A model for T cell differentiation Natasa Miskov-Zivanov University - - PowerPoint PPT Presentation

A model for T cell differentiation

Natasa Miskov-Zivanov University of Pittsburgh

000 001 101 011 111 010 100 110

p1 p2 p3

PI Meeting University of Maryland April 2011

Acknowledgements

 Faeder Lab:

 Department of Computational and Systems Biology, School of Medicine, University of Pittsburgh  John Sekar, James Faeder

 Morel Lab:

 Department of Immunology, School of Medicine, University of Pittsburgh  Michael Turner, Penelope Morel

 Clarke Lab:

 Computer Science Department, Carnegie Mellon University  Paolo Zuliani, Haijun Gong, Qinsi Wang, Edmund Clarke

PI Meeting, April 2011

2

Timeline

PI Meeting, April 2011

3

Kickoff October 2009 NSF Meeting March 2010 PI Review October 2010 PI Meeting April 2011

• .#

• .EF:

Statistical model checker

Today’s talk

PI Meeting, April 2011

4

System Methodology

Model simulations

Model design

Model elements Influence sets (Interaction map) Set of discrete values for each element

Experiments Expert knowledge Literature

Influence table Model rules

Circuit design methods Model analysis

Antigen presenting cell (APC) Naïve T cell

Regulatory T (Treg) cells Helper T (Th) cells

Origins of Regulatory T cells (Treg)

 Treg cells mediate antigen- specific suppression of T cell activation

 Play a key role in maintaining tolerance

 Naïve T cells can be converted into Treg cells in the periphery

 High therapeutic potential

PI Meeting, April 2011

5

Role in cancer

PI Meeting, April 2011

6

Release cytokines that stimulate the immune response Release cytokines that inhibit the immune response

Tumor cell

Antigen presenting cell (APC) Naïve T cell

Tumor secreted cytokines (e.g., TGFβ) Regulatory T (Treg) cells Helper T (Th) cells

Determinants of differentiation

 Determine whether known mechanisms are sufficient to explain experimental

• bservations

 Foxp3 transcription factor is critical for Treg function

PI Meeting, April 2011

7

Huehn et al. Nat. Rev. Immunol. 9, 83-89 (2009)

Challenges for Modeling

 Large number of components and interactions  Rapidly evolving list of important components and interactions

 structural uncertainty in the model

 Involvement of multiple processes

 signaling  gene regulation  protein expression  (cell division)

8

PI Meeting, April 2011

Network model

 Receptors:

 T cell receptor (TCR)  Co-stimulation through CD28  IL-2 receptor (IL-2R)  TGFβ receptor (TGFβR)

 Transcription factors:

 AP-1, NFAT, NFκB, SMAD3, STAT5

 Genes:

 IL-2, CD25, Foxp3

 Other important elements:

 PTEN, PI3K, PIP3, PDK1,  Akt, mTORC1, mTORC2,  TSC1-TSC2, Rheb, S6K1, pS6

9

!"#

$%# &'(! &'(!) #*+, #*-+ &#) " ! # "./0 1,02 "&34 1$!*/ 51, !67 "*02 &1#28&1#- )%39 :&;)#2 :&;)#- "."/ .<8-) =!0/ 7>?@ABCD?7EC@ ?D7EF?7EC@ E@GEHE7EC@ 4(8!& 4(8$9 !"82 )?A 3)0 (CA =I@ =40 1&!&J "0#8% K#?-L

!"82 4(8!& 4(8$9

!"#$

1$!*/ 4(8!& 1&!&J !"82 4(8!& 1&!&J

%&'()

(CM5/ !"82 4(8!& 1&!&J

!"#$*!+ +

4(8$9

(CM5/ $30- $30N )?O $!002 .<8-K 3P .<8- .<8-)# 4(8!& 4(8$9 1$!*/ 1&!&J

Model elements Influence sets (Interaction map)

PI Meeting, April 2011

Influence sets

PI Meeting, April 2011

10

Element Influence set Element Influence set

PI3K TCR, CD28, IL-2, IL-2R AP-1 Fos, Jun Akt PDK1, mTORC2 ERK Ras mTORC1 Rheb, PKC-θ JNK Ras mTORC2 PI3K, S6K1 Fos ERK Foxp3 NFAT, AP-1, STAT5, Smad3 Jun JNK IL-2 NFAT, AP-1, NFκB, Foxp3 NFAT Ca CD25 NFAT, AP-1, NFκB, STAT5, Foxp3 Ca TCR STAT5 IL-2, IL-2R PDK1 PIP3 NFκB PKC-θ, Akt TSC1-TSC2 Akt Smad3 TGFβ, Akt, mTORC1 Rheb TSC1-TSC2 PIP3 PI3K, PTEN S6K1 mTORC1 Ras TCR, CD28, IL-2, IL-2R pS6 S6K1 Model elements Influence sets (Interaction map)

Circuit design: Variables

 Number of values for variables

 Example: three levels for modeling TCR necessary  No antigen  Low antigen dose  High antigen dose

Model elements Influence sets (Interaction map) Set of discrete values for each element

11

PI Meeting, April 2011

Circuit design: Variables

 Number of values for variables

 Example: three levels for modeling TCR necessary  No antigen (TCR_LOW = 0, TCR_HIGH = 0)  Low antigen dose (TCR_LOW = 1, TCR_HIGH = 0)  High antigen dose (TCR_LOW = 0, TCR_HIGH = 1)  encoded with two Boolean variables

Model elements Influence sets (Interaction map) Set of discrete values for each element

12

PI Meeting, April 2011

Circuit design: Variables

 Number of values for variables

 Example: three levels for modeling TCR necessary  No antigen (TCR_LOW = 0, TCR_HIGH = 0)  Low antigen dose (TCR_LOW = 1, TCR_HIGH = 0)  High antigen dose (TCR_LOW = 0, TCR_HIGH = 1)  encoded with two Boolean variables  Example: three levels for modeling PI3K necessary  Low and high level of PI3K have different impact

• n mTORC2

Model elements Influence sets (Interaction map) Set of discrete values for each element

13

PI Meeting, April 2011

Low Antigen Dose Trajectory

Trajectory Summary

!"#

$%# &'(! &'(!) #*+, #*-+ &#) " ! # "./0 1,02 "&34 1$!*/ 51, !67 "*02 &1#28&1#- )%39 :&;)#2 :&;)#- "."/ .<8-) =!0/ 7>?@ABCD?7EC@ ?D7EF?7EC@ E@GEHE7EC@ 4(8!& 4(8$9 !"82 )?A 3)0 (CA =I@ =40 1&!&J "0#8% K#?-L

!"82 4(8!& 4(8$9

!"#$

1$!*/ 4(8!& 1&!&J !"82 4(8!& 1&!&J

%&'()

(CM5/ !"82 4(8!& 1&!&J

!"#$*!+ +

4(8$9

(CM5/ $30- $30N )?O $!002 ;4K8KGEPGKBQFQB ;4K8KBCRKBQFQB ;(( .<8-K 3S 4(8!& 4(8$9 1$!*/ 1&!&J

TCR PI3K PTEN PIP3 AKT MTORC1 S6K1 MTORC2 STAT5 IL-2 CD25 FOXP3

value = ON_HIGH value = ON_ LOW value = OFF

14

PI Meeting, April 2011

High Antigen Dose Trajectory

Trajectory Summary

value = ON_HIGH value = ON_ LOW value = OFF

!"#

$%# &'(! &'(!) #*+, #*-+ &#) " ./0-1 23 ! # ".45 6,57 "&28 6$!*4 96, !:; "*57 &6#70&6#- )%2< =&>)#7 =&>)#- "."4 ./0-) ?!54 ;@ABCDEFA;GEB AF;GHA;GEB GBIGJG;GEB 8(0!& 8(0$< !"07 )AC 2)5 (EC ?KB ?85 6&!&L "5#0% 1#A-M

!"07 8(0!& 8(0$<

!"#$

6$!*4 8(0!& 6&!&L !"07 8(0!& 6&!&L

%&'()

(EN94 !"07 8(0!& 6&!&L

!"#$*!+ +

8(0$<

(EN94 $25- $25O )AP $!557 >8101IGQI1DRHRD >8101DES1DRHRD >(( 8(0!& 8(0$< 6$!*4 6&!&L

TCR PI3K PTEN PIP3 AKT MTORC1 S6K1 MTORC2 STAT5 IL-2 CD25 FOXP3

15

PI Meeting, April 2011

Circuit design: Influence tables

PI Meeting, April 2011

16

PI3K S6K1 1 2 1 1 1 1

Example 2: 3-level PI3K, 2-level mTORC2 Example 3: 3-level Foxp3

Rheb PKC-Θ 1 1 1 1

Example 1: 2-level mTORC1

STAT5,mTOR NFAT, Smad3 00 01 02 10 11 12 20 21 22 00 1 2 1 2 01 1 1 or 0 1 02 10 1 2 1 2 2 1 or 2 2 2 11 1 1 1 0 or 1 1 1 12 1 or 0 1 20 1 2 2 2 2 2 2 2 2 21 1 1 1 1 1 1 1 2 22 1 1

*: Rule 1, *: Rule 2

Model elements Influence sets (Interaction map) Set of discrete values for each element Influence tables Model rules

mTORC1’ = Rheb and PKC-θ ‘and’ rule means both are necessary for activation mTORC1’ = Rheb mTORC1’ = Rheb or PKC-θ ‘or’ rule means either one is sufficient for activation

Rheb PKC-Θ 1 1 1

Example 1: 2-level mTORC1

Rheb PKC-Θ 1 1 1 1

17

PI Meeting, April 2011

Rheb PKC-Θ 1 1 1 1 1

Model elements Influence sets (Interaction map) Set of discrete values for each element Influence tables Model rules

mTORC1’ = Rheb and PKC-θ ‘and’ rule means both are necessary for activation mTORC1’ = Rheb mTORC1’ = Rheb or PKC-θ ‘or’ rule means either one is sufficient for activation

Rheb PKC-Θ 1 1 1

Example 1: 2-level mTORC1

Rheb PKC-Θ 1 1 1 1

18

PI Meeting, April 2011

Rheb PKC-Θ 1 1 1 1 1

Rheb is the activator, PKC-Θ

• nly strengthens the signal

Model elements Influence sets (Interaction map) Set of discrete values for each element Influence tables Model rules

mTORC1’ = Rheb and PKC-θ ‘and’ rule means both are necessary for activation mTORC1’ = Rheb mTORC1’ = Rheb or PKC-θ ‘or’ rule means either one is sufficient for activation

Rheb PKC-Θ 1 1 1

Example 1: 2-level mTORC1

Rheb PKC-Θ 1 1 1 1

19

PI Meeting, April 2011

Rheb PKC-Θ 1 1 1 1 1

Rheb is the activator, PKC-Θ

• nly strengthens the signal

Model elements Influence sets (Interaction map) Set of discrete values for each element Influence tables Model rules

mTORC1’ = Rheb and PKC-θ ‘and’ rule means both are necessary for activation mTORC1’ = Rheb mTORC1’ = Rheb or PKC-θ ‘or’ rule means either one is sufficient for activation

Rheb PKC-Θ 1 1 1

Example 1: 2-level mTORC1

Rheb PKC-Θ 1 1 1 1

20

PI Meeting, April 2011

Rheb PKC-Θ 1 1 1 1 1

Rheb is the activator, PKC-Θ

• nly strengthens the signal

CASE I: include this rule in the model CASE II: increase the number of values to represent mTORC1

Model elements Influence sets (Interaction map) Set of discrete values for each element Influence tables Model rules

Example 2: 3-level PI3K, 2-level mTORC2

PI Meeting, April 2011

21

PI3K S6K1 1 2 1 1 1 1 PI3K_HIGH PI3K_LOW 1 1 1 1 X PI3K_HIGH PI3K_LOW 1 1 1 X S6K1 = 0 S6K1 = 1

mTORC2’ = PI3K_HIGH or (PI3K_LOW and not S6K1)

Example 3: 3-level Foxp3

PI Meeting, April 2011

22

STAT5,mTOR NFAT, Smad3 00 01 02 10 11 12 20 21 22 00 1 2 1 2 01 1 1 or 0 1 02 10 1 2 1 2 2 1 or 2 2 2 11 1 1 1 0 or 1 1 1 12 1 or 0 1 20 1 2 2 2 2 2 2 2 2 21 1 1 1 1 1 1 1 2 22 1 1

FOXP3_HIGH’ = (STAT5_LOW and AP1NFAT_HIGH and not MTORC1_HIGH and not MTORC1_LOW) or (STAT5_HIGH and AP1NFAT_HIGH and not MTORC1_HIGH and not MTORC1_LOW) or (STAT5_LOW and AP1NFAT_LOW and SMAD3_LOW and not MTORC1_HIGH and not MTORC1_LOW) or (AP1NFAT_HIGH and SMAD3_LOW and not MTORC1_HIGH and not MTORC1_LOW) or (AP1NFAT_HIGH and SMAD3_HIGH and not MTORC1_HIGH and not MTORC1_LOW) or (STAT5_LOW and SMAD3_HIGH and not SMAD3_LOW and not MTORC1_HIGH and not MTORC1_LOW) or (STAT5_HIGH and SMAD3_HIGH and not SMAD3_LOW and not MTORC1_HIGH and not MTORC1_LOW) or (AP1NFAT_LOW and SMAD3_HIGH and not MTORC1_HIGH and not MTORC1_LOW) or (STAT5_HIGH and not STAT5_LOW and AP1NFAT_HIGH and not AP1NFAT_LOW and SMAD3_HIGH and not SMAD3_LOW and MTORC1_LOW) or (STAT5_HIGH and AP1NFAT_LOW and SMAD3_LOW and not MTORC1_HIGH and not MTORC1_LOW) Model elements Influence sets (Interaction map) Set of discrete values for each element Influence tables Model rules

Circuit Model of T Cell Differentiation

!"#

$%# &'(! &'(!) #*+, #*-+ " ! # ./0-)

"&12 ".345 %.'% "."3 ! 6 7 8&9)#: &;# ;,4: 8&9)#- )<=> ?;, ;&!&@ ;$!*3 !"0: 2(0$A ./0- ./0-)# (BC"3 !2* 9) .2D1)&1) .2D1)&1*E D!/F1 AF((1) 2(!& &#)5/9G ./0-E1H &#)5%.'% ".345 /9G "."35 /9G "."35 %.'% I!43 #J )JK $!44: )JL $14- $14M 1)4 I24 (BK INO 8&9) "4#% &#)

• Computable

model of cell dynamics

• Two simulation

modes  Synchronous

• Variables

updated simultaneously

• Deterministic

 Asynchronous

• Variables

updated one at a time in random order

• Stochastic

23

PI Meeting, April 2011

Four scenarios

 High antigen dose  Low antigen dose

PI Meeting, April 2011

24

Experiments Simulations

Four scenarios

 High antigen dose  Low antigen dose

PI Meeting, April 2011

25

Experiments Simulations

Model recapitulates experimental results

Four scenarios

 High antigen dose  Low antigen dose

PI Meeting, April 2011

26

Experiments Simulations Prediction(?) - currently tested with experiments

Four scenarios

 High antigen dose  Low antigen dose  High antigen dose, then removed

PI Meeting, April 2011

27

Experiments Simulations

Remove TCR after 18 hrs

Foxp3 CFSE

Four scenarios

 High antigen dose  Low antigen dose  High antigen dose, then removed  High antigen dose, then inhibitors added

PI Meeting, April 2011

28

Experiments Simulations

Remove TCR after 18 hrs

Foxp3 CFSE

Four scenarios

 High antigen dose  Low antigen dose  High antigen dose, then removed  High antigen dose, then inhibitors added

PI Meeting, April 2011

29

Experiments Simulations

Remove TCR after 18 hrs

Foxp3 CFSE

Model recapitulates experimental results

Analysis of Circuit Delays

 Model simulations point to the importance of timing in Foxp3 activation and fate selection

30

PI Meeting, April 2011

Analysis of Circuit Delays

Ras Raf MEKK1 MEK2 MEK4 ERK JNK Fos Jun Ca2+ PKC- NFAT NFB d1 IL-2 IL-2R JAK3 STAT5 PI3K PIP3 PDK1 Akt TSC1-TSC2 Rheb mTORC1 Foxp3 AP-1 TCR CD28 IL-2 Foxp3

d2

AND INVERTER INVERTED VALUE BUFFER MEDIUM case SHORT case

d3

Race determines whether Foxp3 will be expressed with high dose stimulation

31

PI Meeting, April 2011

Analysis of Circuit Delays

Ras Raf MEKK1 MEK2 MEK4 ERK JNK Fos Jun Ca2+ PKC- NFAT NFB d1 IL-2 IL-2R JAK3 STAT5 PI3K PIP3 PDK1 Akt TSC1-TSC2 Rheb mTORC1 Foxp3 AP-1 TCR CD28 IL-2 Foxp3

d2

AND INVERTER INVERTED VALUE BUFFER MEDIUM case SHORT case

d3

Shorter delay High dose stimulation

32

PI Meeting, April 2011

Analysis of Circuit Delays

Ras Raf MEKK1 MEK2 MEK4 ERK JNK Fos Jun Ca2+ PKC- NFAT NFB d1 IL-2 IL-2R JAK3 STAT5 PI3K PIP3 PDK1 Akt TSC1-TSC2 Rheb mTORC1 Foxp3 AP-1 TCR CD28 IL-2 Foxp3

d2

AND INVERTER INVERTED VALUE BUFFER MEDIUM case SHORT case

d3

Shorter delay High dose stimulation Low dose stimulation

33

PI Meeting, April 2011

Further system studies

 Low antigen those query:

Does IL-2 always go to 1? Property: F[20] (IL2 == 1) Test: BEST 0.001 0.999 Result: estimated probability close to 1

PI Meeting, April 2011

34

Statistical model checker

Model Simulations Trace statistics More traces? New model New experiments

Further system studies

 Low antigen those query:

 Probability that IL-2 stays at 0 before Foxp3 becomes 1? Property: (IL2 == 0) U[15] (FOXP3 == 1) Test: BEST 0.0001 0.999 Result: estimated probability = 0.00147– rare event

PI Meeting, April 2011

35

Statistical model checker

Model Simulations Trace statistics More traces? New model New experiments

Further system studies

 More queries:

 High antigen dose: Probability of STAT5 being activated before mTORC2  Low antigen dose: Number of steps IL2 stays active before Foxp3 activation  Antigen removal: Probability of initial CD25 oscillations Probability of PTEN activation Probability of initial PTEN and Foxp3 oscillation

PI Meeting, April 2011

36

Further system studies

 Next step: Multi-valued model

 Studying simulation results complex and time consuming  Many interesting properties to test, for example: Effects of different stimulation vs. co-stimulation levels Effects of PKC-Θ on mTORC1 Dumped oscillations in negative mTORC1/mTORC2 loop

PI Meeting, April 2011

37

Conclusions

 Logical modeling approach allows development of comprehensive models of cell fate

 Model of peripheral T cell differentiation recapitulates a wide range of experimental observations  Circuit analysis reveals key elements of the mechanism for Foxp3 expression  Timing of STAT5 vs. mTOR  Critical role of PTEN  Negative feedback between mTORC1 and mTORC2

 Logical modeling + Statistical model checking

 Gain further insights about the systems

38

PI Meeting, April 2011

Thank you!

PI Meeting, April 2011

39