THE ANATOMICAL RECORD 231:407-415 (1991)

Quantitative Morphology of the Nervous System: Expanding Horizons R.P. BOLENDER, J. CHARLESTON, K. MOTTET, AND J.T. McCABE Departments of Biological Structure (R.P.B) and Pathology (J.C., K.M.), University of Washington, Seattle, Washington; and Department of Anatomy (J.T.M.), Uniformed Services University of the Health Sciences, Bethesda, Maryland

ABSTRACT In this review, we show how some of the recent developments in quantitative morphology (QM) are creating exciting new opportunities for studying the structure of the nervous system. We begin with a brief overview of QM, focusing on the problems neurobiologists are likely to encounter when collecting and interpreting data from tissue sections. Many of these problems, which range from selecting a sampling method to learning the latest methods, are being solved by creating a new generation of research tools. We describe several of these new tools and show how they can be used to assemble new quantitative methods for in situ hybridization, immunocytochemistry, and camera lucida drawings. The review includes examples of how QM is being used to study the brain and concludes with a brief discussion of diagnostic pathology and its need for new quantitative approaches. Biology is rapidly becoming a computational science (Gilbert, 1991). This persistent shift in biology from descriptive to quantitative approaches is being driven by recent breakthroughs in experimental methods and the widespread application of computers. Quantitative morphology has undergone dramatic changes in recent years (Gundersen et al., 1988a; Cruz-Orive and Weibel, 19901, and today virtually all biological structures can be quantified. The purpose of our review is to summarize some of the new methods and, a t the same time, explore some of the advantages of electing a computational approach to neuroanatomy. BASIC QUESTIONS What Is Quantitative Morphology?

Quantitative morphology (QM) includes all data describing biological structures in n-dimensional space, ranging in size from molecules to organisms. Quantitative methods are well established in the basic and clinical sciences and embrace a wide range of imaging modalities. Stereology, serial section reconstruction, computerized tomography (CT), and magnetic resonance imaging (MRI) typically uses different strategies for accessing two- and three-dimensional data, but they share similar problems of sampling and data analysis. Although our discussion focuses on data collected from sections viewed with microscopes, the underlying principles of QM apply to all these imaging methods. What’s Wrong With the Two-DimensionalData of Sections?

The information in sections presents a cruel paradox. Sometimes section data give correct views of biological structures, whereas at other times the views are faulty. Sections allow us to describe microscopic structures qualitatively, but without stereology or serial section reconstruction, the quantitative data of sections are confined to two-dimensional interpretations. This is a 0 1991 WILEY-LISS, INC.

most unfortunate situation because we usually wish to interpret structural changes in three-dimensional space, often in connection with a structural hierarchy (Weibel, 1979; Bolender, 1991a,b). Sectioning a structure to view it in a microscope subtracts one dimension of the structural information, resulting in a narrow two-dimensional view. In fact, the starting point for much of quantitative morphology is a two-dimensional “flatland” lacking the dependability of the familiar three-dimensional world. Once data have been collected from the two-dimensional sections, the investigator is faced with a critical choice. Should the data be interpreted in flatland or should they first be returned to three-dimensional space? Stereology provides a good answer to the question by demonstrating statistically that biological data are far more dependable when interpreted in three-space. Two methods, serial section reconstruction and stereology, offer many ways of returning section data to three-space, thereby recovering most of the information lost by sectioning. Serial sectioning accomplishes this feat by concatenating the raw data of adjacent sections, whereas stereology uses equations and the raw data of random sections. Recall that serial sectioning provides information about individual objects and structures, whereas stereology estimates mean values for structures that may contain one or more objects. What Are QM Data?

The structural plan of biology is quite straightforward. Biological structures consist of many compart-

Received June 15, 1991; accepted July 15, 1991. Address reprint requests to Robert P. Bolender, Dept. of Biological Structure, SM-20, School of Medicine, University of Washington, Seattle, WA 98195.



ments within compartments. A compartment is defined by what it contains and typically ranges in size from molecules to organisms. Each compartment represents a volume defined by a limiting boundary or surface. For example, cells have a volume surrounded by a plasma membrane and the extracellular space is defined by the cells surrounding it. Compartments may include several separate objects (e.g., nuclei) or consist of a single, continuous object (e.g., connective tissue). The data of QM typically include volumes, surfaces, lengths, numbers, shapes, diameters, and thickness. What Are the Critical QM Data?

Usually, three types of data are needed to interpret a structural change in the brain or in some well-defined brain compartment: (1)the volume of the structure, (2) the number of cells in the structure, and (3) the morphological concentrations. The new QM tools provide ready access to all three data types. What Are QM Hierarchies?

The hierarchical organization of biology, described above as compartments within compartments, has several noteworthy characteristics. It provides a descriptive infrastructure for organizing biological data (Morowitz and Smith, 1987). Moreover, it creates opportunities for viewing QM data across the multiple levels of a hierarchy by providing quantitative links among compartments. For example, changes in the mitochondria of a specific cell type can be viewed and interpreted hierarchically by following the changes at each of several levels, including mitochondria, cells, tissues, and organs. Stereological equations, which include the volumes of compartments as variables, define the quantitative links of the hierarchy. Biological complexity seems closely tied to this hierarchical organization. For example, changes that occur in mitochondria-from one hierarchical level to another-probably reflect unique but overlapping sets of conditions and mechanisms. Mitochondria, however, represent only one of the many compartments that may be involved in hierarchical interactions. QM data organized as a hierarchy define a simple and direct strategy for viewing these interactions and offer natural links to a variety of other biological data types (Bolender and Bluhm, 1991). NEW QM TOOLS

In recent years, remarkable progress has been made in our understanding of how to quantify biological structure. The arguments for applying modern methods of QM to the study of the nervous system have never been stronger. Key developments in theory have led to radically improved methods, computer software assists technology transfer, and a comprehensive support structure for biologists is being maintained by professional societies. In this section, we describe some of the new tools of QM and show how they can be used in the lab. The current trend in designing QM methods is to develop new tools from a small, but well-integrated theoretical foundation. As a result, the methods are becoming increasingly modular. When designing an experiment, an investigator can either select single modules to solve typical problems or combine several

modules to “invent” new solutions. The advantage of this approach is that the investigator can find ready solutions to new problems as they arise, without having to wait for the development of new theory. To illustrate how one might use this approach, we describe a few well-established modular methods (sampling, estimating volumes, cell counting) and then try to “invent” some new QM modules for methods commonly used in the neurosciences (in situ hybridization, camera lucida drawings of neurons, immunocytochemistry). QM Modules

Sampling a structure All methods of quantitative morphology are subject to the same rule of unbiased sampling. We define an unbiased sample as one that is selected such that all parts of the structure have the same chance of being sampled. When sampling a highly anisotropic structure such as the brain, stereological methods have had only a modest record of success because it was practically impossible to satisfy the unbiased sampling rule. The recent developments in sampling theory have largely solved this problem, and unbiased sampling methods can be applied to the brain with excellent results. The considerable advantages of unbiased sampling are not widely known in the neurosciences community, and this is most unfortunate. Consider a traditional sampling method, for example, wherein a brain structure is sampled at exactly the same place from animal to animal. Such a sampling scheme provides information about the “same place” but nothing more. It cannot provide an unbiased estimate for the structure being studied, because the unbiased sampling rule is not followed. Of course, we could always argue that it probably doesn’t make any difference whether or not we use an unbiased sampling technique. Today, an expected response to such an argument might be one of the following questions: “Why would you decide not to use unbiased sampling when it gives more reliable results with less work?’ or “Why take the chance of getting a biased result when you no longer have to?” or “Why collect data that are ill-suited for hierarchical interpretations when such interpretations may provide the critical path for detecting complex structural patterns?” A convenient way of making the transition from biased to unbiased sampling is to try out both approaches with the aid of computer simulations (Bolender, 1991b). The lesson to be learned from such an exercise is that biased sampling methods force you to make unrealistic assumptions about the compartments making up a structure, whereas the unbiased ones are free of such assumptions. Determining structural volume The volume of a structure serves as a reference for interpreting a structural change. Currently, the Cavalieri method is the most powerful because it gives the volume of a structure, as well as the volumes of compartments within the structure (Gundersen and Jensen, 1987). Counting objects The task of counting objects in sections becomes much simpler by counting with three-dimensional


probes (DeHoff, 1983). By far the easiest counting method for light microscopy consists of focusing through a thick section (T-50 pm) and counting the objects as they appear in the focal plane (Gundersen et al., 1988b). This method, called the optical disector, is ideal for counting cells that can be easily identified. Physical disectors, which require pairs of sections for the counting can be used with both light and electron microscopy (Sterio, 19841, but they are somewhat harder and more expensive to use. A recent modular solution to the problem of cell counting combines the optical disector with the fractionator (Gundersen 1986; Ogbuihi and Cruz-Orive 1990). The result-the optical fractionator (West et al., 1991)-is a remarkably simple and efficient counting method. A brief example can be used to underscore the importance of counting objects in three-space. In images of brain sections, profiles of neurons and glial cells can be identified and counted in two-space. What do such counts mean? Do counts of cell profiles in twospace equal the counts of whole cells in three-space? We can answer these questions quite easily by restating them as a stereological equation. To wit: Does the number of cell profiles in a unit of tissue section area ( N A , e e l l prof es1 equal the number of cells in a unit of tissue volume (NV(cells))? The expression is given as -

NA,celld = N”ieells)* D,,,,,,,,

where is the mean tangent diameter-a single mean value accounting for all the sizes and shapes of the cells in the population being counted. Now we are ready to do a simple experiment. Imagine three distinct populations of spherical neurons having diameters of 10 pm, 20 pm, 30 pm. If all three populations have exact1 the same number of cells, for example, 1,000 cellskm , then can we expect to find the same number of profiles in the two-dimensional sections? By substituting these data into the above equation, we have our answer.



Fia. 1. itora Dgram from in situ hj Pidization to identify specific cellular messenger ribonucleic acid (mRNA). An oligonucleotide probe (20 nucleotides in length: nucleotides 985-1004, see Ivell and Richter, 1984) complementary to a portion of the untranslated region of the mRNA transcript encoding the neuropeptide oxytocin was endlabeled with 3H-dATP and hybridized to rat brain tissue (method: McCabe et al., 1990). Shown here is an isolated magnocellular neuron in the midlateral hypothalamus. Note the autoradiography grains over the cytoplasm and the process extending from the lower left quadrant of the cell. ~

of total tissue mRNA derived from solution or dot-blot hybridizations, this approach does a reasonable job a t measuring average tissue mRNA across treatment groups (Kelsey et al., 1986; McCabe et al., 1988; Roman0 et al., 1989; Lindfors et al., 1990; White and KerNAieell profiles) = 1,000 cells/cm3 * 10 pm = 1 cell profile/cm2 shaw, 1990). NAieell profiles) = 1,000 cells/cm3 * 20 pm = 2 cell profiles/cm2 The images from in situ hybridization are derived NAi,,,, profilesl = 1,000 cells/cm3 * 30 pm = 3 cell profiles/cm2 from sections in which cell profiles contain labeled cells Clearly, the counts of cell profiles in two-space do not usually identified by autoradiography (see Fig. 1). We equal the counts of the whole cells in three-space. The would like to have a QM method for estimating the results show that the cell diameter determines the like- number of grains (mRNA molecules) in three-dimenlihood of it being included in a section. Larger cells sional cells to account for transcript level. Using the appear in sections more often than smaller ones; they modular approach, we can estimate the mean volume simply present a larger target for the knife. This ex- of a specific cell V(cell)with the nucleator (Gundersen, plains why the three-dimensional probes of stereology 1988) or from cell counts (Bolender, 1991b), count the or serial section reconstruction is needed to count cells grains over the cell(s) to get the grain density NA(gra,ns, cell profile), and finally relate the grain density to a cell in sections. volume that represents the amount of the section reIn situ hybridization sponsible for the radioactive source. This section volIn situ hybridization identifies viral and endogenous ume is calculated as the area of the cell profile times nuclear and cytoplasmic nucleic acids in cells (McCabe, the depth of the radioactive source. Although this may 1990). In neurobiology, a major application is to quan- result in a relatively crude estimate of grains per sectify specific mRNAs following experimental treatment. tion volume, it should be sufficient for detecting experSeveral reports outline what are essentially brute force imental changes for a given experiment. We estimate methods where autoradiography grains are counted the total number N of grains (molecules) in a cell as over all single cells or over complete fields of labeled v i c e l l ~ * [email protected]~nsieell profile) * lit,depth of radioactive soureel N

Quantitative approaches to neuroscience research. 1990 annual meeting of the Neuroscience Society.

THE ANATOMICAL RECORD 231:407-415 (1991) Quantitative Morphology of the Nervous System: Expanding Horizons R.P. BOLENDER, J. CHARLESTON, K. MOTTET, A...
1MB Sizes 0 Downloads 0 Views