Organization Becomes Cause in the Matter
Nature Biotechnology 18, 575 - 576 (2000)
The commentary last month from Huang (1) and two recent reviews from physicists Laughlin (2) and Pines and colleagues (3) offer a view of how experimental biology might address the problems of complexity that have been revealed (but as yet unsolved) by 50 years of otherwise unprecedented progress in molecular biology.
In physics, this view is now referred to as "the middle way," or the search for laws operating at levels and scales of organization (mesoscopic realms) intermediate between the microscopic state of fundamental particles and the macroscopic state of higher levels of organization (2). According to Laughlin(2)and Pines and coworkers(3), the mesoscopic world is abstracted away from our concept of complex matter by a reductionist view that attempts to derive (unsuccessfully) macroscopic behavior in the world of everyday life from a consideration merely of the microscopic details obeying a "theory of everything"(2).
Similarly, in biology, molecular genetic reductionism has mostly distracted us from study of mesoscopic realms between genotype and phenotype where complex organizational states exist and where, as we now realize, there also exist networks of regulatory proteins capable of reorganizing patterns of gene expression, and much other "emergent" cellular behavior, in a context-dependent manner. It is the mesoscopic organization of matter (living or dead) that harbors as yet undiscovered principles lying behind emergent features. As such, this view is a more recent expression of the ideas of Anderson (4) and Polanyi (5) expressed 30 years ago that relate, respectively, with hierarchical organization of matter in physical science and the irreducibility of emergent behavior in biology.
In the newly formulated "middle way," one looks for rules of self organization that "...at macroscopic-length scales were not self-evident at the time of their discovery and were accepted as true only after repeated confrontations with experiment left no alternative." We call the search for the existence of mesoscopic (quantum) protectorates—the proof or disproof of organizing principles appropriate to the mesoscopic domain—the middle way (2, 3). In this search, one tries to find useful approximations to how things work. Such approximations have led to major insights in physics in the past several decades and are not first-principle deductions (from theory and equations), but are derived through "art keyed to experiment" (2, 3).
Huang(1) provides a number of instances where we might arrive at these approximations in biology, and I want to expand on two of these because they seem to me to emphasize similarities and important differences in how one approaches the analysis of biological complexity.
Chemotactic behavior and the control of metabolism are two examples of complex phenotypes that have been illuminated by many years of reductionist work, revealing complicated networks or pathways of signaling and other proteins that control their robust behaviors. Recent attempts have been made to analyze both of these phenotypes using a "systems approach," in which one seeks the controlling mesoscopic rules of a "network as a whole" rather than looking for controls that might be vested in single or even additive effects of several key genes or their "rate-limiting" enzymes.
The initial questions to be asked are why these behaviors are so robust and why they exhibit such invariant behavior over a wide variation in internal and external parameters. The answer in both cases is that robust behavior is not defined by particular proteins, but that control is distributed over the entire network. The next question is, what are the rules of the network that describe a distributed control where, when any one parameter is varied over a considerable range, the system produces an invariant phenotype.
In the case of chemotaxis, present-day researchers have turned to analysis of the network in terms of feedback, circuit diagrams, modeling, and attempts to capture the behavior with mathematical equations(6, 7). Not much attention is paid to the fundamental chemistry of the molecules involved, other than to know how the phenotype is altered when the relative abundance of individual proteins is changed through genetic engineering. Ignoring the contribution from existing laws of chemistry might be in accord with an outdated view in biology that "we already know what we need to know from chemistry," but it would also be a grave error, as demonstrated by the approach and results garnered by those studying metabolic control8.
In the case of metabolism, the rules describing complex, context-dependent outcomes come directly out of chemistry constrained by thermodynamics and kinetics. Using these classical rules together with labor-intensive measurements of kinetic and thermodynamic changes, one makes approximations of changes in flux through metabolic pathways characterized by high connectivity. One finds that a change in any one component (increase or decrease in activity of a pathway enzyme) results in changes in all other components so that the overall behavior remains constant or robust (distributed control)(8). Of equal interest, it is found that each enzyme activity changes by a small amount in the same direction, so that any changes in metabolic flux (phenotype) will be brought about by many small changes acting in concert. Importantly, distributed control is known to predict phenotype in mammalian as well as in bacterial systems as a function of changes made in environmental input.
For example, Veech and colleagues (9, 10) were able to provide the necessary data set to unambiguously prove that the general principle of distributed control applies to glycolysis in mammalian tissue. To accomplish this they had to obtain concentrations of all intermediates involved in glycolytic flux in a perfused rat heart, as well as overall metabolic flux and work output. In addition, the kinetic and thermodynamic parameters of all enzymes in the pathway were determined. In this single, superb, reductionistic study, requiring four years, they concluded that control points were context dependent; "...control of the metabolic flux in glucose metabolism of rat heart is ... variously distributed among enzymes depending on substrate availability, hormonal stimulation, or other changes of conditions."
These conclusions help explain the robust nature of metabolic pathways—for example, the inability of overproduction of any one glycolytic enzyme in yeast to alter glycolytic flux— and help illuminate the concept of "network architecture." In addition, this kind of research appears to have much to offer in the way of improving disease phenotypes, not through controlling individual genes or molecules but through understanding the deeper aspects of bioenergetics and the subtle nature of metabolic control—both of which have been deemphasized within the genetic paradigm (11). The metabolic predictions remain approximations and are limited because branchpoints within discrete pathways present special problems. In addition, every metabolic or signaling pathway seems to be interconnected with all others, a view supported by experiments that show changes in activity in distant pathways when any one enzyme in another pathway is altered. Still, there is much to learn and many benefits from approximations (2, 3).
For understanding and predicting other intracellular physiological phenotypes, evidence also indicates that the principles of metabolic control analysis may also be useful. Thus, constraints arising from thermodynamics and energetics (e.g., chemical potentials of NAD) may be useful in understanding DNA stability through gene silencing controlled partly by NAD "sensing" proteins (12), and it is also being made clear that the mathematical treatments used in metabolic control analysis may be applicable to a variety of other networks, including those involved in DNA supercoiling (13), and to new ways of looking at cancer as a pathology related to genomic instability (14).
Although analysis of metabolism may have far-reaching impact, even in the absence of new complexity rules, at still higher levels of cell−cell interaction, differentiation and histogenesis of tissues, complex disease etiology, and beyond, new rules appear to be needed. Certainly at the level of populations of organisms and their relationship to the environment and to evolution, clear evidence exists for new and useful applications of complexity science: for mathematical modeling of genetic interactions (15), for self-similarity in the analysis of biodiversity (16), and for emphasis on "attractor" states rather than on stable equilibria for the analysis of species diversity (17, 18)—to name a few. In medicine, there are already many reviews of the use of complexity thinking, broadly defined, to predict and diagnose human diseases. Of course, embryology has for many years been an obvious (but unfunded) candidate for application of complex systems analysis (19, 20, 21) and one must expect that the number of different mesoscopic levels between genotype and phenotype in living beings will be high and that each will be defined by different and irreducibly complementary rules (5, 22). But whether it is classical bioenergetics or the new complexity, the principle of organization as "cause in the matter" emerges as a dominant theme.
1. Huang, S. Nat. Biotechnol. 18, 471–472 (2000)
2. Laughlin, R.B. & Pines, D. Proc. Natl. Acad. Sci. USA 97, 28–31 (2000)
3. Laughlin, R.B. et al. Proc. Natl. Acad. Sci. USA 97, 32–37 (2000).
4. Anderson, P.W. Science 177, 393–396 (1972).
5. Polanyi, M. Science 160, 1308–1312 (1968). |
6. Baraki, N. & Leibler, S. Nature 387, 913–917 (1997
7. Alon, U., Surette, M.G., Barkai, N. & Leibler, S. Nature 397, 168–171 (1999).
8.Fell, D. Understanding the control of metabolism. (Portland Press, London; 1997).
9. Kashiwaya, Y. et al. J. Biol. Chem. 269, 25502–25514 (1994).
10. Veech, R.L. & Fell, D.A. Cell Biochem. Funct. 14, 229–236 (1996).
11. Sato, K. et al. FASEB J. 9, 651–658 (1995).
12. Imai, S-i., Armstrong, G.M., Kaeberlein, M. & Guarente, L. Nature 403, 795–799 (2000)
13. Jensen, P.R. et al. Eur. J. Biochem. 266, 865–877 (1999).
14. Duesberg, P. et al. Anticancer Res. 19, 4887–4906 (2000).
15. Waxman, D. & Peck, J.R. Science 279, 1210–1213 (1998)
16. Hart, J., Kinzig, A. & Green, J. Science 284, 334–336 (1998).
17. Polis, G. Nature 395, 744–745 (1998).
18. McCann, K., Hastings, A. & Huxel, G.R. Nature 395, 794–798 (1998).
19. Goodwin, B.C., Murray, J.D. & Kauffman, S.A. J. Theoret. Biol. 163, 135–144 (1993).
20. Gordon, R. The hierarchial genome and differentiation waves, Vols 1, 2. (World Scientific Publishing, Singapore; 1999).
21. Gilbert, S., Opitz, J.M. & Raff, R.A. Dev. Biol. 173, 357–372 (1996).
22. Pattee, H. Appl. Math. Computat. 56, 111–130 (1993).
Richard Strohman is emeritus professor of the University of California at Berkeley, Department of Molecular and Cell Biology. He is a former research director of the American Muscular Dystrophy Association.