled
equivalent improvements to generate high-throughput and valu- environment), but rather in characterizing the plasticity of the
able phenotypic information are urgently needed [8]. This is the plant phenome when exposed to a range of environmental con-
object of the field of plant phenomics, which we define here as ditions. In contrast to most animals, which essentially retain the
the development and application of the suite of tools and same structure regardless of their environment, a plant can form
methods used for three major goals — capturing information on very different architectures depending on environmental condi-
structure, function and performance of large numbers of plants, tions. For example, the same variety of Arabidopsis thaliana
together with their environment; analysing, organizing and stor- can exhibit a large 30-leaf plant or a small 8-leaf plant, after being
ing the resulting datasets; and developing models able to disen- exposed to either short or long day conditions, respectively
tangle and simulate plant behaviour in a range of scenarios. (Figure 1A,B) [15]. Similarly, water deficit, nitrogen deprivation
Over the past decade, plant phenomics has made impressive and low light have major effects on the number and size of plant
progress, developing novel sensors and imaging techniques for organs (Figure 1B,C). As a consequence, plant phenomics
a wide range of traits, organs and situations [8–10]. However, research dedicates a large amount of effort to the study of vari-
data handling and processing remain major challenges when ation in organism structure, whereas animal phenomics essen-
translating sensor information into knowledge. This Review tially focuses on metabolism.
focuses on that translation. We first discuss the reasons why Plant water status, temperature, fluxes or growth rate vary within
the challenges differ between plant and animal phenomics, minutes. Indeed, unlike mammals and birds, plants are not ho-
which are largely due to the strong interaction between environ- meostatic for temperature and water under rapidly fluctuating
mental conditions and plant structure, function and meta- environmental conditions. Plants transpire 50–200% of their own
bolism. We suggest the need for a multi-scale strategy that links weight daily (vs. 1–2% for humans) [16], while their temperature fol-
physiological mechanisms with plant performance across ge- lows their energy balance, resulting in rapid variations [17]. During
notypes and environments from the molecular to field scales, a summer day, a plant can be at 11 C with a favorable water status
based on a series of novel approaches and techniques. We in the early morning, but then experience 36 C and suffer severe

R770 Current Biology 27, R770–R783, August 7, 2017 ª 2017 Elsevier Ltd.
Current Biology

Review

Box 1. Glossary of terms

Convolutional neural nets (CNNs): CNNs are a variant of traditional artificial neural networks (ANNs), machine-learning methods
inspired by biological neuronal systems. Traditional neural nets take a pre-determined set of measurements or features as their
input and learn to perform various tasks; classification is by far the most common. CNNs extend scope of ANNs, learning both
how to achieve the task and what measurements are needed. CNNs operate over raw sensor data, and learn how to extract
the necessary features.
Genome-wide association studies (GWAS): this consists in associating markers in the genome with phenotypes (omic, traits or
yield) through a statistical analysis. The values of alleles at one genome position are associated with a quantitative increase or
decrease of phenotypic values.
Genotype x environment interaction (GxE) : the ranking of a set of genotypes differs between experiments for every trait or yield.
GxE can be extracted from a statistical model, or can be analysed in detail using regressions of the considered trait with environ-
mental variables. GxE is then analysed through the variability of slopes of these regressions.
Genomic selection (GS): represents a novel approach to marker-assisted breeding where, rather than attempting to identify in-
dividual loci significantly associated with a trait, GS uses all marker data as predictors of performance to deliver more accurate
predictions.
Laser scanning systems: a 3D reconstruction method in which a known pattern of light (a line, grid or array of dots) is projected
onto the target object by a laser light source. A camera, often fitted with a filter making the laser pattern easier to detect, views the
reflected pattern. 3D is recovered from differences in the projected and viewed patterns of light.
Magnetic resonance imaging (MRI): MRI is a 3D imaging modality in which the target sample is placed in a strong magnetic field.
Under these conditions some atomic nuclei, particularly hydrogen nuclei, absorb and emit radio frequency energy. Pulses of radio
waves excite the hydrogen atoms, which emit signals that are detected by nearby antenna. The magnetic field allows these signals
to be localised, mapping hydrogen atoms and so water.
Multi-view stereo: a 3D reconstruction technique in which multiple, usually colour, images are taken of a target object from
different viewpoints. Features of interest are identified by independent analysis of each individual image. These features are
then matched between images — features are matched if they are considered to depict the same point on the target object.
The cameras’ viewpoints are obtained by calibration and the 3D location of each object feature is recovered by triangulation.
Phenotype: here, we mean the profiling of the structures and functions associated with allelic variants, at the scales of cells (omic
phenotyping), organs (main plant functions), whole plant (controls of these functions) and canopy (plant performance).
Quantitative trait loci (QTL): QTLs are regions of the genome containing one or more genes, associated with variation of a quan-
titative trait (phenotype). QTLs are identified by showing a statistical association between polymorphic markers and the measured
phenotype.
Unmanned airborne vehicle (UAV): helicopters, drones or small planes able to fly over a field experiment, carrying a diversity of
sensors. Their trajectory is programmed using GPS.
X-ray micro-computed tomography (mCT): X-ray CT produces a 3D image in which each element (voxel) contains a value pro-
portional to the density of the imaged object. The target object is placed on a rotating stage inside the imaging device. An emitter
projects X-rays through the rotating sample to a detector on the other side of the device. The detector records the X-ray energy
passing through the object. Density can be estimated from the difference in projected and detected X-ray energy.

water stress six hours later (Figure 1E), triggering spectacular different organs that exchange information through, for example,
changes in plant morphology (Figure 1D–F). Displacement trans- hormonal or hydraulic messages [25–27]. Such exchanges of in-
ducers reveal that, under these conditions, plants exhibit rapid formation operate at short-term scales at the cell or organ levels,
fluctuations in growth [18]. Leaf elongation can occur at a rate of and translate into long-term plant or canopy behaviours through
4 mm per hour at dawn versus 0 at 2pm [18]. Hence, although whole-plant mechanisms that are highly non-linear. Hence, plant
some degree of homeostasis exists at the cellular level [19,20], phenomics requires analyses at spatial scales ranging from
this is not the case at the organism level. Many molecular events single cell to canopy, and temporal scales ranging from minutes
occur during transitions between different environmental condi- (for metabolism and hydraulics) to months (for yield) (Table 1).
tions [21], so phenomic analysis of non-stable states is essential. Modelling is, therefore, an intrinsic part of plant phenomics,
The low degree of homeostasis in plants also results in large func- aimed at connecting these scales. Indeed, while often non-intu-
tional consequences of the spatial variability of conditions a plant itive, feedback mechanisms are predictable using mathematical
is exposed to [17]. For example, root system architecture exhibits models [28].
large spatial variation reflecting local adaptation to highly hetero-
geneous soil water content [22,23]. Hence, the analysis of pheno- Analysing the Plant Phenome across Spatial and
typic datasets needs consistent time course information on Temporal Scales
environmental conditions as sensed by plants and organs, Given the issues raised above, plant phenomics needs to
together with growth and physiology-related processes. capture and interpret a multi-dimensional matrix of functional
Because there is no central ‘orchestrator’ organ in plants [24], and architectural variables measured at different scales (or-
the control of most functions relies on feedback loops involving gan, plant, canopy), developmental stages and environmental

Current Biology 27, R770–R783, August 7, 2017 R771

 Current Biology

Review

A B C Figure 1. Illustrations of phenotypic

Short day Long day Long day plasticity.
Low ev. demand Low ev. demand High ev. demand Arabidopsis plants under low evaporative demand
with short (A) or long (B) day, or under high evap-
orative demand (C) from [15]. Note the differences
in leaf number and leaf size. (D,E) Maize plants in
the morning and early afternoon and time courses
of leaf temperature (T, from 11 to 36 C) and leaf
water potential (J, MPa) during the day. A leaf
water potential of 0 MPa means free water,
whereas 1.5 MPa is close to lethal values in many
species. In the lower panel of E, symbols are
measurements, lines are an interpolation using a
model. In (F), the change in canopy aspect is due
D E F to leaf rolling, a symptom of water stress. Panels
8 am 2 pm (D), (F) kindly provided by C. Fournier, INRA LEPSE
30 Montpellier France.
Leaf T °C

20

10

elemental composition. Many of these

-0.5
mutants were later shown to be defective
Ψ MPa

-1
for an impermeable barrier in roots,
-1.5 termed the Casparian strip, that regulates
0 6 12 18 0
loading of elements into vascular tissues
Time of day Current Biology
[35]. Metabolomic-based methods profile
the compounds involved in major meta-
scenarios. To address this inherent complexity, researchers bolic pathways. This approach has helped discover how
have developed three categories of phenotyping platforms different genotypes cope with environmental cues, uncovering
that have distinct objectives and employ different approaches the dialogue between the circadian clock and changes in light
and methods (Table 1). availability that allow plants to optimize the use of starch re-
High-Precision Platforms serves [36]. Fluxomics (i.e., in situ imaging of the concentration
These platforms operate at the organ level, most often over short and fluxes of elements [37]) have provided important insights
time scales (Table 1). They aim at the identification of physiolog- about where, when and how water and nutrients are transferred
ical mechanisms allowing plants to respond to changes in envi- in the plant [38]. For example, MRI-PET-based imaging has
ronmental conditions, and lead to the elucidation of their genetic enabled researchers to map carbon flow from leaves to individ-
control. ual roots [39]. Imaging-based phenomic approaches can also be
Profiling organ growth and architecture is used in high-preci- employed in cell-scale profiling studies. For example, eGFP- and
sion platforms to uncover adaptive mechanisms associated FRET-based sensors have proved highly effective for monitoring
with environmental signals. For example, X-ray micro-computed the spatio-temporal dynamics of hormones and elements like
tomography (mCT; Figure 2A–D) of roots growing in soil macro- zinc in Arabidopsis root cells [40–42] (Figure 3D) and uncover
pores revealed the importance of direct contact with soil water new mechanistic insights into their homeostatic regulation.
to determine where new lateral root branches are positioned The examples above highlight how high-precision platforms
(Figure 2E) [23]. This translates into improved water and nutrient are effective for discovering new physiological mechanisms,
uptake (Figure 2F). Similarly, time-lapse 3D imaging of leaves and also for upscaling them from organ to plant level. Neverthe-
combined with computational modelling allows identification of less, at their current stage of development, these platforms
where and when tissue expansion and cell division occur [29]. cannot analyse the many thousands of plants needed to perform
In the case of leaves or sepals, this approach has revealed genetic studies across a range of environments over a whole
how new buds with few cells result in reproducible shapes life-cycle timescale. Hence, they are not directly relevant for
through feedback between patterns of oriented growth and tis- upscaling mechanisms to predict important traits such as yield
sue deformation (Figure 3A) [30,31]. Analyzing the leaf elongation (Table 1).
rate of maize plants with displacement transducers at high tem- Field Multi-Environment Networks
poral resolution (i.e., minutes) in contrasting and fluctuating con- At the other extreme of plant phenomics, ‘field multi-environ-
ditions allowed identification of a novel mechanism of drought ment networks’ (Table 1) are series of field experiments distrib-
adaptation. This mechanism involves regulatory interactions be- uted in a geographical region, aimed at uncovering the genetics
tween circadian control of plant hydraulic properties, daily time of yield stability. They probe the genetic control of plant perfor-
course of evaporative demand and hydraulic properties of the mance in a range of environmental scenarios, without pre-
rhizosphere (i.e., roots and the adjacent soil) [32]. conceived reference to a particular mechanism.
The composition of plant tissues and the metabolite fluxes The yield of a given genotype often differs between field sites,
through organs can be characterized in ‘omics-based’ platforms as does the ranking of genotypes (genotype by environment
for thousands of plants [33]. For instance Ionomics employs ICP- interactions; GxE) [43,44]. Indeed, the relationship between yield
MS to perform elemental profiling [34]. This has allowed identifi- and environmental conditions results from trade-offs between
cation of Arabidopsis mutants whose leaves have altered mechanisms that have distinct optima [45]. The relationship

R772 Current Biology 27, R770–R783, August 7, 2017

Current Biology

Review

Table 1. Phenotyping at different scales of organization.

Typical phenotyping platforms High-precision platforms Whole-plant, multi-environment Field multi-environment networks
(omic, anatomy, organ) platforms (field or controlled)
Level of plant organization Organ Plant or canopy Canopies in a range of environments
Typical methods Omics, 4D organ imaging, 4D plant/canopy imaging fluxes, Yield, sensor network, remote
fluxes sensors sensing
Typical mechanism Hydraulics, metabolism, Light interception, water Trade-offs between processes
signalling transfer, whole-plant signalling
Ratio biology/other processes

Relevance for yield prediction

Methods for trans-scale

communication Gene editing, plant simulation

GWAS, model-assisted dissection

between genotype and phenotype therefore needs to be ana- function of environmental conditions [52–56]. More elaborate
lysed in clusters of microclimatic conditions referred to hereafter traits such as the utilisation efficiency of water, light or nutrients
as environmental scenarios [46,47]. For example, a network of 29 can be calculated from these data using functional/structural plant
field experiments across Europe was used to grow a maize di- models (Figure 2F; Figure 4J) [57–59]. This is illustrated in Figure 4,
versity panel and identify genomic regions associated with yield in which 4D imaging of whole plants and the mapping of inci-
(quantitative trait loci, QTLs) under heat or water stresses [48]. dent light in a greenhouse makes it possible to disentangle the
Nearly all QTLs had conditional effects, positive, negative or biomass accumulation of thousands of plants into well-defined
null, depending on environmental scenarios. For instance, an processes, such as the amount of light intercepted by each
allele at one QTL that controls the biosynthesis of the stress plant (a function of leaf area and geometry) and the photosynthetic
hormone abscisic acid (ABA) was favourable in drought but ability of each plant [58]. Crop models can then be used to connect
detrimental in well-watered situations. Hence, a large number the genetic variability of these processes to yield [60].
(typically 20–40) of experiments needs to be conducted under High-throughput field phenotyping has progressed rapidly in
diverse environmental conditions to explore such allelic effects. the last five years, based on the use of multi-spectral 4D analyses
Genomic selection (GS) extends the former approach to estab- with sensors mounted on mobile systems such as gantries
lish predictions of the best combinations of alleles for yield [61,62], ground vehicles or drones [63,64] (Figure 3C,F). They offer
[49]. GS requires phenotyping of hundreds/thousands of geno- the possibility of estimating the genetic variability of yield, biomass
types (the ‘training population’), in some cases together with accumulation and underlying processes in a variety of environ-
the effects of environmental conditions [50]. The best combi- mental scenarios. For example, canopy temperature provides a
nations of alleles are then used to select, in silico, tens of proxy for genetic differences in transpiration, which is often due
thousands of plants, thereby avoiding direct phenotyping of to variation in root system architecture [63,64] (Figure 2F).
these plants [50]. Cross-Scale Meta-Analyses
Whole-Plant, Multi-Environment Platforms Currently, joint analyses of field experiments have been
Given the complex interactions between QTL and environment performed across years and sites [43,65]. While cross-scale
and between QTLs [51], the interpretation of results generated approaches are also beginning to appear [66], they need to be
by networks of field experiments is typically challenging, developed further.
making it difficult to relate gene alleles with physiological No single plant phenomic platform can analyse every scale,
mechanisms. To achieve this goal, a third category of plant throughput or environment. For example, it would be misleading
phenomic platforms, ‘whole-plant, multi-environment plat- to measure yield in greenhouse experiments, as the amount and
forms’, has been developed. These platforms are highly instru- spectrum of light available to plants in a greenhouse and the distri-
mented greenhouses or fields allowing one to follow and dissect bution of roots in the soil in pots would make any attempt irrelevant
variables such as the growth or transpiration of thousands of [11]. Reciprocally, phenotyping of thousands of varieties in tens
plants or small canopies, thereby allowing their genetic analysis of field experiments is not compatible with costly and labour-
(Table 1). intensive methods. A combination of approaches is therefore
Highly automated platforms in greenhouses enable researchers necessary, which we term ‘cross scale meta-analyses’. For
to perform 4D characterisation of the architecture of shoot, root instance, the plasticity of yield can be analysed in ‘field multi-envi-
or canopy systems of hundreds of genotypes (Figure 3B,E; ronment networks’. The underlying genetic variability of trait adap-
Figure 4A). They allow genetic analyses of traits such as root tation can then be analysed in ‘whole-plant, multi-environment’
and shoot topology, angles, branching and growth rate as a platforms for the same panels of genotypes, thereby associating

Current Biology 27, R770–R783, August 7, 2017 R773

 Current Biology

Review

A X-ray CT phenotyping platform B CT cross-section data

Roots in soil; 25 µm resolution

C D Quantification of root traits

Root system recovery

Quantification of global and local root system traits for plant

Root system of maize
phenotyping

E Response to environmental signals F Simulate uptake efficiency

Soil–root and root–root interaction CT data embedded in a finite element mesh

Current Biology

(legend on next page)

R774 Current Biology 27, R770–R783, August 7, 2017

Current Biology

Review

HP organ Whole plants Canopy Figure 3. Novel imaging techniques at organ

scale with high-precision (HP) platforms,
A B C at plant scale with whole-plant multi-
environment platforms and at canopy scale.
(A) Heat map denoting areal leaf expansion rate
using time-lapse imaging and computer modelling
Leaf, (red to green, rapid to slow growth), from [30].
leaf area Reprinted with permission from AAAS. (B) 3D
representation of a maize plant from multiple im-
ages, at a throughput of 1000s plants/day. Colors
indicate the amount of light received by each pixel
of plant. (C) Multi-spectral (NDVI) image of a can-
opy; increasingly red colors represent increasing
leaf area per unit m2 of soil. (D) Image of an auxin
biosensor in the Arabidopsis primary root obtained
D E F by confocal imaging [122]. (E) Whole-plant root
(i) system imaged in a rhizotron at throughput of
1000s plants/day. Inset, zoom on root nodules.
From [53]. (F) Image of a canopy in the thermal
infrared; increasingly red colors indicate lower
(ii) transpiration rate, often linked to an unfavorable
Roots, root system. Horizontal regions with distinct colors:
(i) non-irrigated plot, (ii) irrigated plot. Note in (i) the
transpiration superposition of spatial patterns with specific ef-
(i)
fects of genotypes in different plots. Panel B kindly
provided by C. Fournier, INRA LEPSE Montpellier
France. Panels (C) and (F) kindly provided by F.
(ii) Baret, INRA CAPTE Avignon France.
Current Biology

QTLs affecting yield in specific environments to allelic variations of in wireless networks to characterize the micro-environment of
traits. The resulting alleles can be tested for their effects on mech- many organs in a plant and many plants in a canopy (Figure 5,
anisms of plant adaptation in ‘high precision’ platforms [67]. Such arrows 1 and 2). This progressively applies to the characteriza-
meta-analyses are particularly vital in the case of root studies, in tion of the soil environment by combination of soil sensors with
which the root architecture or growth can only be analysed in modelling [70]. This local information can be scaled up to
‘high precision’ using ‘whole plant multi-environment platforms’, whole-platform, field or regional levels using local, UAV or satel-
whereas only consequences can be observed in the field, for lite imaging, respectively. This allows efficient mapping of envi-
instance through differences in canopy temperature (Figure 3D–F). ronmental variables, thereby characterizing and capturing the
effects of the spatial and temporal variation of growth conditions
Employing Trans-Scale Analyses to Link Sensors with sensed by individual plants or fields (Table 1).
Knowledge Consistent Analysis of Images and Time Series
Cross-scale meta-analyses, as defined above, require consis- Imaging systems have progressed exponentially in recent
tent methods for recovering data across all platforms, time years, with a variety of non-invasive and information-rich tech-
scales and levels of plant organization (Table 1 and Figure 3). niques (e.g., laser microscopy and rangefinders, X-ray mCT,
We discuss below the major challenges researchers face to multi- and hyper-spectral cameras, isotope tracing methods).
achieve this ambitious, yet essential, next step in plant phe- These techniques have recently been reviewed in detail [8,71]
nomic research. and can be used at a variety of scales to support the 4D
Environmental Characterization, Sensor Networks functional analysis of root or shoot systems, and capture the
In our own experience, the analysis of datasets originating from structure and physiological status of plants (Figure 3). How-
different experiments and groups often suffers from a lack of ever, imaging devices and protocols perform photography, not
consistent environmental information, which makes it impossible phenotyping: traits need to be recovered from raw image
to analyse and model the differences in plant behaviour between data via image analysis (Figure 5, arrows 1 and 2). We discuss
experiments. To that end, several research consortia have pro- some of the key issues and solutions below.
posed ‘minimum environmental datasets’ with the necessary Many software tools dedicated to image analysis of shoots [72],
environmental variables and protocols for data analysis and roots [73,74], canopies [75], leaves [76], seeds [77] and fruit [78]
modelling at any scale [68,69]. Furthermore, a full environmental have been developed in recent years. An increasing number of
characterisation is now being facilitated by rapid progress in these tools offer realistic and non-invasive 3D reconstructions
sensor technology. Cost-effective sensors can now be placed of plant organs [79], based on the combination of multi-view

Figure 2. Plant root phenotyping pipeline using X-ray micro-computed tomography (mCT).
(A) mCT scanning system to non-invasively image columns of soil-grown plants (ranging in resolution from 0.5–150 mm). (B) Example 2D cross-sectional image
generated with mCT scanner showing root material (in red) and the heterogeneous structure of soil (soil and water in grey, air spaces in black). (C) Image analysis
software [74] can be used to recover root system of maize from the mCT volume data after segmenting roots from thousands of 2D image slices, (D) and quantify
root system traits, (E) to discover new root responses to environmental signal, like how soil water distribution patterns the positioning of lateral root branches [23],
and (F) parameterise models to simulate growth and foraging for natural resources by root systems.

Current Biology 27, R770–R783, August 7, 2017 R775

 Current Biology

Review

Phenotyping platform Data aquisition

A B C
Environmental data (sensors) Leaf area/biovolume
Transpiration rates (1000s plants)
12 images/plant

10 20 30 40 50
Time (d)

Spatial distribution of incident light 3D plants and light interception

D E F G
0

70
60 50 40 30 20 10

60
Y Position

50
40
30
20
10
0
6 12 18 24
X Position
Functional-structural plant model,
Modelling sunbeam trajectories
distribution in the greenhouse
{
From radiation-use efficiency to photosynthesis

H n
Biomass = Σ PPFDi × εi × RUEi
i=1

I J “
RUE (g FW mol-1)

11
(g plant-1)
Biomass

2
r 0.54
P 0.001
5
15 25 35
Intercepted light Photosynthesis
(mol plant-1) -2 -1
(μmol m s ) Current Biology

Figure 4. Light interception, photosynthesis and radiation use efficiency, from images to function.
(A) Phenotyping platform (PhenoArch) where 1680 plants can be grown in controlled conditions of soil water status and temperature, imaged and assessed for
transpiration rate. Sensors measure light, relative humidity and air and leaf temperature and transpiration. (B) Twelve images per plant are captured every day
allowing 3D reconstruction. (C) Time courses of leaf area and biovolume are calculated in real time. (D) Spatial distribution of incident light. Images are captured
every m2 in the greenhouse, oriented to the vertical. Blue, sky; black, obstacles (lamps, beams, etc.). The path of sunbeams is modelled every day of the year
(yellow line). This allows calculation of direct and diffuse light in every position of the greenhouse [58]. (F) Virtual digital plants are placed at their positions in a
virtual greenhouse. (G) This allows calculation of light interception by competing plants, in the whole greenhouse [58]. (H) The above steps allow dissection of
biomass accumulation into incident light on day i (PPFDi), the proportion of light intercepted by plants (εi) and radiation use efficiency (RUEi, ratio of biomass
production to intercepted light). (I) RUE is presented for three plants in (F), pink, green and black. Bars near the x and y axes represent the amounts of cumulated
biomass and intercepted light, the slope of regression lines is RUE. (J) RUE closely correlates with photosynthesis rate in a series of genotypes denoted by
different colors. Note that it would be impossible to directly measure gas exchanges for 1680 plants. Adapted from [58] with permission.

R776 Current Biology 27, R770–R783, August 7, 2017

Current Biology

Review

Information Figure 5. Flow chart of operations during

Steps phenotyping; roles of information systems
systems
Images, sensors and modelling.
Trace sensors and plants
Which ones? Where? (position, timings) The left panel represents steps from image/sensor
What to measure? How often? Ontologies to knowledge; the right panel represents the
rationale for information systems at each step
1 2 (green: tools). Red text represents questions at
each step. Dark blue arrows and text: modelling
tools. Purple arrows: connection between steps.
Time courses: Visible phenotype (1) Transforming raw data into time courses for
Environment or Trace methods environmental data, fluxes, growth rates etc.
(interpreted image) (calibrations,
physiological rates (2) Image analysis to transform a series of images
Which objects? parameters) into a phenotype. (3) and (4) Data analysis with
Variables? Time constants? Workflows
Correlations? Correctly represented? statistical and modelling tools, reproducibility.
(5) Extraction of mechanisms or composite vari-
Model-assisted dissection

3 4 ables encapsulating the genotype x environment

interaction, genetic analysis (6) association of
yields to environmental scenarios, genetic anal-
Multi-scale analysis of phenotypes ysis. (7) Prediction and inference of mechanisms
Reproducible? Trace rationale vs scenario-dependent yields using models.
Spatial/temporal variability? and methods (8) Theory, test using meta-analysis and/or new
Compatible with known mechanisms? Workflows experiments.
Can be ascribed to cues?

5 6
images are acquired. This was done
Composite variables Yield per environmental Trace variables
or Traits and methods with success in a multi-laboratory study
or mechanisms scenarios
Causal polymorphisms Causal polymorphisms Trait ontologies using Arabidopsis thaliana accessions
grown in controlled chambers [91], but
Model prediction Trace models
7 (equations, requires further attention. A phenotyp-
Model inference parameters) ing platform might give different assess-
Workflows ments of the same genotype at two
Theory, novel hypotheses
8 different sites, either because of envi-
New meta-analyses or experiments
Current Biology
ronmental changes or as the result of
variations in the phenotyping process.
Image analysis methods need to be
stereo [80] and modelling [81,82], or use laser-scanning systems both understood [92] and evaluated by comparing their results
[83,84], time-of-flight sensors [85,86], X-ray [74,87] or magnetic with pre-obtained ground truth data [93], allowing identification
resonance imaging [88]. Because plants are structurally complex of the limitation of each method [94].
and highly variable, a given set of sensor or camera viewpoints at The wealth of methods used in phenomics (Figures 3 and 4)
fixed positions cannot provide all the data needed to reconstruct a raises the question of how to jointly analyse image and sensor
complete 3D model of a plant or a canopy. Partial descriptions outputs (Figure 5, arrows 3 and 4). Mixed model approaches
recovered from an initial set of camera views can be used, by solv- have progressed rapidly, allowing genetic analysis of datasets
ing a next-best-view problem, to guide a robot to acquire the data involving different sources of information [95,96]. Novel develop-
needed to complete the model. Indeed, robot-assisted imaging ments allow identification of genotypic means of any variable,
allows a loop to be established between image acquisition, from omics to yield, which are isolated from the noise created
analysis and de novo positioning of sensors at the most insightful by the spatial variability in field or platform experiments (Figures
places in plants [61,83,89]. This opens the way for a dia- 3C,F and 4E), the effect of experimental co-variables (e.g., site,
logue between models, sensors and imaging, enabling high- or persons who performed experiments) and environmental vari-
throughput, high-performance phenotyping of plants or canopies. ables [97]. These ‘best linear unbiased estimates’ (BLUEs) are
Interpretation of sensor or camera outputs requires the mil- then analysed genetically either for individual traits or in multi-
lions of raw data points to be organized into environmental or trait analyses [98].
phenotypic time courses. This first requires the identification of Model-Assisted Phenotyping: Connecting Scales
dubious points due to sensor malfunction or computational Models naturally partner with phenotyping (Figure 5, arrow 7).
errors, inevitable when thousands of sensors are involved, or For example, dynamic models offer the possibility of scaling
when thousands of images are automatically processed. Such up the effects of a short-term mechanism at the organ scale,
data cleaning can now be performed based on statistical or ma- identified in ‘high-precision platforms’, to biomass accumula-
chine-learning methods for the large datasets originating from tion after several time steps in ‘whole-plant, multi-environment
high-throughput platforms [90]. platforms’, or to yield in field networks. Dynamic models are
Data Analysis and Reproducibility Tests based on the discretization of a process into time steps
Making reproducible measurements of the same plants or ac- (e.g., minutes or days). Calculations are iterative, with short-
cessions over time and across platforms requires standardized term effects taken into account at each time step (e.g., the ef-
protocols, including camera calibration, careful selection of fect of light on photosynthesis, with different effects between
number and position of viewpoints and the time of day at which genotypes), and long-term effects emerging from feedbacks

Current Biology 27, R770–R783, August 7, 2017 R777

 Current Biology

Review

(e.g., the uptake of water or nutrients by the plant at a given volumes of temporal data in real-time, thereby allowing one to
time step reduces their availability for the next time step) detect potentially incorrect sensors and to act accordingly.
[99]. Models have been used in plant phenomics in two Secondly, these information systems help organize datasets in
ways [60]. such a way that they can be re-analysed by different groups.
Firstly, the dissection of a phenotype observed on a given day Data identification and annotation involves organizing outputs
into the most likely set of mechanisms (model inference; Figure 5, in such a way that a scientist not involved in the original experi-
arrows 5 and 7). For example, the biomass on a given day can be ments can trace the history of plants, re-analyse images with
dissected into the amount of light received by the plant, multi- new methods of his/her own and a posteriori check the calibra-
plied by the proportion of light intercepted by plants every day, tion of each sensor in case of inconsistencies, possibly years
multiplied by the efficiency with which intercepted light is con- after the experiment was performed.
verted into biomass (Figure 4). Similarly, leaf area can be ana- This requires protocols describing content and format of
lysed as the result of time courses of leaf growth over time, phenotypic information [110], and a formalised description of
resulting from environmental conditions and intrinsic traits of all involved objects (i.e., plants, organs, sensors, phenotyping fa-
the considered genotype [100]. cilities) using ontologies [111,112]. Such ontologies may seem
Secondly, the prediction of a given phenotype from environ- unnecessary in simple experiments where unique correspon-
mental conditions and hypothetical mechanisms observed in dences exist between, for example, each plant and its position
high-precision or whole-plant platforms (Figure 5, arrow 7). in a greenhouse. They become indispensable, however, when
Model prediction operates in the opposite direction compared plants are transferred from one platform to another during an
with dissection, and serves as a test for the proposed mecha- experiment for better multi-scale characterization. In the same
nisms based on their ability to account for an observed pheno- way, sensors are replaced, so calibrations of devices located
type. The set of mechanisms taken into account are written at a given position change with time. Keeping track of these
as equations which result in a phenotype after several time- changes requires open and extensible database schemas based
steps [101]. on ontologies and semantics [111]. This also requires keeping
Hence, modelling is an essential tool for phenomics because it track of all operations, including parameters, used in analyses
helps to develop hypotheses allowing multi-scale interpretations that produce an elaborate result from raw data. Such scientific
of results obtained in the three types of phenotyping infrastruc- workflows are being developed [110], thereby allowing any
tures presented in Table 1. Reciprocally, multi-scale phenomics user to perform the same analysis and obtain the same results
represents a major challenge for modelling. Indeed, phenomic as those published.
technology allows multiple traits that contribute to yield to be Finally, these systems help organise data to facilitate genetic
measured at high temporal resolution, providing a rich data set analyses. Correspondence between phenotype and genotype
against which models can be tested [101]. This avoids compen- requires connection of matrices of genotypic data, consisting
sation of errors associated with each trait underlying yield, a of millions of marker data items or genomic sequences, with
common feature of many current crop models that are parame- associated phenotypic data that synthesize time courses or
trized based on yield only [102]. spatial variation into single figures supporting the genetic ana-
Tracing and Storing All Steps from Data to Knowledge in lyses [113]. Because of the complexity of the information sys-
Information Systems tems reviewed above, and of the need for high calculation
Phenomic experiments are not directly reproducible because power, this is performed in dedicated information systems
of the variability of environmental conditions. It is essential that are physically distinct from those managing dataflow and
that any scientist, including those in 30 years, be able to re- object identification. Hence, maintaining consistency of infor-
use phenotypic data and reproduce the data-flows presented mation across multiple information systems will remain a major
above to perform meta-analyses of the effects of alleles or issue.
mechanisms in a range of environmental conditions. This
has led to the definition of new norms named FAIR (findable, The ‘Big-Data’ Challenge of Plant Phenomics
accessible, interoperable and reusable) [103], primarily for Big data approaches can enhance phenotyping pipelines. Image
tracing data, but also protocols, methods and workflows. analysis methods have typically employed fixed sequences of
They involve information systems capable of managing thou- image processing and measurement processes, crafted by their
sands of data points and images captured during an experi- designers to suit specific procedures. As a result, moving a given
ment, together with the necessary metadata, parameters tool to a slightly different problem or environment often requires a
and methods of data analysis (Figure 5). Such information sys- near-complete rewrite of the software. Recently, deep machine-
tems serve three distinct purposes with different requirements learning methods, and particularly convolutional neural networks
[104–107]. (CNNs), have produced impressive results and been widely
The first purpose is real-time management of the dataflow to adopted in the computer vision community [114,115]. CNNs offer
optimise data quality. Real-time access to images, environ- the potential to provide generic solutions to plant image analysis
mental conditions and metadata is required when managing problems [116] and, rather than requiring tuning to their environ-
the quality of an experiment, in particular for testing (typically ment, benefit most from access to training data spanning multiple
every day) the validity of outputs. This may seem trivial in environments. This brings its own challenges — maximum
small-scale experiments but it is not when thousands of plants benefit can only be gained from deep-learned tools if large-scale
and hundreds of sensors are involved. Protocols [108,109] and datasets (input images and required outputs) capturing shared
management tools [90] have been developed to visualize large problems are made available.

R778 Current Biology 27, R770–R783, August 7, 2017

Current Biology

Review

Big data approaches may also help in making full use of tens of plant adaptation to changing environments and underpin
of experiments with thousands of accessions carried out each secure food security efforts.
year. The formalized meta-analysis of phenotypic data, allowed
by the pipelines reviewed above, is critical to the pathway ACKNOWLEDGEMENTS
from sensors to knowledge, and would be a huge source of infor-
mation if data were open, with all necessary metadata and envi- F.T., L.C.B., T.P. and M.J.B. thank the EU-I3 project EPPN (28443) that
partly funded this work. F.T. and L.C.B. thank the EU project DROPS
ronmental conditions included [117]. Indeed, the discussion (FP7-KBBE-244374) and ANR-PIA projects PHENOME FPPN (ANR-11-
above suggests that the combination of datasets collected by INBS-0012) and Amaizing (ANR-10-BTBR-01) for funding works reported
distinct groups from different phenotyping platforms and fields here. T.P. and M.J.B. thank ERC for supporting the FUTUREROOTS proj-
ect, the University of Nottingham for Institute of FUTURE FOOD funding,
could result in unprecedented information that may build up and the BBSRC for responsive mode, partnering award and iPLANT fund-
year after year. Recent papers present ‘proofs of concept’ of ing. T.P. and M.J.B also thank Stefan Mairhofer, Craig Sturrock, Brian
the meta-analysis of large datasets combining environmental Atkinson, Larry York, Marcus Griffiths and Darren Wells for kindly providing
and phenotypic data [118–120], and discuss their role in multi- CT related images. F.T. and L.C.B. thank Christian Fournier for providing
maize images, Frederic Baret for field images and Claude Welcker for
environment quantitative genetics [121]. important discussions.
Finally, the meta-analyses of tens of experiments may be com-
bined with data collected by farmers on yield and environmental
conditions, thereby allowing large-scale association genetics. REFERENCES
The sensor networks that are appearing in farmers’ fields,
1. Field, C.B., Barros, V.R., Mastrandrea, M.D., Mach, K.J., Abdrabo, M.A.K.,
multi-layer maps of climate and soil characteristics and progress Adger, W.N., Anokhin, Y.A., Anisimov, O.A., Arent, D.J., Barnett, J., et al.
in remote sensing may soon provide the environmental data (2014). Summary for policymakers. In Climate Change 2014: Impacts,
necessary to interpret the diversity of yield corresponding to Adaptation and Vulnerability. (Cambridge: Cambridge University Press),
pp. 1–32.
each variety in each field. If large-scale collections of yield and
environmental conditions in farmer’s fields were organized, as- 2. Tester, M., and Langridge, P. (2010). Breeding technologies to increase
crop production in a changing world. Science 327, 818–822.
sociation genetics at the level of countries or continents would
become possible. This type of approach is already operational 3. Barrangou, R., and Doudna, J.A. (2016). Applications of CRISPR technol-
in big-data analyses of, for example, human social media behav- ogies in research and beyond. Nat. Biotechnol. 34, 933–941.
iour, and its adoption in phenomics is of interest to a range of 4. Schiml, S., and Puchta, H. (2016). Revolutionizing plant biology: multiple
stakeholders. ways of genome engineering by CRISPR/Cas. Plant Methods 12, 8.

5. Huang, X.H., and Han, B. (2014). Natural variations and genome-wide

Concluding Remarks association studies in crop plants. Annu. Rev. Plant Biol. 65, 531–551.
Plant phenomics research faces a conceptual challenge. To 6. Nunes-Nesi, A., Nascimento, V.D., Silva, F.M.D., Zsogon, A., Araujo,
date, researchers have focused on employing and/or developing W.L., and Sulpice, R. (2016). Natural genetic variation for morphological
novel sensors and imaging techniques [8–10]. However, meth- and molecular determinants of plant growth and yield. J. Exp. Bot. 67,
2989–3001.
odological advances in terms of data acquisition, handling and
processing are becoming increasingly important. Indeed, the 7. Cobb, J.N., DeClerck, G., Greenberg, A., Clark, R., and McCouch, S.
(2013). Next-generation phenotyping: requirements and strategies
challenges of translating sensor information into knowledge for enhancing our understanding of genotype-phenotype relation-
have been grossly underestimated during the first years of plant ships and its relevance to crop improvement. Theor. Appl. Genet.
phenomics research. Facing this challenge involves taking 126, 867–887.
into account the intimate interaction between environmental 8. Fiorani, F., and Schurr, U. (2013). Future scenarios for plant phenotyping.
conditions and plant structure, functions and metabolism, which Annu. Rev. Plant Biol. 64, 267–291.
require environmental characterization to be part of all steps of 9. Furbank, R.T., and Tester, M. (2011). Phenomics - technologies to relieve
phenotyping, from data collection to meta-analyses. It also re- the phenotyping bottleneck. Trends Plant Sci. 16, 635–644.
quires the use of both dynamic and statistical models allowing 10. Brown, T.B., Cheng, R.Y., Sirault, X.R.R., Rungrat, T., Murray, K.D., Trti-
multi-scale analyses across experiments and platforms, which lek, M., Furbank, R.T., Badger, M., Pogson, B.J., and Borevitz, J.O.
are essential to deal with the plant peculiarities reviewed at the (2014). TraitCapture: genomic and environment modelling of plant phe-
nomic data. Curr. Opin. Plant. Biol. 18, 73–79.
beginning of this paper. Finally, the most recent advances in
information technology must be employed to face the big-data 11. Poorter, H., Fiorani, F., Pieruschka, R., Wojciechowski, T., van der Putten,
W.H., Kleyer, M., Schurr, U., and Postma, J. (2016). Pampered inside,
challenge associated with multi-image processing, of meta-
pestered outside? Differences and similarities between plants growing in
analysis of heterogeneous data and of the deployment of controlled conditions and in the field. New Phytol. 212, 838–855.
phenomics beyond the strict world of research. For obvious
12. Reynolds, M., and Langridge, P. (2016). Physiological breeding. Curr.
budget issues, it will not be possible to monitor all temporal Opin. Plant. Biol. 31, 162–171.
and organization scales in every environment, but we believe
13. Araus, J.L., and Cairns, J.E. (2014). Field high-throughput phenotyping:
that the rapid progress in modelling and information systems the new crop breeding frontier. Trends Plant Sci. 19, 52–61.
will allow identification of adequate cocktails of equipment,
methods and meta-analyses allowing optimization of resources. 14. Ghanem, M.E., Marrou, H., and Sinclair, T.R. (2015). Physiological
phenotyping of plants for crop improvement. Trends Plant Sci. 20, 139–144.
Hence, we propose that phenomics has reached a stage at
which the limiting step is the design of methods and approaches 15. Mishra, Y., Jankanpaa, H.J., Kiss, A.Z., Funk, C., Schroder, W.P., and
Jansson, S. (2012). Arabidopsis plants grown in the field and climate
allowing one to take into account different temporal and spatial chambers significantly differ in leaf morphology and photosystem com-
scales and perform meta-analyses for addressing the challenges ponents. BMC Plant Biol. 12, 6.

Current Biology 27, R770–R783, August 7, 2017 R779

 Current Biology

Review
16. Vadez, V., Kholova, J., Medina, S., Kakkera, A., and Anderberg, H. 33. Langridge, P., and Fleury, D. (2011). Making the most of ‘omics’ for crop
(2014). Transpiration efficiency: new insights into an old story. J. Exp. breeding. Trends Biotechnol. 29, 33–40.
Bot. 65, 6141–6153.
34. Baxter, I. (2009). Ionomics: studying the social network of mineral nutri-
17. Chelle, M. (2005). Phylloclimate or the climate perceived by individual ents. Curr. Opin. Plant Biol. 12, 381–386.
plant organs: What is it? How to model it? What for? New Phytol. 166,
781–790. 35. Hosmani, P.S., Kamiya, T., Danku, J., Naseer, S., Geldner, N., Guerinot,
M.L., and Salt, D.E. (2013). Dirigent domain-containing protein is part of
18. Caldeira, C., Bosio, M., Parent, B., Jeanguenin, L., Chaumont, F., and the machinery required for formation of the lignin-based Casparian strip
Tardieu, F. (2014). Rapid changes in leaf elongation rate are compatible in the root. Proc. Natl. Acad. Sci. USA 110, 14498–14503.
with a hydraulic control under soil water deficit or high evaporative de-
mand. Plant Physiol. 164, 1718–1730. 36. Stitt, M., and Zeeman, S.C. (2012). Starch turnover: pathways, regulation
and role in growth. Curr. Opin. Plant. Biol. 15, 282–292.
19. Band, L.R., Úbeda-Tomás, S., Dyson, R.J., Middleton, A.M., Hodgman,
T.C., Owen, M.R., Jensen, O.E., Bennett, M.J., and King, J.R. (2012). 37. Salon, C., Avice, J.-C., Colombie, S., Dieuaide-Noubhani, M., Gallardo,
Growth-induced hormone dilution can explain the dynamics of plant K., Jeudy, C., Ourry, A., Prudent, M., Voisin, A.-S., and Rolin, D. (2017).
root cell elongation. Proc. Natl. Acad. Sci. USA 109, 7577–7582. Fluxomics links cellular functional analyses to whole-plant phenotyping.
J. Exp. Bot. http://dx.doi.org/10.1093/jxb/erx126.
20. Middleton, A.M., Úbeda-Tomás, S., Griffiths, J., Holman, T., Hedden, P.,
Thomas, S.G., Phillips, A.L., Holdsworth, M.J., Bennett, M.J., and King, 38. Kim, H.K., Choi, Y.H., and Verpoorte, R. (2011). NMR-based plant metab-
J.R. (2012). Mathematical modeling elucidates the role of transcriptional olomics: where do we stand, where do we go? Trends Biotechnol. 29,
feedback in gibberellin signaling. Proc. Natl. Acad. Sci. USA 109, 7571– 267–275.
7576.
39. Jahnke, S., Menzel, M.I., van Dusschoten, D., Roeb, G.W., Buhler, J.,
21. Baerenfaller, K., Massonnet, C., Walsh, S., Baginsky, S., Buhlmann, P., Minwuyelet, S., Blumler, P., Temperton, V.M., Hombach, T., Streun,
Hennig, L., Hirsch-Hoffmann, M., Howell, K.A., Kahlau, S., Radziejwoski, M., et al. (2009). Combined MRI-PET dissects dynamic changes in plant
A., et al. (2012). Systems-based analysis of Arabidopsis leaf growth re- structures and functions. Plant J. 59, 634–644.
veals adaptation to water deficit. Mol. Syst. Biol. 8, 606.
40. Larrieu, A., Champion, A., Legrand, J., Lavenus, J., Mast, D., Brunoud,
22. Sebastian, J., Yee, M.C., Viana, W.G., Rellan-Alvarez, R., Feldman, M., G., Oh, J., Guyomarc’h, S., Pizot, M., and Farmer, E.E. (2015). A fluores-
Priest, H.D., Trontin, C., Lee, T., Jiang, H., Baxter, I., et al. (2016). Grasses cent hormone biosensor reveals the dynamics of jasmonate signalling in
suppress shoot-borne roots to conserve water during drought. Proc. plants. Nat. Commun. 6, 6043.
Natl. Acad. Sci. USA 113, 8861–8866.
41. Lanquar, V., Grossmann, G., Vinkenborg, J.L., Merkx, M., Thomine, S.,
23. Bao, Y., Aggarwal, P., Robbins, N.E., Sturrock, C.J., Thompson, M.C., and Frommer, W.B. (2014). Dynamic imaging of cytosolic zinc in Arabi-
Tan, H.Q., Tham, C., Duan, L.N., Rodriguez, P.L., Vernoux, T., et al. dopsis roots combining FRET sensors and RootChip technology. New
(2014). Plant roots use a patterning mechanism to position lateral root Phytol. 202, 198–208.
branches toward available water. Proc. Natl. Acad. Sci. USA 111,
9319–9324. 42. Brunoud, G., Wells, D.M., Oliva, M., Larrieu, A., Mirabet, V., Burrow, A.H.,
Beeckman, T., Kepinski, S., Traas, J., and Bennett, M.J. (2012). A novel
24. Tardieu, F., Granier, C., and Muller, B. (2011). Water deficit and growth. sensor to map auxin response and distribution at high spatio-temporal
Co-ordinating processes without an orchestrator? Curr. Opin. Plant resolution. Nature 482, 103–106.
Biol. 14, 283–289.
43. Maccaferri, M., Sanguineti, M.C., Corneti, S., Ortega, J.L.A., Ben Salem,
25. Chew, Y.H., Wenden, B., Flis, A., Mengin, V., Taylor, J., Davey, C.L., Tin- M., Bort, J., DeAmbrogio, E., Fernando Garcia del Moral, L., Demontis,
dal, C., Thomas, H., Ougham, H.J., de Reffye, P., et al. (2014). Multiscale A., El-Ahmed, A., et al. (2008). Quantitative trait loci for grain yield and
digital Arabidopsis predicts individual organ and whole-organism adaptation of durum wheat (Triticum durum Desf.) across a wide range
growth. Proc. Natl. Acad. Sci. USA 111, E4127–E4136. of water availability. Genetics 178, 489–511.

26. Vandeleur, R.K., Sullivan, W., Athman, A., Jordans, C., Gilliham, M., Kai- 44. Vargas, M., van Eeuwijk, F.A., Crossa, J., and Ribaut, J.M. (2006). Map-
ser, B.N., and Tyerman, S.D. (2014). Rapid shoot-to-root signalling reg- ping QTLs and QTL x environment interaction for CIMMYT maize drought
ulates root hydraulic conductance via aquaporins. Plant Cell Environ. stress program using factorial regression and partial least squares
37, 520–538. methods. Theor. Appl. Genet. 112, 1009–1023.

27. Visentin, I., Vitali, M., Ferrero, M., Zhang, Y., Ruyter-Spira, C., Novák, O., 45. Tardieu, F. (2012). Any trait or trait-related allele can confer drought toler-
Strnad, M., Lovisolo, C., Schubert, A., and Cardinale, F. (2017). Low ance: just design the right drought scenario. J. Exp. Bot. 63, 25–31.
levels of strigolactones in roots as a component of the systemic signal
of drought stress in tomato. New Phytol. 212, 954–963. 46. Chenu, K., Chapman, S.C., Tardieu, F., McLean, G., Welcker, C., and
Hammer, G.L. (2009). Simulating the yield impacts of organ-level quanti-
28. Tardieu, F., and Parent, B. (2017). Predictable ’meta-mechanisms’ tative trait loci associated with drought response in maize: a ‘‘gene-to-
emerge from feedbacks between transpiration and plant growth and phenotype’’ modeling approach. Genetics 183, 1507–1523.
cannot be simply deduced from short-term mechanisms. Plant Cell
Environ. 40, 846–857. 47. Harrison, M.T., Tardieu, F., Dong, Z.S., Messina, C.D., and Hammer, G.L.
(2014). Characterizing drought stress and trait influence on maize yield
29. Walter, A., Silk, W.K., and Schurr, U. (2009). Environmental effects on under current and future conditions. Glob. Chang. Biol. 20, 867–878.
spatial and temporal patterns of leaf and root growth. Annu. Rev. Plant
Biol. 60, 279–304. 48. Millet, E.J., Welcker, C., Kruijer, W., Negro, S., Coupel-Ledru, A., Nicolas,
S.D., Laborde, J., Bauland, C., Praud, S., Ranc, N., et al. (2016). Genome-
30. Kuchen, E.E., Fox, S., de Reuille, P.B., Kennaway, R., Bensmihen, S., wide analysis of yield in Europe: allelic effects vary with drought and heat
Avondo, J., Calder, G.M., Southam, P., Robinson, S., Bangham, A., scenarios. Plant Physiol. 172, 749–764.
et al. (2012). Generation of leaf shape through early patterns of growth
and tissue polarity. Science 335, 1092–1096. 49. Houle, D., Govindaraju, D.R., and Omholt, S. (2010). Phenomics: the next
challenge. Nat. Rev. Genet. 11, 855–866.
31. Hervieux, N., Dumond, M., Sapala, A., Routier-Kierzkowska, A.L., Kierz-
kowski, D., Roeder, A.H.K., Smith, R.S., Boudaoud, A., and Hamant, O. 50. Technow, F., Messina, C.D., Totir, L.R., and Cooper, M. (2015). Inte-
(2016). A mechanical feedback restricts sepal growth and shape in grating crop growth models with whole genome prediction through
Arabidopsis. Curr. Biol. 26, 1019–1028. approximate Bayesian computation. PLoS One 10, e0130855.

32. Caldeira, C.F., Jeanguenin, L., Chaumont, F., and Tardieu, F. (2014). 51. Bassi, F.M., Bentley, A.R., Charmet, G., Ortiz, R., and Crossa, J. (2016).
Circadian rhythms of hydraulic conductance and growth are enhanced Breeding schemes for the implementation of genomic selection in wheat
by drought and improve plant performance. Nat. Commun. 5, 5365. (Triticum spp.). Plant Sci. 242, 23–36.

R780 Current Biology 27, R770–R783, August 7, 2017

Current Biology

Review
52. Nagel, K.A., Putz, A., Gilmer, F., Heinz, K., Fischbach, A., Pfeifer, J., 67. Uga, Y., Sugimoto, K., Ogawa, S., Rane, J., Ishitani, M., Hara, N., Kitomi,
Faget, M., Blossfeld, S., Ernst, M., Dimaki, C., et al. (2012). Y., Inukai, Y., Ono, K., Kanno, N., et al. (2013). Control of root system ar-
GROWSCREEN-Rhizo is a novel phenotyping robot enabling simulta- chitecture by DEEPER ROOTING 1 increases rice yield under drought
neous measurements of root and shoot growth for plants grown in soil- conditions. Nat. Genet. 45, 1097–1102.
filled rhizotrons. Funct. Plant Biol. 39, 891.
68. Tardieu, F. (2013). Plant response to environmental conditions: assess-
53. Jeudy, C., Adrian, M., Baussard, C., Bernard, C., Bernaud, E., Bourion, ing potential production, water demand, and negative effects of water
V., Busset, H., Cabrera-Bosquet, L., Cointault, F., Han, S.M., et al. deficit. Front. Physiol. 4, 17.
(2016). RhizoTubes as a new tool for high throughput imaging of plant
root development and architecture: test, comparison with pot grown 69. Poorter, H., Fiorani, F., Stitt, M., Schurr, U., Finck, A., Gibon, Y., Usadel,
plants and validation. Plant Methods 12, 31. B., Munns, R., Atkin, O.K., Tardieu, F., et al. (2012). The art of growing
plants for experimental purposes: a practical guide for the plant biologist
54. Walter, A., Scharr, H., Gilmer, F., Zierer, R., Nagel, K.A., Ernst, M., Wiese, Review. Funct. Plant. Biol. 39, 821–838.
A., Virnich, O., Christ, M.M., and Uhlig, B. (2007). Dynamics of seedling
growth acclimation towards altered light conditions can be quantified 70. Vereecken, H., Schnepf, A., Hopmans, J.W., Javaux, M., Or, D., Roose,
via GROWSCREEN: a setup and procedure designed for rapid optical T., Vanderborght, J., Young, M.H., Amelung, W., Aitkenhead, M., et al.
phenotyping of different plant species. New Phytol. 174, 447–455. (2016). Modeling soil processes: review, key challenges, and new per-
spectives. Vadose Zone J. 15, 5.
55. Jansen, M., Gilmer, F., Biskup, B., Nagel, K.A., Rascher, U., Fisch-
bach, A., Briem, S., Dreissen, G., Tittmann, S., and Braun, S. (2009). 71. Li, L., Zhang, Q., and Huang, D.F. (2014). A review of imaging techniques
Simultaneous phenotyping of leaf growth and chlorophyll fluorescence for plant phenotyping. Sensors 14, 20078–20111.
via GROWSCREEN FLUORO allows detection of stress tolerance in
Arabidopsis thaliana and other rosette plants. Funct. Plant Biol. 36, 72. Hartmann, A., Czauderna, T., Hoffmann, R., Stein, N., and Schreiber, F.
902–914. (2011). HTPheno: An image analysis pipeline for high-throughput plant
phenotyping. BMC Bioinformatics 12, 148.
56. Granier, C., Aguirrezabal, L., Chenu, K., Cookson, S.J., Dauzat, M., Ha-
mard, P., Thioux, J.J., Rolland, G., Bouchier-Combaud, S., Lebaudy, 73. Lobet, G., Pages, L., and Draye, X. (2011). A novel image-analysis
A., et al. (2006). PHENOPSIS, an automated platform for reproducible toolbox enabling quantitative analysis of root system architecture. Plant
phenotyping of plant responses to soil water deficit in Arabidopsis thali- Physiol. 157, 29–39.
ana permitted the identification of an accession with low sensitivity to soil
water deficit. New Phytol. 169, 623–635. 74. Mairhofer, S., Zappala, S., Tracy, S.R., Sturrock, C., Bennett, M.,
Mooney, S.J., and Pridmore, T. (2012). RooTrak: automated recovery
57. Coupel-Ledru, A., Lebon, E., Christophe, A., Gallo, A., Gago, P., Pantin, of three-dimensional plant root architecture in soil from X-ray micro-
F., Doligez, A., and Simonneau, T. (2016). Reduced nighttime transpira- computed tomography images using visual tracking. Plant Physiol.
tion is a relevant breeding target for high water-use efficiency in grape- 158, 561–569.
vine. Proc. Natl. Acad. Sci. USA 113, 8963–8968.
75. Korhonen, L., and Heikkinen, J. (2009). Automated analysis of in situ
58. Cabrera-Bosquet, L., Fournier, C., Brichet, N., Welcker, C., Suard, B., canopy images for the estimation of forest canopy cover. For. Sci. 55,
and Tardieu, F. (2016). High-throughput estimation of incident light, light 323–334.
interception and radiation-use efficiency of thousands of plants in a phe-
notyping platform. New Phytol. 212, 269–281. 76. Mielewczik, M., Friedli, M., Kirchgessner, N., and Walter, A. (2013). Diel
leaf growth of soybean: a novel method to analyze two-dimensional
59. Coupel-Ledru, A., Lebon, É., Christophe, A., Doligez, A., Cabrera- leaf expansion in high temporal resolution based on a marker tracking
Bosquet, L., Pechier, P., Hamard, P., This, P., and Simonneau, T. approach (Martrack Leaf). Plant Methods 9, 30.
(2014). Genetic variation in a grapevine progeny (Vitis vinifera L. cvs
Grenache3Syrah) reveals inconsistencies between maintenance of day- 77. Joosen, R.V.L., Kodde, J., Willems, L.A.J., Ligterink, W., van der Plas,
time leaf water potential and response of transpiration rate under L.H.W., and Hilhorst, H.W.M. (2010). GERMINATOR: a software package
drought. J. Exp. Bot. 65, 6205–6218. for high-throughput scoring and curve fitting of Arabidopsis seed germi-
nation. Plant J. 62, 148–159.
60. Tardieu, F., and Tuberosa, R. (2010). Dissection and modelling of abiotic
stress tolerance in plants. Curr. Opin. Plant. Biol. 13, 206–212. 78. Brewer, M.T., Lang, L., Fujimura, K., Dujmovic, N., Gray, S., and van der
Knaap, E. (2006). Development of a controlled vocabulary and software
61. Vadez, V., Kholova, J., Hummel, G., Zhokhavets, U., Gupta, S.K., and application to analyze fruit shape variation in tomato and other plant spe-
Hash, C.T. (2015). LeasyScan: a novel concept combining 3D imaging cies. Plant Physiol. 141, 15–25.
and lysimetry for high-throughput phenotyping of traits controlling plant
water budget. J. Exp. Bot. 66, 5581–5593. 79. Fahlgren, N., Gehan, M.A., and Baxter, I. (2015). Lights, camera, action:
high-throughput plant phenotyping is ready for a close-up. Curr. Opin.
62. Virlet, N., Sabermanesh, K., Sadeghi-Tehran, P., and Hawkesford, M.J. Plant Biol. 24, 93–99.
(2016). Field Scanalyzer: An automated robotic field phenotyping plat-
form for detailed crop monitoring. Funct. Plant Biol. 44, 143–153. 80. Furukawa, Y., and Ponce, J. (2010). Accurate, dense, and robust mul-
tiview stereopsis. IEEE Trans. Pattern. Anal. Mach. Intell. 32, 1362–
63. Tattaris, M., Reynolds, M.P., and Chapman, S.C. (2016). A direct com- 1376.
parison of remote sensing approaches for high-throughput phenotyping
in plant breeding. Front. Plant. Sci. 7, 1131. 81. Burgess, A.j., Retkute, R., Pound, M.P., Preston, S.P., Pridmore, T.P.,
foulkes, j., Jensen, O., and Murchie, E.H. (2015). High-resolution 3D
64. Rattey, A.R., Shorter, R., and Chapman, S.C. (2011). Evaluation of structural data quantifies the impact of photoinhibition on long term car-
CIMMYT conventional and synthetic spring wheat germplasm in rainfed bon gain in wheat canopies in the field. Plant Physiol. 169, 1192–1204.
sub-tropical environments. II. Grain yield components and physiological
traits. Field Crop. Res. 124, 195–204. 82. Pound, M.P., French, A.P., Murchie, E.H., and Pridmore, T.P. (2014).
Automated recovery of three-dimensional models of plant shoots from
65. Bonneau, J., Taylor, J., Parent, B., Bennett, D., Reynolds, M., Feuillet, C., multiple color images. Plant Physiol. 166, 1688–1698.
Langridge, P., and Mather, D. (2013). Multi-environment analysis and
improved mapping of a yield-related QTL on chromosome 3B of wheat. 83. Paulus, S., Schumann, H., Kuhlmann, H., and Le on, J. (2014). High-pre-
Theor. Appl. Genet. 126, 747–761. cision laser scanning system for capturing 3D plant architecture and an-
alysing growth of cereal plants. Biosys. Eng. 121, 1–11.
66. Obata, T., Witt, S., Lisec, J., Palacios-Rojas, N., Florez-Sarasa, I., Yousfi,
S., Araus, J.L., Cairns, J.E., and Fernie, A.R. (2015). Metabolite profiles 84. Dornbusch, T., Lorrain, S., Kuznetsov, D., Fortier, A., Liechti, R., Xenar-
of maize leaves in drought, heat, and combined stress field trials reveal ios, I., and Fankhauser, C. (2012). Measuring the diurnal pattern of leaf
the relationship between metabolism and grain yield. Plant Physiol. hyponasty and growth in Arabidopsis – a novel phenotyping approach
169, 2665–2683. using laser scanning. Funct. Plant Biol. 39, 860–869.

Current Biology 27, R770–R783, August 7, 2017 R781

 Current Biology

Review
85. McCormick, R.F., Truong, S.K., and Mullet, J.E. (2016). 3D sorghum 102. Martre, P., Wallach, D., Asseng, S., Ewert, F., Jones, J.W., Rotter, R.P.,
reconstructions from depth images identify QTL regulating shoot archi- Boote, K.J., Ruane, A.C., Thorburn, P.J., Cammarano, D., et al. (2015).
tecture. Plant Physiol. 172, 823–834. Multimodel ensembles of wheat growth: many models are better than
one. Glob. Chang. Biol. 21, 911–925.
86. Che ne
, Y., Rousseau, D., Lucidarme, P., Bertheloot, J., Caffier, V., Morel,
P., Belin, E., and Chapeau-Blondeau, F. (2012). On the use of depth cam- 103. Wilkinson, M.D., Dumontier, M., Aalbersberg, I.J., Appleton, G., Axton,
era for 3D phenotyping of entire plants. Comput. Electron. Agric. 82, M., Baak, A., Blomberg, N., Boiten, J.-W., da Silva Santos, L.B., Bourne,
122–127. P.E., et al. (2016). The FAIR guiding principles for scientific data manage-
ment and stewardship. Sci. Data 3, 160018.
87. Dhondt, S., Vanhaeren, H., Van Loo, D., Cnudde, V., and Inze, D. (2010).
Plant structure visualization by high-resolution X-ray computed tomogra- 104. Klukas, C., Chen, D., and Pape, J.-M. (2014). Integrated analysis
phy. Trends Plant Sci. 15, 419–422. platform: an open-source information system for high-throughput plant
phenotyping. Plant Physiol. 165, 506–518.
88. Van As, H. (2007). Intact plant MRI for the study of cell water relations,
membrane permeability, cell-to-cell and long distance water transport. 105. Fabre, J., Dauzat, M., Negre, V., Wuyts, N., Tireau, A., Gennari, E., Neveu,
J. Exp. Bot. 58, 743–756. P., Tisne, S., Massonnet, C., Hummel, I., et al. (2011). PHENOPSIS DB:
an Information System for Arabidopsis thaliana phenotypic data in an
89. Jahnke, S., Roussel, J., Hombach, T., Kochs, J., Fischbach, A., Huber, environmental context. BMC Plant Biol. 11, 77.
G., and Scharr, H. (2016). phenoSeeder - A robot system for automated
handling and phenotyping of individual seeds. Plant Physiol. 172, 1358– 106. Merchant, N., Lyons, E., Goff, S., Vaughn, M., Ware, D., Micklos, D., and
1370. Antin, P. (2016). The iPlant collaborative: cyberinfrastructure for enabling
data to discovery for the life sciences. PLoS Biol. 14, e1002342–
90. Salehi, A., Jimenez-Berni, J., Deery, D.M., Palmer, D., Holland, E., Ro- e1002342.
zas-Larraondo, P., Chapman, S.C., Georgakopoulos, D., and Furbank,
R.T. (2015). SensorDB: a virtual laboratory for the integration, visualiza- 107. Jones, A.R., Miller, M., Aebersold, R., Apweiler, R., Ball, C.A., Brazma, A.,
tion and analysis of varied biological sensor data. Plant Methods 11, 53. Degreef, J., Hardy, N., Hermjakob, H., Hubbard, S.J., et al. (2007). The
Functional Genomics Experiment model (FuGE): an extensible framework
91. Massonnet, C., Vile, D., Fabre, J., Hannah, M.A., Caldana, C., Lisec, J., for standards in functional genomics. Nat. Biotechnol. 25, 1127–1133.
Beemster, G.T.S., Meyer, R.C., Messerli, G., Gronlund, J.T., et al.
(2010). Probing the reproducibility of leaf growth and molecular pheno- 108. Rebetzke, G.J., Chenu, K., Biddulph, B., Moeller, C., Deery, D.M., Rattey,
types: a comparison of three Arabidopsis accessions cultivated in ten A.R., Bennett, D., Barrett-Lennard, E.G., and Mayer, J.E. (2013). A multi-
laboratories. Plant Physiology 152, 2142–2157. site managed environment facility for targeted trait and germplasm phe-
notyping. Funct. Plant Biol. 40, 1–13.
92. Pridmore, T.P., French, A.P., and Pound, M.P. (2012). What lies beneath:
underlying assumptions in bioimage analysis. Trends Plant Sci. 17, 109. Junker, A., Muraya, M.M., and Weigelt-Fischer, K. (2015). Optimizing
experimental procedures for quantitative evaluation of crop plant per-
688–692.
formance in high throughput phenotyping systems. Front. Plant. Sci.
93. Mairhofer, S., Zappala, S., Tracy, S., Sturrock, C., Bennett, M.J., 5, 770.
Mooney, S.J., and Pridmore, T.P. (2013). Recovering complete plant
110. Pradal, C., Artzet, S., Chopard, J., Dupuis, D., Fournier, C., Mielewczik,
root system architectures from soil via X-ray m-computed tomography.
M., Nègre, V., Neveu, P., Parigot, D., Valduriez, P., et al. (2017).
Plant Methods 9, 8.
InfraPhenoGrid: A scientific workflow infrastructure for plant phenomics
on the Grid. Future Generation Computer Systems 67, 341–353.
94. Scharr, H., Minervini, M., French, A.P., Klukas, C., Kramer, D.M., Liu, X.,
Luengo, I., Pape, J.-M., Polder, G., Vukadinovic, D., et al. (2016). Leaf 111. Li, Y.-F., Kennedy, G., Ngoran, F., Wu, P., and Hunter, J. (2013). An
segmentation in plant phenotyping: a collation study. Machine Vision
ontology-centric architecture for extensible scientific data management
and Applications 27, 585–606. systems. Future Generation Computer Systems-the International Journal
of Grid Computing and Escience 29, 641–653.
95. Bustos-Korts, D., Malosetti, M., Chapman, S., and van Eeuwijk, F. (2016).
Modelling of genotype by environment interaction and prediction of com- 
112. Cwiek-Kupczy ska, H., Altmann, T., Arend, D., Arnaud, E., Chen, D.,
n
plex traits across multiple environments as a synthesis of crop growth Cornut, G., Fiorani, F., Frohmberg, W., Junker, A., Klukas, C., et al.
modelling, genetics and statistics. In Crop Systems Biology (Springer In- (2016). Measures for interoperability of phenotypic data: minimum infor-
ternational Publishing), pp. 55–82. mation requirements and formatting. Plant Methods 12, 44.
96. van Eeuwijk, F.A., Bink, M., Chenu, K., and Chapman, S.C. (2010). Detec- 113. Steinbach, D., Alaux, M., Amselem, J., Choisne, N., Durand, S., Flores,
tion and use of QTL for complex traits in multiple environments. Curr. R., Keliet, A.O., Kimmel, E., Lapalu, N., Luyten, I., et al. (2013). GnpIS:
Opin. Plant Biol. 13, 193–205. an information system to integrate genetic and genomic data from plants
and fungi. Database (Oxford). bat058.
97. Malosetti, M., Ribaut, J.M., and van Eeuwijk, F.A. (2013). The statistical
analysis of multi-environment data: modeling genotype-by-environment 114. Long, J., Shelhamer, E., and Darrell, T. (2015). Fully convolutional net-
interaction and its genetic basis. Front. Physiol. 4, 44. works for semantic segmentation. In Proceedings of the IEEE Confer-
ence on Computer Vision and Pattern Recognition. pp. 3431–3440.
98. Wang, H.G., Paulo, J., Kruijer, W., Boer, M., Jansen, H., Tikunov, Y., Usa-
del, B., van Heusden, S., Bovy, A., and van Eeuwijk, F. (2015). Genotype- 115. Krizhevsky, A., Sutskever, I., and Hinton, G.E. (2012). Imagenet classifi-
phenotype modeling considering intermediate level of biological varia- cation with deep convolutional neural networks. Adv. Neural Inf. Process.
tion: a case study involving sensory traits, metabolites and QTLs in ripe Syst. 25, 1097–1105.
tomatoes. Mol. Biosyst. 11, 3101–3110.
116. Pound, M.P., Burgess, A.J., Wilson, M.H., Atkinson, J.A., Griffiths, M.,
99. Tardieu, F., and Parent, B. (2017). Predictable ‘meta-mechanisms’ Jackson, A.S., Bulat, A., Tzimiropoulos, G., Wells, D.M., Murchie, E.H.,
emerge from feedbacks between transpiration and plant growth and et al. (2016). Deep machine learning provides state-of-the-art perfor-
cannot be simply deduced from short-term mechanisms. Plant Cell Envi- mance in image-based plant phenotyping. bioRxiv. http://dx.doi.org/
ron. 40, 846–857. 10.1101/053033.

100. Welcker, C., Sadok, W., Dignat, G., Renault, M., Salvi, S., Charcosset, A., 
117. Krajewski, P., Chen, D., Cwiek, H., van Dijk, A.D.J., Fiorani, F., Kersey, P.,
and Tardieu, F. (2011). A common genetic determinism for sensitivities Klukas, C., Lange, M., Markiewicz, A., Nap, J.P., et al. (2015). Towards
to soil water deficit and evaporative demand: meta-analysis of quantita- recommendations for metadata and data handling in plant phenotyping.
tive trait loci and introgression lines of maize. Plant Physiol. 157, J. Exp. Bot. 66, 5417–5427.
718–729.
118. Poorter, H., Niinemets, U., Walter, A., Fiorani, F., and Schurr, U. (2010).
101. Parent, B., and Tardieu, F. (2014). Can current crop models be used in the A method to construct dose-response curves for a wide range of environ-
phenotyping era for predicting the genetic variability of yield of plants mental factors and plant traits by means of a meta-analysis of phenotypic
subjected to drought or high temperature? J. Exp. Bot. 65, 6179–6189. data. J. Exp. Bot. 61, 2043–2055.

R782 Current Biology 27, R770–R783, August 7, 2017

Current Biology

Review
119. Poorter, H., Niklas, K.J., Reich, P.B., Oleksyn, J., Poot, P., and Mommer, 121. Adam-Blondon, A.F., Alaux, M., Pommier, C., Cantu, D., Cheng, Z.M.,
L. (2012). Biomass allocation to leaves, stems and roots: meta-analyses Cramer, G.R., Davies, C., Delrot, S., Deluc, L., Di Gaspero, G., et al.
of interspecific variation and environmental control. New Phytol. 193, (2016). Towards an open grapevine information system. Hortic. Res. 3,
30–50. 16056.

120. Parent, B., and Tardieu, F. (2012). Temperature responses of 122. Band, L.R., Wells, D.M., Fozard, J.A., Ghetiu, T., French, A.P., Pound,
developmental processes have not been affected by breeding in M.P., Wilson, M.H., Yu, L., Li, W.D., Hijazi, H.I., et al. (2014). Systems
different ecological areas for 17 crop species. New Phytol. 194, analysis of auxin transport in the Arabidopsis root apex. Plant Cell 26,
760–774. 862–875.

Current Biology 27, R770–R783, August 7, 2017 R783