Imke Thormann, Bioversity International scientist, presented at the international conference Enhanced genepool utilization - Capturing wild relative and landrace diversity for crop improvement, in Cambridge, UK, 16-20 June 2014. Novel approaches to enhance characterization of plant genetic resources are being developed, as traditional phenotypic characterization techniques have shown to be insufficient to fully harness crop wild relative (CWR) and landrace diversity. These are genomics, transcriptomics, metabolomics, high-throughput phenotyping, as well as less resource intensive predictive characterization techniques. The latter build on the hypothesis that the environment influences gene flow and natural selection, and thus spatial genetic differentiation of organisms. CWR populations growing in a specific environment will possess a suite of adaptive traits shaped by selection pressures unique to these environments. Thus information about a CWR occurrence site can be used to approach the utilization of genetic resources in a more rational way. Two predictive characterization methods for CWR were developed within the PGR Secure project, using an agro-ecological approach for optimizing the search for populations and accessions with targeted adaptive traits: The ecogeographical filtering method combines spatial distribution of the target species with the ecogeographical identification of those environments that are likely to impose selection pressure for the selected trait. Edaphic, geophysic and bioclimatic variables most relevant for adaptation are identified and used together with ecogeographic land characterization maps to identify promising occurrences. The calibration method bases the criteria to filter accessions on existing evaluation data for the trait of interest. Ecogeographical data specific to the environment at collecting sites evaluated for the trait are used as input to identify existing relationships between trait and environment. This relationship is then used to calibrate a model through which other non-evaluated accessions can be assessed. The methods were applied to the four project genera, Avena, Beta, Brassica and Medicago to identify subsets of potentially interesting accessions or occurrences, investigating the following abiotic stress factors: aluminium toxicity for Avena, drought for Beta, drought and salinity for Brassica, and frost for Medicago. Find out more about our work on crop wild relatives http://www.bioversityinternational.org/research-portfolio/conservation-of-crop-diversity/crop-wild-relatives/

1. Predictive characterization methods for
accessing and using CWR diversity
Thormann I, Parra-Quijano M, Iriondo JM, Rubio-Teso ML, Endresen DT,
Dias S, van Etten J, Maxted N
ENHANCED GENEPOOL UTILIZATION, Cambridge 16-20 June 2014

2. 2
One aim of PGR-Secure: Research novel characterization techniques for CWR + LR
 high throughput phenotyping
 metabolomics
 transcriptomics
 predictive characterization through FIGS
FIGS (focused identification of germplasm strategy) carries out a predictive
characterization of yet uncharacterized germplasm by assigning potential phenotypic or
genotypic properties using environmental information from collecting sites or C/E data
from already characterized samples as predictor.
Environmental profiles are used as filters to increase the likelihood of finding trait of
interest when selecting accessions for field trials.
Assumption: different environments generate different selective pressures and genetic
differentiation of adaptive value.
PGR-Secure context
WP1
WP2

3. 3
• Predictive association between trait data and ecogeographic data for Nordic barley landraces
• Predictive association between biotic stress traits and ecogeographic data for wheat and barley
• Ug99 wheat rust:
– Traditional characterization: 4563 wheat LR screened
for Ug99 in Yemen 2007  10.2 % resistant accessions
– FIGS predictive characterization: 500 accessions selected from
3728 accession  25.8% resistant accessions
• Net blotch - barley
• Boron toxicity - wheat
• Sunn pest - wheat
• Powdery mildew - wheat
• Russian wheat aphid
• Drought – faba bean
Bari et al 2012; El Bouhssini et al 2011; Endresen 2010; Endresen et al 2011, 2012; Khazaei et al 2013; Mackay and Street 2004; Street et al 2008
Examples of predictive association studies and
identification of resistant material through the use of FIGS

4. 4
• Predictive association between trait data and ecogeographic data for Nordic barley landraces
• Predictive association between biotic stress traits and ecogeographic data for wheat and barley
• Ug99 wheat rust:
– Traditional characterization: 4563 wheat LR screened
for Ug99 in Yemen 2007  10.2 % resistant accessions
– FIGS predictive characterization: 500 accessions selected from
3728 accession  25.8% resistant accessions
• Net blotch - barley
• Boron toxicity - wheat
• Sunn pest - wheat
• Powdery mildew - wheat
• Russian wheat aphid
• Drought – faba bean
Bari et al 2012, El Bouhssini et al 2011; Endresen 2010; Endresen et al 2011, 2012; Khazaei et al 2013; Mackay and Street 2004; Street et al 2008
Examples of predictive association studies and
identification of resistant material through the use of FIGS

5. 5
Two FIGS methods were adapted to optimize the search for populations
and accessions with targeted adaptive traits in LR and CWR in the
PGR-Secure genera
 Ecogeographical
filtering method
 Calibration method
The various existing methods
mainly differ in the way in which
the environmental profile used
as filter is developed and embedded
in the process
FIGS methods used in PGR-Secure project
Target traits identified in PGR Secure project in
collaboration with breeders and crop experts

6. 6
Major steps
1) Compile + clean occurrence data
• Data sources: GRIN, SINGER, EURISCO,
GBIF
• Data cleaning
• Georeferencing
• Quality check of existing geographic coordinates (now through online tool developed
in CAPFITOGEN)
 passport data set of occurrences of the target taxon, with a minimum of duplicate
records, and with verified geographic coordinates
Ecogeographical filtering method
spatial distribution
of the target species
ecogeographical identification of those
environments that are likely to impose selection
pressure for the target trait
Genus LR all
records
CWR all
records
Avena 3855 3900
Beta 1614 1596
Brassica 3606 886
Medicago 149 2153

7. 7
2) Develop ecogeographical land
characterization map
• ELC maps represent the adaptive scenarios
that are present over the territory studied
• Requires to identify the
bioclimatic, edaphic and/or geophysical
variables that determine
the spatial distribution of the species
• Map development now supported by CAPFITOGEN
tools
Ecogeographical filtering method
Variables identified based on literature and expert knowledge as relevant for the geographical distribution of Avena
Avena ELC map

8. 8
Ecogeographical filtering method
Beta ELC map
Variables
Bioclimatic Geophysic
BIO3 Isothermality (BIO2/BIO7)
(* 100)
NORTHNESS Northness
BIO6 Min temperature of coldest
month
ELEVATION Elevation
BIO12 Mean annual precipitation SOLRADOP Global irradiation on an optimal inclination
PRECIP2 Average February precipitation
PRECIP6 Average June precipitation Edaphic
PRECIP7 Average July precipitation MINERALOGY Mineralogical profile of soil
PRECIP8 Average August precipitation WRBCODESTU World reference base for soil resources
(WRB) coder for soil typological unit (STU)
TMED1 January mean temperature DEPTHTOROC Depth to rocks
TMED3 March mean temperature DOMPARMAT Dominant parent material (obstacle to
roots)
TMED11 November mean temperature
TMIN1 Average January minimum
temperature
TMIN12 Average December minimum
temperature
Variables identified based on literature and expert knowledge as relevant for the
geographical distribution of Beta

9. 9
3) Identify the most appropriate variables that
describe the environment profile (EP) of
sites where the target trait may evolve, and
threshold values
• Based on literature research and expert consultations
• Data for identified variables is added to the occurrence
data file
Iar-DM value Zone classification
0 - 5 Extremely arid (desert)
5 - 10 Arid (steppic)
10 - 20 Semiarid (mediterranean)
20 - 30 Subhumid
30 - 60 Humid
> 60 Perhumid
Ecogeographical filtering method
De Martonne aridity index, threshold value
for Beta: < 10

10. 10
4) Filtering in R – environment using the
R – script developed for this method
• The script first produces an optimized
subset based on ELC map
• Then records are selected based on the
EP threshold value
Ecogeographical filtering method
Genus LR all
records
CWR all
records
LR
identified
subset
CWR
identified
Subset
Avena 3855 3900 103 171
Beta 1614 1596 133 33
Brassica 3606 886 121 275
Medicago 149 2153 4 54
Results for PGR Secure project genera: Number of total records
and number of selected records
Using the R script developed in PGR Secure
Distribution of Beta CWR – selected records in pink

11. 11
Major steps
1) Compile occurrence and climate data of uncharacterized
accessions (= test set)
2) Compile C/E and climate data for training and calibration set
3) Run R – script on training set to calibrate model based on
relationship identified between trait and environment
4) Fine tune model with calibration set
5) Run test set through model to select occurrences
Insufficient C/E data available for LR and CWR of Avena, Beta, Brassica, Medicago
Calibration method
Existing evaluation
data for trait of interest
Climate data specific to the
environment at collecting
sites
Model relationships
between trait and
environment
Builds a computer model explaining the crop trait score from the climate
data

12. 12
Implemented assumption: different environmental conditions generate
different selective pressures and genetic differentiation of adaptive value
 accurate georeferenced information about accessions/populations is
required to allow extraction of climate, edaphic and geophysic data
 interest in making use of the increasing number of environmental
variables and their quality that are made available globally
 ELC maps and calibration models correctly reflect the different
environmental conditions
 EP: correctly assigning an environmental variable (for which we have
data on the territory) that is strongly linked to the environmental
conditions that promote a particular targeted trait
 Useful for LR + CWR, but not for improved varieties (complex pedigree)
Critical aspects and limitations

13. Next steps
Publication of guidelines on how to use these FIGS
methods, including
• Detailed steps
• Example data
• R – scripts
Application of FIGS methods in new EU – ACP
funded project SADC Crop Wild Relatives
Project objective: Enhance link between
conservation and use of CWR through
• Scientific capacity building
• Development of National Strategic Action Plans
for the conservation and use of CWR

14. Thank you