The physics of autonomous biological information

Published in Biological Theory, Summer 2006, Vol. 1, No. 3: 224–226. The Physics of Autonomous Biological Information H. H. Pattee 1611 Cold Spring Road Williamstown, MA 01267 [email protected] The general concept of information does not belong in the category of universal and inexorable physical laws but in the category of initial conditions and boundary conditions. Boundary conditions formed by local structures are often called constraints. Informational structures such as symbol vehicles are a special type of constraint. It should be clear that the concepts of initial conditions and constraints in physics make no sense outside the context of the law-based physical dynamics to which they apply. This is also the case for the concept of information. There are many different origins, functions, and hierarchical levels of informational constraints. Physicists require the type of information that begins by an observer choosing to perform a measurement. This passive information is necessary to establish the initial conditions of a chosen dynamical system at a particular time. The behavior of the system can then be derived by integration over time. Initial conditions include the positions of all the microscopic units (configuration), their masses and rates of change of position (momenta). Constraints are macroscopic structures that require additional information for their description. Biological information begins not by observer’s choice but by chance. Chance produces variation in the structures of potential informational constraints. Only by self-replication and natural selection do these constraints become functional information that controls the dynamics of chemical syntheses of the organism. This distinction between physical and biological information is essential because it is the undirected origin of biological information that is one of several necessary conditions for open-ended evolution. Other conditions were presented by von Neumann (1966) who first gave a plausible logical argument that information in the form of non-dynamic symbolic constraints (“quiescent” descriptions) must be distinguished from the dynamics they control in order to allow open-ended evolution, but he did not address the physical conditions necessary for symbolic control of physical systems. As a matter of practice, symbolic expressions do not appear to take place by physical necessity, nor do physical laws appear to restrict symbol sequences (e.g. Hoffmeyer and Emmeche, 1991). Because of this, some mathematicians and physicists believe in the reality of true Platonic symbol systems independent of physical laws. Nevertheless, it is the case that no symbol vehicle or symbolic operation can evade physical laws. This means that in spite of the apparent autonomy of biological information, physical laws impose several conditions on the material embodiment of the different forms of information. The most general relation between information and physical laws is the well-known statistical condition that any controlled reduction of entropy of a system by measured information must at some point be compensated by enough increase of entropy by energy dissipation to satisfy the 2nd law of thermodynamics. Specific forms of information have additional conditions. For efficient heritable storage of information, there must exist distinguishable but nearly equiprobable stable structures or states. Since the probabilities of physical states depend on energy, differing informational constraints should have stable but equivalent energetic configurations. In order to achieve high information capacity this energy degeneracy is often realized by one-dimensional discrete sequences as in copolymers and other types of symbol strings. Higher dimensional memories are also possible but require more complicated reading and writing processes than do linear sequences. In addition to energy degeneracy, permanent, long-term, and random access informational structures must be relatively time and rate independent, by definition. More precisely, the temporal events of such memory have no coherent relation to the dynamical time of the systems they control. Furthermore, the interpretation or meanings of symbolic memory structures, like genes, texts, logical and mathematical expressions, and computer programs have no essential rate-dependence as do physical laws. In other words, the rate of reading such information is largely arbitrary. For example, the same protein results from a given gene whether its synthesis is fast or slow. The same is the case for writing a mathematical proof or computing a function. In contrast, physical laws are fundamentally rate dependent. Consequently, in addition to von Neumann’s logical conditions for open-ended self-replication, informational constraints must also satisfy these physical conditions: (1) free variation in memory, (2) energy dissipation, (3) energy degeneracy, and (4) rate-independence. These conditions allow symbolic descriptions and physical constructions to form a nearly autonomous or self-controlled unit, a property called semiotic closure (Pattee, 2001; Rocha, 2001). Without this semiotic closure that includes the grounding of symbols in the material world, all our symbol systems, all languages, all mathematics and formal systems could appear to exist outside the physical world as if they were pure Platonic forms. Addition conditions have evolved for the different forms of biological information as in (1) a memory structure (storage), (2) the transmission and replication of this structure (communication), (3) the syntactic restatement of information (coding), (4) the activation of information (folding), and (5) the specific, local rate control of a dynamical system by the informational constraints (function or semantics) (cf. Shannon and Weaver, 1949, p. 96). Cells require different structures to realize these distinct informational processes, presently embodied by DNA, polymerases, messenger RNAs, transfer RNAs, coding enzymes, and ribosomes, although much simpler processes must have existed at the origin of life. Replication and communication of information require recognition of individual informational units that does not remove the energy degeneracy or the ordering of these units. In cells, the translation or decoding requires associating passive stable storage units, the base triplets, with the amino acids that will form the active control structures, the enzymes. It is only after this last syntactic step that the sequence degeneracy is removed by the folding of the linear string. Conceptually and physically, the folding process requires strong bonds to maintain the one-dimensional informational sequence while weaker bonds must largely determine the folded three-dimensional shape. Folding is the crucial process that removes the energy degeneracy of one-dimensional memory structures and allows the symbolic information to reenter the three-dimensional world of energy and rate dependent physical laws. To understand evolution we need much better understanding of the mapping from the informational sequence space to the physical function space (e.g., Schuster, 2002). At this folding process we lose any simple definition or measurement of information in the molecule because the non-dynamic informational constraints of the linear copolymer become inseparably incorporated into the physical dynamics. The detailed dynamics of folding is enormously complex because it involves the concurrent energetic interactions of many hundred constrained units (e.g., Wolynes, et al., 1995). Computer modeling of folding is still only approximate even on the largest computers. The initial function of genetic information is the control of the rates of synthesis of all the metabolic components, as well as the control of the rates of expression of the genetic information itself. Because control constraints are temporally incoherent the dynamics they control are generally not integrable over time in the usual sense. Most enzyme-catalyzed reactions increase rates so drastically that their description can be simplified by representing them as discrete switching events. The same is the case for the control of gene expression. This property adds to the appearance of autonomy of the informational constraints because the details of the dynamics become functionally less crucial (Pattee, 1974). This is fortunate because the detailed ab initio dynamics of enzyme-substrate binding and catalysis is also too complex for present computers. The extreme case of autonomy of informational constraints is the behavior of the programmed computer. A computer program makes no reference at all to the physical dynamics of the hardware because all events are approximated as switches and no significant synthesis or physical construction of components takes place. Something like an inverse process occurs in the brain’s production of speech and writing. Here an enormously complex concurrent dynamic network of neurons with both genetic and learned constraints produces relatively simple linear strings of discrete symbols that have little significant intrinsic dynamics or rate dependence. Until the production of the discrete symbolic expression, there is no simple concept or measure of the “information in the brain.” Unfortunately, two controversial schools of thought about symbols and dynamics have developed. This controversy has entered the fields of artificial intelligence, cognitive theory, artificial life, and even physics. One school emphasizes dynamical models; the other emphasizes symbolic computational models. The former argue that symbols are a derived concept reducible to dynamical systems. The latter argue that all events are informational and that continuous dynamics is itself a derived concept (e.g., Wheeler, 1990). Controversy over this difference obscures the central point: Life and evolvability require the complex interaction of rate-independent symbol constraints and rate-dependent physical dynamics. Like the physicist’s laws and initial conditions, the meaning of biological information cannot make sense if taken out of this inseparable context. References Hoffmeyer, J. and Emmeche, C., (1991) Code duality and the semiotics of nature. In On Semiotic Modeling, M. Anderson and F. Merrell, eds. Berlin: Mouton de Gruyter, pp.117-166. Pattee, H. H. (2001) The physics of symbols: bridging the epistemic cut, BioSystems, 60, 5-21. Pattee, H.H. (1974) Discrete and Continuous Processes in Computers and Brains. In The Physics and Mathematics of the Nervous System, M. Conrad, W. Guttinger & M. DalCin (eds.) New York: Springer, pp.128-148. Rocha, L. M., (2001) Evolution with material symbol systems. BioSystems, 60, 95-121. Schuster, P., (2002) A testable genotype-phenotype map: Modeling evolution of RNA molecules. In Biological Evolution and Statistical Physics, Lessig, M. and Valleriani, A. eds. Berlin: Springer-Verlag, pp. 56-83. Shannon, C. E. and Weaver, W., (1949) The Mathematical Theory of Communication. Urbana, IL: University of Illinois Press, p. 96. von Neumann, J., (1966) Theory of Self-reproducing Automata, A. W. Burks, ed. Urbana, IL: University of Illinois Press. Lecture 5, pp. 74-87. Wheeler, J. A., (1990) Information, physics, quantum: the search for links. In Complexity, Entropy, and the Physics of Information. Zurek, W. H. ed., Redwood City, CA: Addison-Wesley, pp. 3-28. Wolynes, P. G., Onuchic, J. N., and Thirumalai, D., (1995) Navigating the folding routes, Science, 267, 1619-1620. PAGE 1