Cerebral Cortex, Vol. 11, No. 4, 298-311,
April 2001
© 2001 Oxford University Press
Where is ‘Dorsal V4’ in Human Visual Cortex? Retinotopic, Topographic and Functional Evidence
Nuclear Magnetic Resonance Center, Department of Radiology, Massachusetts General Hospital, Charlestown, MA 02129, USA
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
In flattened human visual cortex, we defined the topographic homologue of macaque dorsal V4 (the ‘V4d topologue’), based on neighborhood relations among visual areas (i.e. anterior to V3A, posterior to MT+, and superior to ventral V4). Retinotopic functional magnetic resonance imaging (fMRI) data suggest that two visual areas (‘LOC’ and ‘LOP’) are included within this V4d topologue. Except for an overall bias for either central or peripheral stimuli (respectively), the retinotopy within LOC and LOP was crude or nonexistent. Thus the retinotopy in the human V4d topologue differed from previous reports in macaque V4d. Unlike some previous reports in macaque V4d, the human V4d topologue was not significantly color-selective. However, the V4d topologue did respond selectively to kinetic motion boundaries, consistent with previous human fMRI reports. Because striking differences were found between the retinotopy and functional properties of the human topologues of ‘V4v’ and ‘V4d’, it is unlikely that these two cortical regions are subdivisions of a singular human area ‘V4’.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
It is difficult to be certain whether a given cortical area in humans is homologous to a specific cortical area in a different species, such as macaque monkeys. In this example, the two species have evolved independently from each other over ~30 million years. The evolution of the cortical maps during this time cannot be reconstructed, since the cortical maps left no fossil record. Thus any assertion of homology between two such candidate cortical areas is ultimately inferential.
Despite this uncertainty, certain cortical areas are widely accepted as homologous across species, based on multiple lines of circumstantial evidence (Baker et al., 1981
; Kaas and Krubitzer, 1991; Sereno and Allman, 1991; Rosa et al., 1993; Kaas, 1995; Rosa and Krubitzer, 1999) (D.C. Van Essen et al., submitted for publication). In the visual cortex (one of the most well-mapped cortical regions), the kinds of evidence used to infer homology between candidate areas in two species include similarities in: (i) functional properties; (ii) retinotopy; (iii) patterns of intra-cortical connections; (iv) histological and biochemical features; and (v) topography. Based on three of these criteria (functional properties, retinotopy and topography), several visual cortical areas have already become accepted as homologous, in comparisons between macaque maps and functional magnetic resonance imaging (fMRI)-based human maps (Sereno et al., 1995a; Tootell et al., 1995, 1997; DeYoe et al., 1996; Wandell, 1999). Such human areas include areas V1, V2, V3/VP, V3A and MT+. Earlier positron emission tomography (PET) data first suggested a likely human homologue of monkey area MT (Lueck et al., 1989; Zeki et al., 1991; Watson et al., 1993).
Although this list includes homologues of most of the well-studied areas in macaque visual cortex (e.g. V1, V2 and MT), one glaring omission is macaque area V4. Macaque V4 has been well-studied with regard to spatial filtering properties (Desimone and Schein, 1987; Gallant et al., 1993; Cheng et al., 1994; Connor et al., 1996; McAdams and Maunsell, 2000), color (Zeki, 1973, 1977, 1978; Van Essen and Zeki, 1978; Fischer et al., 1981; Schein et al., 1982; Schein and Desimone, 1990), extraretinal modulation (Moran and Desimone, 1985; Haenny et al., 1988; Motter, 1993; Connor et al., 1996, 1997), and models of visual processing (Gochin, 1994; Niebur and Koch, 1994; Courtney et al., 1995; Salinas and Abbott, 1997; Bar and Biederman, 1999; Wilson et al., 2000). V4 also occupies a critical position in the cortical hierarchy; it is often regarded as a ‘gatekeeper’ in the chain of ‘ventral stream’ areas extending into inferotemporal cortex (Maunsell and Newsome, 1987; Felleman and Van Essen, 1991b; Van Essen and Gallant, 1994). Unlike neighboring area MT, V4 receives roughly balanced inputs from both magnocellular and parvocellular streams (Ferrera et al., 1992, 1994).
Macaque V4 is often subdivided into dorsal and ventral sub-areas (V4d and V4v, respectively), which include retinotopic representations of the lower and upper visual field, respectively (see Fig. 1). In humans, retinotopic and topographic evidence from fMRI has revealed the apparent homologue of ventral V4 (‘V4v’), in several laboratories (Sereno et al., 1995a; DeYoe et al., 1996; Tootell et al., 1997; Grill-Spector et al., 1998b; Hadjikhani et al., 1998; Baseler et al., 1999; Wandell, 1999). However, the human homologue of macaque dorsal V4 (‘V4d’) has not yet been systematically described.
|
This is potentially quite important, because almost all of the data from macaque ‘V4’ has in fact been acquired from dorsal V4, not ventral V4. Experimentally, dorsal V4 is much more accessible than ventral V4. This distinction between V4d and V4v would be irrelevant, if these two cortical regions are in fact two retinotopic subdivisions of the same functional area, as currently assumed. However, at least in human visual cortex, our data suggest that these two cortical regions (‘V4v’ and ‘V4d’) appear to be distinctly different cortical areas, rather than two retinotopic subdivisions of the same area.
Here we describe these and related fMRI results, which were originally designed to reveal a homologue of macaque ‘dorsal V4’ in human visual cortex. As described above, the results of these tests turned out quite different than we expected. Our fMRI results strongly suggest that the human cortical visual area located where V4d should be (the V4d ‘topologue’) does not correspond to previous descriptions of dorsal V4 in macaques, neither retinotopically nor functionally. Unlike human V4v, the human V4d topologue is not retinotopically well differentiated, although it can be subdivided into two eccentricity-based subdivisions.
These unexpected results in the human V4d topologue may indicate that a significant species difference exists between macaque and humans. This would not be terribly surprising, since cortical maps have evolved into quite different forms, based on comparisons between other primates — especially in the higher-tier areas beyond V1 and V2. Many of the cortical visual areas in other well-mapped primate species (e.g. aotus monkey and galago) have no obvious counterpart in macaque (Kaas, 1995; Rosa and Krubitzer, 1999). It is arguable whether a ‘V4’ homologue even exists in non-human primates other than macaque (Felleman and Van Essen, 1991b; Kaas and Krubitzer, 1991; Sereno and Allman, 1991; Rosa et al., 1993; Kaas, 1995; Rosa and Krubitzer, 1999).
Alternatively, it is possible that human-like retinotopic and functional distinctions also exist between V4v and V4d in macaques, which have not yet been widely recognized (Steele et al., 1991; Stepniewska and Kaas, 1996). Irrespective of how this matter is eventually resolved, it is significant, since the topographical organization of macaque visual cortex is often used as a model for that in human visual cortex.
V4 and Color Processing
The localization of V4 has special importance because it has been claimed that V4 processes color information selectively. The original report was that all (Zeki, 1973) or a majority (Zeki, 1977, 1978) of cells in (dorsal) V4 are wavelength-selective. However, when more systematic studies began considering the effects of stimulus variations in luminance as well as color, and measuring color responses quantitatively, the reported percentage of wavelength-selective cells in V4 decreased to levels which were similar to those in neighboring cortical areas (Fischer et al., 1981; Schein et al., 1982). A more recent report (Schein and Desimone, 1990) confirmed that the percentage of classic opponent-color cells was small in V4, but it also allowed for some higher-order processing of wavelength information over wide regions of the visual field (e.g. color constancy). On the other hand, similar responses to wide-field color changes have been reported in macaque V1 (Wachtler et al., 1997, 1998, 1999a, 1999b), so there may be nothing unique about the wide-field color responses in V4.
A similar conclusion was reached by lesion studies. Across a range of laboratories, lesions of macaque V4d had little effect on behavioral tests of wavelength discrimination (Schiller, 1993; Walsh et al., 1993) (W. Merrigan, personal communication). Even lesions that apparently included ventral V4 did not produce achromatopsia (Heywood et al., 1992). In one lesion study including V4 (Walsh et al., 1993), deficits were reported in the processing of color constancy. However such effects could also (or instead) reflect secondary effects of brain lesions on higher-tier areas to which V4 projects (e.g. pIT; see below), rather than a direct effect in V4.
Other evidence suggests that a different area of macaque visual cortex is homologous to the area implicated in human achromatopsia. That revised color-selective area is located anterior and ventral to V4, in-or-near posterior inferotemporal (pIT) cortex, perhaps in areas PIT or TEO (see Fig. 2). The evidence for this includes the following. First, when pIT is lesioned, prominent deficits are seen in wavelength sensitivity (Heywood et al., 1995) (W. Merrigan, personal communication), unlike the lack of such deficits following V4 lesions. Secondly, this pIT region shows higher brain activity in neuroimaging experiments, when macaques are performing wavelength discriminations (Takechi et al., 1997). Finally, pIT corresponds to the region which should be color-selective in macaques, based on the topography of color-selective areas in human visual cortex (see below; cf. Figs 2 and 3).
|
|
Overall, these data do not constitute a persuasive case for color selectivity in macaque V4. Despite this, a small color- selective region which was discovered in human visual cortex was initially named ‘human V4’ (Lueck et al., 1989
). Subsequent retinotopic studies revealed that this color-selective region (‘V8’, ‘VO’ or ‘V4’, in different accounts) is actually located in an area anterior to human area V4v (Hadjikhani et al., 1998) (see Fig. 3
).
A subsequent controversy developed about the details of the retinotopy and the localization of this color-selective region, relative to human V4v (Tootell and Hadjikhani, 1998; Zeki et al., 1998; Bartels and Zeki, 2000). However, these arguments are misleading, since the original claims for color selectivity in single unit reports were acquired from dorsal V4, not ventral V4 (see Figs 2 and 3). Thus the real issue is whether the small color- selective region described in ventral human visual cortex could possibly be homologous to dorsal V4, not ventral V4. This appears unlikely, based on the cortical topography (see Figs 2 and 3) and other factors (see below). As described above, this distinction becomes especially important if V4d and V4v are actually distinct cortical areas with inappropriately similar names, rather than two retinotopic subdivisions of a single visual area. Such tests were the focus of this report.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
The methods in this study were similar to those described elsewhere (Sereno et al., 1995a
; Tootell et al., 1997, 1998; Hadjikhani et al., 1998). Informed written consent was obtained from each subject prior to the scanning session, and all procedures were approved by Massachusetts General Hospital Human Studies protocol no. 96-7464. Normal human subjects, with (or corrected to) emmetropia, were scanned in a high field (3 T) General Electric magnetic resonance (MR) scanner retrofitted with echo-planar imaging (ANMR Corp.). Head motion was minimized by using bite bars with deep, individually molded dental impressions. The subject's task was to continuously fixate the center of all visual stimuli throughout the scan acquisition. Magnetic resonance images were acquired using a custom-built quadrature surface coil, shaped to fit the posterior portion of the head. Magnetic resonance slices were 3–4 mm thick, with an in-plane resolution of 3.1 x 3.1 mm. Each scan took either 8 min 32 s (retinotopy), or 4 min 16 s (all other scans), using a TR of either 4 or 2 s, respectively. Each scan included 2048 images, consisting of 128 images per slice in 16 contiguous slices.
Phase-encoded retinotopic maps were obtained from 41 subjects (115 scans polar angle, 115 scans eccentricity, 471040 images total). Additional area-labeling scans (49 scans, 100352 images) were also acquired to clarify the location of MT+ and other visual areas. Among these subjects, 14 were also tested for sensitivity to color-versus-luminance, color afterimages, and sensitivity to kinetic boundary stimuli (216 scans, 442368 images total).
All stimuli were projected onto a screen located ~24 cm from the subjects eyes, using a LCD projector (NEC, model MT 800; 800 x 600 pixel resolution). Stimuli for retinotopic mapping were slowly moving, phase-encoded thin rays or rings comprised of counterphasing checks, which varied semi-randomly in both luminance and color. For a given subject, retinotopic information was signal-averaged from 4–12 scans (8192–24576 images) of polar angle or eccentricity. Data were also combined from different slice prescriptions on the same cortical surface, to reduce intervoxel aliasing. The retinotopic stimuli extended from 0.2° to 18°/25°/30° into the periphery, along the vertical/horizontal/oblique axes, respectively (thus up to 60° in visual extent).
Cortical Flattening and Spatial Filtering
For each subject, a first step was to acquire the structural MR images needed for reconstruction. Such acquisitions were optimized for contrast between gray and white matter in brain, and these procedures are described in full elsewhere (http://www.nmr.mgh.harvard.edu/freesurfer) (Sereno et al., 1995b; Dale et al., 1999; Fischl et al., 1999). This structural scan was acquired only once per subject, in a head coil for full head coverage.
From these three-dimensional data, image components caused by the skull were stripped off automatically by ‘shrink-wrapping’ a stiff deformable template onto the brain images. Then the gray–white matter boundary was estimated for each hemisphere with a region-growing method. The result was tessellated to generate a surface (~130 000 vertices) that was refined against the MRI data with a deformable template algorithm. Then it was inspected for topological defects (e.g. ‘handles’) and reconstructed without surface defects, if necessary.
The resulting surface was aligned manually with the functional scan by direct iterative comparisons in three orthogonal planes between the echo-planar imaging (EPI) inversion recovery scan (1.5 x 1.5 x 3–4 mm) and the high-resolution data set (1 x 1 x 1 mm) used to construct the cortical surface. By choosing a surface centered on the gray–white matter boundary, we effectively sampled activity most in cortical layers 4–6, rather than near the surface where the macrovascular artifact is maximal. Thus it was possible to assign activity more accurately to the correct bank of a given sulcus. The lower resolution activation data (3 x 3 x 3–4 mm) was interpolated smoothly onto the higher-resolution surface polygon (one polygon ~1 x 1 mm). Then the surface was unfolded by reducing curvature while adding an additional area-preserving term. For a completely flattened cortical surface, the inflated brain was cut along the calcarine fissure and just posterior to the sylvian fissure. The resulting surface was pushed onto a coronal plane in one step and unfurled on the plane. A relaxation algorithm was applied to minimize areal and linear distortion, weighted equally. The vertex update rule for flattening was further modified to include a shear-minimizing term, because maintaining only the original local areas allows substantial distortion (e.g. rectangular and rhombic distortions of an original square). After flattening, the data was spatially smoothed using a kernel of ~2.5 mm (half-amplitude at half-maximum).
In some analyses, we sampled values of retinotopic phase across the flattened cortical surface, in one-dimensional plots. In such analyses, data were plotted for each vertex crossed in the flattened cortical surface, along a line which was as straight as possible between adjacent vertices.
The V4d Topologue
We tested for a ‘V4d’ in human visual cortex by first delineating the cortical region in which V4d should be located (i.e. by creating a V4d ‘topologue’), in each human subject tested. This ‘topologue’ approach assumes that neighborhood relationships between cortical areas tend to be conserved during evolution. For example, it assumes that area V2 in a given species does not arbitrarily disengage itself from adjacent areas V1 and V3 during evolution, then move a long way across cortex, to re-settle itself in some random location far anteriorly. This assumption is well supported by empirical comparisons of cortical visual maps across multiple species of non-human primate (Baker et al., 1981; Sereno et al., 1994; Kaas, 1995; Rosa and Krubitzer, 1999), and by theories of cortical development (Van Essen, 1997). The location of our V4d topologue was not constrained by the location of specific sulci and gyri, since the existence of specific gyri and sulci (and their relationship to corresponding visual areas) can vary greatly between species.
According to all published accounts, macaque V4d is located: (i) superior to the cortical region called ‘V4v’; (ii) anterior to V3A; and (iii) posterior to MT (and the small transition area ‘V4t’) (see Figs 1 and 2). Therefore, we defined our human V4d topologue so that it was also located: (i) superior to human V4v; (ii) anterior to V3A; and (iii) posterior to human area ‘MT+’ (see Fig. 3). Human ‘MT+’ is presumed homologous to human MT plus small adjacent motion-selective areas (DeYoe et al., 1996).
This was essentially a topographic analysis. Thus, the V4 topologue was defined on a two-dimensional cortical surface which could be displayed in either normal, inflated or flattened mode (e.g. Figs 2 and 3).
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
Stereotaxic Localization
The Talairach coordinates for the center of our V4d topologue were: +/–41.25, –81, 8 (SD 5.85, 3.46, 2.71) (Talairach and Tournoux, 1988).
Retinotopy
Next we tested whether the fMRI retinotopy in our V4d topologue matched the retinotopy predicted above (e.g. Fig. 1). At first glance, the human maps of retinotopic eccentricity in the V4d topologue seemed to match the retinotopic predictions of macaque V4d, and V4v in both humans and macaques. As illustrated in Figure 4, the near-foveal stimuli produced preferential activity in the inferior and posterior corner of our V4d topologue, near the confluent foveal representation of V4v, V3/VP, etc. More peripheral retinotopic stimuli activated the V4d topologue further anterior and superiorly, between MT+ and V3A.
|
However, closer inspection revealed that the representation of retinotopic eccentricity was unusual in this V4d topologue. The representation of parafoveal eccentricities (rendered in blue) here was unusually (sometimes vanishingly) thin. In contrast, activation produced by stimuli at central eccentricities (rendered in red) and more peripheral eccentricities (rendered in green) was robust and extensive (see Fig. 4
). By comparison, in the classical retinotopic areas (e.g. V1, V2, V3/VP, etc.), there was a significantly larger representation of the middle eccentricities (see blue band in those areas in Fig. 4), as one would expect from previous measurements of the cortical magnification in other cortical areas (Engel et al., 1994, 1997; Sereno et al., 1995b) and other primate species (Daniel and Whitteridge, 1961; Hubel and Wiesel, 1974; Van Essen et al., 1984; Tootell et al., 1988). This was our first hint that the eccentricity representation is anomalous in the V4d topologue.
For convenience, we describe the cortical zone which responds preferentially to central visual stimuli as ‘LOC’ (Lateral Occipital Central), and the zone which responds preferentially to more peripheral stimuli as ‘LOP’ (Lateral Occipital Peripheral). The anomalous nature of the eccentricity map in LOC/LOP was revealed more clearly by measuring the representation of optimal stimulus eccentricity, in one-dimensional plots measured across the cortical surface. As shown in Figure 5B (bottom row), responses in the V4d topologue were biased for either central or peripheral stimuli (in LOC or LOP, respectively), without the typical monotonic gradations of eccentricity sensitivity seen in V1, V2, V4v and other classically retinotopic areas (Fig. 5B, top row). In most of these one- dimensional samples, we found little systematic variation in the maps of stimulus eccentricity within either LOC or LOP. Instead, there was a pronounced discontinuity at the border between LOC and LOP.
|
Two alternative hypotheses arise from this data. First, LOC and LOP could be two parts of the same visual area, like centrally and peripherally driven regions in classical retinotopic areas, but (for unknown reasons) the representation of eccentricities from ~0.5 to 4° is systematically compressed or absent.
An alternative hypothesis is that LOC and LOP are two distinct cortical areas, located adjacent to each other, but lacking internal retinotopy in either area. This second hypothesis requires furthermore that LOC is driven most effectively by stimuli from the central visual field, and that LOP is driven preferentially by stimuli in more peripheral visual field regions. Similar eccentricity-based biases have been reported previously in other visual areas (Allman and Kaas, 1976; Cusick and Kaas, 1988; Stepniewska and Kaas, 1996).
The first hypothesis is a complicated modification of classical retinotopic maps, like these in V1, V2, V3, V3A. The second hypothesis is driven more by the actual data—such as the relative lack of internal retinotopy within each of LOC and LOP, and the pronounced retinotopic discontinuity between them (e.g. Figs 4 and 5).
This choice between these two hypotheses can be narrowed down by testing the mapping of retinotopic polar angle in the V4d topologue. If there is a lack of eccentricity retinotopy within LOC and LOP (the second hypothesis), we would expect a noisy, inconsistent polar angle selectivity—or none at all—in LOC and LOP. Hypothesis no. 1 would instead predict a systematic, classical polar angle representation of the lower visual field spanning the V4d topologue, as in Figure 1.
In contrast to the mapping of retinotopic eccentricity, the polar angle retinotopy in the V4d topologue was variable across subjects (e.g. Fig. 6). In this region, we did not find any consistent relationship between the human fMRI maps, relative to the polar angle retinotopy predicted in Figure 1. On balance, we conclude that the polar angle retinotopy supported hypothesis no. 2: LOC and LOP appear to be separate cortical areas, each with little systematic internal retinotopy, rather than something akin to macaque V4d.
|
It might be argued that the retinotopic differences illustrated in Figures 4–6
are based on idiosyncratic differences between individual maps, rather than true biological variations that remain consistent across individuals. In order to test this quantitatively, and across subjects, we sampled values of both retinotopic eccentricity and polar angle, from all 12 of the hemispheres (six subjects) in which the retinotopic maps were statistically most robust. The variations in eccentricity were measured in the V4d topologue along lines which were topographically similar to those shown in Figure 5 — approximately perpendicular to the lines of iso-eccentricity (or more accurately in this region, the two zones of foveal or peripheral retinotopic bias). This would be the appropriate way to measure such eccentricity variations, in classically retinotopic areas (e.g. Fig. 5B, top row of graphs). For comparison, we also measured variations in retinotopic polar angle in the V4d topologue, along lines which were oriented approximately parallel to the lines of constant retinotopic eccentricity. To maximize consistency, the end-points of these sampling lines were set at the inferior V3A–V7 intersection on the posterior side, and at the inferior intersection of V4d with MT+ on the anterior side.
As shown in Figure 7A, these averaged values of retinotopic eccentricity show the same behavior as the individual samples shown in Figure 5. The central and peripheral biases were statistically different (average standard deviation = 24.6°). On the other hand, our averaged values of retinotopic polar angle showed no coherent pattern (i.e. the mean phase angles appear randomly distributed), with standard deviations which were correspondingly larger (SD = 104.5°). Thus this group-averaged data confirmed the retinotopic conclusions we reached earlier, from examination of the individual maps (e.g. Figs 4–6).
|
Functional Specialization
Using fMRI, we could also test whether the human V4d topologue responded preferentially to functional stimuli like those suggested by earlier electrophysiological reports in macaque V4d. For instance, does the human V4d topologue respond preferentially to color-varying stimuli?
In previous fMRI experiments (Hadjikhani et al., 1998), we tested for color-selective activity throughout human visual cortex, using two types of color-selective tests. Such tests did produce preferential color-selective activity in some expected locations, including a site in the collateral sulcus that had been described previously (Lueck et al., 1989), which may be involved in the syndrome of achromatopsia. However, such tests did not reveal preferential color-selective activity in the V4d topologue (e.g. Fig. 8). At least in this set of tests, the V4d topologue did not appear to be color-selective.
|
Previous fMRI reports suggested a quite different kind of functional specialization in the human V4d topologue. Orban and co-workers have reported that an area which is apparently co-extensive with our V4d topologue (which they named ‘KO’) responds selectively to kinetic motion boundaries (Orban et al., 1995
; Dupont et al., 1997; Van Oostende et al., 1997).
To test whether our V4d topologue would respond selectively to kinetic motion boundaries, we used copies of the same stimuli used by the Orban group to reveal where that activity was located, relative to our phase-mapped retinotopic boundaries on the flattened cortex. As shown in Figure 9, our V4d topologue did respond selectively to the kinetic motion boundaries, consistent with earlier reports (Orban et al., 1995; Dupont et al., 1997; Van Oostende et al., 1997). The kinetic motion comparison also produced additional activity in human area MT+ (see Fig. 9).
|
Although the human topologue of dorsal V4 did respond selectively to the kinetic motion, ventral V4 did not (see Fig. 9
). This is important because again, in this test, V4v and V4d appear to be distinctly different areas, rather than two retinotopic subdivisions of a single inclusive area. |
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
We began this study with specific and limited aims: to confirm previous descriptions of macaque V4d, which we expected to find in corresponding parts of the human visual cortical map (i.e. the V4d ‘topologue’). Instead, the fMRI evidence suggested either that (1) a V4d homologue does not exist in human visual cortex, or that (2) previous descriptions of function and retinotopy are inaccurate in macaque V4d.
Neither of our fMRI measures of retinotopy (neither eccentricity nor polar angle) matched the reported retinotopy of V4d in macaques. The retinotopic data in the V4d topologue was not even an appropriate lower field counterpart to the classical upper field representation in V4v—in neither humans nor macaques.
Furthermore, certain stimulus comparisons (e.g. kinetic motion boundaries) activated V4d but not V4v. Such functional differences between V4d and V4v are frankly incompatible with the definition of a single inclusive area ‘V4’. As an analogy, if credible new studies revealed that motion selectivity were present in one half of area MT but not the other half, then the cortical maps would likely be redrawn to include two corresponding areas within the original area ‘MT’.
Thus overall, there was very little support for (and much against) the hypothesis that a V4d homologue exists in the expected location in human visual cortex. However, is it possible that we somehow missed the ‘real’ human homologue of V4d? Does an actual homologue, with properties more similar to macaque V4d, lie near (but not in) our predicted human V4d topologue? To answer this question, we systematically reexamined the retinotopy of the surrounding cortical areas V3A, V7, V4v and V8. The goal was to consider whether any of those neighboring areas could be the ‘missing’ V4d homologue, which could have been misinterpreted and/or misnamed previously.
V3A
Here and in previous reports (Tootell et al., 1997, 1998; Culham et al., 1998; Hasnain et al., 1998; Baseler et al., 1999; Boynton et al., 1999; Mendola et al., 1999; Somers et al., 1999; Sunaert et al., 1999; Wandell, 1999), human V3A encompasses a classical, contiguous representation of the entire contralateral visual field (i.e. 180° of polar angle). The inferior vertical meridian is mapped posteriorly (bordering V3), and the superior vertical meridian is mapped at the anterior border of V3A. The foveal representation is mapped inferiorly, and the periphery is mapped superiorly. These properties of human V3A are generally consistent with those described previously in macaque V3A (see Fig. 1).
Human V3A cannot be a misnamed homologue of V4d, for several reasons. First, V3A includes a contiguous representation of both the upper and lower quadrants of the contralateral visual field (180°), whereas macaque area V4d instead represents just half of that extent (90°). Secondly, macaque V4d is located anterior to V3A. Thus if human V3A was instead the actual homologue of V4d, one would have to somehow explain the absence of human V3A.
V7
Anterior to V3A lies another representation of polar angle that includes (at least) the upper visual field, and is mirror-symmetric to that in anterior V3A. This has been called ‘V7’ (Tootell et al., 1998; Press et al., 1999). The representation of eccentricity is not yet clear in V7.
Human ‘V7’ cannot be a misnamed homologue of macaque V4d because the definitive part of V7 represents the upper visual field, whereas macaque V4d is a representation of the lower visual field. It would only compound the problem of the ‘missing V4d’ to propose that human V7 (the upper field representation) is the retinotopic counterpart of human V4v (another upper field representation).
Cortical areas can also be distinguished by whether their overall internal retinotopy matches that in the visual field, or whether it is mirror-symmetric to that. This property has been called the field sign, and it is very resistant to evolutionary change (Sereno et al., 1994, 1995b). The field sign in V7 is opposite to that found in macaque V4d—so again, V7 cannot be a misnamed V4d.
V4v
The retinotopy and topography of human V4v has been described in several previous reports, with good consensus (Sereno et al., 1995a; DeYoe et al., 1996; Tootell et al., 1997; Van Essen and Drury, 1997; Grill-Spector et al., 1998b; Hadjikhani et al., 1998; Baseler et al., 1999; Wandell, 1999) but see Zeki et al. and Bartels and Zeki (Zeki et al., 1998; Bartels and Zeki, 2000). In terms of polar angle retinotopy, human V4v is a classical representation of the contralateral upper visual field, with a field sign which is mirror-reversed relative to that in the visual field (i.e. equivalent to that in V2, but opposite to that in V1 and VP) (see Fig. 10). The vertical meridian is represented posteriorly in V4v, along the border with VP. The horizontal meridian forms the anterior boundary of V4v, adjacent to V8. The fovea is represented superiorly in V4v, with increasingly peripheral eccentricities mapped at progressively more inferior locations in cortex. Unlike the anomalous representation of eccentricity in LOC/LOP, the variations in eccentricity representation are quite continuous and orderly in V4v (e.g. Figs 4 and 5).
|
All these retinotopic features of human V4v are consistent with area V4v as described in macaques (Gattass et al., 1988
; Van Essen et al., 1990; Boussaoud et al., 1991; Felleman and Van Essen, 1991a). Thus human ‘V4v’ cannot be a misnamed ‘V4d’, for multiple and obvious reasons. For instance, V4v is a representation of the upper visual field, whereas a representation of the lower visual field is required in V4d. Furthermore, if V4v were instead a misnamed ‘V4d’, then it would be necessary to explain the absence of V4v. V8
The evidence for an additional retinotopic area which is located immediately anterior to human V4v is described in detail elsewhere (Hadjikhani et al., 1998; Wandell, 1999). This area has several names: either ‘V8’ (Hadjikhani et al., 1998); or ‘V4’ (Lueck et al., 1989), or ‘VO’ (Wandell, 1999). The polar angle retinotopy in V8 is crude (consistent with relatively large receptive fields), but it is consistent across subjects (e.g. Figs 6 and 10).
V8 cannot be an unrecognized or misnamed V4d, for multiple reasons. First, V8 is located ~5.0 cm (center-to-center) from our V4d topologue, based on neighborhood relations in the flattened cortical maps. This is a very long distance across cortex. By comparison, other human retinotopic visual areas (e.g. V2, V3, VP, V4v) are only ~1 cm wide. Thus it could be argued that V8 is separated from our V4d topologue by about four or five ‘cortical area equivalents’.
Secondly, V8 includes a coherent representation of both the upper and the lower visual fields, whereas macaque V4d includes a (much larger) representation of the lower visual field. Thirdly, the foveal representation of V8 lies at its anterior edge, several centimeters from the foveal representation in V4v. Fourth, the field sign in V8 is opposite to that in V4v, so logically V4v and V8 cannot be parts of the same visual area (see Fig. 9).
The original rationale for referring to this small human area (V8) as ‘V4’ is that this area shows some color selectivity (e.g. Fig. 8). However, since the original claims of color selectivity in macaque V4d have not been confirmed, this original rationale is moot.
The border between ‘V8’ and ‘V4v’ is somewhat unusual, since it is not a simple mirror-reversal of polar angle. In data with adequate signal-to-noise, this representation of the (horizontal) meridian is perfectly straightforward (e.g. Fig. 10). However with less robust data, and/or lower polar angle resolution (e.g. pseudocolor only), the horizontal meridian at the anterior border of V4v can be somewhat subtle.
Talairach Coordinates
The Talairach coordinates of the original ventral color-selective region (‘V8’ or ‘V4’ or ‘VO’) were never in dispute, although this has been a matter of apparent confusion (Zeki et al., 1998; Bartels and Zeki, 2000). In fact, there has been fairly good agreement in these coordinates across different studies, with average coordinates near +/–26, –67, –9. However, the averaged coordinates of our V4d topologue were quite different (+/–41.25; –81; 8). This supports all the other evidence that the ventral color-selective area is not the human homologue of area V4(d).
Functional Tests in V4d
In prior fMRI studies, the V4d topologue was not explicitly defined as we have done here. Nevertheless, the approximate location of the V4d topologue can be estimated post hoc in those studies that illustrated the boundaries of V3A, V4v and MT+. Such retrospective analysis reveals that the V4d topologue responds well in a wide range of stimulus comparisons that require information processing across relatively large regions of the visual field. Such examples include tests of illusory contours (Mendola et al., 1999), object recognition (Grill-Spector et al., 1998b; ., 1998b), and visual symmetry (Tyler and Baseler, 1997). A specialization for kinetic motion processing (e.g. Fig. 9) may also require processing across a relatively large region of the visual field. Such a conclusion is consistent with other evidence that the receptive fields in the V4d topologue (i.e. LOC and LOP) are relatively large and non-retinotopic (Tootell et al., 1997).
Neither of the human ‘V4’ subdivisions had unusually high color sensitivity, in the tests we performed so far (e.g. Fig. 8). In that sense, the functional activity in human ‘V4’ was unlike that described in early reports from macaque V4 (Zeki, 1973, 1977, 1983).
Implications for Macaque V4
The present fMRI data brings up another possibility: perhaps macaque V4d is actually more like its human counterpart than previously recognized. This would be easy to rationalize post hoc. The receptive fields in macaque V4d tend to be large and sometimes poorly defined, and electrophysiological retinotopic maps are (by necessity) highly under-sampled. Similarly, retinotopic maps based on neural tracers are usually extrapolations based on one or just a few injections, often across animals. If one is trying to extrapolate maps of an expected continuous, and classical retinotopic map from such data, one could easily miss the unusual retinotopic features revealed by the fMRI in LOC/LOP.
In fact, there is some support for this idea. Even the early studies of Zeki (Zeki, 1977) concluded that ‘this [V4d] is not a homogeneous region but [sic] can be subdivided into separate anatomical and functional domains’, and described it as a ‘complex’ of different sub-regions such as the transiently proposed ‘V4A’ (Zeki, 1977). The early retinotopic mapping studies of Maguire and Bazier (Baizer and Maguire, 1983; Maguire and Baizer 1984) also concluded that multiple retinotopic areas could be distinguished, along a border which lay approximately where the LOC/LOP border would lie in macaque. Even the most well-documented electrophysiological mapping study of V4d (Gattass et al., 1988) shows clear receptive field discontinuities, also near where the LOC/LOP border should be located in macaque.
Subsequent tracer studies by Kaas and co-workers strongly suggest that area DL [which was originally considered homologous to V4d (Baker et al., 1981)], is in fact subdivided into two adjacent areas, based partly on retinotopic eccentricity (DL rostral and DL caudal) along a border similar to that separating human LOC from LOP (Cusick and Kaas, 1988; Steele et al., 1991; Weller et al., 1991; Stepniewska and Kaas, 1996). Cumulatively, all this data suggests that subdivisions similar to LOC/LOP may exist in some species of non-human primates, as well as in humans.
With hindsight, it can even be argued that ‘V4v’ should never have been considered a match to ‘V4d’, even in the macaque. To begin with, the shape and size of the two subdivisions are quite different: ‘V4d’ is large and irregularly circular in shape, whereas ‘V4v’ is topographically long and thin. Thus these two ‘V4’ subdivisions are not mirror-symmetrical, in the way that the more classic quarter-field representations are laid out in V2d/V2v, and in adjacent area(s) V3d/V3v (also known as V3/