Strategies for the Physiome Project

Abstract

The physiome is the quantitative description of the functioning organism in normal and pathophysiological states. The human physiome can be regarded as the virtual human. It is built upon the morphome, the quantitative description of anatomical structure, chemical and biochemical composition, and material properties of an intact organism, including its genome, proteome, cell, tissue, and organ structures up to those of the whole intact being. The Physiome Project is a multicentric integrated program to design, develop, implement, test and document, archive and disseminate quantitative information, and integrative models of the functional behavior of molecules, organelles, cells, tissues, organs, and intact organisms from bacteria to man. A fundamental and major feature of the project is the databasing of experimental observations for retrieval and evaluation. Technologies allowing many groups to work together are being rapidly developed. Internet II will facilitate this immensely. When problems are huge and complex, a particular working group can be expert in only a small part of the overall project. The strategies to be worked out must therefore include how to pull models composed of many submodules together even when the expertise in each is scattered amongst diverse institutions. The technologies of bioinformatics will contribute greatly to this effort. Developing and implementing code for large-scale systems has many problems. Most of the submodules are complex, requiring consideration of spatial and temporal events and processes. Submodules have to be linked to one another in a way that preserves mass balance and gives an accurate representation of variables in nonlinear complex biochemical networks with many signaling and controlling pathways. Microcompartmentalization vitiates the use of simplified model structures. The stiffness of the systems of equations is computationally costly. Faster computation is needed when using models as thinking tools and for iterative data analysis. Perhaps the most serious problem is the current lack of definitive information on kinetics and dynamics of systems, due in part to the almost total lack of databased observations, but also because, though we are nearly drowning in new information being published each day, either the information required for the modeling cannot be found or has never been obtained. “Simple” things like tissue composition, material properties, and mechanical behavior of cells and tissues are not generally available. The development of comprehensive models of biological systems is a key to pharmaceutics and drug design, for the models will become gradually better predictors of the results of interventions, both genomic and pharmaceutic. Good models will be useful in predicting the side effects and long term effects of drugs and toxins, and when the models are really good, to predict where genomic intervention will be effective and where the multiple redundancies in our biological systems will render a proposed intervention useless. The Physiome Project will provide the integrating scientific basis for the Genes to Health initiative, and make physiological genomics a reality applicable to whole organisms, from bacteria to man.

INTRODUCTION

The time has arrived in biological science to put it all back together. The Genome Project is not finished, nor are the wondrous advances of molecular biology slowing down. Quite the contrary, the advances are so rapid, the knowledge so detailed, and the scope of the new information so broad, that the consequences of the discoveries are often obscured by the complexity of the systems being elucidated.

A new informatics is required, going beyond the genome, beyond the cataloging of proteins, and beyond the mapping of biochemical pathways—going beyond all those into defining how the intact system works. For the industrialist who uses E. coli or yeast as a protein factory, understanding the regulated biology at the single-cell level will suffice. For the evolutionary biologist, the relationships among the genomes of the various species will suffice, without getting into function. For the physician and the pharmaceutical manufacturer, the needs are for understanding the relationships among genome, proteome, morphome, and physiome to a deep level: for the physician, to understand the genomic and environmental influences on health and disease, for the drug developer, to understand the effects of a proposed drug in diseased and healthy people. As Michael Gresser of Merck-Frosst puts it, “My strategy is to find the drug with the fewest and least noxious side effects that hits the defined target.” (Gresser, personal communication).

Edward O. Wilson, foremost sociobiologist, has captured the essence of the philosophy in his book, Consilience: The Unity of Knowledge.⁴⁰ What he writes about bringing together social science, ethics, environmental policy, and biology applies directly to the narrower realm comprised by modern biological science, computing, informatics, and health care. The physiome, the virtual cell, and the virtual human are all a part of this next challenge.

Consilience is a “jumping together” of knowledge by the linking of facts and fact-based theory across disciplines to create a common groundwork of explanation. Accordant. To quote the Oxford Dictionary’s quote of William Whewell: “The Consilience of Inductions takes place when an Induction, obtained from one class of facts, coincides with an Induction, obtained from another different class. This Consilience is a test of the truth of the Theory in which it occurs.”

Wilson goes on to say, “Scientists are professionally focused. Their education does not orient them to the wide contours of the world. They acquire the training to travel to the frontier and make discoveries of their own, as fast as possible.” … “grants and honors are given for discoveries, not for scholarship or wisdom” …. ”productive scientists have no time to think about the big picture, and see little profit in it.” This is pessimistic, but basically on target. In contrast, forward-looking molecular biologists are trying to become integrative.

Consilience is more than a conciliation of ideas: to conciliate is to demonstrate that ostensibly different ideas are really either compatible or the same. In contrast, “consilience” implies “emergent properties,” a merger of ideas and principles that creates something with behavior or composite properties qualitatively different from any of the parts or the mere sum of the parts. This is where it pays off to pull the pieces together.

THE SIMPLICITY AND COMPLEXITY OF BIOLOGY

The simplicity of biology is that there are few levels in the hierarchy of organization of an individual organism (Fig. 1). From genome to proteome to cells, tissues, organs, and endocrine and neural systems are only a few layers. The genome appears simple, a four letter alphabet forming a string of words, the genes, and their spacers and helpers. But even there, although the human genome has been almost completely unraveled this year, and figuring out which parts are genes may be straightforward, the questions of which genes do what, and when and how they play their roles is sure to take yet a few decades.

FIGURE 1.

The multicellular organism is only a few steps from the genome. Organismal behavior and the environment influence gene expression on a time scale of seconds to minutes. The connectedness across the hierarchical levels occurs via a multitude of signaling pathways.

The proteome, and the route to it, are both more complex. Transcription to form the ribonucleic acid (RNA) or messenger sequence may use two or more regions of the deoxyribose nucleic acid (DNA). Post-transcriptional modification of the mRNA complicates the prediction of protein from genome. The next stage is even more complex: post-translational modification of the protein is variegated. The simple modifications, like phosphorylation and glycosylation, usually modify function and serve as regulatory mechanisms in normal physiology. Much larger modifications, slicing off parts of the protein (Fig. 2, left) or splicing together parts (Fig. 2, right), result in proteins with functions differing from the original. In humans there are perhaps 10 proteins from each gene; in yeast there are 3 per gene. If the genome contains 100,000 genes there will be a million proteins. The beliefs⁷ in 1 gene: 1 enzyme, and later, 1 gene: 1 amino acid sequence are gone. Complexity wins again. Can the complexity be reduced to understanding by going from genome to physiome? Reductionist approaches have gone the other way, delving deeper to gain understanding of mechanism.

FIGURE 2.

Variants in Protein Kinases. Left panel: From a protein tyrosine phosphatase there are two major post-transcriptional modifications, one to splice in a 740 amino acid sequence, then a second to cut a plasmatic or cytosolic segment free from the membrane-spanning parent protein (figure from J. Schlessinger, New York University, with permission). Right panel: A protein tyrosine phosphatase 1b which may be formed from one of two homologous domains of a protein CD45 (from E. H. Fischer, University of Washington, with permission).

Doing the Two Step

Two centuries of reductionist tactical effort have produced an immense, even if uncoordinated, body of knowledge. The breadth of information is so great that grasping it within a single mind is impossible. Wilson⁴⁰ defines the two step. Step 1: work top down across two or three levels of organization by reductionist analysis; Step 2: work bottom up across the same levels by synthesis. Good examples of this are to be found in cardiac biophysics.

One is the excitatory process in the heart. Over the course of five decades of research the nature of the kinetics of individual ionic channels in the cardiac sarcolemma were elucidated through voltage clamp and then patch clamp studies, and in some cases through studies of the sequence and structure and conformational changes in the channel protein. The “top down” went from studies of tissue (small trabecular fibers), to whole cell clamp to a patch of membrane containing a single channel and down to studies of the isolated proteins with specific amino acid substitutions. The yield from this, starting at one level and digging two deeper, laid the groundwork for the return surge, to put it together.

Efforts to put these channels and their time- and voltage-dependent currents back together into a comprehensive scheme began even as the reductionist probing was proceeding, following the lead of Hodgkin, Huxley, and Katz^21,22 on the squid giant axon. Weidmann³⁹ and Noble and Tsien³³ pushed the synthetic analysis of cardiac currents. Beeler and Reuter¹⁰ composed the first cardiac action potential model with Na, Ca, and K currents, and the evolution progressed as more of the currents were characterized. Comprehensive models of the ionic currents during each action potential now represent over a dozen time- and voltage-dependent channels, concentration-driven ionic exchangers, and ATP driven pumps.^16,29,30,41

A similar evolution occurred with excitation-contraction coupling following a century’s work on the relationships between extracellular concentrations, the timing of stimuli, and the contractile responses.^35,43 Experiments using calcium-sensitive photoprotein aequorin (Ref. ¹¹, and others) revealed that free cytosolic calcium concentration rose before force was generated and the muscle shortened, and fell before the muscle relaxed. The time course of cytosolic free Ca levels is almost independent of the extracellular Ca²⁺ concentration. But some extracellular Ca²⁺ is important, and the necessity for a small influx led to the recognition that “trigger calcium” entering via Ca channels into the junctional region between the sarcolemma and the cisternal sarcoplasmic reticulum (SR) initiates a wave of secondary release of Ca from the SR into the cytoplasm and then Ca binding to troponin, initiating contraction. The early models¹ did include these phenomena although they now have been superseded by refined, detailed models^23,41 which now set the standards for the field. These comprehensive models provide quite precise exploration of the behavior of excitatory, contractile cells; what is being recognized is also that each cell type and each species has its own peculiarities: while the model structures differ little, if at all, the parameters are rather widely variable, even among cells of a given heart. Pacemaker cells, atrial myocardial cells, Purkinje cells, and ventricular myocytes have widely different action potentials and contractile capabilities; there are further differences within each cell type, depending on physiological status.

The electrical events have now been taken up a level, namely to the whole heart level, e.g., by Refs. ^{24, 14, 41}, and ⁴². Comprehensive, integrative models reveal emergent behavior; in this case that behavior is the basis for arrhythmias, for ventricular fibrillation, and for the formation of three-dimensional patterns of depolarization in the form of scroll waves, in other words, chaotic dynamical behavior. We can remind ourselves at this point that fractal and chaotic behavior occur in integrated systems, but not in their simple isolated components.⁴ A challenge for modelers of the physiome is to make use of these integrating principles, where fractals indicate correlation and chaos denotes predictability, at least in the short range.

THE PHYSIOME AND THE PHYSIOME PROJECT

The goal of the Physiome Project is to define the physiome, the ultimate vehicle for understanding biology, in its broadest terms. The name “physiome” is coined from “physio,” life, and “ome,” as a whole entity. The term is consistent with genome, proteome, morphome. The physiome is the quantitative integrated description of the physiological dynamics of the normally functioning intact organism. It has its recipe in the genome together with the environment in which the organism develops and is maintained.

The genome and the proteome are a part of the morphome, the structure of the organism, but it is useful to designate these critical components not only because of their special places in the history of biology, but also because they are, at least to a good approximation, components for which we expect to determine complete descriptions for a given cloned organism.

Let us define the genome as the DNA sequence and the identification of each of the genes in the chromosomes of the organism. The genotype is the summary of the information contained in the genome. The phenotype is a little more vague; I interpret it to be the description of the identifiably distinct anatomic and physiologic features of the organism associated with the particular genotype.

The proteome is also a little vague, it starts with the amino acid sequence of the proteins, then must include the structure, not just its crystalline form from x-ray crystallography or nuclear magnetic resonance (NMR), but its folds, its active sites at which it has biological effects, and something of its varied conformational states and energetics. Since the form of proteins is dependent upon the presence of other solutes, of pH, of charge and hydration, the picture is not of just a single form. Nevertheless, though proteins are at the center of the phenomena of life, they alone do not define life, proteins act upon small solutes which are not defined by the genome, and do so in the context of local conditions and geometric constraints. Composite metabolic enzymes like lactate dehydrogenase allow more than one reaction; larger protein complexes can support a series of reactions without releasing intermediates into the cytosol, a molecular assembly line.^26,27

The “phenome” might be a word to define the next higher level of integration, for it implies action, as in phenomena. The Oxford defines “phenomenon” too closely for our broader intentions: it is “a thing that appears, or is perceived or observed; an individual fact occurrence, or change as perceived by any of the senses, or by the mind.” Whewell’s 1840 phrasing relative to “Descriptive or Phenomenal geology” helps: “We must have a phenomenal science preparatory to each aetiological one.” Thus we must pass up phenome as defining our goal since this word carries no implications with respect to either mechanism, etiology, or quantitation. Nevertheless, without observations of the phenome we would have no inkling of biology; for it is the observations of changes that take us far beyond the morphome. Data, observations, are what provoke us toward understanding.

The physiome, like the genome, is an object to be defined through discovery. There is a physiome for each species and for individuals of a species, all species from viruses, bacteria, to monkey and man. In man there are about 10⁵ genes, 10⁶ proteins, each with a few too many conformational states, and perhaps 10¹⁴ cells. Given that there are many states of each protein and many different degrees of expression of each of these, and that the body is dynamic, not static, the physiome defies simple description. Its complexity fulfills our visions of a “complex system.”

The “Physiome Project” is the concerted effort to define the physiome, through the databasing of observations and analyses and through the development of integrated descriptive, logical, and quantitative modeling. The models, ultimately quantitatively descriptive at any particular hierarchical level, should be supported by mechanistic, explanatory or descriptive models at the next deeper level. The style of “explanation” depends on the level, being physiological at the higher levels, biochemical and pharmaceutical at deeper levels, biophysical and physicochemical below that.

The models, via iteration with experimentation, remove contradiction and demonstrate emergent properties, i.e., structure, or functional behavior, that is not evident from their component parts. Hypothesis-driven research as well as fortuitous discovery serve this “reverse engineering” of biological systems.

The “Project” is more than science: it is education, archiving, disseminating, and databasing. The goal is that the data, the concepts, the descriptions of biological elements, processes, and models will be accessible publicly via the Internet.

The completion of the Physiome Project lies in the distant future. This is no surprise when one considers the immense amount of work that is needed to complete the Genome Project following the completion of the sequencing of the DNA. The Genome Project could be designated as complete with the identification of each of the genes, the RNA, mRNA, the tRNA and the proteins and their post-translational modifications. But there will be much to do after that, still at the genomic level, such as the control of expression and the regulation of development.

The genome, a magnificent resource, will become a part of the physiological systems that make up the physiome. Changes in physiological activity, in environment state, the stage of life, and the deformations of disease all lead to changes in gene expression. The rates of transcription and translation must vary continuously, even minute by minute: the dynamics of gene regulation should be considered as a part of normal physiological regulation.

Phenotype can then be considered as emanating from genome and physiome together with environment. Is the cardiac hypertrophy of the athlete fundamentally different from that the adult with a hemoglobin FA hemoglobinopathy, an arteriovenous shunt, or early hypertension? Undertaking long distance running or competing in the Tour de France are responses to a social or cultural environment that encourage hard work and high cardiac output, made possible via the cardiac hypertrophy induced by hourly and daily stimuli modifying the rates of expression. At what point do “physiological” responses turn into “pathological” responses?

The genome and the physiome are inseparable. Having the complete genome is insufficient. Since expression and function are affected by activity, the environment, and disease, we must account for what we eat, our substrate supply, and the demands put on our bodies and minds.

The Early Goals of the Project

The three essential, early goals of the Physiome Project all involve developing strategies and methods for putting information together: (1) to develop comprehensive methods for the acquisition and databasing of very large sets of information on all aspects of biology, and (2) to construct descriptive and quantitative models that integrate the available knowledge so as to determine the inadequacies, inconsistencies, and insecurities in that knowledge, and (3) to organize collaborations at the national and international levels to target particular areas of integrative biology. Given the end goals, models, how should these goals be developed to serve the purpose of advancing and integrating knowledge of biology?

Models Begin as Diagrams of Relationships

The Boehringer–Mannheim chart of biochemical reaction is a popular example, on the walls of many laboratories. Websites nowadays give updated information, usually for particular species, recognizing that these maps are not at all the same throughout eukaryotic or even mammalian phylogeny. Likewise there are charts of physiological regulatory systems, e.g., for endocrine or signaling systems or for immune responses. The charts summarize the perceptions of what substances react to form others or which elements of a system are interacting. Sometimes the connecting lines have an arrow to indicate direction of the action or information flow. There is no indication of the traffic density along these pathways, nor of how the traffic fluctuates. Neither the nature nor the kinetics of the reactions are defined in the diagrams.

The next model level is one where the nature of the reactions are defined. This is where the information set enlarges. The facilitating enzymes, the descriptions of the conditions, substrates, energetics, etc., for the reactions are folded into the information set. Good examples are Ref. 36, developed over many years. There should be similar databases for transporters, though some are included in Selkov’s. Gilman et al. (swnt240.swmed.edu/gradschool/webrib/gilman.htm) and others (e.g., www.stke.org) are constructing the signaling pathway databases. Many others are being developed, some linked to the genome databases, some not. As with the genome efforts, multi-institutional collaborations are blossoming, simply because these problems are too big to be tackled within a single lab and the requisite knowledge set is too complex, covering experimental data and technologies, information storage and retrieval, mathematics, computation, and the biology itself, in health and disease.

Focused Collaborations

Thus with regard to step (1), the collaborations tend to grow together in focused scientific target areas, and yet require a diversity of the most effective technologies. A group targeted on a problem focused on a particular phenotypic disorder (e.g., Allen Cowley’s and Howard Jacob’s program with others in hypertension,¹⁵ must have collaborators with experience ranging from genomics and animal colony management, physiological measurement, clinical assessments of populations, and advanced statistics. The general rule is that one starts the collaborative effort at any chosen level of the hierarchical system (Fig. 1) and then works up and down from the starting level.

Shared Databases

The vehicle for collaboration is information flow and archiving, so that all of the collaborating investigators have instantaneous access allowing entry and retrieval. These too must be in targeted, relatively focused fields. Data to be acquired and retrieved are growing in complexity, and include everything from the relevant genes, expression patterns, proteins, biochemistry, physiology, images of three-dimensional moving objects, to mathematical models. Mechanisms for entry, curation, retention, correction and for graded access are yet to be worked out. Nor are there funds for their support yet designated in the budgets of federal agencies. New methods of databasing need to go beyond the standard, though powerful, relational databases now being used for the genome and the proteome.

The integrative systems analysis of large scale systems is perhaps the most technically advanced of this triad of needs, but has not gone very far, except in the cardiac arena. Modeling requires databases: data sets against which to evaluate models, collections of parameter values from different circumstances and different species, the models themselves, the tests of model validation (to see if they reflect the physiology correctly) and verification (to see if they have been computed correctly), and full descriptions of what they do and how to use them. Particularly critical are clear definitions of model simplifications, assumptions, the experimentation on which a model is based, the sensitivity of its parameters to noise and error in the data, and so on. Even more important is that the models must be accessible over the internet for evaluation broadly, first by the collaborating group and then by scientists at large.

Given a target area, a set of colleagues around the globe, must each group begin its own database or are there major resources generally available? The answer is that beyond the genome and proteome and some selected subsets of the proteome, there are no generally useful databases for physiological systems. So this is a high priority.

There is growing awareness of this. The needs were outlined in an initial outline of the Physiome Project in an on-line report (www.physiome.org/history/). More emphasis is placed on the combination of computation and databasing in the report on “The Biomedical Information Sciences and Technology Initiative” prepared by an Advisory committee to Harold Varmus, formerly Director of NIH (www.nih.gov/about/director/060399.htm).

DATABASES FOR THE PHYSIOME

Proteome databases can be built upon genomic databases, for the polypeptide sequences emerging from the genes are prescribed precisely. Of course there are many alleles: the CFTR gene (for the cystic fibrosis transporter) has now been found to have over 700 variants, though the regions most specifically responsible for its function tend to have fewer variations than other parts of the protein. Now, midyear 2000, as the sequencing of the human genome is being “completed,” fewer than 5000 of the genes have been identified, out of a total as yet unknown, but estimated at 45,000–140,000. Call it 100,000 in round numbers. The number in E. coli is 3600, in yeast 6000, and 19,000 in the worm C. elegans: all of these are sequenced now, and even though E. coli and yeast are single celled and rather thoroughly studied for decades, about one-third of the genes are not yet associated with functions. In C. elegans, though its 960 cells are explicitly identified, half the genes’ functions are unknown.

Proteins are not synthesized simply. Often the sequence comes from several parts of the genome, usually neighboring parts. Post-transcriptional modification of messenger, mRNA, changes the plan before the mRNA leaves the nucleus. Ribosomal translation from mRNA to polypeptide in the cytosol is usually followed by post-translational modification which produces only a few variants in E. coli and yeast but an estimated ten variant proteins per gene in humans, for example as for CD45 and protein tyrosine phosphorylation (Fig. 2). Splicing of post-transcriptional fragments is probably rather common. The variants often do not have similar functions, and it is fairly common to see the functions diverge through the phylogeny.

Data of the types suggested in Fig. 3, right upper, are to be found in databases like SwissProt (expasy.hcuge.ch/sprot/sprot-top.html). For the most part the data are not really complete. A problem for the integrative modeler is to find values for parameters such as rates of association and dissociation, equilibrium dissociation constants, reaction rates, and so on. The values available tend to be those for strictly in vitro observations, rather than in vivo, and in relatively dilute solutions rather than for concentrated protein solutions of the nature of the cytosol of the particular cell type in a given species. The classification of function is available for most of the identified proteins. What is more difficult to find out are those features listed in the left lower bubble of Fig. 3, the least known of which are protein-protein interactions.

FIGURE 3.

Elements of proteome databases.

Morphome databases are more difficult to capture in relational databases as both their dimensionality and their size are greater. The morphome is defined in modern quantitative terms as the quantitative description of anatomical structure, chemical and biochemical composition, and material properties of an intact organism, including its genome, proteome, cell, tissue, and organ structures up to those of the whole intact being. The morphome has its descriptive beginnings in the textbooks of anatomy, histology, and cell biology. It includes the National Library of Medicine’s Visible Human Project, a slice by slice representation of the whole bodies of one man and one woman (at 1 mm thickness for the man and 1/3 mm thickness for the more refined woman). The Visible Anatomist Program at the University of Washington (sig.biostr.washington.edu/projects/da/) is one of a set of programs going a step further by attempting to “segment” the sliced data sets to define the organs separately so that nerve tracts, vascular systems, individual muscles, lobules of glands, and ducts can all be identified and color-coded for identification. The same thing needs to be done at the microscopic level for organs, tissues, cells, and subcellular structures. When adequate generic structures are available by download from the web, then these can be used as the basis for anatomically appropriate models at all levels. Anatomic shapes for adult organisms is a start, but the morphome databases should cover all stages of development (right upper bubble, Fig. 4).

FIGURE 4.

Elements of the morphome databases.

Anatomy is just the start for the morphome. Tissue composition, the concentrations of the whole set of proteins, of salts and amino acids, and substrates and reactants of all sorts are really a part of the morphome. Well defined data on these are needed before one can model biochemical reactions or contractile or excretory processes correctly. Composition is much more than water, fat, carbohydrate, inorganic salt and protein, and the details are essential. As one looks forward to the advances of tissue engineering, it becomes obvious that a different factor dominates, namely the material properties, the ability to withstand shear, to maintain tissue integrity under mechanical stress are more important than having an exactly correct “milieu interieure” or intracellular composition (left upper bubble, Fig. 4).

The Physiome databases, covering all of biology, are a grand challenge effort, in the style of the genome databases, but much larger and more variegated in content. The four-letter alphabet of the gene sequence and the twenty letter alphabet of the amino acid sequences cannot be used as backbones for the physiome database. Consider an individual’s database to consist of:

1. The individual’s genome.

2. The individual’s proteome, including the concentrations of all proteins and other solutes in plasma, all cells, interstitial spaces, and body spaces. The state of the functioning proteins and signaling systems.

3. The individual’s morphome, including the quantitative aspects of the anatomy, the material properties, the morphometry of typical cells. The volumes, surface areas, mechanical, acoustic, and thermal properties of the elements.

4. The physicochemical status of all systems components from organelles to the whole body.

5. Schema of interactions among the components.

6. Models describing the functional relationships and the kinetics of the interactions. These range from gene expression and its regulation to the relationships between the organism and the environment.

7. Functional models of the intact organism must include the stage of hormonal cycling, exercise and training states, fight or flight conditions, and eventually become predictive of the health issues for an astronaut on a three-year Mars mission.

The literature in biology is traditionally in the style of the right upper bubble of Fig. 5. The data are gathered in table or plotted as Y on X plots, or of plots of Y versus time. They give summaries of the data for a study and examine the statistical associations between variables. A traditional physiological approach was to control all of the variables in a living animal as closely as possible and then manipulate one while observing the response in another variable. These, while not necessarily attempting to determine causality, were designed to allow statistical evaluations.

FIGURE 5.

Databases for the Physiome.

From these, inferring causality wherever reasonable, the schema of relationships emerged, the right lower bubble. When very large amounts of data were obtained on many varied aspects of the system one could begin to examine biochemical, pharmaceutic or physiological systems in a quantitative way.

Modeling, as in the left upper bubble of Fig. 5, had two kinds of history. One was to use the models simply as summaries of data or as parametric descriptors allowing distinction between two classes of behavior, as in sickness versus health. The second way was as an hypothesis, a vehicle for designing experiments to distinguish between alternative hypotheses.³⁴ For the Physiome Project, the modeling is the ultimate goal, to summarize information succinctly, to attempt to resolve contradictions amongst data and amongst hypotheses, to provide explanation for system and cell behavior, and eventually, when the model is very good, to allow prediction of the effects of intervention.

The archiving of models, model interfaces, subroutines, tutorial examples, and the sources of the scientific information on which they are based is a prime goal. For any given model, there would be possibly many different parameter sets, for different species, for different physiological states or stages of development, or different disease states or different environmental conditions.

Databases are the most critical elements needed for the Physiome Project right now. NIH and NSF are unlikely to fund grants to individual investigators to do this. A possibility is that the National Center for Research Resources, an institute of NIH, can take a leadership role: of all the agencies this may be the one whose mandate is closest. The National Library of Medicine is another possibility, but this project is so much larger than the Genome Project that putting it under the NLM is unlikely. The Genome Institute is established to further the aims of the Genome Project and looks to have its hands full for the next decades. In any case, database development requires political decision making, requiring strategy development that goes far beyond this essay.

Databases are being developed in a variety of particular areas of cell biology, but at this point are primitive. Most investigators are dependent on searches of the biological literature, a notoriously difficult approach, for a variety of reasons. One is that raw data are seldom preserved in published works; though journals such as the American Journal of Physiology and those of the American Institute of Physics (the producer of this journal) are making archival space available to authors for the permanent storage for their raw data. Few make use of it, so what is usually published are summaries of observations or parameters describing the observations or the analyses.

Search engines for published literature and for databases are being developed in the hope of automating searches, allowing them to cover vast areas of the literature without manual intervention. A search might allow one to find the set of pharmaceutical agents which increase the expression of protein A by a factor of three while reducing the expression of proteins B and C. When the semantics for such search engines are well defined and recognized by those who are reporting their data, then the searches will become more effective. Web searches of high quality allow Boolean search commands; one such as Google (google.com) may be a good starter. A troubling fact is that such automation is unlikely to retrieve much from published papers, even when it is there. We are going to be dependent for some time on personal searches by trained biologists, who can then put the information into well-constructed databases.

Searching databases will not be easy until the databases contain more than sets of tables or listings of numbers associated with particular features of proteins or cells. Responses to physiological interventions expressed in terms of time course data strings will not be easily searchable and will not immediately yield meaning in terms of parameters for governing equations. Clearly needed are well-defined semantics for databases and ways for defining the relationships between variables, and then expressing them in terms of quantitative expressions for computation. John MacNeil and Alan Goates of Isis Pharmaceuticals are developing a database query language, Metagraph, with carefully defined semantics; at this point the relationships cannot be automatically translated into equations for quantitative relationships. Perhaps qualitative relationships should be defined so that they are directionally correct even if there is no mechanism inherent to the description. If this too fails, can a stochastic relationship be defined? Failing this, can a logical relationship be useful? My prejudice is that purely logical relationships all fail when the system contains more than a very few elements.

These databasing aspects of the Physiome Project are summarized in the top part of the table in Fig. 6. The database will become the record of the information yield from the reductionist efforts to work downward by a few levels, the first step of the “two step.” The consilience⁴⁰ is the next step, developed via the database, the network, and ultimately by the integrative modeling.

FIGURE 6.

Main components of the Physiome Project.

MODELING: FROM GENOME TO HUMAN? OR VICE VERSA? OR FROM THE MIDDLE….?

The model, per se, is not the objective. The objective is develop and improve over time a comprehensive and integrated understanding of a complex system. A model is a vehicle for bringing together information, for retaining an understanding, for removing inconsistencies, and for testing ideas not yet subjected to experimental examination. A model is formulated to represent the state of the art, and is doomed to fall victim to newer ideas, broader viewpoints, and to corrections in the knowledge base. The models of Copernicus and Kepler became mere stepping stones when Newton’s better explanation replaced them, but Newton stood upon the shoulders of his great predecessors. And his model too has been replaced, though all of these models remain useful even when known to be incomplete. Children are still taught that the earth circles the sun and that the moon circles the earth…. not exact but close enough.

Big models should be better than little ones. They should have a greater predictive capacity, and encompass a wider diversity of information. But inevitably they have some gaps; the effort to put models together usually reveals the gaps. Recognition of gaps leads to studies to fill them, but the usual story is that much information remains missing since no one wants merely to fill in the chinks and no agency funds one for being compulsive about it. Compulsive behavior, fortunately common among scientists, helps. The models of physiology which are firmly based on physics, chemistry, anatomy, biochemistry, and which fulfill requirements for mass and energy balance do the best, for even when all is not known, strong inference prevails.³⁴ These models will early on contain alternative hypotheses, a pair or more of alternative explanations for unexplained or unresearched parts of the system. These are the guides or at least inspiration for the next experiments.

“From Genome toward Function!” is the current battle cry. Physiological genomics appears to evoke an image of pulling the proteins out of the genome, finding out what each one does, and then learning how it affects the function of the animal. This approach can work if one actually knows what the physiology is. Given a million proteins in humans, this is going to be a long slow path.

Selecting a disease target focuses the effort: starting from the top down instead of bottom up is going to be more effective. In practice, most of the steps between genome and physiome are ignored. The gene for the anion transporter, CFTR, in cystic fibrosis, is identified, the association of DNA sequence abnormality with the disease is statistically very clear, and the mechanisms by which a deficiency in the expression of an ion channel disrupts function are understood at a qualitative level. It is hard to make a case that every step along the route from transcription to epithelial fluid secretion needs to be spelled out in detailed quantitative terms. It is equally difficult to argue that no further research need be done; discovering and sequencing the gene has not led to a cure, and won’t for some time. Of the over 700 alleles identified, it is not obvious that there is a “right” one.

The natural inclination of scientists to want to understand, exactly if possible, has the same power in biology as in physics and chemistry. That the blood circulated through capillaries was a well defined conceptual model before one could see capillaries. Models for blood flow, rheology, for blood–tissue exchange, for cardiac pumping are all over a century and a half old. The steps that are remembered are the quantitative ones: Fick’s laws of diffusion and his method for estimating cardiac output, Hodgkin and Huxley’s action potential, Nernst’s expression for the membrane potential, Starling’s law of the heart.

These are relatively simple models, the building blocks. Will larger models have the same persuasive clarity and be similarly regarded as fundamental? Only with some difficulty. Arthur Guyton’s pioneering modeling of the regulation of body salt and water and blood pressure²⁰ did not garner much appreciation, though it should have. It was a detailed, conscientiously constructed model of much of what was then known about the physiological regulation of blood pressure and volume and salt and water in the body. It represented years of experimentation and thoughtful analysis. The problems were twofold, I think. The models were incomplete and a good many components were descriptive of relationships between a governing variable and a dependent one, e.g., pressure-volume curves, which one could accept for particular arteries where the measurements had been made, but maybe could not accept so readily for the whole venous system, where measurements were less direct. Inevitably there was felt to be some shortcoming or other for any component of the model with its three hundred parts. And though there were real shortcomings, recognized by Guyton and colleagues, that was not the real issue. After all, the model did work rather well for describing data and in predicting the results of interventions. It has certainly proven to be a useful teaching model.

I think the real issue was that it could not be intuitively understood, and was therefore mistrusted. One could argue logically that it should not be trusted since it was made up of components that were each not wholly correct. In later years Tom Coleman’s version of the model was released for use in teaching as QPC or Human,³² and proved useful, but as a vehicle for science the original model and its derivatives were pretty well ignored. Too bad! Like any starting model it could have been developed further, step by step, from its original form of a mixture of concepts and data representations into a masterpiece of solid scientific understanding.

I am optimistic enough to believe that the fate of Guyton’s model will not be suffered by all future large scale all-embracing models. Even now it is too early to say that its fate is known. While it is fair to say that the physiome will not reach the stage of forming a mathematically complete and realistic model for decades or even centuries, substantial parts of it are now being brought together. Our examples are the pieces of the cardiome.

Strategies Toward Constructing Large Models

Organization, strategies, and tactics, are needed to go the next steps. The breadth of scientific knowledge, even in a narrow field, is so diverse, extending from genomics to whole organism function, that no one lab group can call itself expert in every aspect. The consequence is that large scale modeling requires the involvement of many lab groups and the development of attitudes of trust and collaboration. American science, particularly molecular and genomic science, has exhibited much collaborative behavior, but just as much fierce competitiveness.

Funding of this kind of biological science has been difficult, but the scenario is changing. The granting mechanisms that fostered individuality, even secretiveness and hostility, are only now being replaced by the development of interinstitutional grant programs within NIH, NSF, and other agencies. NIH and NSF staffers have been surprised by the dearth of submissions for cooperative programs on complexity and integration of biological systems from the research community. Naturally it takes a while to develop a broad collaborative effort, and it takes longer to write a grant application when the several investigators are at different institutions. The new biomedical information science and technology initiatives (www.nibib.nih.gov/becon/becon2.htm) recognize this and offer support for a planning phase preceding the definitive group application. Investigators are aware that if an agency’s offerings are not attracting applications the programs may eventually be withdrawn, so it is important to work hard to respond to these opportunities.

My recommendation for initiating the Physiome Project is to start with the databasing systems and with modeling projects simultaneously. Databasing requires serious efforts, including major research into how to do it well, Dao and Dewey (in this issue, 2000) define many of the issues, from semantics for biology to the details of databasing technologies.

The integrating efforts must begin before the databases are well developed. They will serve to drive the databasing effort, and help to identify missing information. The experimental data are critical, and in any particular narrow area are pretty well known by a relatively small cadre of investigators. A set of these investigators forging a collaboration can lead the pack. Within each target area are definable subtargets, allowing a strategy for partially independent development by the individual participants.

1. Break the system into small components. Choose these so that there are a minimal number of interactions between them. These may be at any desired level in the hierarchy. The components or modules each will go through the stages of modeling: the schema of relationships, the definitions of the types and characteristics of the relationships, the quantitation of the kinetics and the mechanisms for the relationships, and eventually the quantitative mathematical models of the subsystem or submodule.

2. For each submodule, define the variables that are used or defined by other submodules. Define all other inputs and outputs. These are the “hooks” that are essential to linking the submodules. Ideally the variables used in two or more submodules would vary slowly with time, so that the submodules are only loosely coupled kinetically to one another even if tightly coupled spatially. The influences of one submodule upon another are wielded through these variables, the observable hooks, which must be carefully compared to and matched to physiological data. There are advantages and disadvantages to a modular approach, only some of which are computational; these are listed in the Table 1.

3. Bring the submodules together into a composite model. This necessitates going through the same sequence of modeling stages, now from the broader point of view. In actuality, the beginnings will have been accomplished by the time the collaborative endeavor was started, at least to the stage of the overall schema of interactions. Because each submodule may be quite complex, it makes sense to consider the original developer of the submodule to be its caretaker, the group that maintains it over years, even while improving it in the light of newly acquired data and broadened insight. This introduces serious questions concerning curation, caretaking, maintenance, for databases, schema, models, etc., that we return to later.

4. Distribution and provision of free access to the models and to the data. These require another level of effort. In our NCRR-supported resource facility this is a major component of effort. Taking a model to the stage where it can be used for analysis of data is actually only about 5%–10% of the effort it takes to make it available to others. The distribution effort requires writing in standard code (no small effort to assure this), writing the test sets and checking out the solutions under various operating systems, preparing files of test solutions so that new users can determine if their implementation is correct, writing manuals and tutorials, and finally setting these forth on a website so they can be downloaded. While we have a good many general models for transport available on our website (nsr.bioeng.washington.edu/Models), a substantial number of more advanced ones have not reached the stage of general release because of the extensive effort required. A new Java-based approach to making models platform independent, described later, facilitates model development with interaction between groups but does not reduce the need for tutorials and manuals to introduce the models to users.

TABLE 1.

Modular design.

Advantages

Disadvantages

♦ Single group is central source of expertise    for the module. ♦ Group can be expert in experiments and in    analysis. ♦ Code is written, refined, validated and verified,    documented, archived, and disseminated    through the single coordinating    group. ♦ Modifications, expansion, and maintenance    provided at one place. ♦ This assigns the decisions about “right”    vs “wrong” versions of the module to    this group, and their oversight committee.

♦ Group knowledge is limited. ♦ Collaboration, national and international,    may be inhibited. ♦ The group’s tool kits may not continue to    meet advancing standards. ♦ Modules are not connected to other modules,    and therefore are incomplete. ♦ One module in an assemblage of modules    may block a critical path to “completion”    of a system. ♦ Responding to requests for modifications    from others competes with time for the    modules development. ♦ Numerically less accurate.

MANAGEMENT OF THE PHYSIOME PROJECT

A good plan for implementing the goals should be as clear as possible. A distant goal, the complete computational representation of all organisms, may be seen to have adequate clarity, but is so large an undertaking that it easily dissuades entry into the effort. Defining lesser goals and defining milestones is therefore mandatory. In this section we suggest an approach to the efforts needed at the whole organism level. It is conceivable that one group could put together a practical representation of E. coli but for larger organisms many groups will need to work together.

A Draft Version of a Vision Statement

Understanding human biology is critical to linking genomics to health. Integration of knowledge allows intervention to alleviate or cure disease and reduce the ravages of aging.

Descriptive and quantitative modeling are the keys to integrating what is known about anatomy, physiology, metabolism, and behavior. It identifies what needs to be understood. The goal for medicine and health is to use the human physiome as a predictive tool. To do this requires producing conceptual, descriptive, and ultimately quantitative models of many organisms, each composed of an integrated set of submodels describing the morphology (genome, proteome,… organ, organism, through embryonic development to adult anatomy), molecular and cellular dynamics, organ function, system regulation, and adaptation.

The project is the natural successor to the genome project. Because it is larger than previous scientific missions for mankind, it should be started now and organized so as to minimize missteps in its development, and to make progress maximally efficient.

Steering the Physiome Project (Management Strategies)

Four kinds of efforts are needed here: (1) define the overall direction, (2) provide management by devising strategies and organizing communication, (3) organize scientific working groups focused on achievable targets, some of which should have relatively short term benefits to the science community, and (4) define a set of technology working groups to provide the tools for the endeavor.

There are aspects of this effort which would be best accomplished via leadership at the level of the national science agencies, in the USA by NIH, NSF, and other federal agencies. Templates for this kind of effort are the High Performance Computing Initiative and the Genome Project. An example of a particularly effective support style is provided by the National Center for Research Resources, NCRR, an institute of NIH, which supports about fifty facilities providing special technical and scientific expertise in target areas of national consequence and which are not readily supportable simply to satisfy the needs of a single university.

As an example, our University of Washington Resource Facility for Circulatory Mass Transport and Exchange focuses on tracer exchange and metabolism, providing tools for basic research and for clinical image analysis, as in Table 2. The research is in methods of analysis of data on solute and water transport, on enzyme and receptor activity, cellular metabolism in cells, organs and organisms. The applications are to clinical imaging by positron emission tomography (PET), NMR, x-ray-computed tomography (CT), and ultrasound, to multiple tracer studies, pharmacokinetic systems and drug metabolism, tracer studies in cells and organs, cardiovascular integrative physiological dynamics, and fractal and chaotic dynamical systems. Just as important as the scientific projects are the other aspects of the resource responsibilities: the development of tools for simulation analysis (Core Technology), the dissemination and training, and the support for collaborative programs within the U.S. and elsewhere. The (numbers) in Table 2 indicate the number of projects of each type.

TABLE 2.

NCRR Simulation Resource in Circulatory Mass Transport and Exchange.

■ Core technology projects  ♦ Simulation software   ● XSIM interface   ● J2XSIM   ● JSIM  ♦ Functional imaging  ♦ Visualization  ♦ Optimization  ♦ Identifiability ■ Dissemination, training  ♦ Courses in transport analysis  ♦ Software distribution  ♦ On-line web computation  ♦ Training pre-, post-M.D./Ph.D.

■ Projects: Transport, Exchange, Metabolism  ♦ Cellular metabolic systems (3)  ♦ Energy regulation (2)  ♦ Receptors, signaling (3)  ♦ Blood-tissue exchange (4)  ♦ Facilitated flux systems (2)  ♦ Fractal vascular structure and heterogeneous system (4)  ♦ Dynamics of osmotic balance (1) ■ Collaborative programs  ♦ ROI projects (12)  ♦ PPG’s in Resp. Med., Nuclear Med., Cardiology (3) ■ Training via websites  ♦ Kinetic processes  ♦ Integrative biology

http://nsr.bioeng.washington.edu

One overall strategy for the Physiome Project would be to set up a large number of such resource facilities, each targeted to a particular aspect. The supercomputer centers (San Diego Supercomputer Center: www.sdsc.edu; Pittsburgh Supercomputer Center at Carnegie-Mellon: www.psc.edu/biomed/biomed.html) might in turn serve a number of the more focused resources. Informatics resources as starting points are: NCBI (www.ncbi.nlm.nih.gov) and CMS-MBS (www.sdsc.edu/ResTools/cmshp.html).

An Administrative Structure

Again the question arises as to whether this should be organized at the federal level or regarded as a grass roots, spontaneously organized effort coming from the investigative community. While the NIH generally tries to support efforts which originate outside of the agency, other agencies have a stronger history of being proactive in efforts to achieve national goals. NSF’s Engineering Research Centers are exemplary of this approach. What needs doing are:

1. Devise organizational structure, and provide templates for subprojects.

2. Establish quality control programs, provide guidelines, and define standards.

3. Enhance funding opportunities at local and federal levels.

4. Establish special interest, coordinated working groups on circumscribed target topics.

5. Define the semantics of the field of biological sciences by establishing a basic uniform vocabulary.

6. Define technical support structures and working groups for tool developers.

7. Encourage the view that anointed web publication, in the public domain, with experimental data, schema of systems and complete and documented models for free use is as reputable and is more valuable to the research community than is mere journal publication.

Scientific Working Groups for Specific Targets

Believing that there is virtue in beginning with the systems for which there are plentiful data and much incentive to develop models for practical applications, one might target a few areas for attention. These include some which are formally embarked upon by self-assembled groups of investigators and some which are highly likely to form: Les Loew’s Virtual Cell (www.nrcam.uchc.edu/index.html), Cardiome (National Simulation Resource, nsr.bioeng.washington.edu; Cardiac Mechanics Research Group, cmrg.ucsd.edu; Computational Medicine and Biology Laboratory, perspolis.bme.jhu.edu; National Biomedical Computation Resource, nbcr.sdsc.edu; the Bionome Resource, bionome.sdsc.edu), Microcirculation (The Microcirculation Physiome Project www.jhu.edu/popel), Lung (e.g., Entelos Inc., www.entelos.com; Respiratory and Physiological Systems Identification Laboratory, lungs.bu.edu; Roger Kamm, me.mit.edu/research/rdkamm.html), Glucose homeodynamics (e.g., Arthur Sherman http://mrb.niddk.nih.gov/sherman/), Intermediary metabolism, G-protein coupled signaling systems, Kidney (Raymond Mejia, mrb.niddk.nih.gov/ray), Neuroscience (Genesis, http://sourceforge.net/projects/genesis-sim/), and Musculoskeletal system (Dava Newman, web.mit.edu/aeroastro/www/people/dnewman/bio.html). The lung is especially targeted because of the close relationship between structure and function^17,18 and the growing epidemic of asthma in our inner cities. For the kidney there is a long history of detailed quantitative modeling for over 40 years. Much of this was compiled in a Renal Parameter Database, formerly available via Gopher, but transiently off the web. Only a part of this is currently available (mrb.niddk.nih.gov/cddb/ which concerns microarray probes).

Technical Working Groups to Develop and Refine the Toolkits

Each of the technical areas requires organization and development to foster the integrated development of the Physiome Project. Although developments are occurring rapidly in all of these areas, the complexity of the project and the need to facilitate interactions between investigators in academia and in industry gives rise to many special considerations that go beyond those practices currently established. These include: (1) database design, (2) modeling methods, (3) web management, (4) communications, (5) funding, and (6) intellectual property.

AN EXAMPLE, A CARDIOME EFFORT

Technical and Psychological Challenges to Modeling as an Objective

Biologists currently are drastically undertrained in the quantitative, analytical and integrative aspects of their field. Success in reductionist fields of molecular and cell biology required observation and deduction. Now many feel that analysis and modeling are of no consequence. How can we knock down the barriers to quantitative analysis in biology? Here are some questions relating to how biologists will work with integrated modeling.

1. Can the technical aspects of the modeling process be simplified?

2. Why are the psychological barriers to quantitative approaches so formidable?

3. Can one easily integrate submodels into large-scale models that function correctly?

4. Can databases for physiological and biological data be developed early so that investigators become used to storing their data in them?

5. Can databases be integrated with simulation systems so that models use the data and so the interfaces serve as entry points into the databases?

6. Can we network investigative groups together more efficiently with Internet II?

The Cardiome Modeling Effort

As one of the areas of biomedical science with a long history of quantitative data gathering, the cardiovascular system lends itself particularly well to serving as a focus of integration. A comprehensive picture is growing; while earlier books emanating from the Cardiome-oriented investigators, e.g., Ref. 19, featured the many excellent individual efforts, the interaction among investigators is becoming more evident in later works edited by McCulloch et al.³¹ and Bassingthwaighte et al.,⁶ in accord with expectations of Ref. 12. Divisions of labor might be established along such lines as: (1) electro-physiology, (2) mechanics, (3) excitation-contraction coupling, (4) flows, transport and exchanges, (5) metabolism of substrates (glucose, fatty acids, TCA cycle, phospho-energetics), (6) signaling, regulation of metabolism, contractile force, responses to injury, etc., and (7) regulation of gene expression in response to local and global activity of myocardium.

What we consider here are approaches for the set of steps beginning with a set of acquired data through to describing the data in terms of model parameters, i.e., reducing the data to minimal form, dependent upon the model as the vehicle for the parsimony. Since models are always wrong, incomplete or incorrect, the parsimonious description is only good if the model is adequate to the situation, is well developed and thoroughly understood, and is mathematically and numerically correct.

This is not a new perspective and needs no stretch from current practice since readers are quite willing to accept the results of measures of blood glucose, of cardiac output, of cardiac ejection fraction, of glucose tolerance tests, PET imaging of glucose metabolism, and many other tests of physiological function which are derived from observations only through model based analyses. For cardiac output, for example, the model most commonly used is the Fick measurement: steady-state, mass conservation, with measures of whole body oxygen consumption and the difference in concentration of oxygen in inflow and outflow from the lung. The basic assumptions cannot actually be attained, but the model results are close enough for clinical purposes. As another example of generally accepted model-based reporting, estimating cardiac ejection fraction by x-ray computed tomography, the modeling is in the differences in geometric form of the left ventricular cavity between the beginning and end of systole. The glucose tolerance test is to interpret the time course of blood glucose levels after ingesting a large dose of glucose in terms of the ability to secrete insulin in response to the load. Positron emission tomographic imaging of fluorodeoxyglucose uptake in brain is interpreted as measuring glucose consumption based on a model of glucose transport and metabolism and of the relationship between deoxyglucose retention and glucose usage. The assumptions inherent to these models are not completely valid, but the answers provided are still highly valued.

At the cellular level several groups are targeting different aspects of myocardial cellular behavior, as suggested in Fig. 7.

FIGURE 7.

Diagram of components of a set of submodels for a contractile cell. Phosphoenergetics is central. Each model or submodule might come from a different institution

A particular achievement is the excitation-contraction coupling modeling of Ref. 41. In a nicely described set of equations, 93 in all at the published stage, Winslow et al.⁴¹ captured the essence of ionic exchange through the pumps and leaks in a myocardial cell during the cardiac cycle. This model, with their permission and following their inspection, is available for exercising on our website. Its elements, added to by Michailova and McCulloch, are shown in Fig. 8. It serves as an important basis for further development, to put this model together with the other elements shown in Fig. 7.

FIGURE 8.

Components of a single cell model for the ionic currents during an action potential, for example, the composite of the models of [29,30], and [41]. Transmembrane voltage is V. (Figure from Michailova and McCulloch, with permission.)

Another set of important vehicles for advancing the cardiome project are representations of the anatomy. The works of Peter Hunter’s group in Auckland on the dog (e.g., Ref. 28) and Vetter and McCulloch on the rabbit³⁸ provide detailed mechanical models based on the three-dimensional descriptions of cardiac anatomy, including the details of fiber directions and of myocardial sheet structure.³⁷ The studies of coronary artery anatomy by Kassab et al.²⁵ and by Zamir and Chee⁴⁴ over several years will when combined with the cardiac structure provide the basis for oxygen delivery to the tissue. Our initial attempts to do this using Kassab’s data show some success in providing the right kinds of spatial profiles in regional flows and in the time course of tracer through the heart^8,9 but much more refinement is needed. Model simplification can be based on the fractal nature of myocardial flows and functions.² The reason that fractal behavior in space allows parsimony in modeling is that near-neighboring regions have similar flows and substrate uptake,¹³ allowing the use of coarser spatial resolution while accounting for regional heterogeneity. Fractal behavior in time, power-law self-similarity of washout processes⁵ likewise allows simplification of the kinetic descriptions of the transport processes.³

The Cardiome and Other Targets

The cardiome is still just beginning as a multi-institutional collaborative project. Others are making headway. The Microcirculation Physiome Project, coordinated by Popel (Johns Hopkins), Pries (Berlin), and Greene (Medical College of Wisconsin) is focusing initially on microcirculatory network structure and flow distributions, setting a framework for physiological processes. Other organ or system projects are in earlier stages.

The cell level efforts can be expected to achieve success earlier. A natural sequence of effort would be quantitative description of cellular structure and function for erythrocytes (no nucleus or mitochondria), for prokaryocytes such as E. coli (no nucleus) and for a eukaryocyte such as yeast, then more specialized cell types, epithelial secretory cells, contractile cells, and others. These could initially be kept to a minimal level of complexity by assuming the absence of protein degradation and of mechanisms for cell maintenance. A critical goal is to relate the regulation of expression to cellular function: these simplistic cell models can serve as a basis for the complex representations with adaptive behavior.

THOUGHTS ON NEXT STEPS

More often than not, one can gain efficiency by going for the specific rather than the general objective. If the questions are such that obtaining answers leads to the development of tools that are general, then getting to some answers demonstrates the utility of those general tools. In this essay I have used the cardiome and the cardiac myocyte as specific examples. For the very limited set of functions shown in Figs. 6 and 7, the “virtual cell” can be tuned to match reality, at least roughly. Being incomplete as a biological model, it will fail in many ways, especially since the level described is still disconnected from the genome. As a model from which to predict long-term pharmacological effects, it is useless and in fact at this point cannot even account for cardiac remodeling. Systems relating to these will have to be added.

So the long-range questions go far beyond anything computable at the moment. There are both biological and technological questions that might be raised. Here are some examples:

1. How have organisms evolved to maximize operating efficiency? By spatial and temporal distribution of its regulatory systems? By high adaptability?

2. Can reverse engineering (knock-out or knock-in or block of enzyme or receptor) reveal functional behavior?

3. Can knowledge of the current state be used to predict outcomes following interventions?

4. Is the complexity of the multipath heterogeneous system so constrained by redundancies in its regulatory components that the dynamics are so reduced so as to produce “homeostasis?”

5. Every reader will think of many additional questions. With good effort and with support for databasing, more and more of them will become answerable.