A stereological approach to human cortical architecture: identification and delineation of cortical areas

https://doi.org/10.1016/S0891-0618(00)00076-4Get rights and content

Abstract

Stereology offers a variety of procedures to analyze quantitatively the regional and laminar organization in cytoarchitectonically defined areas of the human cerebral cortex. Conventional anatomical atlases are of little help in localizing specific cortical areas, since most of them are based on a single brain and use highly observer-dependent criteria for the delineation of cortical areas. In consequence, numerous cortical maps exist which greatly differ with respect to number, position, size and extent of cortical areas. We describe a novel algorithm-based procedure for the delineation of cortical areas, which exploits the automated estimation of volume densities of cortical cell bodies. Spatial sampling of the laminar pattern is performed with density profiles, followed by multivariate analysis of the profiles‘ shape, which locates the cytoarchitectonic borders between neighboring cortical areas at sites where the laminar pattern changes significantly. The borders are then mapped to a human brain atlas system comprising tools for three dimensional reconstruction, visualization and morphometric analysis. A sample of brains with labeled cortical areas is warped into the reference brain of the atlas system in order to generate a population map of the cortical areas, which describes the intersubject variability in spatial conformation of cortical areas. These population maps provide a novel tool for the interpretation of images obtained with functional imaging techniques.

Introduction

Stereological procedures are efficient tools, if quantitative analyses of the microstructural organization of the cerebral cortex are to be performed (e.g. Haug, 1987, Pakkenberg and Gundersen, 1988, Pakkenberg and Gundersen, 1997; for review see Schmitz et al., 2000). A prerequisite for such analyses is the precise and reliable definition of a reference space in which the stereological parameters will be estimated (Howard and Reed, 1998). This definition is not only necessary for calculating absolute cell numbers from densities and volumes, but also for adequate sampling, since each volume element of the reference space must be sampled with equal probability. If sampling is focused on a putative ‘core’ volume of the reference space while neglecting the more peripheral parts, the estimate will be biased, if the structure of interest (e.g. a certain cell type or synapses) is not distributed homogeneously throughout the reference space. The well known structural and functional heterogeneity of the cerebral cortex frequently requires the definition of structurally and/or functionally distinct reference spaces, such as the precise and reliable parcellation of the cerebral cortex into subunits (areas). Thus, stereological tests of specific hypotheses can only be performed if the borders between cortical areas are unambiguously defined.

Cortical areas are delineated by the pial surface, the cortex/white matter transition and their borders to neighboring areas. The pial surface can easily be defined in histological sections. Depending on the local curvature, the cortex-white matter border may be more difficult to define, but this boundary is obvious by a steep decrease in cell density at the transition from cortical layer VI to the white matter in most cortical areas. The crucial step in defining a cortical area is the localization of borders to neighboring areas. The precise and reliable localization of such borders is the main topic of the present article.

Since the pioneering cytoarchitectonic studies of Brodmann, 1909, Brodmann, 1914, and von Economo and Koskinas (1925), cytoarchitectonically defined cortical areas have been correlated with specific functions. Vogt and Vogt (1919) and numerous following studies demonstrated by combining architectonical, electrophysiological and tracing studies in non-human primates that this is a valid hypothesis. Recent combined functional imaging and cytoarchitectonic studies using a common spatial reference system demonstrated such correlations in the human brain (Larsson et al., 1999, Naito et al., 1999, Bodegård et al., 2000a, Bodegård et al., 2000b).

In the context of a stereological study, a cortical area is defined as a cortical tissue volume characterized by a homogeneous microstructural organization (architecture). Thus, an areal border has to be established at locations where the architecture of the cortex changes significantly. All anatomical maps of the human cerebral cortex are based on this axiom, but they suffer from at least one of three shortcomings:

  • The maps do not reflect the intersubject variability in size of cortical areas and location of their borders. Several architectonic studies, however, indicatethe existence of a considerable degree of intersubject variability (Filimonoff, 1932, Geyer et al., 1996, Geyer et al., 1999, in press; Roland et al., 1997, Zilles et al., 1997, Roland and Zilles, 1998, Amunts et al., 1999, Amunts et al., 2000).

  • The areal borders were established using pure visual inspection of histological sections, and thus, were dependent from varying observer-dependent conditions, e.g. individual abilities in pattern recognition (for discussion see Lashley and Clark, 1946, Bailey and von Bonin, 1951).

  • In most cases, the cortical maps are based on different projection or unfolding procedures and are frequently two-dimensional (2D), highly schematic drawings. These drawings, however, do not provide the spatial information necessary for correlations with the recently developed functional imaging techniques in the living human brain.

As a result, striking differences between the maps of different authors in terms of number, localization, extent and contour of cortical areas can be found in the literature.

It would be a straightforward approach to locate areal borders using macroscopical landmarks as defined by topographical features of the cortical surface (Rademacher et al., 1993). But do macroscopical landmarks as indicated by the sulcal and gyral pattern yield reliable clues to the location of the borders of a cortical area? Recent investigations (Geyer et al., 1996, Geyer et al., 1999, Geyer et al., 2000, Amunts et al., 1999, Amunts et al., 2000) clearly demonstrated that areal borders do not consistently coincide with sulcal bottoms or other macroscopical landmarks. Consequently, the definition of areal borders must be performed in each individual brain using a standardized, reproducible and statistically testable observer-independent method.

One of the key features of cortical microstructure is its organization in layers running parallel to the cortical surface. In histological specimens stained for cell bodies, this feature is depicted by a sequence of layers differing in cell density and cell size (laminar pattern). Areal borders are located at the transition of the laminar pattern of one area to that of the neighboring area, assuming that each area has a unique, specific laminar pattern (architecture). In classical cytoarchitectonic investigations, this change in laminar pattern and cellular composition was localized by purely qualitative, light microscopical inspection (e.g. Brodmann, 1909).

An early approach to the quantification of a cytoarchitectonic laminar pattern in the cerebral cortex was described by Hudspeth et al. (1976). These authors analysed optical density (OD) profiles to describe the distribution of cell density across cortical layers in the human primary visual cortex and introduced the term ‘fingerprint’ for the quantified layering pattern of a cortical area. Density profiles based on OD, however, suffer from a major shortcoming, i.e. the OD value as measured with microdensitometry in a small measuring field is dependent on the staining intensity of the background and of the cellular elements. In addition, OD measurements in histological specimens using microdensitometry do not fulfil the requirements of the Lambert–Beer law which relates the concentration of a homogeneously dissolved chromophore to the measured OD value (Bitensky, 1980). Therefore, we used the most basic stereological parameter, the volume density, to quantify the laminar pattern. This parameter has a long tradition in quantitative neurobiology. It is defined as the volume fraction of a phase (i.e. the neuronal cell bodies) in relation to the reference volume. A first quantitative application to cortical microstructure using the ‘gray cell coefficient’ was described by Haug (1956), who extended a previous concept of von Economo and Koskinas (1925). Volume density offers several advantages (Schleicher et al., 1986). One of these advantages is that estimates of volume density are not affected by variations of staining intensity. In addition, it is independent on the degree of anisotropy (deviation from directional randomness in three dimensions; Weibel, 1979). These are major requirements for stereological analysis of the cerebral cortex, which takes place in histological specimens with varying staining intensity and in a highly anisotropic structure. Moreover, using volume density measures, already existing and — because of the enormous time and efforts necessary for their production — extremely valuable collections of serially sectioned whole human brains can be analyzed. New, design-based procedures such as vertical sectioning (Braendgaard and Gundersen, 1986) are not required.

Light microscopical estimates of the volume density of neurons from relatively thick histological sections are biased due to overestimation caused by projection (Howard and Reed, 1998). Using TV-based image analyzers (for a first application see Adhami, 1973), volume densities of neurons are further biased by the contribution of non-neuronal elements like glia and endothelial cells, which cannot be reliably identified using automated image analyzing procedures in Nissl-stained sections. Both effects were studied by Wree et al. (1982). According to their findings in various brain regions and cortical areas, the areal fraction of cell bodies as measured with an image analyzer is highly correlated with the volume density of neurons, since the density of endothelial and glial cells does not vary significantly throughout the cortical layers, and therefore represents a relatively constant, additive contribution to the volume density of neurons. This volume density of neurons measured as an areal fraction of all stained cellular profiles in square measuring fields of 20–30 μm was defined as grey level index (GLI; 0≤GLI≤100%).

Using the GLI-procedure, cortical maps of various species were defined (e.g. Zilles et al., 1978, Zilles et al., 1980, Zilles et al., 1982, Fleischhauer et al., 1980, Wree et al., 1981, Wree et al., 1983). Its main advantage is a substantially improved imaging of the cortical laminar pattern as compared to conventional, visual inspection. The most crucial step, however, the identification of cortical borders, was still left to the interpretation of the investigator in these early studies. A substantial progress was introduced by combining the GLI-approach with the analysis of cortical density profiles.

The extraction of density profiles from GLI images reduces information from 2D images to one dimensional signals. Density profiles describe the course of the GLI along a measuring line or traverse through the cortex. The laminar pattern of the neocortex is best described by traverses running from the pial surface to the cortex-white matter crossing all cortical layers perpendicularly. GLI profiles were used to analyze various aspects of cytoarchitecture, such as the postnatal development of the human primary motor cortex (Amunts et al., 1995). Profiles extracted from different cortical areas differ in shape and have been used to compare cortical areas (Sauer, 1983) and to analyze left-right asymmetries in architecture of cortical areas (Amunts et al., 1996, Amunts et al., 1997). The concept of density profiles can be applied to all image modalities which reflect the laminar architecture of the cerebral cortex, such as the density of myelinated nerve fibers (Zilles and Schleicher, 1993) or the laminar distribution of transmitter receptors (Zilles, 1991, Zilles, 1992, Zilles and Schleicher, 1995, Geyer et al., 1997).

We extended the concept of density profiles to a novel approach for an algorithm-based, observer independent procedure for the localization of areal borders (Schleicher et al., 1995, Schleicher et al., 1998, Schleicher et al., 1999). Profiles sampled from a homogeneous area can be expected to be similar in shape. Significant dissimilarities in shape can be found between profiles on opposite sides of an areal border. The locations of borders are found by sampling cortical regions with equidistant traverses and analyzing the degree of dissimilarity between the GLI profiles by calculating distances between groups of neighboring profiles.

As detailed above, conventional anatomical brain atlases do not provide 3D information on the exact localization of areal borders and do not account for their intersubject variability. Thus, the volume of cortical areas cannot be determined using such atlases, and 3D data sets of different brains cannot be matched to these atlases. Probabilistic atlases are, therefore, needed to take the interindividual variability between human brains into account (Roland and Zilles, 1994, Mazziotta et al., 1995). The value of such an atlas for the structural interpretation of functional imaging studies depends on the number of histologically processed brains integrated in the atlas, on the quality of the spatial normalization procedures applied and on and the precision and reliability of the cortical parcellation method.

Section snippets

Histology

Every 15th section of coronal, sagittal or horizontal, 20 μm thick serial sections through formalin-fixed and paraffin-embedded complete human brains was processed with a silver staining technique (Merker, 1983) for cell bodies. This stain leads to a high contrast between background and cellular profiles. This allows a precise segmentation of cellular profiles using grey-value thresholding (see below). Because each hemisphere was completely sectioned in only one of the three planes, mapping of

The somatosensory cortex

The cortical region of interest indicated in Fig. 1 contains four of Brodmann's (1909, 1914) cytoarchitectonically defined areas and was analysed in ten brains (four males and six females, ages between 39 and 85 years). Details of the histological procedure, sampling and data acquisition can be found in Geyer et al. (1999). The borders between areas 4, 3a, 3b, 1 and 2 were defined using the procedure described in the present paper (Fig. 4 for details and the example of a parcellation pattern in

Discussion and conclusions

Quantitative cytoarchitectonic definition of areal boundaries and comparison of their laminar patterns were based on (i) GLI as a stereological parameter, (ii) central moments as descriptors for the shape of the laminar pattern, and (iii) the Mahalanobis or the Euclidean distances as multivariate measures of dissimilarity.

Acknowledgements

This work was supported by a grant from the Deutsche Forschungsgemeinschaft (SFB 194/A6) and the Human Brain Project (NIMH).

References (79)

  • P.E. Roland et al.

    Structural divisions and functional fields in the human cerebral cortex

    Brain Res. Rev.

    (1998)
  • A. Schleicher et al.

    A quantitative approach to cytoarchitectonics: software and hardware aspects of a system for the evaluation of structural inhomogeneities in nervous tissue

    J. Neurosci. Methods

    (1986)
  • A. Schleicher et al.

    Observer-independent method for microstructural parcellation of cerebral cortex: a quantitative approach to cytoarchitectonics

    Neuroimage

    (1999)
  • R.K. Snider et al.

    Classification of non-stationary neural signals

    J. Neurosci. Methods

    (1998)
  • A. Wree et al.

    Estimation of volume fractions in nervous tissue with an image analyzer

    J. Neurosci. Methods

    (1982)
  • K. Zilles

    Codistribution of receptors in the human cerebral cortex

  • K. Zilles

    Neurotransmitter receptors in the forebrain: regional and laminar distribution

    Prog. Histochem. Cytochem.

    (1992)
  • K. Zilles et al.

    Cyto- and myeloarchitecture of human visual cortex and the periodical GABAA receptor distribution

  • H. Adhami

    Die photometrische Bestimmung des Cortexzell- und Graugehalts auf der Grundlage des Nissl-Bildes

    Acta Anat.

    (1973)
  • K. Amunts et al.

    Postnatal development of the primary motor cortex: a quantitative cytoarchitectonic analysis

    Anat. Embryol.

    (1995)
  • K. Amunts et al.

    Postnatal development of interhemispheric asymmetry in the cytoarchitecture of human area 4

    Anat. Embryol.

    (1997)
  • K. Amunts et al.

    Broca's region revisited: cytoarchitecture and intersubject variability

    J. Comp. Neurol.

    (1999)
  • E. Armstrong et al.

    The ontogeny of human gyrification

    Cereb. Cortex

    (1995)
  • P. Bailey et al.

    The Isocortex of Man

    (1951)
  • P.H. Bartels

    Numerical evaluation of cytologic data II. Comparison of profiles

    Anal. Quant. Cytol.

    (1979)
  • P.H. Bartels

    Numerical evaluation of cytologic data. VII. Multivariate significance tests

    Anal. Quant. Cytol.

    (1981)
  • L. Bitensky
  • A. Bodegård et al.

    Object shape differences reflected by somatosensory cortical activation

    J. Neurosci.

    (2000)
  • A. Bodegård et al.

    Somatosensory areas in man activated by moving stimuli. Cytoarchitectonic mapping and PET

    NeuroReport

    (2000)
  • S.T. Bok et al.

    The size of the body and the size and the number of the nerve cells in the cerebral cortex

    Acta Neerl. Morphol.

    (1939)
  • Brodmann, K., 1909. Vergleichende Lokalisationslehre der Großhirnrinde in ihren Prinzipien dargestellt auf Grund des...
  • K. Brodmann

    Physiologie des Gehirns

  • K.R. Castleman

    Digital Image Processing

    (1979)
  • W.J. Dixon et al.

    BMDP Statistical Software Manual

    (1988)
  • I.N. Filimonoff

    ber die Variabilität der Großhirnrindenstruktur. Mitteilung II-Regio occipitalis beim erwachsenen Menschen

    J. Psychol. Neurol.

    (1932)
  • K. Fleischhauer et al.

    A revised cytoarchitectonic map of the neocortex of the rabbit (Oryctolagus cuniculus)

    Anat. Embryol.

    (1980)
  • F. Gallyas

    Silver staining of myelin by means of physical development

    Neurol. Res.

    (1979)
  • S. Geyer et al.

    Two different areas within the primary motor cortex of man

    Nature

    (1996)
  • S. Geyer et al.

    Receptor autoradiographic mapping of the mesial motor and premotor cortex of the macaque monkey

    J. Comp. Neurol.

    (1998)
  • Cited by (112)

    • Convolutional neural networks for cytoarchitectonic brain mapping at large scale

      2021, NeuroImage
      Citation Excerpt :

      The predictions produced by the proposed model typically include some remaining errors (Figs. 9, 12), which require careful interpretation. Close to brain region boundaries, it must be noted that the GLI-based reference annotations enforce straight lines to model the border, which reflects the vertical arrangement of neurons in columns as a major principle in cortical organization (Schleicher et al., 2000; 1999). This may lead to discrepancies with the present segmentations, that do not enforce such constraints (see for example Fig. 12).

    • Organization of the macaque monkey inferior parietal lobule based on multimodal receptor architectonics

      2021, NeuroImage
      Citation Excerpt :

      The position of all borders, including those with areas not included within the region of interest (e.g., rostrally with somatosensory area 2 or dorsally with intraparietal areas LIPd or AIP), was confirmed by means of an observer-independent and statistically testable method based on the analysis of interareal differences at the laminar level (Palomero-Gallagher and Zilles, 2018; Schleicher et al., 2005). For detection of cytoarchitectonic borders, GLI profiles (Schleicher et al., 2000; Zilles et al., 2002b) oriented vertically to the cortical surface were extracted to quantitatively describe the laminar distribution of the volume fraction of cell bodies. The shape of each GLI profile was parametrized to create a feature vector which can be used to quantify the degree of similarity or dissimilarity in the shape of the corresponding profiles, and the Mahalanobis distance (Mahalanobis et al. 1949) was used as the distance function.

    • Multimodal 3D atlas of the macaque monkey motor and premotor cortex

      2021, NeuroImage
      Citation Excerpt :

      Blocks of profiles were used instead of single profiles, since the latter were affected by local structural inhomogeneities such as blood vessels or cortical columns, which reduced the signal-to-noise ratio of the distance function (Schleicher et al. 2009). To confirm and accept maxima of the distance functions as statistically significant borders, we applied Hotelling's T2 test in combination with a Bonferroni adjustment of the P-values for multiple comparisons, and threshold was set at (p < 0.01) (Schleicher et al., 1999, 2000, 2005; Zilles et al., 2002b). Significant maxima identified with multiple block sizes in one section were biologically evaluated by comparison with maxima at comparable locations in three following sections to exclude maxima caused by artifacts (e.g. ruptures, folds or local discontinues in microstructure due to blood vessels.

    View all citing articles on Scopus
    View full text