Australian Plant Phenomics Facility

Mark Tester

Phenotyping – the new bottleneck in plant science

Genomics is accelerating gene discovery but how do we capitalise on these resources to establish gene function and development of new genotypes?

Physiological characterization of plants is still time consuming and labor intensive

High throughput phenotyping

Phenotyping is essential for– functional analysis of specific genes– forward and reverse genetic analyses– production of new plants with beneficial characteristics

High throughput is essential for phenotyping– in different growth conditions (e.g. watering regimes)– of many different lines

• mutant populations• mapping populations• breeding populations• germplasm collections

The technological opportunity

Relieve phenotyping bottleneck with robotics, noninvasive imaging and analysis using powerful computing

Provide “whole of lifecycle”, quantitative measurements of plant performance from the growth cabinet to the field

Help deliver genomics advances to all plant science - e.g. model systems, cereals, grapevines, natural ecosystems

Accelerate transfer of IP from gene discovery to trait discovery and release of innovative new varieties

Australian Plant Phenomics Facility

Established with NCRIS award of $15.2m to relieve the phenotyping bottleneckTotal package = $53m

Aim: To provide infrastructure based on automated image analysis to enable the phenotypic characterisation of plants- National facility, at the international forefront- Robotics, non-invasive imaging, analysis using powerful computing- ‘Whole lifecycle’ quantitative measurements of plant performance

from the growth cabinet to the field- Ontology-based storage of phenomics data

- Research collaborations, international profile and engagement

Australian Plant Phenomics Facility – two nodes

The Plant Accelerator™ Adelaide

Mark Tester (mark.tester@acpfg.com.au)

High Resolution Plant Phenomics CentreCanberra

Bob Furbank (robert.furbank@csiro.au)

$21 m$32 m

Australian Plant Phenomics Facility – two nodes

Australian Plant Phenomics Facility

The Plant Accelerator™

Mark Tester

ACPFG

The Plant Accelerator™

The Plant AcceleratorTM

High throughput phenotyping of plant populations 4,485 m2 building, 2,340 m2 of greenhouses, 250 m2 for growth chambers Grow >100,000 plants annually in a range of conditions 4 x 140 m2 fully automated ‘Smarthouses’

– Plants delivered on 1.2 km of conveyors to five sets of cameras– High capacity state-of-the-art image capture and analysis equipment– Regular, non-destructive measurements of growth, development, physiology

First public sector facility of this type and scale in the world– Owned by University of Adelaide, opened 29 Jan 2010– National facility to support Australian plant research– Full GM and quarantine status

UniSA and ACPFG established a Chair and Assoc Prof in Plant Phenomics and Bioinformatics ($1.5m)

Measuring techniques relevant for drought research

Colour imaging– biomass, structure, phenology– leaf health (chlorosis, necrosis)

Near infrared imaging– tissue water content– soil water content

Far infrared imaging– canopy/leaf temperature

Fluorescence imaging– physiological state of photosynthetic machinery

Automated weighing and watering– water usage, control of drought conditions

Image acquisition modes

Top View Side View Side View 90°

Technical Details:

Camera: 1280 x 960 PixelOptic: 17 mm technical optic

Plant skeleton analysisKey to growth dynamics and

morphology

• separation of stem and leaves• information about nodes, length of leaves• morphology• plant growth phase

 

 

Color classification of leaves

User defined color classification e.g. to characterise plant fitness under optimum or draught conditions or to distinguish herbicide/genetically modified from other plants

0

20

40

60

80

100

120

140

A01 A02 A03 A04 B01 B02 B03 B04 C01 C02 C03 C04 D01 D02 D03 D04

Pot position

Are

a in

mm

²

yellow areadark green arealight green area

Quantitative morphology to characterise plants

050

100150200250300350400

width

height

area top

area wide side

area small side

Dist B-1

Dist 1-2

Dist 2-3Dist 3-4Dist 4-5

Dist 5-6

Angle 1

Angle 2

Angle 3

Angle 4

Angle 5

Angle 6

0

100

200

300

400width

heightarea top

area wide side

area small side

Dist B-1

Dist 1-2

Dist 2-3Dist 3-4Dist 4-5

Dist 5-6

Angle 1

Angle 2

Angle 3

Angle 4

Angle 5Angle 6

0

100

200

300

400w idth

height

area top

area w ide side

area small side

Dist B-1

Dist 1-2

Dist 2-3Dist 3-4Dist 4-5

Dist 5-6

Angle 1

Angle 2

Angle 3

Angle 4

Angle 5

Angle 6

0

100

200

300

400w idth

height

area top

area w ide side

area small side

Dist B-1

Dist 1-2

Dist 2-3Dist 3-4Dist 4-5

Dist 5-6

Angle 1

Angle 2

Angle 3

Angle 4

Angle 5

Angle 6

Fingerprinting of morphological data

•Areas

•Node distances

•Leaf-stem angle

•Height

•width

Plant colour classificationKey to plant health

LemnaTec 2004

Identification of active male flower

But wheat is not as neat….

Growth measurements – counting pixels

0 5 10 15 20 25 30 350

50000

100000

150000

200000

250000

300000

Berkut Krichauff

Time post transplant [days]

Proj

ecte

d sh

oot a

rea

[pix

el]

mean±SE;n=8

Estimation of shoot biomass

The projected shoot area of the RBG images gives a good correlation with shoot biomass

Tested for various plant species– wheat, barley– rice– cotton– Arabidopsis …

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.60

50000

100000

150000

200000

250000

300000

f(x) = 154154.220319601 x + 19065.3925892172R² = 0.920516039741044

Dry weight [g]

Proj

ecte

d sh

oot a

rea

[pix

el]

5wk old barley plants, 8 cultivars

Estimation of shoot biomass

0 0.05 0.1 0.15 0.2 0.250

10000

20000

30000

40000

50000

60000

f(x) = 167806.785972392 xR² = 0.993976565757912

f(x) = 216362.953029618 xR² = 0.995820614282722

control

Dry weight [g]

proj

ecte

d sh

oot a

rea

[pix

el]

20d old barley

But control and salt stressed plants have different area-weight ratios

Measured shoot dry weight [g]

Pred

icte

d sh

oot d

ry w

eigh

t [g]

Golzarian et al. (2010) IEEE Proceedings Signal Processing, in review

Estimation of shoot biomass

Improved estimate of biomass when age of the plant is taken into account

Y = a0 + a1×(G+B+Y)+ a2×(G+B+Y)×H(H = number of days after seed

preparation date)

(Correction for leaf colour did not greatly improve weight estimates)

(Cross validation run 10x)

Colour classified image

Line Green area Necrosis area % Necrosis

Sahara 30739 4232 12%

Clipper 11640 15321 57%

Treated with 100 mM GeO2, 8 d

Julie Hayes, Margie Pallotta and Tim Sutton, ACPFG

Use of colour informatione.g. boron toxicity screen

Original image

B toxicity - leaf symptoms Ge toxicity - leaf symptoms

Jefferies et al. 1999. TAG 98, 1293-1303 Hayes et al., unpubl., using LemnaTec

QTL for Ge tolerance identified using colour imaging overlaps QTL for B tolerance (1999)

Salinity tolerance - trait dissection

Breeding for overall salt tolerance difficult due to low heritability

Dissection into individual traits suitable for forward genetics approach

Use of The Plant AcceleratorTM to perform high throughput phenotyping

Na+ exclusion

tissue tolerance

osmotictolerance

Munns & Tester (2008) Annu Rev Plant Biol 59: 651-681

Osmotic tolerance screen in bread wheat

Mapping population of Berkut x Krichauff– Berkut – CIMMYT– Krichauff – Australian cultivar– Berkut higher overall

tolerance despite higher tissue [Na+]

Parents– Berkut – 0.65– Krichauff – 0.33

Range of progeny– 0.13 to 0.96 (day-1)

Berkut

Krichauff

Karthika Rajendran

QTL mapping of osmotic tolerance

Significant QTL on chromosome 1D

QTL1D.9 explains 21% of phenotypic variation in the population

Favourable allele comes from Berkut

Chromosome 1D

Karthika Rajendran

Hardware purchased

• Currently– IBM BladeCenter Chassis– 3 x HS21 blade servers– 6 x HS22 blade servers– 2 x DS4700 storage controller– 8 x DS4000 storage expansion units– 140 x 1TB hard drives– $510K (2008 & 2009)

• Virtualisation with VMware

• Expansion, room for– 5 additional servers– 20 additional hard disks

• TPA acknowledges– Lachlan Tailby (ACPFG)

• Picked up by IBM’s Smarter Planet campaign

LemnaTec Data System

FLUO

1392 x 1040

RGB

2056 x 2454

IR

320 x 256 320 x 256

NIRSnapshot

Smarthousedatabase

Imaging configurations

Conveyor tasks

Watering tasks

Smarthouse operations

• Single database stores acquired data, SmartHouse operation configurations and tasks and analysis results

• No project level data management– Backup, archive, delete– Access control

• Around 30MB per snapshot– 72 GB per day, 0.5 TB per week Analysis results

James Eddes

Data flow / management

SH1

Smarthouse 1 (South)

OPE

RATI

ON

& A

CQU

ISIT

ION

DAT

A PR

OCE

SSIN

G &

MIN

ING

Smarthouse 2 (North)

Plant Accelerator Project DBs

LemnaTec Production DB

SH2

SH2SH1

Project 1 Project 2 Project 3 Project 4 Project 5

buffe

red

tran

sfer

buffe

red

tran

sfer

Project 6

MID

DLE

DaemonDaemonLemnaLauncher

LemnaLauncher LemnaLauncher

LemnaMiner

Plant Accelerator servers

James Eddes

Building databases, managing export of data from LemnaTec, returning data to LemnaTec for further analyses

Image analyses – LemnaTec image processing grids, quality control, basic statistics

Data service – image directories, processing, analysis spreadsheets, metadata, PODD

Data dissemination Embargo Offsite back-up

Data management issues

James Eddes

• Data acquisition• Data management• Image analysis

• Counting pixels• 3-D modelling – computer vision, machine intelligence

• Statistical analyses• Modeling and biological interpretation

• Plugging numbers back in to the plant• Genetics – aligning phenomics data with genomics data to allow

quantitative genetics

Wider computational issues

• Raise money, hire people, collaborate• NCRIS ALA (Bogdan!) - Systems manager, feeding PODD• NCRIS ANDS - 2 data architects for 1 yr, feeding PODD• EIF programming - Image analysis• - Computer vision (Anton Vandenhengel)• ARC Linkage (LemnaTec) - Image analysis, computer vision• HFSP - Computer vision• - Machine intelligence• Collaboration with - PODD, ALA, etc within IBS• - UniSA node of ACPFG • - Desmond Lun• - Computer vision group of UniAdl• - Anton Vandenhengel

Plans to address issues

The Plant Accelerator™ team to date

Mark Tester Geoff Fincher Helli Meinecke – business manager Bettina Berger – postdoctoral scientist James Eddes, Bogdan Masznicz, Jianfeng Li – computer programmers Robin Hosking – horticulturalist Richard Norrish – electrical engineer Lidia Mischis, A.N. Other – technicians Karthika Rajendran – PhD student Brett Harris – Honours student Desmond Lun, Irene Hudson, Mahmood Golzarian

– UniSA /ACPFG maths, stats Anton van den Hengel – UA computer vision + three programmers in UQ to construct the database repository

www.plantaccelerator.org.au www.plantphenomics.org.au