INTRODUCTION

TOP ABSTRACT INTRODUCTION CONCLUSION REFERENCES

We will discuss four major issues--the first, a general point, is an obvious need for innovation and discovery in all areas of science. We will then explore the nature of simple versus complex systems and consider the limitations of reductionism as a way of analyzing complex phenomena. Living systems will be considered as a special case of complexity. In the next two sections, we will focus on issues that are specifically relevant to genomics, taking examples from our current work on the complex interrelated diseases of asthma and atopy. We will then explore an area of complexity theory developed by Kauffman and others.

	INNOVATION AND DISCOVERY IN SCIENCE

In any scientific endeavor, one can fall into the dangerous trap of being unquestionably sure of the "established facts." For example, in the 1970s, suppressor T cells were all the rage, and page after page of the Journal of Immunology was taken up with trying to characterize the properties of these cells. Subsequently, there turned out to be a flaw in this theory: the suppressor or J region of the T cells could not be found, and the suppressor T cell theory fell out of favor (1). This is an example of scientists becoming so confident in a particular idea that they ignore the exploration of other possibilities and, by so doing, become stuck in a theory that becomes difficult to prove.

In the field of physical chemistry, consider buckminsterfullerene, or "C60." This molecule was named after the discoverer of the geodesic dome because of its shape, a truncated icosohedron. Kroto, Curl, and Smalley (2), three physical chemists, announced their discovery in 1985, which led to their Nobel Prize in December 1996. In his book, The Most Beautiful Molecule: The Discovery of the Buckyball (3), Hugh Aldersey-Williams remarks, "Chemistry is supposed to be one of the more developed sciences. Within chemistry, carbon is the element about which we are supposed to know most. It is the element around which life is built." The discovery of the buckyball was surprising in view of the fact that the whole story of carbon's structure and propertiestwo known crystalline forms, diamond and graphitewas thought to be complete. But here was a totally new form, reddish-brown crystals. "The remarkable thing about the discovery of buckminsterfullerene, then, is not how clever we are to have found it, but how unobservant and unimaginative we have been not to have found it sooner. The discovery is a potent reminder that science is not, as it is sometimes portrayed, on the verge of reaching an absolute knowledge but that after three centuries of modern chemistry, we have only just begun the exploration. It highlights not how much we know, but how very little."

The physicist, Richard Feynman, a remarkable man, famous for his Feynman Diagrams (4) and for figuring out the problem that caused the Challenger shuttle disaster, is the origin of many thought-provoking quotes: "It will not do you any harm whatever to think in an original fashion. The odds that your theory will be in fact right, and that the general thing that everybody is working on will be wrong, is low. But the odds you . . . will be the guy who figures a thing out is not smaller. . . It's very important that we do not all follow the same fashion," as quoted in Genius, The Life and Science of Richard Feynman by James Gleick (5).

A final example of original thinking is the PCR, the polymerase chain reaction, worked out by Kary Mullis in 1985. It has revolutionized the way molecular biologists think about manipulating genes, resulting in a tremendous acceleration in the rate of gene discovery and genome sequencing (6). Mullis put together some basic ideas of computer theory, particularly regarding iterative processes, with the known action of DNA polymerase, which led to the PCR.

In citing each of these discoveries, the message that comes through loud and clear is "Be inquisitive; don't take anything for granted." As scientists, we are under a continuing obligation to try out new ideas, to formulate new and searching hypotheses that can be tested and retested. We must ask the right questions in their proper contexthow else can science advance?

	SIMPLE VERSUS COMPLEX SYSTEMS

Simple systems involve only a small number of components and very few interactions between the components. If one knows the inputs acting on the system and the environmental influences are negligible, one can usually predict the result. The assumption is that the world is composed of nested, deterministic systems that one can analyze in a step-wise fashion (7, 8); one can understand the whole by examining the parts. The long-favored approach of scientific reductionism encompasses much of the world of physics, but, in fact, reductionism does not work well except in the simple systems usually studied by physicists. Gödel, and later Turing, proved that, in any mathematical formalization, including the rules of ordinary arithmetic, there are statements that cannot be categorized as true or false (the Entscheidungsproblem) (9). This proof showed that reductionism does not work outside defined narrow limits. In general, to understand the behavior of the parts, you must understand the behavior of the whole. Therefore, the usefulness of reductionism is limited.

Over the last 100 years, the world-view of determinism that is, certainty of behaviorhas been upset by several theories, including Einstein's special and general theories of relativity, and Heisenberg's uncertainty principle. The work of Heisenberg, Schrödinger, Dirac, and others pointed to the nondeterministic nature of the universe.

The assumption had been that if one could only determine the initial conditions which describe a system accurately enough, one could then deduce its characteristics and its laws based on straightforward, deterministic principles. With the development of complexityor chaos theoryin the 1960s, the nondeterministic nature of the universe became still more apparent. In the 1960s, Lorentz (10) described what came to be known as "The Butterfly Effect"a butterfly flapping its wings in the Amazon may tomorrow bring a snowstorm in Kansas. We realized that complex systems like the weather exhibit extreme sensitivity to initial conditions and their behavior is, therefore, essentially unpredictable except over short periods of time.

The science of complexity, which emphasizes uncertainty and unpredictability, is seen as an inherently subjective concept tied up with meaning and context (7). Complex systems should generally be viewed more in a descriptive sense rather than as hard science. Deduction of laws from facts becomes problematic; it seems better to consider matches between models and phenomena (7, 8).

Finally, the concepts of reductionism and complexity should not be seen as a battleground. We need to think about these approaches as two different and especially valid ways of looking at the world, just as Newton's and Einstein's theories provided different views of the universeviews that are not mutually exclusive. The reductionistic approach, appropriate in view of its formal scientific rigor, is often an approximation. We need to take a balanced approach to solving different types of complex problems, using all of the tools of science at our disposal.

	LIVING SYSTEMS

Complex nonlinear systems that operate far from equilibrium (like natural systems) are driven by sensitivities to initial conditions that reflect the subtle, pervasive forces at work. The biological world is extemely complex and full of novelty and surprise, involving chance, self-organization, and selection (11, 12).

Although the reductionist approach as applied in the Human Genome Project seems to be leading to spectacular success, actual understanding of the complex genome as a whole is going to be much more difficult, except for narrowly defined problems. In very simple systems (such as a two-particle or two-body system), it is possible to give a complete mathematical solution for what happens, so it is tempting to try to solve a three-body problem by piecing together solutions of several two-body problems. But this in inappropriate because the essence of the three-body problem lies in the linkages (i.e., forces) among all of the particles. The same can be said of genes. As soon as one starts ignoring any of the connections, one ends up throwing out the problem with the bath-water, so to speak (7). With the human genome, we cannot simply put together the jigsaw puzzle and get the picture at the end!

	GENOMICS STUDIES

The term "complex trait" refers to any phenotype that does not exhibit classic Mendelian recessive, dominant, or codominant inheritance attributed to a single genetic locus. Often, many genes, as well as environmental factors, interact to produce the complex phenotype. Therefore, a gene that influences the expression of a complex disease in one population may show little effect in another, making replication of experimental findings problematic. The historical context in which each gene finds itself plays a vital role in disease expression and results in an inherent plasticity in the outcome, requiring a continued assessment of multiple parameters over time. Thus, rather close parallels can be drawn between the studies of simple versus complex systems on the one hand, and the studies of simple versus complex diseases on the other.

From this point of view, it should not be surprising that geneticists have made enormous strides in identifying the genetic basis of many simple single-gene diseases, but progress in the complex diseases like asthma (13), schizophrenia (17), and insulin-dependent diabetes (18) has been slower and the results more difficult to interpret and reproduce. In large part, this lack of progress results from the pronounced influence of both nongenetic (environmental) factors and the effects of multiple genes. Used as part of a reductionistic strategy, linkage analysis has limited power to detect genes having modest effects. Therefore, we need to think seriously about the more focused use of candidate genes in syndromes associated with asthma. As emphasized by Risch and Merikangas (19), association analysis has a greater statistical power than linkage studies. However, a limitation of the use of association studies is the need for the identification of the actual gene(s) or very close markers, usually in tight linkage disequilibrium with the particular disease-causing alleles of these gene(s). A further limitation relates to population stratification, where a disease association may be observed in the absence of linkage as the result of the admixture of two populations (20).

	ASTHMA AND ALLERGY STUDIES

Historically, studies of associations between human leukocyte antigen (HLA) and disease have proved to particularly important in analyzing many conditions with a strong immunological basis (21). These studies involving HLA are examples of gene interaction (epistasis); many other examples have been found for other genes (22). Studies of specific immune responsiveness to highly purified allergens have been especially useful in understanding the genetics of the human immune system and have provided a model for immunoglobulin E (IgE)-mediated atopy, including sensitivity toward allergens associated with asthma (21). The most definitive and reproducible findings have been for associations of immune responsiveness to simple allergens which contain only one major epitope (e.g., ragweed Amb a 5, M_r = 5,000 [23]) or a very limited number of epitopes (e.g., Amb a 6 [24] and rye grass Lol p 3 [25]). We have demonstrated a clear causal relationship between the expression of specific T cells (and associated immune responses) to the specific antigenic epitope on Amb a 5 and the presence of the HLA-DR2 type, specifically DR(1*1501) (26). This finding provides the earliest definitive evidence for the existence of an allergen-specific, HLA-linked immune-response gene involved in atopic disease. We call this gene, which is located on human chromosome 6p21.3, the atopy-and-asthma-1 gene, or AA1.

Moving up the molecular complexity scale of allergens, roughly approximating to increasing log molecular mass (Amb a 6, M_r = 9,900; Lol p 3, M_r = 11,000; Der p 2, M_r = 14,100; Der p 1, M_r = 24,200; Amb a 1, M_r = 37,800), we find evidence for an increasing number of T-cell epitopes on these allergens; further, the associations between immune responsiveness and specific HLA alleles become weaker or nonexistent, especially at higher allergen doses. An interesting example is found in the analysis of immune responsiveness to the Lol p 3 allergen (25). In this case, both the total IgE level and the HLA-DR3 specificity appear to be important interacting factors in determining whether or not a specific IgE response to Lol p 3 is observed. At low total serum IgE levels, DR3 is the more important factor; conversely, at high IgE levels, the requirement for DR3 is not apparent, its presence being no more than the frequency of this HLA allele in the study population (Figure 1). A similar finding which demonstrates the association between immune responsiveness and specific HLA alleles was subsequently made in Bet v 1, a major birch-tree pollen allergen.

View larger version (24K): [in this window] [in a new window]   Figure 1.   Frequency of HLA-DR3 positive subjects in consecutive, overlapping groups of 11 Lol p 3 IgE-Ab positive subjects and Lol p 3 IgE-Ab negative subjects. The presence of DR3-positive subjects in each group was plotted at the median total IgE level of that group. (Also see Reference 25.)

Not surprisingly, numerous particulate allergens for which exposure is abundant and persistent have been strongly implicated as causative agents in atopic asthma. Such allergens are often derived from indoor sources such as house dust mites, cat dander, molds, etc. (27). Because of the enormous array and variety of allergens that can be studied, HLA-association studies clearly have great potential in genetic analysis and in molecular and cellular studies of allergic disease and inflammation.

Moving into the clinical arena, pulmonologists and allergists have been wrestling with an appropriate definition of asthma for many years. A landmark clinical review of the field in 1959 organized by the CIBA Foundation provided a focus for modern thinking on the subject (28). Since then, there have been frequent updates. Today, asthma is broadly described as a complex disease associated with bronchial hyperreactivity, inflammation, and atopy, with narrowing of the bronchial airways that changes in severity, either spontaneously or with treatment. Whereas this description may portray an appropriate clinical picture of asthma, it is not sufficiently precise for molecular genetic studies employing a reductionistic strategy. In such case, one might look either at asthma-associated immune phenotypes such as the overall upregulation of total serum IgE level and specific IgE responsiveness as noncognate and cognate arms of IgE responsiveness; conversely, one might study responsiveness to defined irritants such as methacholine as measures of bronchial hyperreactivity to provide narrower asthma-associated phenotypes (29).

One should also investigate and quantify the perturbation of the asthmatic response by specified epidemiologic factors; i.e., one should look for causality. Because the disease outcome will likely vary according to where the gene finds itself in the organism at a given time, there is a need to standardize the epidemiologic factors associated with the expression of asthma, such as different racial/ethnic groups, different environments, and different levels and types of allergen exposure. Epidemiologic tools such as the standardized questionnaires developed by the International Study of Asthma and Allergy in Children (ISAAC) and the Collaborative Study on the Genetics of Asthma (CSGA) and standardized asthma algorithms under development by Togias, Horowitz, Barnes, Marsh, and others are proving to be very useful in such analyses.

In the meantime, our group has been searching for evidence of asthma-associated genes on each of the human chromosomes in several different populations. For example, using markers in chromosome 12q15-q24.1, we have recently accumulated evidence for linkage to asthma and the asthma-associated trait of high total IgE from several populations. In an inbred caucasian population, the Pennsylvania Amish, selected on the basis of detectable specific IgE Ab in at least one child, we found evidence for linkage of high total IgE to 12q by both sibling-pair and transmission disequilibrium analyses (13). Sibling-pair analyses in 29 multiplex Afro-Caribbean families, a population from Barbados living in the tropics and selected for asthma, provided both further evidence for linkage to 12q with high total IgE and, as importantly, evidence for linkage to 12q with asthma. Applying the transmission disequilibrium test to unrelated children of bilateral German ethnicity with persistently high IgE and their parents, our multicenter study of the genetics of high IgE responsiveness (with Drs. Renate Nickel, Ulrich Wahn, and colleagues, 30) provided still further evidence of linkage of high IgE with 12q15- q24.1. The CSGA, a multicenter research effort of which we are a part, used multipoint genetic analyses to provide further evidence for linkage of asthma to 12q14-q24.2, in caucasian sibling pairs from families with  2 asthmatic siblings (16).

An earlier set of studies searching for evidence of atopy and asthma-associated genes in several different populations exists for chromosome 5q31-q35, a region containing multiple candidate genes for allergy and asthma, including a clustered family of cytokine genes (interleukin-4 [IL-4]) that play important interactive roles in the allergic inflammatory response. In the Pennsylvania Amish population, we found evidence for linkage of total IgE, but not specific IgE, within the 5q31.1 region (31). Along with further studies, these studies suggest that IL-4 and/or nearby gene(s) in 5q31.1 regulate IgE production in a noncognate fashion. It seems likely that a generalized upregulation of IL-4 could induce B cells, which are precommitted to make IgG Ab to a broad array of antigens, to switch to IgE. Under normal physiology, such polyclonal IgE would not be specific for common environmental allergens.

Rosenwasser and colleagues (32) obtained evidence in families with asthma that a polymorphism in the IL-4 promoter is associated with elevated total IgE levels: subjects with the T variant at position -590 had significantly higher geometric mean IgE levels than subjects with the C polymorphism. Consistent with these data, analysis revealed higher levels of IL-4 gene expression in individuals having the T polymorphism.

Linkage of total IgE to 5q31-q35 is supported by studies of Meyers and colleagues (33). However, in families selected for asthma, Postma and co-workers (34) showed the strongest evidence for linkage of both total IgE and bronchial hyperresponsiveness. Interestingly, this marker has been shown to map within about 100 kb of the glucocorticoid receptor GRL and about 11 Mb telomeric of IL-4 (personal communication, E. Rubin, 1995). Finally, it is interesting to note that Marquet and colleagues (35) report results indicating the presence of a major gene within chromosome 5q31-q33 that controls for the intensity of infection by schistosomes in subjects from 20 Brazilian pedigrees. They report finding, in several populations, that immune protection against schistosomes is dependent on IgE. These findings further support our evidence for linkage of asthma-associated genes to chromosome 5q31-q35.

To define the genes involved in asthma more clearly, it will be appropriate to look for particular mutations, or series of mutations, that are often associated with expression of different subsets of inflammatory responses. Here, we are looking for genes that influence the pathophysiological versus normal functioning of the cells and the tissues involved in asthma.

Common disease-causing genetic variants, which Lander has aptly termed "isotopes" (36), may hold the key to understanding the susceptibilities toward complex diseases such as asthma. In the future, we will need to create a comprehensive catalog of these common variants to facilitate mapping of asthma-susceptibility genes. Turki and colleagues (37) have been searching with some encouraging success in the ₂-adrenergic receptor (5q34) for genetic variants involved in asthma subsets. For instance, although several variants within the ADRB2 have been identified, one variant in particular, which results in a downregulation of ADRB2, shows a clustering in nocturnal asthmatics.

For genetic analysis of the factors involved in asthma, both conventional linkage studies and association studies with candidate genes are needed (20, 36). For the candidate-gene approach, a saturation of genetic markers close to genes associated with inflammation would be appropriate. An alternative approach is a genome-wide search using equally spaced genetic markers. Multiple diallelic polymorphisms will soon be available for the further analysis of asthma-associated genes (36).

It is important to continue to limit and narrow the context of our analyses, not only in terms of the genes but also the epidemiology of the disease. But however we may manipulate, narrow, or subdivide the asthma phenotype to provide definitions that may allow more rigorous molecular genetic analysis, the essence of asthma itself, like many other complex diseases, will likely remain elusive. Despite recent advances in immunology and molecular biology, paradoxically, asthma seems to be moving away from us: we seem to be farther from understanding the disease. We face a conundrum: on the one hand, we can try to analyze the elements of asthma by strict reductionism, looking for ever more precise answers to increasingly narrow questions through the power of molecular genetic analysis. On the other hand, since the expression of asthma is so context-dependent, we can try to understand in a more holistic sense the totality of the clinical condition in all its complexity. Together, both approaches are needed; but the question is, how can such analysis be implemented? One interesting approach was developed over the past 20 years by Stuart Kauffman of the Santa Fe Institute, New Mexico.

	KAUFFMAN NETWORKS

In moving toward an understanding of the molecular and genetic basis of human disease, we need to integrate both reductionist and complexity theories with the theory of evolution. Natural selection had previously been viewed as the sole source of genomic variation in living organisms. However, the genome itself can be thought of as a network, and the conditions under which order can emerge depends on the way the network is constructed. Kauffman has tried to understand such complex systems using Boolean logic (11, 12). In a random Boolean network, the status is usually chaotic, but Kauffman has suggested that such networks do, in fact, self-organize around a small number of attractors under limiting circumstances. When, for instance, the number of inputs to an element of the network (i.e., a gene) is two, the median number of attractors is approximately the square root of the number of elements. For the human genome, with approximately 100,000 elements or genes, the square root is 317, which corresponds remarkably with the 256 known cell types (Figure 2). This is "stunning order" which Kauffman understandably says "blows his socks off" (12).

View larger version (24K): [in this window] [in a new window]   Figure 2.   Logarithm of the number of cell types in organisms across many phyla plotted against the logarithm of the DNA content per cell. Plot is linear with a slope of 0.5, indicating a power-law relation in which the number of cell types increases as the square root of the amount of DNA per cell. If the total number of structural and regulatory genes is assumed to be proportional to DNA content, then the number of cell types increases as a square-root function of the number of genes (reprinted by permission from Reference 11).

In Kauffman networks, stable cell types arise spontaneously as the attractors of a dynamic system. One randomly chooses an input that determines whether a gene (governing disease expression, for example) is on or off at the next instant. In complex reaction networks, the concentrations of substrates and products can change rapidly from high to low within very short times, leading to an approximation of gene-switching. Usually, the pattern of switching eventually falls into an ever-repeating cycle and exhibits homeostasis. Independent of the starting status of the network, gene expression eventually reaches a stable, repeating pattern, and the cellular genome organizes itself spontaneously; the on genes remain on and the off genes remain off. The network falls into a limited number of state cycles, known as "basins of attraction," which mimic the behavior of complex biological systems. Indeed, Kauffman has observed that "order seems to be provided for free."

There is evidence to argue that the dysfunctioning of such highly involved, interactive genetic networks is implicated in many complex diseases (11, 12). The interactive pathways probably look something like the diagram shown in Figure 3, which illustrates a theoretical case of a system containing 20 genes with 40 interactions. In reality, even more genes and expressed protein products interact to form a largely self-regulating system, whose order is said to be emergent. The first point to emphasize is the marked redundancy of this type of network. Second, such a self-organizing genetic ensemble is remarkably resistant to perturbation: a potentially destructive mutation in a gene involved in one of the multiple redundant pathways will not usually upset the system unduly, because other pathways can help to compensate for the loss of function. This implies that the system is buffered from the influence of mutations, which are allowed to accumulate.

View larger version (38K): [in this window] [in a new window]   Figure 3.   A theoretical interactive model containing 20 genes (indicated by numbers) and 40 interactive pathways (arrows) connected at random to the genes. As the ratio of arrows to genes increases, more and more interlocking cycles form, to produce the complex webbed structure indicated in the figure (reprinted by permission from Reference 11).

While reductionism has proved to be extraordinarily successful in many scientific endeavors, its use becomes limited as one steps outside the context of the narrowly defined hypotheses implicit in the gene-by-gene approach. Genome sequencing is a reductionistic phase in our steps toward understanding complex behavior. We need to understand gene function much better, but the question is, How can we use the new information provided by Kauffman and others to further our understanding of the genome as a whole? Just as Einstein was able to make the intuitive leap onto a light wave to understand the concept of relativity, we need to make a similar leap to understand how and why Nature works at some new level.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION CONCLUSION REFERENCES

We have seen dramatic advances in the analysis and interpretation of the structure of the human genome over the last several years. It seems difficult to appreciate that the PCR, invented just over ten years ago, is now regarded as an absolutely essential tool in modern genomics research. On the way to sequencing several "simple" genomes of a few million base pairs, researchers are now moving rapidly towards sequencing the entire three billion base pairs of the human genome by the year 2005, or before. These developments have coincided with remarkable advances in our knowledge of how to manipulate genes through knock-out experiments and gene therapy. As the technology needed to hunt for and characterize the estimated 75,000-100,000 human genes improves, we must consider how these advances in genetics and genomics will impact biomedical research into the next millennium and beyond. More important, we need to develop new paradigms that are more broadly based than simply applying the reductionist theories of the past, which involved breaking all scientific information into its elements and analyzing each one independently. The very enormity of the problems associated with genomics revolution would seem to demand a broader array of approaches. We cannot confine ourselves merely to the old reductionist strategies, but must embrace the complexity inherent in the natural world, with its tendency toward self- organization. This is, indeed, an exciting time in the development of the biological sciences as reductionist theories become integrated with the theories of natural selection and complexitywhere we can truly learn a new understanding of ourselves.

". . . with the probability revolution behind us and the complexity revolution ahead of us, the Newtonian world in whose last days Darwin lived seems more distant to us than ever before, and the Aristotelian world it displaced positively archaic."

Darwinism Evolving by D. J. Depew and B. H. Weber (8) p. 430.