Transcript Part 2

Modelling and experimentation of the spatio-temporal spread of
soilborne pathogens: Rhizoctonia solani on sugar beet as an
example pathosystem
Leclerc Melen
PhD defence
1st February 2013 – Agrocampus Ouest
UMR – IGEPP
Cifre I.T.B
Reporters: Joël Chadoeuf, Christian Lannou
Examiners: Yannick Outreman, Marc Richard-Molard
Supervisors: Philippe Lucas, Thierry Doré, João Filipe
Introduction
General context – current problems
• 50% reduction in pesticide use (Grenelle de l’environnement  Ecophyto)
• Find alternatives to pesticide use
• Keep crop production levels and growers’ incomes
2
Introduction
Proposed approaches
• Necessity of considering the complexity of agro-ecosystems (e.g. using system approaches)
• Understand ecological and epidemiological processes involved in the dynamics of pathogens
• Use ecological and epidemiological knowledge to improve pest management and design
efficient crop protection strategies
• Combine several controls with partial effects (there is no ‘one fits all’ solution)
• SysPID Casdar Project: Reduce the impact of soilborne diseases in crop systems towards an
integrated and sustainable pest management
• Action n°2: Epidemiological processes in field crop systems
3
Introduction
Soilborne disease epidemics
Soilborne diseases
• Wide range of pathogenic organisms (fungi, bacteria, viruses, nematodes, protozoa)
• Cause substantial damage to crops worldwide (up to 50 % of crop loss in the US (Lewis &
Papaizas, 1991) )
• Pathogens often survive for many years in soils (5-7 years for Pythium)
• Difficult to detect, predict and control
Epidemiology
• External source of inoculum X(t) : dynamic pathogen population
• Primary infections : external inoculum epidemic initiation
• Secondary infections : spread of the pathogen within the crop/population
Susceptible
hosts
S
X
I
Infected/Infectious
hosts
4
Rhizoctonia solani on sugar beet as an example pathosystem
Introduction
The host
The pathogen
5
Introduction
The host : sugar beet
The plant
• Cultivated Beta vulgaris
• High production of sucrose
The crop
• Grown for sugar production
• France is one of the largest producer (33 Mt in 2009)
6
Introduction
The pathogen : Rhizoctonia solani
R. solani fungi
• Basidiomycetes
• Polyphagous saprotrophic fungi
• Anastomisis Groups (AG)
R. solani AG2-2 IIIB
• Parasites maize, rice, sugar beet, ginger …
• Important optimal temperature range for growth
• Develops late in the growing season
• Infects mostly mature plants
7
Aoyagi et al., 1998
Introduction
The disease : the brown root rot disease of sugar beet
8
Introduction
Rhizoctonia root rot disease control
Current management strategies
• Think crop rotations (host & non-host crops)
• Resistant varieties
• Biological controls ?
Antagonists (e.g. Trichoderma fungi)
Biofumigation
9
Introduction
Our system
Visible epidemic:
symptomatic plants
Above-ground
Hidden epidemic:
Cryptic infections
Source of inoculum :
R. solani
Temporal scale : growing season of sugar beet
Belowground
 Spatial scale : field
10
Introduction
Research questions and structure of the presentation
1. How does R. solani spread in field conditions ?
(Understand epidemiology of R. solani in field conditions)
Study 1 : R. solani spread
Experimentation
Modelling
2. How to infer hidden infections from observations of the disease ?
Study 2 : Incubation period
Experimentation
Modelling
3. How does biofumigation affect epidemic development ?
Study 3 : Effects of biofumigation
2007 data (Motisi, 2009)
Modelling
11
Part 1
How does R. solani spread in field conditions ?
12
Part 1
Pathozone concept and spread of soilborne pathogens
• Difficult to asses R. solani growth in soils  use of pathozone concept
• “Pathozone means the region of soil surrounding a host unit within which the centre of a
propagule must lie for infection of the host unit to be possible” (Gilligan, 1985)
Probability of
infection
Placement experiment
Inoculum(donor)-host(recipient)
n replicates at distance x
ni number of infected recipients
at time t
 P(x,t) = ni / n
13
Part 1
Results of the experiments (Pathozone profiles)
• Field experiments in 2011 (Le Rheu)
Primary inoculum
Secondary inoculum
(5 infested barley seeds)
(an infected plant)
• Localised spread (nearest neigbour plants) (Filipe et al., 2004, Gibson et al.,2006)
• Infections occurs further with secondary inoculum ( Kleczkowski et al.,1997)
The fungus translocates nutrients from the parasited host to other parts of the mycelium
14
Part 1
Host growth may increase pathogen transmission at individual level
 Host-plant growth decreases the contact distance between neighbouring plants
 Host growth may increase pathogen
transmission at individual level
15
Part 1
Host growth can trigger the development of epidemics
Static contact distance
Dynamic contact distance
xcc= 11 cm
Non-invasive behaviour (linear trend)
xHost
can cause a switch from non-invasive to invasive behaviour
= 14 growth
cm
cc
xcc= 17 cm
Invasive behaviour (non-linear trend)
16
Part 1
Conclusion (Part 1)
• First Pathozone profiles measured in a real soil
• Locality of pathogen spread in field conditions
 Importance of considering space : mean-field approximation/homogeneous mixing
assumption may fail when predicting the spread of soilborne pathogens (Dieckmann et al.,
2000 ; Filipe & Gibson, 2001)
• Host growth can trigger the development of epidemics by decreasing contact distances
 Need to take into account host growth for epidemic thresholds – conditions for invasive
spread of plant population by pathogens (Grassberger, 1983 ; Brown & Bolker, 2004)
17
Part 2
How to infer hidden infections from observations ?
18
Part 2
Problem
19
Part 2
A problem of incubation period
• Time between hidden infection and appearance of detectable symptoms of pathology
 Incubation period (Kern, 1956 ; Keeling & Rohani, 2008)
• Incubation period distributions are often described by non-negative probability distributions
with a pronounced mode (Keeling & Rohani, 2008 ; Chan & Johansson, 2012)
• Incubation period data are rare …
20
Part 2
Compartmental models an incubation period distributions
• How to describe cryptic infections with compartmental models ?
Susceptible hosts
Infected/infectious hosts
Detectable/symptomatic hosts
• Compartmental Markovian models
The time spent in each state is exponentially distributed
 In a simple SID model the incubation period is exponentially distributed
• How to introduce realistic distributions in classical compartmental models ?Realistic distribution
A tractable way: by subdividing compartments (i.e. introducing transient states)
Distribution in
classical Markovian
Sum of exponentialy distributed random variables
models
=
Erlang (or Gamma) distributed random variable
21
Part 2
Working hypothesis and methodology
• Hypothesis: the distribution (e.g. mean and range) of the incubation period is age-dependent
• Methodology:
 Experimental measures for various ages of infection
 Statistical analysis (is Gamma distribution robust enough ?)
 Build a model for age-varying distribution of the incubation period
 Incorporate it into an SID compartmental model
22
Part 2
Experimental measures of the incubation period
Experiments
• Plant inoculated with 3 infested barley seeds (inoculum)
• 9 ages of plants (14, 32, 46, 60, 74, 88, 102, 116, 130 days)
• For each individual the time of first above-ground symptom was recorded
• At least 45 individual observations for each age  distributions of the incubation period
23
Part 2
Inubation period calculated in degree- days
Raw data (results of the experiments)
24
Part 2
Age-varying model of the incubation period distribution
• Age by age distribution analysis  Gamma distribution (general case of Erlang) can reasonably
describe incubation period distributions
• Age-varying model of the incubation period T(t)
T (t ) ~ Erlang[k ,  (t )] with k an integer and  (t )  aebt  c
k : shape parameter =
λ: time dependent rate parameter
number of transient states (=19)
• Compartmental model with realistic incubation distribution (19 transient non-symptomatic states)
25
Part 2
Hidden infections and observations
• Simulations of cryptic epidemics (individual-based spatial model with stochastic continuous time)
 Infected and detectable/symptomatic individuals have different dynamics
26
Part 2
Conclusion (Part 2)
• One of the first epidemiological model for soilborne disease with data-supported incubation
period
• Link hidden processes and observations of disease
 Estimate rates of infections and cryptic infections from observations
 Test management strategies based on the detection of symptomatic individuals
• This end of the incubation period corresponds to a visual detectability
 May change with other detection/survey methods, e.g. molecular techniques, remote sensing
• Variability ? (soils, human error, strains, environmental conditions …)
27
Part 3
How does biofumigation affect epidemic development?
28
Part 3
Background
Previous work
Susceptible
dS
   (t ) X   (t ) I (t )  S (t )
dt
Cryptic infections
dI
  (t ) X   (t ) I (t )  S (t )
dt
Detectable plants
D I
Rates of infection
 (t )  1 exp( 2t )
Force of infection
• The effect of biofumigation on the root rot disease has been analysed using a simple
epidemiological model (Motisi, 2009 ; Motisi et al., 2012)
 (t )  1 exp(0.5[log(t /  3 ) /  2 ]2 )
• Observations of symptomatic plants for 3 treatments :
1) without control, 2) with complete biofumigation, 3) with partial biofumigation
 Biofumigation affects mostly primary infections
 Biofumigation can affect secondary infections with a variable pattern
29
Part 3
Aim of the current study
1. Integrate new epidemiological knowledge and data
2. Improve existing epidemiological models
3. Re-analyse the effects of biofumigation
4. Investigate the variability of epidemics to estimate uncertainty in the outcome of
treatments
30
Part 3
•
Improved epidemiological model : epidemic predictions
Spatial individual-based model with stochastic spread of the pathogen
 Spatial component : better description of epidemics
 Stochastic model : introduce variability in outcomes  predictions of uncertainty
Stochastic infections
P( St  I t dt )  [ (t )   (t )nI ].dt
Rate of primary infection
 (t )  1 exp( 2 (t  t0 )) if t0  t

if t0  t
 (t )  0
Rate of secondary infection
  1 exp(0.5[log(t / 3 ) /  2 ]²)
31
Part 3
Estimate parameter for each treatment from observations of disease
•
Introduce a more realistic incubation period for inferring epidemiological parameters
•
Statistical inferrence of spatio-temporal can be difficult and time consuming…
 Estimate spatial rates of infection using a semi-spatial model (Filipe et al., 2004)
•
Localized spread of infections (see Part 1)
 Pair approximation (Matsuda et al., 1992 ; Filipe & Gibson, 1998 ; van Baalen, 2000)
•
Need to describe the dynamics of all pairs of the system (i.e. SS, SI II for an SI model)
 Tractability : necessity to simplify the incubation period…
32
Part 3
Model fitting and estimated rates of infection
Rate of primary infection
 (t )  1 exp( 2 (t  t0 )) if t0  t

if t0  t
 (t )  0
Symptomatic plants (2007 data)
Rate of secondary infection
  1 exp(0.5[log(t / 3 ) /  2 ]²)
 Biofumigation reduced rates of primary and secondary
infection in this trial (2007)
33
Part 3
Spatial model predictions
Distributions of infected plants at
harvest (%)
• Biofumigation allows a partial control of epidemics
• Biofumigation seems to reduce the uncertainty in epidemic outcome
• Marginal differences between partial and complete biofumigation in 2007
34
Part 3
Conclusion (Part 3)
• Analyses are consistent with previous results on the effect of biofumigation on the spread of R.
solani, but
• We predict less primary infections and more secondary infections than in the previous
study
 New vision of epidemic : different disease progress curves
• Biofumigation seems to reduce the uncertainty in epidemic outcome
• Take these results with care
 More statistical analyses are required to assess model fitting and conclude on the effects of
treatments on epidemic development
 Assess the effects of incubation period simplification – Pairwise vs temporal model …
 Isotropic space (may overestimate epidemics ?)
• Re-analyse 2008 data
35
General conclusion
1 dPSS
  PSS   ( z  1)[ PSS  PSSS ]
2 dt
dPSI1
  PSS  (    1 ) PSI1   ( z  1)[ PSI1  PSSI1  PSS  PSSS ]
dt
dPSI 2
 1 PSI1  (    2 ) PSI 2   ( z  1)[ PSI 2  PSSI 2 ]
dt
dPSD
 2 PSI 2  (   ) PSD   ( z  1)[ PSD  PSSD ]
dt
1 dPI1I1
 (   ) PSI1  1 PI1I1   ( z  1)[ PSI1  PSSI1 ]
2 dt
dPI1I 2
 (   ) PSI 2  1 PI1I1  (1  2 ) PI1I 2   ( z  1)[ PSI 2  PSSI 2 ]
dt
dPI1D
 (   ) PSD  2 PI1I 2  1 PI1D   ( z  1)[ PSD  PSSD ]
dt
1 dPI 2 I 2
 1 PI1I 2  2 PI 2 I 2
2 dt
dPI 2 D
 2 PI 2 I 2  1 PI1D  2 PI 2 D
dt
dPDD
 2 PI 2 D
dt
4
Prob( S t  I t+dt )   s,k (t  tinf, k , x) dt
k 1
36
General conclusion
Soilborne disease epidemics
• This work provide insights into root rot disease epidemics
 spread of R. solani
 incubation period
• Data-supported studies – field experiments
• We still need to improve knowledge on the epidemiology of this disease
• May apply to others pathosystems : perennial and non-perennial plants
37
General conclusion
Control of soilborne disease epidemics
• Biofumigation
 partial control of the root rot disease (Motisi et al., 2009 , 2010, 2012)
 can reduce the uncertainty in epidemic outcome
• This work points out important epidemiological parameters for disease management
 Design and test new strategies
Plant growth  use crop mixing, precise key phenological stages to select for resistances
Incubation period  improve disease survey
Locality of pathogen spread  optimize the effects of treatments, use local treatments ?
• Combine partial controls (new and conventional)  improve the control of epidemics
• Models may help to test disease management strategies
38
General conclusion
Perspectives
• Mutltiple perspectives (theoretical, applied, epidemiological, ecological…)
• Consider main environmental parameters (temperature, moisture)
• Investigate pathogen dynamic at the crop rotation scale
• Understand ecological functionning of soils (pathogenic and non pathogenic communities)
39
• Doug Bailey
• Philipe Lucas – Thierry Doré – João Filipe
Merci…
• Françoise Montfort
• Les anciens membres de l’équipe EPSOS
• UMR IGEPP
• Les Unités Expérimentales de Dijon et de Le Rheu
• L’Institut Technique de la Betterave
• Christian Lannou – Joël Chadoeuf – Marc Richard Molard – Yannick Outreman (jury)
• Pauline Ezanno – Marie Gosme – Christian Steinberg – Agnès Champeil – Étienne Rivot
(comité de thèse)
Mon bureau…avant-hier
• Chris Gilligan et l’Epidemiology and Modelling Group
• Les membres du projet Casdar SysPID
• Le portakabin (qui a eu chaud…)
• Et tous ceux qui m’ont supporté…
Introduction
Evolution of symptoms
41
Introduction
The pathogen: R. solani
Rate of primary infection
 Maximum rate 
p  

 of infection (a p ) 
 Spatial decline due to

 location of inoculum away

from host ( p )






 Time decline due to 


 source of nutrients 
 decline ( ) 
p


 Delay in 
 onset of 


 infection ( ) 


Rate of secondary infection



(as ) 
Maximum rate
s  
of infection

 Spatial decline due to
 location of inoculum away


from host ( s )






 Delay in 
 onset of 


 infection ( ) 


Rates of infection and pathozones
dPp ( x, t )
  p ( x, t )[1  Pp ( x, t )]
dt
dPs ( x, t )
 s ( x, t )[1  Ps ( x, t )]
dt
Pathozones P(x,t)
x: contact distance
t: time of exposure
Infer parameters from pair experiment data
ninf: number of infected recipients
ninf ( x, t ) ~ Binomial (ntot , P( x, t ))
ntot: number of replicates (25) 42
Part 2
Incubation period ?
• Periods in Natural history of disease in a host – these are incorporated as compartments in
epidemiological models
Latent/Exposed
Incubation
Infectious
Recovered
Disease
Disease
Incubation
Disease
Pathological status
Incubation
time of infection
Infectious status
Susceptible
time since
infection
• Incubation period:
"time required for multiplication of a parasitic organism within a host organism up to the
threshold point at which the parasite population is large enough to produce detectable
symptoms of pathology“ (Kern, 1956)
43
Introduction
Host growth and dynamic contact distances
• Increase in the radius h(t)
• Radial growth measured with Pepista tools (ITB)
• Simple empirical model
h(t )  5 / [1  1000exp(1.18* t 0.4 )]
• Dynamic of the contact distance between
nearest neighbours xee(t)
xee (t )  xcc  h(t )
• Static centre-centre distance xcc
xee (t )  xcc  1.5 h(t ) , if 70  tinf < 90
, if 30 < tinf < 70
xee (t )  xcc  2 h(t ) , if 90  tinf
• Spatial population model
• 30*30 square lattice
• t0= 30 days
• 5% infected
4
Prob( S t  I t+dt )   s,k (t  tinf, k , x) k (t ) dt
k 1
 x  xcc

xee (t )
 x  44
Part 3
First results
Previous epidemiological
model
Detectable/
symptomatic
Force of infection
 dS
 dt    (t ) X   (t ) I (t )  S (t )

 dI   (t ) X   (t ) I (t ) S (t )
 dt


D   I
 (t )   exp( t )
1
2

  (t )  1 exp(0.5[log(t / 3 ) /  2 ]2 )
Infected
Assumptions
• Mean field mass action/homogeneous
mixing assumption
• Epidemics initiated too soon
Rate of primary
infection
• Pre-emergence damping off
• Unrealistic incubation period
Motisi et al., 2012
45