Stochastic single gene expression model · Goodwin’s modelStochastic model Reduced modelInverse...
Transcript of Stochastic single gene expression model · Goodwin’s modelStochastic model Reduced modelInverse...
Stochastic single gene expression model
Romain Yvinec1, Michael C. Mackey2, MartaTyran-Kaminska3, Changjiing Zhuge4, Jinzhi Lei4
1BIOS group, INRA Tours, France.
2McGill University, Canada.
3University of Silesia, Poland.
4Tsinghua University, China.
Goodwin’s deterministic model
Stochastic gene expression modelReduced modelInverse problemFuture work : Taking into account division
Goodwin’s model Stochastic model
Crick (1958) : Central Dogma.Jacob, Perrin, Sanchez, Monod (1960) : Operon.
$
’
’
’
’
’
’
&
’
’
’
’
’
’
%
dM
dt“ λ1pE q ´ γ1M ,
dI
dt“ λ2M ´ γ2I ,
dE
dt“ λ3I ´ γ3E .
(1)
Goodwin (1965), Griffith (1968), Othmer (1976), Selgrade (1979)...
Goodwin’s model Stochastic model
The transcription rate function λ1
§ Inducible Operon : Repressors R interacts with both theOperator O and the Effector E ,
R ` nEK1é REn , K1 “
REn
R ¨ En,
O ` RK2é OR , K2 “
OR
O ¨ R.
With QSSA, and if Otot ! Rtot ,
λ1pE q „O
Otot“
1` K1En
1` K2Rtot ` K1En. (2)
§ Repressible Operon : Similar but
O ` REn K2é OREn , K2 “
OREn
O ¨ REn.
and we get
λ1pE q „O
Otot“
1` K1En
1` pK1 ` K2RtotqEn. (3)
Goodwin’s model Stochastic model
Bifurcation analysis in ODE
0 5 10 15 200
2
4
6
8
10
K
κd
uninduced
bistable
induced
§ Inducible : Mono-stabilityor Bi-stability.
§ Repressible : Mono-stabilityor limit cycle.
$
’
’
’
’
&
’
’
’
’
%
dx1dt
“ γ1rf px3q ´ x1s ,
dx2dt
“ γ2px1 ´ x2q ,
dx3dt
“ γ3px2 ´ x3q .
(4)
Here f pxq “ κd1` xn
K ` xn.
0 1 2 3 4 5 6 70
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
x
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Eldar and Elowitz (Nature 2010)
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
’New’ Central dogma
DNA
mRNA ProteinON
OFF
Novick & Weiner (1957), Ko et al. (1990), Ozbudak et al. (2002), Elowitz et al. (2002), Raser & O’Shea (2004)...
$
’
’
’
’
’
&
’
’
’
’
’
%
dx
dt“ G ptqλ1pyptqq ´ γ1xptq ,
dy
dt“ λ2xptq ´ γ2yptq ,
pG “ 0qαpyptqqÝÝÝÝáâÝÝÝÝβpyptqq
pG “ 1q .
(5)
Rigney & Schieve (1977), Berg (1978), Peccoud & Ycart (1995), Thattai & Van Oudenaarden (2001)...
Outline
Goodwin’s deterministic model
Stochastic gene expression modelReduced modelInverse problemFuture work : Taking into account division
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Can we perform a systematic bifurcation theory on suchsystems ?
§ If the gene switching rates are fast (α, β Ñ8), we recoverthe Goodwin’s model (no gene noise).
§ If the mRNA lifetime si short (γ1 Ñ8), we can perform an
adiabatic reduction (xptq « G ptqλ1γ1pyptqq) :
$
’
&
’
%
dy
dt“ G ptqλ2λ1
γ1pyptqq ´ γ2yptq ,
pG “ 0qαpyptqqÝÝÝÝáâÝÝÝÝβpyptqq
pG “ 1q .(6)
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
$
’
&
’
%
dy
dt“ G ptqλ2λ1
γ1pyptqq ´ γ2yptq ,
pG “ 0qαpyptqqÝÝÝÝáâÝÝÝÝβpyptqq
pG “ 1q .(7)
§ If the Gene active periods are short (β Ñ8), we obtain thebursting model
dy
dt“ Z ptq ´ γ2yptq , (8)
where Z “ř
i ZiδTiis a jump process, of jump rate αpyptqq
and jump size distribution
P
ypT`i q ě z | ypT´i q “ y(
“ exp´
´
ż z
y
γ1β
λ1λ2pwqdw
¯
.
RemarkFor a constitutive gene, b :“ λ1λ2
γ1βis the average number of
proteins produced per Gene activation event.
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Bifurcation analysis inSDE
0 1 2 3 4 5 6 70
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
x
§ Inducible :Uni-modal orBi-modal.
§ Repressible :Uni-modal.
(under technical assumptions...)upt, yqdy “ P
yptq P ry , y ` dyq(
converges as t Ñ8 (in L1) towards u˚,
du˚
dy“
„
αpyq ´ γ2p1`y
bpyqq
u˚pyq
γ2y.
Here αpyq “ κb1` yn
K ` yn, and bpyq “ b .
0 2 4 6 8 10 12 14 150
5
10
15
20
K
The presence of bursting can drastically alter regions of bistability
κd−/+
b*κb=κ
d =K
bκb−/+
; b=0.01
bκb−/+
; b=0.1
bκb−/+
; b=1
bκb−/+
; b=10
n=4
κd ≡ b*κ
bκ
d−/+
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Stationary distribution pα, γ, bq ñ pu˚q
3
3.5
3.8
4
4.44
5
5.5
0 1 2 3 4 5 6 7 86
x
κb
2.8
3
3.15
3.3
3.7
4
4.45
0 1 2 3 4 5 6 7 85
x
κb
Here αpyq “ κb1` yn
K ` ynand bpyq “ b.
Small recap’ on the forward problem (1/2)
§ We performed an adiabatic reduction to make the problemanalytical tractable.
§ We solved the reduced problem for arbitrary coefficients atequilibrium.
§ We performed an (deterministic-analogous) bifurcation study.
0 2.5 50
0.5
1
1.5
X
density
0 2 4 6 8 10
0
0.2
0.4
0.6
Y
density
0 5 100
0.2
0.4
0.6
0.8
1
X
0 2 4 6 8 10
0
0,1
0,2
Y0 2 4 6 8 10
0
0,1
0,2
Y
0 25 500
0.2
0.4
0.6
0.8
1
X
0 2 4 6 8 10
0
0,1
0,2
Y
0 200 4000
0.2
0.4
0.6
0.8
1
X
γ1=0.1 γ
1=1 γ
1=10 γ
1=100
Small recap’ on the forward problem (2/2)
§ Careful ! The two notions of deterministic bistability andstochastic bimodality are in fact quiet different
0 100 200 300 400 500 600 700 800 900 10000
0.5
1
1.5
2
2.5Deterministic Bistabilty
Time
x
κd=2.3
γ=1
K=4n=4
0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000
2
4
6
8
10
12
14Bistability with Bursting Only
Time
x
Outline
Goodwin’s deterministic model
Stochastic gene expression modelReduced modelInverse problemFuture work : Taking into account division
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Inverse Problem :pu˚q ñ pα, γ, bq
For a constitutive gene, we can infer the burst rate (in proteinlifetime unit) α
γ and the mean burst size b from the first twomoments
bα
γ“ E
“
X‰
,
b “VarpX q
E“
X‰ .
For an auto-regulated gene, we can inverse the formula for thestationary pdf :
pxu˚pxqq1
u˚pxq“αpxq
γ´
x
bpxq.
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Simulated data
Density reconstruction by Kernel Density Estimation
x
0 2 4 6 8 10 12 14 16 18 20
pd
f
0
0.05
0.1
0.15
0.2
0.25True solution
Kernel density estimation
Histogram
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Inferred bursting rate
x
0 5 10 15
α(x
)#
0
1
2
3
4
5
6
7
8With the exact b
True solution
With b estimated in the same time
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Resulting Probability Density Function
x
0 5 10 15
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35True solution
With the exact b
With b estimated at the same time
Kernel Density estimation
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Single cell data onself-regulating gene
§ Synthetic Tet-Off in buddingyeast.
§ Feedback modulated by anexternal parameter(doxycycline)
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Single cell data onself-regulating gene
Smoothed pdf obtained bykernel density estimation
x
1 1.5 2 2.5 3 3.5 4 4.5 5
0
1
2
3
4
5
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Inferred bursting rate
x
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
α(x)
0
10
20
30
40
50
60
70
80
90
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Resulting Probability Density Function
x
0 5 10 15 20 25 30
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Does that Fit ? Can do better !
x
0 1 2 3 4 5 6 7 8 9 10
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Small recap’ on the inverse problem
§ With the help of the full solution, we obtained a formula tofind the parameter functions from the stationary density.
§ We applied this on simulated and real data.
§ The inverse problem seems generally ill-posed (cannot findburst size b and burst rate α at the same time).
§ If the resulting pdf does not ’fit’ the data, it may be a goodreason to rule out this model.
§ Work still on progess...
Outline
Goodwin’s deterministic model
Stochastic gene expression modelReduced modelInverse problemFuture work : Taking into account division
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Similar results may be obtained for a continuous growth divisionmodel or a bursting-division model. For instance, with uniformrepartition kernel, constant division rate d and constant burst sizeb,
d
dyu˚ “
„
´α1pyq ` d
αpyq ` d`
αpyq
αpyq ` d
´1
x`
1
b
¯
´xb2
bx ` 1´
1
x
u˚pyq
0
10
20
30
40
50
60
0 5 10 15 20 25
X(t)
time
One stochastic trajectory, λb=10
0
10
20
30
40
50
60
0 5 10 15 20 25
X(t)
time
One stochastic trajectory, λb=10
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
This may be used to predict the long time behavior of a dividingcell population
A cell tree, λb=10 A cell tree, λb=100
§ Inverse problems may be adapted to genealogical tree data(Doumic et al. 2014, Krell 2014)
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
This may be used to predict the long time behavior of a dividingcell population
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0 10 20 30 40 50 60
u*(x)
x
Steady-state profile
λb=1, b=10
λb=5, b=2
λb=10, b=1
λb=100, b=0.1
§ Such framework can be used to study non-linear populationmodel...
Goodwin’s model Stochastic model Reduced model Inverse problem Division model
Vielen Dank !
§ Molecular distributions in gene regulatory dynamics, M.CMackey, M. Tyran-Kaminska and R.Y., Journal of TheoreticalBiology (2011) 274 :84-96
§ Dynamic Behavior of Stochastic Gene Expression Models inthe Presence of Bursting, M.C Mackey, M. Tyran-Kaminskaand R.Y., SIAM Journal on Applied Mathematics (2013)73 :1830-1852
§ Adiabatic reduction of a model of stochastic gene expressionwith jump Markov process, R.Y., C. Zhuge, J. Lei, M.CMackey, Journal of Mathematical Biology (2014)68 :1051-1070