Receptor architecture of macaque and human early visual areas: not equal, but comparable

Further, we used a total of five post-mortem human brains from donors (76 ± 3 years of age; 3 males) without a history of neurological or psychiatric diseases and obtained through the body donor program of the Department of Anatomy, University of Düsseldorf, Germany. Causes of death were sudden cardiac failure, multiorgan failure caused by sepsis and pneumonia, and lung edema.

To examine the laminar and regional distribution patterns of 14 receptor types belonging to the classical neurotransmitters glutamate (AMPA, kainate and NMDA), GABA (GABA_A, GABA_A/BZ and GABA_B), acetylcholine (muscarinic M₁, M₂ and M₃), noradrenaline (α₁ and α₂), serotonin (5-HT_1A and 5-HT₂) and dopamine (D₁), and to enable comparison with cytoarchitectonic features, alternating sections were processed for quantitative in vitro receptor autoradiography according to previously published protocols (Palomero-Gallagher and Zilles 2018; Zilles et al. 2002), or for visualization of cell bodies with a modified silver cell-body staining (Merker 1983) that provides a high contrast between cell bodies and neuropil. The radiolabelled sections were then air dried and exposed against tritium-sensitive films (Hyperfilm, Amersham, Braunschweig, Germany) together with plastic tritium standards of known radioactivity concentrations (Microscales^®, Amersham) for 4–18 weeks. The ensuing autoradiographs reveal the regional and laminar distribution of receptor binding sites.

Histological sections were scanned by means of a light microscope (Axioplan 2 imaging, ZEISS, Germany) equipped with a motor-operated stage controlled by the KS400^® and Axiovision (Zeiss, Germany) image analyzing systems applying a 6.3 × 1.25 objective (Planapo^®, Zeiss, Germany), and a CCD camera (Axiocam MRm, ZEISS, Germany) producing frames of 524 × 524 µm in size, 512 × 512-pixel spatial resolution, with an in-plane resolution of 1 µm per pixel, and eight-bit gray resolution. These digitalized serial images were used for the qualitative cytoarchitectonic identification of distinct occipital areas in the macaque monkey brains.

Autoradiographs were digitized with an image analysis system consisting of a source of homogenous light and a CCD camera (Axiocam MRm, Zeiss, Germany) with an S-Orthoplanar 60-mm macro lens (Zeiss, Germany) corrected for geometric distortions, connected to the image acquisition and processing system Axiovision (Zeiss, Germany), to carry out densitometric analysis of binding site concentrations in the radioactive sections (Palomero-Gallagher and Zilles 2018; Zilles et al. 2002). Spatial resolution of the resulting images was 3000 × 4000 pixels (8-bit gray value resolution). Because these images only code gray values, which represent concentration levels of radioactivity, a scaling (i.e., a linearization of the digitized autoradiographs) was carried out to transform the gray values into fmol binding sites/mg protein using in house developed Matlab (The MathWorks, Inc. Natrick, MA) scripts. To provide a clear visualization of the regional and laminar receptor distribution patterns, digitized autoradiographs were linearly contrast enhanced and pseudo-color coded.

Receptor densities of each area were extracted from the linearized images by computing the surface below profiles defined vertically to the cortical surface using in house developed scripts for Matlab (The MathWorks, Inc. Natrick, MA) as previously described (Palomero-Gallagher et al. 2008; Palomero-Gallagher and Zilles 2018). Location of measuring sites, and assignment to a cytoarchitectonically identified area was ensured by comparison of the autoradiographs with the adjacent cell-body-stained sections.

Selection of regions of interest for extraction of receptor densities in both macaque and human early visual areas was based on the architectonic identification of areas according to previously published criteria, and the analysis of multiple receptors in adjacent sections from the same brains. Specifically, in the macaque brain, areas V1–V3 and their subdivisions, as well as area V3A, dorsally and V4v ventrally were identified according to previously published cytoarchitectonic criteria and cortical maps (Felleman and Van Essen 1991; Niu et al. 2020; Van Essen et al. 1986). Areas marked as V3 and VP (Felleman and Van Essen 1991) correspond to our V3 subdivisions, mV3d and mV3v, respectively. A large portion of macaque V1 is also found on the lateral surface of the hemisphere, whereas human V1 is mainly located within the calcarine sulcus (Schira et al. 2012). Since retinotopic mapping has shown that the lower and upper visual fields are represented on the lateral surface in same manner as within the calcarine sulcus (Rosa 2002), we here concentrated on the sulcal portion of macaque V1.

In the human brain, areas V1 and V2 were identified according to Amunts et al. (2000), areas hOc3v and hOc4v on the ventral occipital cortex according to Rottschy et al. (2007), and areas hOc3d and hOc4d on the dorsal occipital cortex according to Kujovic et al. (2013). Areas hOc3d and hOc4d are the putative anatomical substrates of functionally defined areas V3d and V3A, respectively (Kujovic et al. 2013), and areas hOC3v and hOC4v are those of functionally defined areas V3v and V4v, respectively (Rottschy et al. 2007). Eickhoff et al. (2008) analyzed the receptor architecture of early visual areas in the human brain and besides confirming the cytoarchitectonically defined areas, identified dorso-ventral subdivisions within areas V1 and V2. Since visual inspection of the color coded receptor autoradiographs hinted at a comparable situation in the macaque brain, for each of these areas, densities were quantified in both species in a dorsally located region of interest (i.e., V1d, V2d) and in a ventrally located region of interest (i.e., V1v, V2v).

Note, that to avoid confusion, in the present analysis the prefix m- will be used to identify all monkey areas. Furthermore, to facilitate comparison between species, for human areas we will apply the functionally relevant nomenclature, albeit with the prefix h- (e.g., hV3A for area hOc4d).

To determine if there were significant differences in receptor architecture between paired areas (dorsal and ventral subdivisions of the same visual region, or of adjacent areas from different hierarchical levels), stepwise linear mixed-effects models were performed separately for human and macaque visual areas. To ensure an equal weighting of each receptor in subsequent statistical analyses, receptor density values were normalized within each receptor type separately in human and macaque by applying the min–max scaling (Eq. 1).where x is absolute receptor density, i represents an individual section, and z is the normalized data. Unless otherwise specified, normalization was performed separately in macaque and human data.

Statistical analyses were conducted using the R programming language (version: 3.6.3; Team 2013). For each species, the statistical testing involved three levels. In the first step, an omnibus test was carried out to determine whether there were differences across all areas when all receptor types are considered simultaneously (Eq. 2). The model consisted of fixed effects for area and receptor type, and human/macaque hemisphere was set as a random factor.where D is the receptor density, A is visual area, R is receptor type and B is human/macaque brain.

If the interaction effect between area and receptor type at the first step test was found to be significant, a second set of simple effect tests was performed for each receptor separately (i.e., 14 simple effect tests were performed in total) to determine whether there were significant differences across all areas for each receptor type. To correct for multiple comparisons in the second step tests, the false-discovery rate correction (Benjamini and Hochberg 1995) was performed (i.e., p-values were corrected for 14 comparisons).

Finally, for the receptor types that were found to show significant differences across all areas in the second step tests, a third set of post hoc tests were used to explore the paired areas that drove the statistical difference. For each receptor type, 28 post hoc tests were performed. To correct for multiple comparisons in the third step tests, we performed the false-discovery rate correction (Benjamini and Hochberg 1995) separately for each receptor type (i.e., p-values were corrected for 28 comparisons per receptor type).

For each architectonically defined early visual area, we calculated mean areal densities (i.e., averaged over all cortical layers) of each of the 14 different receptors. To display the densities of multiple receptors within and between different cortical areas more intuitively, the ensuing densities were visualized for each area as a “receptor fingerprint”, i.e., as a polar coordinate plot simultaneously depicting the concentrations of all examined receptor types within that area (Palomero-Gallagher and Zilles 2018; Zilles et al. 2002).

Principal components (PCA) and hierarchical cluster analyses were carried out using in house developed Matlab (The MathWorks, Inc. Natrick, MA) scripts. Analyses were first carried out for macaque and human areas separately to identify the grouping of early visual areas in each species based on similarities in their receptor architecture. Receptor densities were normalized by z-scores for each receptor and species separately to ensure an equal weighting of receptors expressed at overall high and low densities. The Euclidean distance was used in the hierarchical cluster analysis as a measure of (dis)similarity of areas since it best captures the differences in size and shape between fingerprints, and the Ward linkage algorithm was chosen as the linkage method, since in combination with the Euclidean distance it resulted in the maximum cophenetic correlation coefficient as compared to any combination of alternative linkage methods and measurements of (dis)similarity (Palomero-Gallagher et al. 2009). The optimal number of clusters, k, for the K-means algorithm was determined by clustering the data with k from 1 to 9. For each clustering of the data, the squared Euclidean distance between the data points and their respective centroids, i.e., distortion, was calculated and plotted against each k (Rousseeuw 1987). We also sought to determine similarities between receptor types by how their expression levels varied across areas. To this purpose, we transposed the matrices used for the clustering of areas based on differences in their receptor fingerprints, so that receptor densities were normalized by z-scoring for each area and species separately, and carried out a second set of PCA, hierarchical cluster and K-means analyses separately for the macaque and human brains.

Finally, to address the question of homologies between human and macaque visual areas, a species-combined PCA was conducted as previously described (Sherwood et al. 2004). Note that prior to this PCA, a species-combined normalization was performed. That is, within each receptor, the density for all macaque and all human areas were jointly normalized by z-scores.

Figure 6 shows the receptor fingerprints of areas analyzed in the present study in the early visual cortex of both macaque monkey and human brains, and Supplementary Table 6 provides numeric values. The corresponding normalized data are presented in Supplementary Fig. 6 and Supplementary Table 7. The absolute mean receptor concentration varies considerably between the different receptor types in each area. In both species, GABA_A and GABA_B receptors, as well as GABA_A/BZ binding sites are present at the highest absolute densities, whereas lowest absolute densities are reached by the 5-HT_1A and D₁ receptors.

Two sets of hierarchical clustering and principal component analyses were performed with each the macaque and the human data. The first set of analyses aimed to visualize the degree of (dis)similarity in the normalized fingerprints of early visual areas (Fig. 7). The k-means analysis and elbow plots clearly indicated that k = 3 provided the optimal trade-off between number of clusters and distortion for both the macaque and the human brain (Supplementary Fig. 7). The second set of analyses aimed to identify (dis)similarities between receptors in their expression levels across visual areas (Supplementary Fig. 8), and k = 2 was found to be the optimal number of clusters for the macaque brain, whereas that for the human brain was k = 4 (Supplementary Fig. 9).

In the macaque monkey (Fig. 7A), cluster 1 contained the two subdivisions of mV1, which separated very early from the remaining areas (cluster 2). Within cluster 2, areas mV2d, mV2v, mV3d, mV3v and mV3A are found in one group (cluster 2.1), whereas area mV4v is separated from the other early visual areas to form an isolated cluster 2.2. The segregation of clusters 1 and 2 was also confirmed by the 1st principal component of the PCA, and that of clusters 2.1 and 2.2 by the 2nd principal component (Fig. 7A). The analyses aiming to reveal which receptors can be grouped based on how their densities change across the examined areas (Supplementary Fig. 8A) indicated a segregation of AMPA, kainate, M₁, α₁ and 5-HT_1A receptors (cluster 2) from the remaining examined receptor types (cluster 1), which was confirmed by differences along the 1st principal component of the PCA (Supplementary Fig. 8A).

In the human brain, areas hV1d and hV1v also grouped together as a single cluster (cluster 1, Fig. 7B). Areas hV2d, hV2v, hV3d and hV3v are all found in a single cluster (cluster 2.1), and areas hV3A and hV4v are grouped in cluster 2.2. The 1st principal component of the PCA clearly segregated these three clusters, whereas the 2nd principal component more strongly reflected differences between hV2d, hV2v, hV3d and hV3v and the remaining areas (Fig. 7B). Clustering of the receptors according to variations in their distribution patterns across visual areas (Supplementary Fig. 8B) revealed four clusters: Cluster 1.1 contained the NMDA, GABA_A, M₁, M₂, α₂ and 5-HT₂ receptors; cluster 1.2 the M₃, α₁ and D₁ receptors; cluster 2.1 the AMPA and 5-HT_1A receptors; cluster 2.2 the kainate and GABA_B receptors as well as the GABA_A/BZ binding sites. In the PCA, AMPA and 5-HT_1A were separated from the remaining receptors by differences along the 2nd principal component, whereas clusters 1.1, 1.2 and 2.2 were segregated by the 1st principal component.

Finally, to address the issue of comparability between homolog areas in each species, a species-combined PCA was performed (Fig. 8). The 1st principal component clearly segregates human and macaque areas, whereas the 2nd principal component generally reflects differences in the fingerprints associated with the hierarchical processing level of each area.

Area V1 is the cytoarchitectonically most differentiated isocortical area in the primate brain, with a unique sublamination of layer IV (Zilles et al. 2015b). This cytoarchitectonic uniqueness is mirrored by its receptor architecture, which clearly reveals the border to V2, as revealed not only in the macaque (present results; Hendry et al. 1990; Rakic et al. 1988; Rakic and Lidow 1995; Rosier et al. 1991; Zilles and Clarke 1997; Zilles and Palomero-Gallagher 2017a) but also in the vervet brain (Takemura et al. 2020).

Layer-specific differences in receptor densities enabled the definition of qualitative dorsal and ventral components of mV1 within the calcarine sulcus, as well as medio-lateral density gradients within each of these compartments. This heterogeneous receptor distribution probably represents the molecular underpinning of the fact that visual information from the upper and lower, as well as from the peripheral and central, visual fields is known to processed separately in primate V1 (Dougherty et al. 2003; Gattass et al. 2005; Previc 1990; Silson et al. 2018; Van Essen et al. 1986). Furthermore, since mV1d sends topographically organized projections to mV2d and mV3d, whereas mV1v projects to mV2v, but not to mV3v (Van Essen et al. 1986), this retinotopic organization is propagated through early extrastriate visual areas, and also reaches areas of the posterior inferotemporal and dorsal occipitotemporal cortex (Kolster et al. 2014; Zhu and Vanduffel 2019).

The modular distribution of M₂, GABA_A, and 5-HT₂ receptors within macaque V1 resembles the previously described blobs and interblobs revealed by cytochrome oxidase staining (Horton and Hubel 1981; Wong-Riley 1979), as well as the periodical distribution of GABA_A receptors in the human brain (Zilles and Schleicher 1993). Although the functional meaning of blobs and interblobs has been controversially discussed in the literature, they are commonly thought to be associated with differential color domains and orientation-selective processes (Lu and Roe 2008).

Similar to V1, area V2 in the monkey has been described as a cytoarchitectonically homogeneous region (de Sousa et al. 2010), and we detected no significant differences in receptor densities at the mean areal level. However, we found a trend towards higher kainate, NMDA, GABA_B, and M₁ densities in the infragranular layers of mV2d, as well as higher 5-HT_1A but lower M₂ concentrations in its supragranular layers than in the corresponding layers of mV2v. These qualitative differences would be in accordance with the dorso-ventral asymmetry in connectivity patterns of V2. Whereas V2d and V2v project back to V1 and forward to dorsal and ventral parts of V3 (Gattas et al. 1997), output to area V4t was found to originate in the dorsal part of V2, but not in V2v (Gattass et al. 1997, 2005). Furthermore, V2 encompasses dorsal and ventral functional subdivisions in which the inferior and superior contralateral quadrants are represented, respectively (Gattass et al. 1981), and these subdivisions also differ in the length and orientation of their cytochrome oxidase positive stripes (Olavarria and Van Essen 1997).

Cortex located immediately rostral to V2 has been designated as the “third visual complex”, and encompasses our areas V3v, V3d, and V3A (Rosa et al. 2005; Zeki 1978), where area V3v has also been designated as area VP (de Sousa et al. 2010; Hof and Morrison 1995; Zilles and Clarke 1997). The receptor architecture of areas V3d and V3A, which are located at the junction of the intraparietal and parieto-occipital sulci, was comprehensively characterized in a recent mapping study of the macaque intraparietal sulcus, and the same sample was used as for the present analysis (Niu et al. 2020). Our data confirm and expand on a study by Kötter et al. (2001) on the relationship between area-specific differences in receptor densities and connectivity patterns in multiple areas of the macaque monkey brain, including visual areas analyzed here, since their analysis of the visual cortex only included the AMPA, kainate, GABA_A, M₁, M₂ and 5-HT₂ receptors, and they only extracted densities from a single macaque hemisphere (which was not included in the present analysis).

We found the fingerprints of macaque visual areas to differ in shape from those of their human homologs, indicating species-specific differences in the balance between the analyzed receptor types of the GABAergic system. E.g., whereas GABA_A/BZ binding site densities were higher than GABA_A receptor densities in mV1d and mV1v, the opposite holds true for hV1d and hV1v. Human and macaque V1 are also known to differ in their laminar distribution pattern of cytochrome oxidase activity in layers IVa and IVb, and in the organization of input from the magno- and parvocellular projections from LGN (Preuss et al. 1999) which has been interpreted as suggesting an evolutionary shift in the organization of LGN input to the primary visual cortex and reflecting different mechanisms of motion processing in humans than in non-human primates (Orban et al. 2004).

In both species, primary visual area V1 significantly differed from V2 by a higher mean density of M₂ and α₂ receptors, but a lower one 5-HT_1A receptors. These differences are in accordance with previous receptor architectonic reports in the human visual cortex (Eickhoff et al. 2007, 2008; Zilles and Palomero-Gallagher 2017b), and are also supported by qualitative descriptions in the macaque brain (Rakic et al. 1988; Rakic and Lidow 1995). Notably, the hierarchical cluster analysis carried out to identify groupings of receptors based on (dis)similarities in their expression levels throughout visual areas revealed for both macaque and human brains that the 5-HT_1A receptors were located cluster 2, whereas the M₂ and α₂ receptors were in cluster 1. Additionally, in macaques V1 presented significantly higher 5-HT₂ levels than V2, whereas human V1 and V2 differed in GABA_A receptor densities, thus highlighting possible interspecies differences in the molecular mechanisms subserving information transfer between V1 and early visual areas.

Given the differences between V1 and V2, it is not surprising that the hierarchical cluster analysis and the 1st principal component of the PCA clearly segregated the fingerprints of human and macaque primary subdivisions from the rest of the visual areas (Fig. 7). Furthermore, as shown by the combined PCA, both species have in common that segregation along the 2nd principal component reflected differences in fingerprints which are associated with the hierarchical processing level of each area. Thus, the transition that the molecular structure of early visual areas undergoes when moving from the primary visual cortex through V2 and V3, and up to V3A and V4v, is comparable in the macaque and human brains.

There were species differences, however, concerning the segregation pattern of V3A, and they were also confirmed by the hierarchical clustering analysis: in the macaque brain, mV3A clustered with mV3d, but in the human brain it was clearly separated, together with hV4v, from lower level visual areas (Fig. 7). Receptor fingerprints of hV3A and hV4v differ in shape from the rest of areas in the balance between GABA_A/BZ and GABA_B receptors, indicating functionally specific areas, which represent different hierarchical levels within the visual system. Interestingly, differences in kainate receptors were found to be significant in the monkey brain, but not human; i.e., mV3A and mV3d expressed significantly lower kainate densities than the surrounding areas. Pre- and postsynaptic kainate receptors are important for neurotransmission regulation, and seem to be involved in short- and long-term plastic phenomenon, highlighting their crucial role in synaptic signaling (Lerma 2003).

Area V3A represents an intermediate region in visual processing between lower level areas V1–V3 and higher visual areas of the dorsal and ventral streams (Tootell et al. 1997), since it shares connections with areas in the parietal and the temporal cortex (Felleman and Van Essen 1991). Interestingly, functional studies in humans associated area V3A with motion processing (Tootell et al. 1997), while similar studies in monkeys described area V3d as being more sensitive to motion than area V3A (Tootell et al. 1997; Tolias et al. 2001; Vanduffel et al. 2002), suggesting that area V3A plays different roles in humans and monkeys (Orban et al. 2003, 2004; Tootell et al. 1997), and the differing clustering patterns of area V3A in the human and macaque visual systems described in the present study provide further support for this hypothesis. However, monkey V3A has a similar retinotopic organization to that of human V3A, with a complete representation of the visual field separated by the horizontal meridian (Brewer et al. 2002; Fize et al. 2003; Tootell et al. 1997; Gattass et al. 1988), and in both species is associated with the processing of stereoscopic stimuli (Backus et al. 2001; Tsao et al. 2003). The fact that our clustering analyses did not result in a clear segregation of areas V2d, V2v, V3d and V3v could indicate that crosstalk between areas of the dorsal and ventral streams not only occurs at hierarchically higher processing levels (Van Polanen and Davare 2015), but that there is already a strong interconnectivity between both streams at very early stages of the processing of visual stimuli.

In the macaque, area mV4v formed its own cluster, not only due to differences in the shape of fingerprints, but also to the fact that its fingerprint is the smallest of all analyzed areas. However, in humans hV4v was found to cluster with hV3A, indicating that the receptor fingerprint of mV4v differs more from those of the remaining macaque extrastriate visual areas than does hV4v from the remaining human extrastriate visual areas. This latter fact seems to be driven by species-specific differences since the overall receptor balance in hV4v is driven by the high densities of the GABAergic receptors. Primate area V4v constitutes a mid-level visual processing region that receives input primarily from area V2 and sends output to the inferior temporal cortex (Tootell et al. 1997) as well as topographically organized feedback projections to V2 and V3 (Ungerleider et al. 2008). It has been characterized as a color-sensitive area representing the dorsal half of the visual field (Felleman and Van Essen 1991; Gattass et al. 1988; Zeki 1978). A functionally comparable region was defined in the human brain based on in vivo retinotopic imaging (DeYoe et al. 1996; Sereno et al. 1995; Tootell et al. 1996), although a later imaging study showed that only a quarter-field is represented in hV4v (Wilms et al. 2010). However, given that the Euclidean distance between the normalized receptor fingerprints of mV4v and hV4v was the smallest of all interspecies comparisons, it is plausible to consider them homolog areas.

Concluding, we identified and characterized eight receptor architectonically distinct areas in the early visual cortex of the macaque monkey, i.e., V1d, V1v, V2d, V2v, V3d, V3v, V3A and V4v, and compared their fingerprints with those of their homologs in the human brain. Multivariate analyses revealed that although macaque and human early visual areas differ in their molecular architecture, within each species the area-specific differences in receptor fingerprints reflected comparable hierarchical processing levels. Furthermore, in both species the subdivisions of areas V2 and V3 were found to be more closely grouped, i.e., to bear a closer neurochemical resemblance to each other than to remaining areas, and were clearly segregated from the subdivisions of the primary visual cortex and also from V4v. Thus, the macaque monkey early visual cortex can be considered as a good animal model for translational studies.