Multisite prediction of yield based on phenomics, …A tension between genetic variability of -...

Multisite prediction of yield based on phenomics, genomic prediction and environmental information

F. Tardieu

Rationale of EMPHASIS and EPPN2020

A tension between genetic variability of - “mechanisms” (organ, minutes) - yield (canopy, months)

‘Understand’ ‘Dissect’ ‘Predict’

Tardieu, Cabrera, Pridmore and Bennett 2017 Current Biol.

Why so many installations ? different categories of installations are required

A tension between genetic variability of - “mechanisms” (organ, minutes) - yield (canopy, months)

Canopies in a range of environments Weeks to months

Cell- Organ Minute/days

Plant or Canopy Minute to weeks

Level of organization

High precision platforms

Field multi-environment

network

High throughput platforms (controlled

or field)

Tardieu, Cabrera, Pridmore and Bennett 2017 Current Biol.

‘Understand’ ‘Dissect’’ ‘Predict’

Why so many installations ? different categories of installations are required

Affordable phenotyping : yield + sensor networks + modelling

Sensor networks, remote sensing and European grids: One can measure or estimate environmental conditions in any field + combine with simple experiments + databases + modelling Is this affordable phenotyping ? Example: GWAS in 40 European fields. Environment + yield components

252 genotypes 950k polymorphic markers Yield variations from 5 to 12 T ha-1

16 fields X 2 years X 2 W treatments

490 significant SNPs, few in common

Grain Yield (t ha-1)

Fre

qu

en

cy

0

25

50

75

100

0

25

50

75

100

0 5 10 15

0

20

40

60

80 6

4

2

0

6

4

2

0

6

4

2

0 7 6 5 4 3 2 1 8 9 10

WW, cool

WW, hot

WD, hot

Millet et al.2016 Plant Phys


Allelic effect

on grain yield

(t ha-1)

Allelic effects 1 QTL. each square, one scenario

WD

HotDN

HotDays

Cool

Mild Severe

WD

Coll van Eeuwijk Charcosset


Millet et al. 2016 Plant Phys


mild WD

HotDN

HotDays Cool

HotDN

HotDays

Cool

HotDN

HotDays

Cool

mild WD

Allelic effect

on grain yield

(t ha-1)

Coll van Eeuwijk Charcosset

Millet et al. 2016 Plant Phys

QTL1 QTL2

QTL3 QTL4

QTL5 QTL6

Allelic effects 6 QTL. each square, one scenario

WD




Environmental sensor outputs are part of phenotyping:

Environmental scenarios allow dissecting/predicting the GxE interaction

Phenotyping: Sensor networks + yield components + stat analysis

(cheap, but not for data analysis)


Affordable phenotyping : application to climate change

Sensor networks and weather grids are ‘game changers’ - Environment captured in all European fields - Make sense of simple information (yield, durations) in a range of environmental scenarios if combined with statistical and process-based modelling - Progress in remote sensing will bring more information … we are in the era of big data ; breeders already do that.

This is part of EMPHASIS - ‘Simple phenotyping in field networks’ - ‘Modelling’ - ‘Information system’

‘Proper’ Phenomics needs to bring added value to this largely the role of the methodological part of EPPN2020

Combining phenomics and modelling : yields of 100s genotypes

Need to simulate the behaviour of 100s genotypes in 100s scenarios

Where is Phenomics ? Measure genotypic parameters of models

Genomic prediction : simulation of yield based on genomic info environment needs to be taken into account

Cabrera et al. 2016 New Phytologist


PhenoArch 1680 2300 plants

Images analysis Individual 3D reconstruction

Light interception of the canopy

Parameter: Light interception and radiation use efficiency

Genetic dissection


Chen et al 2019 J Exp Bot

0

5

10

15

0 20 40 60

Phenoarch Winter13 Phenoarch Spring13 Phenoarch Spring12

Reference hybrid

Cabrera-Bosquet L. Alvarez Prado S.

Thermal time since emergence (d20°C)

Parameter: phenological stages

Leaf

ran

k


0

5

10

15

0 20 40 60

Ner12 Ner13 Phenoarch Winter13 Phenoarch Spring13 Phenoarch Spring12

Reference hybrid

Thermal time since emergence (d20°C)

Leaf

ran

k Nérac

Gaillac

Karlsruhe

Cadriano

Campagnola

Martonvasar Debrecen

Murony

Craiova

Graneros

14

Parameter: Phenological stages


R²= 0.85*** CV= 13%

Plant gs (mmol m-2 s-1) 30 40 50 60 70 80 90

100

200

300

400 Well-watered Water deficit

R²= 0.54**

Leaf

gs (

mm

ol m

-2 s

-1)

15

Parameter: Stomatal control

Alvarez Prado et al., 2018 Plant Cell Environment


[G, E]

[nG, E]

Training set (200 hyb)

Testing set (46 hyb)

0 50 100 1500

5010

015

0

Observed gs

Pred

icte

d gs

Training set; R² = 0.86New Genotypes; R² = 0.87

0 50

10

0 15

0 Spr16.WW Spr16.WD Spr13.WW Spr13.WD Spr12.WW Spr12.WD Win13.WW Win13.WD

Observed gs

Pred

icte

d g

s

0 50 100 150

0 50

10

0 15

0

[nG, E]

[G, E]

r-coeff

0.57-0.92

0.19-0.75

Genomic prediction Parameter, maximum stomatal conductance

S. Alvarez Prado Alvarez Prado et al., 2018 Plant Cell Environment


APSIM

Model parameters/Traits

Climatic and soil data from 15 field experiments

Nérac

Gaillac

Karlsruhe

Cadriano

Campagnola

Martonvasar

Debrecen

Murony

Craiova

Genomic prediction

Biomass

TE T RUE

Rint

LAI

Grain Number

[nG, E] [G, E]

Yield prediction

S. Alvarez Prado Tardieu et al 2018, annual Rev Plant Biol


Montpellier Serre + Champ

Toulouse Champ

Ouzouer Champ-Semi

Contrôlé

Dijon Serre + Champ

Bordeaux Omic

Controlled conditions

Semi controlled fields

fields

Installations in field and controlled conditions

+ imaging, information system and models

Clermont- Ferrand Champ-Semi Contrôlé

2017-2020 Access to 31 installations in Europe https://eppn2020.plant-phenotyping.eu/

Calls every 6 months (next, February 2019)

INRA

IT

ABER

ALSIA

ASA

AU

WRWU FZJ

IPK

HMGU

IPG

MTA

SUA

Pspex

UCL

UCPH

UHEL

UNOTT

VIB

VSNi

Common effort towards a European community - A European information system - Sensors and imaging - Stastistical applications

https://eppn2020.plant-phenotyping.eu/




Phenomic information system (PHIS)

The cheapest experiment is the one in a database … but data must be organized in such a way that the can be reused

BRAPI and ELIXIR (G.Valle yesterday) help reusing data, not organizing them (common work under progress)


Tracking/ontology of all objects in experiments (plants, pots, sensors…) Where were they (spatial variability), calibrations, samples of which plants ?

P. Neveu

Prefix m3p: <http://phenome-fppn.fr/m3p>

URI of plant<m3p:arch/2017/c17000118>

URI of pot:<m3p:arch/2013/pc13001542>

URI of cabin:<m3p:arch/2018/ac180015>

URI of camera:<m3p:arch/2018/ac180019>

URI of image:<m3p:arch/2017/ic17002295855>

URI of cart:<m3p:arch/2013/ct1300123>

Prefix diaphen: <http://phenome-fppn.fr/diaphen>

URI of plot<diaphen:2017/o1700029>

URI of plant:<diaphen:2017/17000147>

URI of camera:<diaphen:2018/ac180002>

URI of image:<diaphen:2017/ic14001480237>

(a) (b)

URI of leaf:<diaphen:2017/l17000590>

In field and controlled conditions

Neveu et al 2018 New Phyt Ll Cabrera

Relating/ontology of objects via semantic web ‘samples, belong to plants, leaves, genotypes, experiments...’

leaf885

Installation_1

hasEvent

variety_Aleaf884

plant736 moveTo328

leaf883

Exp2016-A

uses

sample333

sample332

sample331

#URIs of the different objects

‘Installation_1’= <http://www.nationalinfrastructure/Installation_1/>

‘Exp2016-A’=<http://www.nationalinfrastructure/Installation_1/Exp2016-A>

‘variety_A’= <http://www.nationalinfrastructure/g/A>

‘plant736’= <http://www.nationalinfrastructure/Installation_1/Exp2016-A/p0736>

‘leaf883’= <http://www.nationalinfrastructure/Installation_1/Exp2016-A/10883>

#RDF examples

#plant736 is of variety A is expressed in RDF as:

<http://www.nationalinfrastructure/Installation_1/Exp2016-A/p0736>

<http://www.nationalinfrastructure/vocabulary/2015#hasVariety>

<http://www.nationalinfrastructure/g/A>

#Informally this relationship can be expressed as:

<plant736> <hasVariety> <variety_A>

isPartOf hasVarietyisPartOf

participates_in

(a)

(b)Creates the metadata in a parsimonious way (transitivity)

Neveu et al 2018 New Phyt


(a)

anno86

plant795

user2

dcterms:creator

dcterms:created

oa:bodyValue

oa:hasTarget

(b)

(d)

Plot 23 Plot 24

(c)

"2017-05-15”

‘Fallen plant during imaging’

anno480

plot23

user1

dcterms:creator

dcterms:created

oa:bodyValue

oa:hasTarget

"2017-06-21”

‘Plots lodged after the storm’

Tracking/ontology all events in experiments (ontologies)



Allows organizing event reports, lost in lab books otherwise

(a)

(b)

(c)

(d)

(f)

(e)

Available upon demand - Under implementation in French infra + Wageningen, - Linked to other information systems via ‘Emphasis layer’, Ghent and Julich - Interfaced with ELIXIR infromation systems and BRAPI MIAPPE: a common initiative ELIXIR and EMPHASIS



Conclusion

Phenomics in the era of big data Sensor networks are the first element of phenomics adopted by breeders Phenomics stricto sensu needs added value

Combination of genomic prediction, phenomics and modelling Potentially allows yield prediction 100s genotypes 100s sites

Phenotyping is an expensive exercise per se if everything taken into account management, sensors, data management) similar costs for field (even affordable) and controlled conditions,

The cheapest experiment is the one from information systems + modelling

5th Annual Meeting, 2nd – 4th March, 2015 - Bologna, ITALY

C Welcker Ll Cabrera

S. Alvarez Prado

Field experiments Genetic analyses

Moulon, A Charcosset S Nicolas

Acknowledgements

RAGT, Euralis MaisAdour

B. Parent E. Millet O. Turc

Platform experiments, modelling, GWAS

N. Ranc T. Presterl S Praud

Wageningen F van Eeuwijk Willem Kruijer

Information system

P. Neveu Ll Cabrera

Multisite prediction of yield based on phenomics, …A tension between genetic variability of -...

Documents

Transcript of Multisite prediction of yield based on phenomics, …A tension between genetic variability of -...