Cytoarchitectonical analysis and probabilistic mapping of two extrastriate areas of the human posterior fusiform gyrus

Ten human postmortem brains (Amunts et al. 2000), five males and five females (Table 1), were obtained from the body donor program of the Institute of Anatomy, University of Düsseldorf. One brain came from a subject with transitory motor disturbance (brain 3), all other donors had no history of neurological or psychiatric diseases. Handedness of the subjects was unknown. We may assume that most of the subjects were right-handed supposing a prevalence of left-handedness of <10 % in the general population (Annett 1973). The brains were removed from the skulls within 8–24 h after death and fixated in 4 % formalin or Bodian’s fixative for at least 6 months. To document brain size and shape before the inevitable distortions by histological processing occurred, each brain was scanned using a T₁ weighted 3D-FLASH sequence (flip angle 40°, TR = 40 ms, TE = 5 ms) implemented on a Siemens 1.5T scanner (Erlangen, Germany). Subsequently, the complete brains were embedded in paraffin and serially cut into coronal sections of 20 μm thickness. Every 15th section was mounted on glass slides and silver stained for cell bodies (Merker 1983) to achieve a high contrast between darkly stained neuronal perikarya and unstained neuropil. Every fourth stained section, i.e. every 60th section of the series of sections, was examined, resulting in a distance of 1.2 mm between the analyzed sections (Fig. 2).

Cytoarchitectonical analysis was performed using a quantitative method for observer independent and statistically testable detection of cortical borders (Zilles et al. 2002; Schleicher et al. 2005, 2009). First, rectangular regions of interest covering the posterior fusiform gyrus and neighboring cortex were defined in the histological sections and digitized using a microscope with a scanning stage (KS400; Zeiss, Oberkochen, Germany) and a CCD camera (Sony, Tokyo, Japan; resolution 1.01 × 1.01 μm²/pixel) (Fig. 2b). The digitized sections were then transformed into gray level index (GLI) images, in which pixel values represent the volume fraction of cell bodies in the corresponding square measuring fields of 20 × 20 μm² each (Schleicher et al. 2000, 2005). If the GLI amounts to 20 %, for example, 80 % of the volume in a measuring field are occupied by neuropil (dendrites, axons, processes of glial cells and blood vessels).

Equidistant GLI profiles were extracted along curvilinear trajectories oriented perpendicular to the cortical layers and running from an interactively defined outer contour between layer I and layer II to an inner contour between layer VI and the white matter (Fig. 2c). These profiles represent the course of the regional cell density from superficial (outer contour) to deep (inner contour). To compensate for variations in cortical thickness, we resampled each profile with linear interpolation to a standard length corresponding to a cortical thickness of 100 %. The shape of each GLI profile was quantified by a vector consisting of ten features based on central moments, which were also used for previous cytoarchitectonical studies (e.g. Amunts et al. 2000; Amunts and Zilles 2001; Zilles et al. 2002). These features were: the mean GLI value, the position of the center of gravity on the profile curve (cortical depth), the standard deviation of the mean GLI (indicating the variability of the GLI throughout all layers), skewness and kurtosis of the profile curve and the respective features from the profile’s first derivative (Schleicher et al. 1999).

Differences in shape between GLI profiles indicate differences in cytoarchitecture and were quantified as the Mahalanobis distance (Mahalanobis et al. 1949; Bartels 1979) between the respective feature vectors of neighboring blocks of profiles (Schleicher et al. 2005) at every position along the cortical ribbon (Fig. 3). To assure reliability, the procedure is repeated for different block sizes ranging from 8 to 24 profiles per block. Areal borders are expected at positions where the distance function shows local maxima corresponding to a great dissimilarity in laminar pattern between adjacent blocks of profiles. These maxima were detected and their significance evaluated by the Hotelling’s T² test with Bonferroni correction for multiple comparisons (Fig. 3b). Cortical borders were confirmed, if they were consistently present at the same position across several block sizes, and if the positions were found at comparable sites in adjacent sections.

3D reconstructions of the histological volumes were computed using the following three datasets: (i) the previously ascertained 3D-MRI scan, (ii) images of the paraffin block face obtained during sectioning for the precise alignment of the histological sections and (iii) the digitized images of the cell body-stained sections (Amunts et al. 2004).

The defined borders of the cortical areas were interactively traced on the corresponding sections of the 3D-reconstructed brains. The histological volumes were then spatially normalized by registration to the stereotaxic space of the Montreal Neurological Institute (MNI) (Evans et al. 1992) using a combination of affine transformations and nonlinear elastic registration (Hömke 2006). To keep the anterior commissure as the origin of the coordinate system and for consistency with previous cytoarchitectonical studies (e.g. Amunts et al. 2005; Rottschy et al. 2007; Caspers et al. 2008; Scheperjans et al. 2008; Kurth et al. 2010), data were shifted by 4 mm caudally (y-axis) and 5 mm dorsally (z-axis) to the ‘anatomical MNI-space’ (Amunts et al. 2005). Corresponding areas of the ten brains were superimposed and a probabilistic map was generated for each area (Zilles et al. 2002). These maps indicate for each voxel of the reference brain the relative frequency with which a respective area was found at that position (Eickhoff et al. 2007; Zilles and Amunts 2010).

The probability maps of all areas in the ventral visual cortex were visualized as a continuous, non-overlapping map of this region by generating a maximum probability map (MPM). The MPM was computed by comparing the probabilities of all areas in each single voxel and by assigning to each voxel the cytoarchitectonical area with the highest probability (Eickhoff et al. 2005, 2006). If two or more areas showed equal probabilities in a single voxel, this voxel was assigned to the area with the highest average probability of the directly adjacent voxels.

The volumes of the cytoarchitectonical areas were calculated for each hemisphere separately based on area measurements in the individual histological sections, section thickness and distance between the measured sections as well as the shrinkage factor of each brain. Shrinkage factors were determined as the ratio between the fresh volumes of the brains and their volumes after histological processing (Amunts et al. 2005). The volumetric data were then analyzed for interindividual and interhemispheric differences using a repeated measurement analysis of variance (ANOVA) with the following design: between-subject factor: gender; within-subject factors: area and side; blocking factor: subject.

Both areas belong to the homotypical isocortex with an inner granular layer IV but with further distinctive cytoarchitectonical features:

FG1 was characterized mainly by a marked columnar arrangement of small pyramidal cells and a lower cell density in layer IV (Fig. 5) when compared to neighboring areas. Layer II showed a rather low density of cells and smooth transition into a layer III of low to moderate cell density. The sizes of pyramidal cells in layer III were small with no superficial to deep increase. These pyramidal cells were arranged in clearly visible cell columns, a feature that was unique to FG1 compared to the adjacent areas. Layer IV of FG1 was thin, but clearly delineable from adjacent layers IIIc and V. Pyramidal cells in layer V were small and showed a columnar arrangement as well. The cell sparse layer VI had no conspicuous border with lower layer V. The border between cortex and white matter was blurred.

FG2 showed large pyramidal cells in lower layer III and a prominent layer IV as characteristic features (Fig. 5). A columnar arrangement of pyramidal cells was not found, neither in layer III nor in layer V. Layer II of FG2 had a rather high cell density with a clearly delineable border to the cell sparse layer IIIa. Large, typical pyramidal cells of layer III were primarily found in sublayers IIIb and IIIc with a slight increase of cell sizes in deep sublayer IIIc. This sublayer often merged with the broad and cell dense layer IV. Layer V consisted of equally distributed pyramidal cells, which were smaller than those in lower layer III but larger than those of FG1. Layer V was clearly delineable from the cell dense layer VI, which again had a distinct border to the white matter.

To demonstrate the variability of these cytoarchitectonical features of FG1 and FG2 along the rostro-caudal extent, Fig. 6 illustrates three representative sections of a single brain. The intersubject variability of cytoarchitectonical features is illustrated by sections of three different brains (Fig. 7). Figures 6 and 7 highlight that the main characteristic features of each area, e.g. visible or not visible columnar arrangement of pyramidal cells, prominent or inconspicuous subdivisions of layer III, and width and cell density in layer IV, could be found as distincitive features at all sectioning levels and in all brains studied.

FG1 and FG2 differed not only by their cytoarchitecture (Fig. 8a) but could also be separated from adjoining cortical areas (see Table 2 for an overview):

Posteriorly, FG1 immediately adjoined hOc4v, which was characterized by a more cell dense layer II with a well-defined border to layer III (Fig. 8b). This latter layer showed larger pyramidal cells with a characteristic superficial to deep increase of cell sizes and no clear arrangement in cell columns. Layer IV provided the most pronounced distinction as it was prominent and had a higher cell density in hOc4v when compared to FG1. Cells in layer V were also larger in hOc4v but not densely packed. Layer VI of hOc4v again showed a higher cell density and a better defined border to the white matter.

In the more rostral parts, FG1 was followed medially by a new, not further analyzed area col.s.*, which was found anterior to hOc4v (Fig. 8c). col.s.* showed medium-sized and densely packed pyramidal cells in layer III, particularly in sublayer IIIc, with no columnar arrangement. Layer IV of col.s.* was more cell dense and slightly thicker than that of FG1, layer V showed a higher cell density, and layer VI had a distinct border to the white matter.

The likewise not yet completely mapped area l.ot.s.* was found lateral to FG2. Area l.ot.s.* extended into the lateral occipitotemporal sulcus and on the inferior occiptal and inferior temporal gyrus. l.ot.s.* showed cell sparse sublayers IIIa and IIIb with large pyramidal cells only in lower IIIc (Fig. 8d). Layer IV of this area, was less prominent than in FG2 and difficult to delineate from layer III. The cell density of layer V in l.ot.s.* was lower than in FG2, and the border between layer V and layer VI was blurred.

The statistical analysis of the histological volumes of FG1 and FG2 (Table 3) revealed a significant size difference between both areas (F = 25.547, p = 0.01) in favor of FG2. There were no significant effects of the factors side (F = 2.990, p = 0.122), gender (F = 1.784, p = 0.218) and no significant effects of the interactions between area and side (F = 0.011, p = 0.920), between area and gender (F = 0.254, p = 0.628), between side and gender (F = 0.380, p = 0.555) and between area, side and gender (F = 0.001, p = 0.981).

The coordinates for the centers of gravity of areas FG1 and FG2 in anatomical MNI space are given in Table 4. A slight difference of the anterior–posterior location was noted as both areas were located more rostrally in the right hemisphere. This interhemispheric shift has already been reported for other striate and extrastriate areas in previous studies (Amunts et al. 2000; Rottschy et al. 2007), and reflects the marked asymmetry in position of the occipital poles in the MNI single-subject template.

Probability maps (PMaps) were generated for each FG area by superimposing the spatially normalized representations of the individual subjects (Fig. 9). The PMaps reflect the probability of observing the respective area in single voxels of the MNI space in our sample of postmortem brains. The probability maps for FG1 and FG2 barely showed regions with high probabilities for one area (overlap of 9 or 10 brains), but broad regions with low probabilities (overlap of 1 or 2 brains) in the periphery of each of the FG areas. This reflects a high intersubject variability of both areas, which is much higher compared to early visual areas hOc1 and hOc2 (Amunts et al. 2000).

Consequently, there was substantial overlap between the probabilistic maps of FG1 and FG2 and also adjoining area hOc4v. To assign the most likely area to each single voxel, an MPM of the visual cortex (comprising areas hOc1, hOc2, hOc3v, hOc4v, FG1 and FG2) was computed (Fig. 10). The MPM represents a contiguous, non-overlapping parcellation of this region. It hence bears a close conceptual resemblance to presentations in classical brain maps. Importantly, however, the MPM does not show the parceling of an exemplary or “typical” hemisphere as do previous architectonic brain maps (Brodmann 1909; von Economo and Koskinas 1925), but reflects the most likely area based on a sample of ten brains and represented in each voxel of the reference space.

Classical anatomical maps of the human brain show a tripartition of the visual cortex (Brodmann 1909; von Economo and Koskinas 1925; Sarkisov et al. 1949). Brodmann (1909) mentioned that his temporal area BA37, which adjoins BA19 in its ventral parts, is a transitional area between temporal and occipital regions. Von Economo and Koskinas (1925) found a roughly comparable area to BA37 adjoining their area O_A, which can be seen as an equivalent to BA19. They describe the cytoarchitecture of this adjoining area as highly heterogeneous but were not able to delimit clearly defined subregions of this area. Because they saw primarily cytoarchitectonical characteristics of the parietal lobe, they assigned this area to the parietal region and labeled it “area parietalis temporooccipitalis” or P_H. Whereas the classical architectonical definitions of the primary and secondary visual cortex (BA17 and BA18) are fully agreed by most observers (Amunts et al. 2000), it became evident by functional neuroimaging and non-human primate data that BA19/O_A should be subdivided into multiple functionally and putatively also architectonically distinct regions (Zeki 1969; van Essen 1979; Braak 1980; Tootell et al. 1996). It seems conceivable that FG1 and FG2 are situated within the border region between BA19 and BA37 or O_A and P_H, respectively.

Detailed maps of the occipital and adjacent temporo-parietal lobe were published by Braak (1977) in his pigmentoarchitectonical study. His report included drawings of identified areas on the cortical surface as well as coronal sections, which makes a comparison less difficult. FG1 and FG2 topographically fit to Braak’s (1977) “area peristriata densopyramidalis”. Its description as a “well-developed bitaeniate cortex with conspicuous pIIIc” (“bitaeniate” = “double-striped”) is likewise in accordance with our observations on the cytoarchitecture of area FG2. Braak (1977), however, identified a single area in this region. Our results indicate clear cytoarchitectonical differences between two areas and hence a sub-parcellation of “area peristriata densopyramidalis” into FG1 and FG2.

Retinotopy is a common organizational principle of early visual areas in the primate visual cortex (e.g. Sereno et al. 1995; DeYoe et al. 1996; Engel et al. 1997). Utilizing this principle, retinotopic mapping has emerged to the gold standard in functional imaging of these early visual areas and yielded robust results for the borders of areas V1, V2 and V3v (Tootell et al. 1998; Wade et al. 2002; Wohlschläger et al. 2005).

However, there has been an intense debate during the last decade concerning the anterior border of human area V4. One view proposed an area V4v, representing the (contralateral) upper-quarterfield, followed anteriorly by a color-selective area, containing a complete hemifield representation termed V8 (Hadjikhani et al. 1998; Tootell and Hadjikhani 2001). This arrangement was challenged by others, who postulated V3v to be the most anterior quarterfield representation, which is then followed by a hemifield representation selective for color perception. They labeled this area the V4-complex, containing V4 and an anterior non-retinotopic subarea V4α. The authors did not see any evidence for a quarterfield representation like in V4v (Bartels and Zeki 1998, 2000; Zeki 2001). A third scheme contained an upper field representation, analogous to V4v, followed anteriorly by its matching lower field termed human V4 (hV4). The authors argued for yet another color-selective hemifield representation, VO-1 anterior to it (Wade et al. 2002; Wandell et al. 2005; Brewer et al. 2005).

It should be mentioned that the hV4/VO-1 model does not contradict the V4-complex model, but can be harmonized with it, while the V8 model cannot (Wade et al. 2002). Furthermore, the hV4/VO-1 model was later confirmed by other laboratories (Kastner et al. 2001; Larsson et al. 2006; Arcaro et al. 2009; Kolster et al. 2010). Our study shows two distinct cytoarchitectonical areas FG1 and FG2 antero-lateral to the cytoarchitectonical area hOc4v (Rottschy et al. 2007), which overlaps with retinotopically defined hV4 (Wilms et al. 2010). The descriptions of VO-1 as an area in the collateral sulcus and on the medial fusiform gyrus (Brewer et al. 2005; Liu and Wandell 2005) would topographically fit to FG1. However, the apparently closer proximity of VO-1 to the collateral sulcus and its location directly anterior to hV4 as well as the reported coordinates (Kastner et al. 2001; Brewer et al. 2005; Liu and Wandell 2005; Arcaro et al. 2009) suggest that VO-1 is more medial than FG1 and probably correlates with the medially adjoining cytoarchitectonical area col.s.* (Fig. 8c).

Recently, another retinotopic region has been discovered lateral to hV4, i.e. the phPIT cluster (Kolster et al. 2010), which refers to macaque areas PITd and PITv on the posterior inferotemporal gyrus (Felleman and van Essen 1991). The phPIT cluster is located on the inferior temporal gyrus at the posterior end of the lateral occipitotemporal sulcus and consists of two hemifield representations, phPITd and phPITv, which share their foveal representation and vertical meridians. The lateral of our identified cytoarchitectonical areas, FG2, is located on the lateral bank of the posterior fusiform gyrus and within the posterior lateral occipitotemporal sulcus. FG2 rarely reaches the inferior temporal gyrus. Thus, the topography and the center of gravity coordinates indicate that phPIT is located posterior and dorsal to FG2, and most probably overlaps with the laterally adjoining cortex (Fig. 8d).

Taken together, the comparison with current retinotopic literature indicates that FG1 and FG2 do not correspond to any hitherto identified retinotopic area. Instead, both cytoarchitectonical areas seem to fill in the “non-retinotopic gap” that is spanned between VO-1 medially and phPITv laterally (see Kolster et al. 2010, Fig. 16A). A minor peripheral overlap between these retinotopic and our cytoarchitectonical areas cannot be completely ruled out. However, such a mismatch between cytoarchitectonically and retinotopically defined cortical units would be in contrast to the correlation of both parcellation approaches in other cortical areas, e.g. in V1 and V2 (Wohlschläger et al. 2005).

Besides its retinotopic organization, the human ventral visual cortex contains a number of apparently category-specific functional modules for visual object processing. Some regions that respond more strongly to visual objects than to scrambled images were identified around the posterior fusiform gyrus and the inferior occipital gyrus. They are called the “lateral occipital complex” (LOC, Malach et al. 1995, 2002; Kanwisher et al. 1996; Grill-Spector et al. 2001). Bilateral LOC-activations receive input from both hemifields (Grill-Spector et al. 1998) and are correlated with recognition performance of objects (Grill-Spector et al. 2000). Conventionally, the LOC comprises two entities, the dorsal “LO” and the ventral and anterior “pFs” on the posterior and mid-fusiform gyrus, which differ in their response to size and position changes of presented objects (Grill-Spector et al. 1999). A later scheme assigned pFs to an object-selective cluster “VOT” of the ventral occipitotemporal cortex directly adjoining early retinotopic visual areas (Malach et al. 2002). In consideration of the pertinent reports of the LOC, a direct correspondence of our areas FG1 and FG2 to a LOC-cluster seems to be unlikely, as they are located between both classical clusters, ventral to LO and posterior to pFs. Indeed, the center of gravity of FG2 is in close proximity to the coordinates of the “branching point” between both LOC-clusters described by Malach et al. (1995). Thus, an overlap of the margins of LOC with FG2 is possible. Based on the VOT-scheme predicting object-responsive areas directly adjacent to early visual areas (Malach et al. 2002), it can be hypothesized that FG1 and FG2 are both situated within the higher order object-related cortex.

Additional to the LOC and partially overlapping with it, some areas selective for specific objects have been identified within the ventral occipitotemporal cortex including the prototypical example of the “fusiform face area” (FFA). First hints on the existence of the FFA were derived from studies on subjects with prosopagnosia. These subjects showed a bilateral or right-hemispheric lesion in the ventral occipitotemporal cortex (Damasio et al. 1982; De Renzi 1986; Landis et al. 1986; Sergent and Signoret 1992). Later PET studies revealed a distinct activation on the fusiform gyrus during face perception tasks in healthy volunteers (Sergent et al. 1992; Haxby et al. 1994). The face specificity and location of the FFA could be demonstrated at higher spatial resolution in numerous fMRI studies (Clark et al. 1996; Puce et al. 1996; Kanwisher et al. 1997; Tong et al. 1998; Halgren et al. 1999; Hasson et al. 2001; Grill-Spector et al. 2004; Kanwisher and Yovel 2006). In addition, a response to images of headless bodies was reported on the fusiform gyrus in the vicinity of the FFA (Cox et al. 2004; Peelen and Downing 2005). High resolution fMRI identified this activation as adjacent yet distinct from the FFA in the fusiform body area (FBA, Schwarzlose et al. 2005). Most functional investigations locate the FFA on the lateral bank of the posterior or mid-fusiform gyrus, about 1–2 cm anterior to the here identified cytoarchitectonical area FG2 (e.g. Gauthier et al. 2000; Hasson et al. 2001; Levy et al. 2001; Avidan et al. 2003; Rossion et al. 2003; Grill-Spector et al. 2004; Bartels and Zeki 2004; Peelen and Downing 2005). However, some reports show peak activations much closer to FG2 (Puce et al. 1996; Kanwisher et al. 1997; Halgren et al. 1999; Ishai et al. 1999). This discrepancy might be explained by a recent investigation of the fusiform gyrus demonstrating a subdivision of the FFA into a posterior and an anterior part (Weiner and Grill-Spector 2010, 2011a), which was also suggested by others (Pinsk et al. 2009; Mei et al. 2010). The posterior face patch named pFus-faces accurately matches the location of our lateral cytoarchitectonical area FG2 situated on the lateral bank of the posterior fusiform gyrus antero-lateral to hV4 and, hence, might be its functional correlate.

Another functional, category-specific area, which is located within the lateral occipitotemporal sulcus extending onto the lateral fusiform gyrus is the visual word-form area (VWFA), which responds specifically to words and letter strings. First hints came from patients suffering from lesions in the ventral occipitotemporal cortex and respective neuropsychological deficits, i.e. pure alexia (Damasio and Damasio 1983; Binder and Mohr 1992). This functionally defined area could later be localized by PET (Petersen et al. 1990; Petersen and Fiez 1993), MEG (Tarkiainen et al. 1999) and fMRI (Wagner et al. 1998; Cohen et al. 2000; Hasson et al. 2002; Dehaene et al. 2002), although the functional specificity of VWFA was also controversially discussed (Price and Devlin 2003). Similar to our area FG2, the VWFA was first described to be located on the lateral fusiform gyrus (Cohen et al. 2000; Dehaene et al. 2002), but more recent findings show that the large portion of VWFA lies within the fundus of the lateral occipitotemporal sulcus and about 1 cm anterior to FG2 (Cohen and Dehaene 2004; Baker et al. 2007; Ben-Shachar et al. 2007; Wandell et al. 2012). An overlap of FG2 with language-related visual areas is possible, since FG2 largely extends into the lateral occipitotemporal sulcus and distances of FG2 to the center of gravity of VWFA are quite small (e.g. Cohen et al. 2002; Vigneau et al. 2005; Dehaene et al. 2010; Mei et al. 2010). Moreover, the processing of words seems to continuously extend from early visual areas to the ventral occipitotemporal locations in a hierarchical manner (Dehaene et al. 2005; Szwed et al. 2011), where FG2 could possibly be involved at an intermediate stage.

Our results indicate that FG1 and FG2 are symmetrically found in all ten brains, showing the same cytoarchitectonical features on both sides and no significant left–right differences in volume. By contrast, the majority of reports show a right lateralization of the FFA that apparently depends on handedness (Willems et al. 2010), while the VWFA is most often left lateralized (e.g. Cohen and Dehaene 2004). This does not exclude a possible overlap of the functional and our cytoarchitectonical areas, since different functional manifestations can be implemented on the same cytoarchitectonical basis. Furthermore, recent investigations imply that lateralization for faces (Weiner and Grill-Spector 2010) and words (Ben-Shachar et al. 2007) is possibly much less pronounced as hitherto assumed.

Thus, our cytoarchitectonical areas FG1 and FG2 probably lie within the object-related higher order visual cortex on the posterior fusiform gyrus. Furthermore, FG2 likely comprises a posterior fusiform face-selective patch. However, distinct functional correlates in this region are rare. This issue might also be affected by the fMRI artifacts evoked by the transverse sinus, a venous vessel, which commonly proceeds directly inferior to the cortex we investigated here (Winawer et al. 2010). Appropriately, Winawer et al. (2010) denote this region as ‘No man’s land’. A more specific-functional characterization of FG1 and FG2 and the relation to object, face and visual word processing remains a topic for future work. It can certainly be encouraged by further improved functional imaging techniques and comprehensive approaches including cytoarchitectonical, retinotopic and category-related investigations. For now, we provide probability maps that can be used to relate functional measurements to our cytoarchitectonical delineations and, hence, can help to understand the functional role of this region rostrally adjoining the early ventral visual cortex.