Maxwell's Demon in Biochemical Signal Transduction

Transcription

1 Maxwell's Demon in Biochemical Signal Transduction Takahiro Sagawa Department of Applied Physics, University of Tokyo New Frontiers in Non-equilibrium Physics July 2015, YITP, Kyoto

2 Collaborators on Information Thermodynamics Masahito Ueda (Univ. Tokyo) Shoichi Toyabe (Tohoku Univ.) Eiro Muneyuki (Chuo Univ.) Masaki Sano (Univ. Tokyo) Sosuke Ito (Titech) Naoto Shiraishi (Univ. Tokyo) Sang Wook Kim (Pusan National Univ.) Jung Jun Park (National Univ. Singapore) Kang-Hwan Kim (KAIST) Simone De Liberato (Univ. Paris VII) Juan M. R. Parrondo (Univ. Madrid) Jordan M. Horowitz (MIT) Jukka Pekola (Aalto Univ.) Jonne Koski (Aalto Univ.) Ville Maisi (Aalto Univ.)

3 Outline Introduction Information and entropy Review of previous results Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary Today s main part!

4 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

5 Thermodynamics in the Fluctuating World Thermodynamics of small systems with large heat bath(s) Thermodynamic quantities are fluctuating! Second law W F Nonlinear & nonequilibrium relations

6 Information Thermodynamics Information System Demon Feedback Information processing at the level of thermal fluctuations Foundation of the second law of thermodynamics Application to nanomachines and nanodevices Review: J. M. R. Parrondo, J. M. Horowitz, & T. Sagawa, Nature Physics 11, (2015).

7 L. Szilard, Z. Phys. 53, 840 (1929) Szilard Engine (1929) Initial State Partition Which? Heat bath T ktln2 B Work Isothermal, quasi-static expansion Feedback Measurement Left Right ln 2 Free energy: Increase F E TS Decrease by feedback Can control physical entropy by using information

8 Information Heat Engine System (Working engine) Information Memory (Controller) Feedback Free energy / work Entropic cost Can increase the system s free energy even if there is no energy flow between the system and the controller

9 Experimental Realizations With a colloidal particle Toyabe, TS, Ueda, Muneyuki, & Sano, Nature Physics (2010) Efficiency: 30% Validation of ( W F) e With a single electron Koski, Maisi, TS, & Pekola, PRL (2014) Efficiency: 75% Validation of ( W F ) I e 1

10 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

11 Shannon Information Information content with event : Average k ln 1 p k Shannon information: H k p k ln 1 p k

12 Mutual Information System S Measurement with stochastic errors Memory M (measurement device) System S I Memory M I( S : M) H( S) H( M) H( SM ) 0 I H( M) Correlation between S and M No information No error I ln 2 ln (1 )ln(1 ) Ex. Binary symmetric channel I

13 Entropy Production Stochastic dynamics of system S (e.g., Langevin system) Work W System S Heat bath B (inverse temperature β) Heat Q Entropy production in the total system: S SB S S Q Change in the Shannon entropy of S Averaged heat absorbed by S If the initial and the final states are canonical distributions: S W F SB Free-energy difference

14 Stochastic Entropy Production Work W System (phase-space point x) Heat bath (inverse temperature β) Q Stochastic entropy production along a trajectory of the system from time 0 to τ s SB s S Q s S ss[ x( ), ] ss[ x(0),0] s S [ x, t] ln P[ x, t] s S S S P [ x, t] : probability distribution at time t If the initial and the final states are canonical distributions: s W F SB

15 Integral Fluctuation Theorem and Jarzynski Equality Integral fluctuation theorem e s SB 1 Seifert, PRL (2005), for any initial and final distributions Second law can be expressed by an equality with full cumulants The second law of thermodynamics (Clausius inequality) s SB 0 S S Q Jarzynski equality Jarzynski, PRL (1997) s SB W F e W e F W F

16 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

17 Setup Heat bath B System X System Y Heat bath B Time evolution of X under the influence of Y with initial and final correlations

18 Special Cases: Measurement and Feedback Measurement X: Memory Y: Engine Feedback X: Engine Y: Memory

19 Stochastic Entropy and Mutual Information Entropy increase in XB s XB s X Q X Initial correlation I i xy ln P[ x, y] P[ x] P[ y] I i xy dxdyp[ x, y]ln P[ x, y] P[ x] P[ y] Final correlation I f x' y ln P[ x', y] P[ x'] P[ y] I f x' y dx' dyp[ x', y]ln P[ x', y] P[ x'] P[ y]

20 Decomposition of Entropy Production Total entropy production in XYB s XYB s XB I s s I XYB XB s s f I x ' y I X XY i xy Q Q X X s XY s s X X s I Y I

21 Fluctuation Theorem Integral fluctuation theorem s XYB s XB I e s XYB 1 e s I XB 1 Second law s XYB 0 s I XB TS and M. Ueda, Phys. Rev. Lett. 109, (2012).

22 Special Case 1: Feedback Control X: Engine Y: Memory Feedback: Control protocol depends on the measurement outcome I I rem I e s XB ( I rem I ) 1 s I XB I rem W ext F k B T I F : equilibrium free energy

23 Special Case 2: Measurement X: Memory Y: Engine I I e s XB I 1 s XB I

24 Minimal Energy Cost for Measurement Memory X Engine Y Heat bath B Energy change E X Heat bath B Work WX s XB I W X E X s X I Change in the nonequilibrium free energy of only X Additional energy cost to obtain information TS and M. Ueda, Phys. Rev. Lett. 109, (2012). Information is not free

25 General Principle of Information Thermodynamics e s I XB 1 s XB I Feedback: Measurement: e s XB ( I rem I ) 1 e s XB I 1 s I XB I rem s XB I Unified formulation of measurement and feedback TS and M. Ueda, PRL 109, (2012).

26 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

27 Problem Heat bath System Memory Heat bath What compensates for the entropy decrease here? For Szilard engine, ssb ln 2

28 Conventional Arguments Measurement process Brillouin Erasure process (From Landauer principle) Bennett & Landauer Widely accepted since 1980 s

29 Total Entropy Production Total entropy production in XYB s XYB s s XY XB I Q X s XB I s XYB 0 Equality: thermodynamically reversible If the mutual information is taken into account, the total entropy production is always nonnegative for each process of measurement or feedback.

30 Revisit the Problem Heat bath System Memory Heat bath What compensates for the entropy decrease here? Mutual-information change compensates for it. For Szilard engine, ssb ln 2 s SMB s SB I ln 2 ln 2 0

31 Key to Resoluve the Paradox Maxwell s demon is consistent with the second law for measurement and feedback processes individually The mutual information is the key We don t need the Landauer principle to understanding the consistency

32 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

33 Thermodynamics of Autonomous Information Processing Second law & fluctuation theorem Allahverdyan, Dominik & Guenter, J. Stat. Mech. (2009) Hartich, Barato, & Seifert, J. Stat. Mech. (2014) Horowitz & Esposito, Phys. Rev. X (2014) Horowitz & Sandberg, New J. Phys. (2014) Shiraishi & Sagawa, Phys. Rev. E (2015) Ito & Sagawa, Phys. Rev. Lett. (2013) Models of autonomous Maxwell s demons Mandal & Jarzynski, PNAS (2012) Mandal, Quan, & Jarzynski, Phys. Rev. Lett. (2013) Strasberg, Schaller, Brandes, & Esposito Phys. Rev. Lett. (2013) Horowitz, Sagawa, & Parrondo, Phys. Rev. Lett. (2013) Shiraishi, Ito, Kawaguchi & Sagawa, New J. Phys. (2015) Toward deeper understanding of information nanomachines

34 Two Approaches Transfer entropy approach Applicable to non-markovian dynamics Second law is weaker in Markovian dynamics But we derived a stronger version! (Poster by Ito) Information flow approach Ito & Sagawa, Phys. Rev. Lett. (2013) Not applicable to non-markovian dynamics Second law is stronger in Markovian dynamics Second law: Allahverdyan, Dominik & Guenter, J. Stat. Mech. (2009) Hartich, Barato, & Seifert, J. Stat. Mech. (2014) Horowitz & Esposito, Phys. Rev. X (2014) Horowitz & Sandberg, New J. Phys. (2014) Fluctuation theorem: Shiraishi & Sagawa, PRE (2015) Poster by Shiraishi

35 Transfer Entropy ), ( 0 X 0 Y ), ( 1 1 n X n Y ), ( n X n Y Time Transfer entropy: Directional information flow from X to Y during time n and n+1 n y n y x n n n n n n n n n y y y p y y x p y y y x p y y x p,, ), ( ), ( ),, ( )ln,, ( : ) ( Y Y Y X I Y X T n n n n n Conditional mutual information T. Schreiber, PRL 85, 461 (2000) Directional information transfer between two systems

36 Many-body Systems with Complex Information Flow Time Node: Event Arrow: Causal relationship Characterize the dynamics by Bayesian networks Sosuke Ito & TS, PRL 111, (2013).

37 Second Law on Bayesian Networks Time S XB I fin I ini l I l tr S. Ito & T. Sagawa, PRL 111, (2013) S XB : Entropy production in X and the bath I ini : Initial mutual information between X and other systems I fin : Final mutual information between X and other systems I l tr : Transfer entropy from X to other systems

38 Information Flow VS Transfer Entropy Infinitesimal transition of coupled Langevin system x' x( t dt) y' y( t dt) x ( t) f ( x( t), y( t)) ( t) x y ( t) g( x( t), y( t)) ( t) y ( t) ( t) x y 0 : independent noise x x(t) y y(t) Stronger: s( x') s( x) Q I( x': y) I( x : y) Information flow Weaker: s( x') s( x) Q I( x': y') I( x : y) I( x : y' y) s( x' y') s( x y) Q I( x : y' y) Transfer entropy

39 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

40 Toward Biological Information Processing What is the role of information in living systems? Mutual information is experimentally accessible ex. Apoptosis path: Cheong et al. Science (2011). There is no explicit channel coding inside living cells; Shannon s second theorem is not straightforwardly applicable Application of information thermodynamics Barato, Hartich & Seifert, New J. Phys. 16, (2014). Sartori, Granger, Lee & Horowitz, PLoS Compt. Biol. 10, e (2014). Ito & Sagawa, Nat. Commu. 6, 7498 (2015). Our finding: Relationship between information and the robustness of adaptation

41 Signal Transduction of E. Coli Chemotaxis E. Coli moves toward food (ligand) The information about ligand density is transferred to the methylation level of the receptor, and used for the feedback to the kinase activity.

42 Adaptation Dynamics 2D Langevin model Y. Tu et al., Proc. Natl. Acad. Sci. USA 105, (2008). F. Tostevin and P. R. ten Wolde, Phys. Rev. Lett. 102, (2009). F. G. Lan et al., Nature Physics 8, 422 (2012). : kinase activity : methylation level : average ligand density : time constants : stationary value of at Negative feedback loop: Instantaneous change of a t in response to l t Memorize l t by m t a t goes back to the initial value

43 Second Law of Information Thermodynamics (Weaker version with transfer entropy) : Transfer entropy : Change in the conditional Shannon entropy : Robustness against the environmental noise Upper bound of the robustness is given by the transfer entropy S. Ito & T. Sagawa, Nature Communications 6, 7498 (2015).

44 Stationary State Fluctuation (inaccuracy of information transmission) induced by environmental noise Transfer entropy Without feedback:

45 Exact Expression of Transfer Entropy If the Langevin equation is linear: Signal-to-noise ratio : power of the signal from a to m : noise of m Analogous to the Shannon Hartley theorem

46 Information-Thermodynamic Efficiency Input ligand signal: a, step function. b, sinusoidal function. c, linear function. Numerical simulation: Red: robustness of adaptation Green: information-thermodynamic bound Blue: conventional thermodynamic bound Information thermodynamics gives a stronger bound! The adaptation dynamics is inefficient (dissipative) as a conventional thermodynamic engine, but efficient as an information-thermodynamic engine.

47 Information-Thermodynamic Figure of Merit

48 Comparison with Shannon s Information Theory Second law of information thermodynamics Information di tr t gives the bound of robustness J a t Well-defined in living cells Shannon s second theorem Information (channel capacity) C gives the bound of achievable rate R How to define in living cells??

49 Outline Introduction Information and entropy Information thermodynamics: a general framework Paradox of Maxwell s demon Thermodynamics of autonomous information processing Application to biochemical signal transduction Summary

50 Summary Unified framework of information thermodynamics T. Sagawa & M. Ueda, Phys. Rev. Lett. 109, (2012). T. Sagawa & M. Ueda, New J. Phys. 15, (2013). Fluctuation theorem for autonomous information processing S. Ito & T. Sagawa, Phys. Rev. Lett. 111, (2013). Review: S. Ito & T. Sagawa, arxiv: (2015). N. Shiraishi & T. Sagawa, Phys. Rev. E 91, (2015). Information thermodynamics of biochemical signal transduction S. Ito & T. Sagawa, Nature Communications 6, 7498 (2015). Review: J. M. R. Parrondo, J. M. Horowitz, & T. Sagawa, Nature Physics 11, (2015). Thank you for your attention!