Strides in the technology of systems physiology and the art of testing complex hypotheses

Abstract

Scientific understanding in physics or physiology is based on models or theories devised to describe what is known, within the limits imposed by observation error. Carefully integrated models can be used for prediction, and the inferences assessed via further experiments designed to test the adequacy of the theory summarizing the state of knowledge. This is the systems approach, the basis of theoretical physiology; the models, like those of theoretical physics, should be firmly based on fundamental reproducible observations of a physical or chemical nature, held together with the principles of mathematics, logic, and the conservation of mass and energy. Modern computing power is such that comprehensive models can now be constructed and tested. For this approach data sets should include as many simultaneously obtained items of information of differing sorts as possible to reduce the degrees of freedom in fitting models to data. By taking advantage of large memories and rapid computation, modular construction techniques permit the formulation of multimodels covering more than a single hierarchical level, and thereby allow the investigator to understand the effects of controllers at the molecular level on overall cell or organism behavior. How does this influence the research and teaching practices of physiology? Because the computer also allows a new type of collaboration involving the networking of ideas, data bases, analytical techniques, and experiment designing, investigators in geographically distributed individual laboratories can plan, work, and analyze in concert. The prediction from this socioscientific model is therefore that networked computer-based modeling will serve to coalesce the ideas and observations of enlarging groups of investigators.

The contributions of the five scholars and teachers comprising this symposium cover a wide range of topics from subcellular biochemical physiology to therapeutic strategy at the bedside. These scholars come from not only departments of physiology, but also departments of computer science, chemical engineering, and clinical medicine. The composite effort shows that modern physiology is diverse, comprehensive, and yet integrative (see Fig. 1).

Figure 1.

Conceptual relationships surrounding physiological studies.

The tasks that these scholars have undertaken had small beginnings, and grew in magnitude and complexity, for example, to encompass large sets of biochemical reactions, or multilevel systems for transport, accumulation, and regulation of the internal ionic milieu. That each of these teacher-investigators understands computational methodology is perhaps a foregone conclusion. But skillful use of computers is not the fundamental common thread that links these investigators. It is rather their desire to integrate bodies of knowledge into comprehensive self-consistent form (see Fig. 2). This is the heart and soul of physiology, the study of function of living species.

Figure 2.

Integrative analysis requires multilevel conceptualization.

An integrated viewpoint defines a model, whether it be qualitative or quantitative, verbal, or mathematical. Any composite model is made up of a set of submodels, each of which is made up of submodel components. A general strategy in composing large-scale models is that the form of the components and submodels is not changed by incorporating them into the composite model, as emphasized by Garfinkel et al. in this symposium. This is to suggest that in an idealized situation, components, submodels, and composite models are at three different hierarchical levels in the structure of the system, and further, that the component or submodels can be defined explicitly and completely without regard to the form of the composite model.

In the span of hierarchical levels of systems ranging from the formation of galaxies to quarks and charms, we in physiology concentrate on a narrow portion of the total spectrum (see Fig. 3). The left column identifies a hierarchical level in broad terms, and the two right columns illustrate cardiovascular functional hierarchical trees, one related to the heart’s function to produce flow with each beat, and the other to substrate exchange in metabolism. Generally speaking, in this scheme, the level of the hierarchy is associated with the physical size of the unit, but as we approach the lowest level, in the illustrated cases, ion vs. substrate transport, energy utilization vs. production, we reach the stage of commonality between the two chains at the level of molecular transformation.

Figure 3.

Hierarchical levels in cardiac mechanical function (middle column) and substrate transport (right column) merge at the level of molecular reactions. The range of hierarchical levels is so broad that one can work normally over only a narrow range.

This unidirectional hierarchical structuring appears to be straightforward enough until you try to apply it. Taking the illustration of glucose consumption in the myocyte (Fig. 4), one can easily argue that membrane transport kinetics and intracellular biochemical reactions are at an appropriate submodel level, and that transport affinities, the regulation of transporter numbers, and the regulation of glucose formation as well as its metabolism are all at the level of regulated enzymatic transformations. The problem comes when one looks at the elements of the regulatory processes, namely, that glycogenolysis is related to cyclic AMP formation, which in turn may be regulated by events quite outside the cell. Thus the nice, simple structure breaks down, because it requires modulation of the behavior of a model subcomponent via an agonist that may be introduced via neural stimulation. The lesson here is that the model’s subcomponents should be built into our multihierarchical level models so that the modulation can be accomplished without making fundamental changes in the submodel component itself.

Figure 4.

Regulatory components traverse multiple hierarchical levels.

Another way of looking at this is shown in Fig. 5. At the lowest levels of modeling physical and chemical processes, we may be able to write explicit equations with fairly precisely known energies of activation and so on. At the next level, however, it is usually recognized that the local environments for these physical chemical models are different, even at different places within the same cell, and so their behavior may be different. If we take this to the subcellular-cellular functional level where aggregates of processes are involved, heterogeneity within the single cell, apart from subcellular compartmentalization, is usually not an issue because the chemical controllers will tend to be similar throughout all the regions of the cell. For multicellular preparations, the highest one illustrated in the diagram, for an intact organ, the processes have to be recognized to be generally heterogeneous because now the cells will have external environments that are different in different parts of the organ, with different demands on them, and different blood flows, and therefore at the multicell or organ level the controllers would have to be regarded as heterogeneous. The strategy for dealing with heterogeneous systems is twofold. One is to simplify the system description so that it appears homogeneous; the other is to gather as much data as possible to define the heterogeneity and then deal with it explicitly. The latter is almost impossible when multiple heterogeneities are involved, for example, in any or all of flows, local volumes, local reaction rates, etc., and therefore the compromise is to account for heterogeneity where one can, and to consider the others as homogeneous. This viewpoint can be strengthened if it can be shown what the error in assuming homogeneity might be for each one of the simplifications involved. Our purpose in this symposium is to demonstrate that highly complex systems can be managed with modern technology, and that it is important to consider the systems as realistically as possible to obtain correct and enduring ideas of how they work.

Figure 5.

Generality for hierarchical schema relating to physiological functions of an organ. Because the high-level systems are inherently very complex, for practical considerations models of them must be simplifications.

FUTURE NEEDS

As the complexity of the systems begins to emerge, it becomes more difficult for individual investigators in their own laboratories to cover all aspects of the systems. To solve this problem we should first have another look at the training of young physiologists, and second, plan for a national support scheme to aid physiological investigation.

1. The two aspects of physiological training that go beyond traditional training are the following:

   1. More intensive and extensive training in using quantitative tools for solving problems

      1. Formulating precise hypotheses
      2. Defining systems descriptions
      3. Using computational tools
      4. Learning control systems theory
   2. Training for teamwork: the other aspect is a broader sociological one, namely, training for undertaking research in teams. Because physiologists, even though they may believe that their training is broad compared with many disciplines, are still relatively narrowly trained, it is difficult for many to collaborate equally well with clinicians, engineers, and mathematicians. The training should therefore involve

      1. Developing cross-disciplinary collaborations
      2. Sharing responsibilities in solving problems
      3. Focusing on goals and packaging completed tasks cleanly
      4. Maintaining wide peripheral vision while probing in depth

      It is not that these features are neglected wholly in traditional training, but if these are kept as overt goals rather than covert or subsidiary goals in the training, our programs will be strengthened.
2. Resource facilities for service to physiological investigation. Although it is true that computational capabilities in individual laboratories and universities are growing at a miraculous rate, putting the power of the new technology into the hands of the new or the experienced investigator is not accomplished by merely delivering a new desktop Cray Multiprocessor. I therefore advocate the establishment of a set of simulation resources covering the fields of physiology and investigative medicine, each of which can help investigators to develop their programs in particular areas. Each such resource facility might provide the following:

   1. Consultant services for simulation analysis

      1. How to use models in experiment design
      2. Strategies for model fitting to data
      3. Model reduction techniques
   2. Resources for simulation analysis

      1. Facilities with special focuses
      2. Developers of models in limited fields
      3. Exporters of cleanly coded, well-documented subroutines as submodels
   3. Networking the various resource facilities for the purposes of

      1. Communication with respect to general problems
      2. Porting of subroutines of overlapping interest
      3. Maintenance of standardized approaches
      4. Use by the physiological community

These recommendations in training and in service to physiological investigation are really suggestions for getting across the technology barrier. The goal is to integrate facts into comprehensive, self-consistent schema describing cellular organ and body functions. The training and the tools are not easily come by, but are much needed for future investigators and for physiology as an integrating discipline.