Main

Species-specific behaviors are represented in cortical specializations (such as the rodent barrel cortex)1,2,3. Human and nonhuman primates also have behavioral specializations, but whether this is reflected in cortical organization is unknown. Here, we focus on a well-known and defining specialization for visual behavior in primates called the retinal fovea (center of gaze). Of particular interest is the foveola, the centralmost (central 1°) region of the fovea. This structure is characterized by an extremely high density (ten times the density of the nonfoveal retina) of photoreceptors needed for high-acuity vision and color vision4. The foveola is also important for visually directed attention in primates, which is mediated by foveating (moving the eye) to objects of interest5. The essential role of central vision in common tasks such as reading and facial recognition predicts the existence of foveal brain circuits underlying learning and social behavior6,7. While there are extensive studies of extrafoveal brain areas, relatively little is known about foveal and, in particular, foveolar brain organizations.

Study of the foveolar cortex has been challenging because of the need for excellent fixation and fine precision mapping8,9,10. Electrophysiological studies in foveolar locations are difficult because of the extremely tiny (~0.1°) receptive fields and ever-present small eye movements (microsaccades). While eye movements can be largely removed in anesthetized animals, it is difficult to precisely locate (to within <0.5°) where on a screen the fovea is viewing. Thus, because of the precision required, basic questions remained unanswered or controversial. These included whether the foveolar representation is a single confluence or multiply represented across cortical areas9,11,12, what the foveolar cortical magnification factor (CMF) is (a measure of the behavioral importance), and how distinct visual features such as color, shape and motion are represented.

Early electrophysiological studies proposed a ‘foveal confluence’ shared by several visual areas (V1, V2, V3 and V4) and the posterior inferior temporal cortex (TEO)13. The advent of neuroimaging and phase-encoded functional magnetic resonance imaging (fMRI) mapping10,14 provided a revised view, revealing several foveolar loci along the anteroposterior extent of striate and extrastriate areas (both in humans11,12,15 and in nonhuman primates16,17,18). fMRI phase-encoding procedures10 and fMRI population receptive field mapping11 methods for mapping foveolar representation in humans revealed visuotopic representations that converged to a single central foveal confluence. However, closer inspection of these studies revealed inhomogeneity in the central 1° of representation and horizontal and vertical meridian (HM and VM) representations that fall short of the centermost region. The importance of these details was not recognized at the time.

Here, using ultrahigh field fMRI methods, we examined the retinotopy and functional organization of the foveolar cortex in awake, behaving macaque monkeys trained to perform stable foveal fixation in high-field 7T MRI11,12,19. Using a custom 16-channel surface coil20 for enhancing the signal-to-noise ratio (SNR), which achieved 0.1° precision functional imaging (0.6-mm in-plane precision) and using very fine visual stimuli (foveolar stationary visual stimuli and phase-encoding methods), we show that the foveolar representation consists of multiple cortical loci (eight loci per hemisphere). In addition, we identify a previously unknown central area termed the ‘foveolar core’, which responds poorly to small foveolar stimuli, appears to fall outside visuotopic regions of visual cortex and exhibits responsiveness to larger stimuli, very high spatial frequencies (SFs), motion and color. We also support these findings with intrinsic signal optical imaging. Our results suggest the presence of a cortical specialization for foveolar vision.

Results

Stability of acquisition

One major challenge of studying foveolar representation is minimizing eye movements. Two macaque monkeys were trained for several months to maintain stable performance with little body motion (Fig. 1a) and few saccades (Fig. 1b, left) during scanning. Only data in which fixations remained within a 1° radius window >85% of the time and the center of mass of the eye trace remained within a 0.5° radius of the fixation point were used (Fig. 1b, right). With sufficient trial numbers, consistent and reproducible fMRI results were generated across different sessions (Fig. 1c). These procedures made direct retinotopic and foveolar mapping possible (Fig. 1d,e).

Fig. 1: Acquisition stability.
figure 1

a, Left, head motion in six dimensions: roll, pitch, yaw, A–P (anterior–posterior), R–L (right–left) and I–S (inferior–superior). Right, head displacement: disp, displacement relative to the initial frame (mean = 120 µm); Δdisp, difference between two sequential frames (mean = 80 µm). Data were averaged across eight sessions in monkey E. b, Eye fixation behavior. Left, eye position trace in one run. Red and blue lines represent the x and y eye position coordinates, respectively. Two red arrows denote blinks. Note: eye blinks did not cause disruptions of the BOLD signal. Right, all eye positions in four runs (250 Hz, 4 × 280 s). Eye fixations were within a 0.5° radius 94% of the time. Right, despite occasional saccades, across four runs (each trial: 20 s blank and 20 s visual stimulus; seven trials per run; a total of 280 s), the center of mass of the eye trace was within a 0.5° radius of the fixation point (central dotted red circle). With sufficient training, the monkeys maintained stable performance with little body motion and few saccades during scanning. c, BOLD signal time course. Left, time course of activated voxels during eye fixation in b (n = 348 voxels). Right, average BOLD time course. Yellow shading denotes the stimulus on period. d, Map stability. Maps obtained from 3, 7 and 28 trials. Left inset, visual stimulus: 1° and 5° eccentricity rings. Activation clusters remained stable across trials, indicating stable eye fixation. Yellow arrows, V1; red arrows, V2. Right, each dot is a measurement of the shift in center of mass between activations from 7 and 28 trials. For each of the activation clusters, six measurements were taken from six sequential slices. e, Increased trial number, from 7 (middle panel) to 28 (right panel), reveals additional visual areas. Left, section from D99 atlas. Activations in V1 (1° and 5°), V2 (yellow circles), V3 (green arrow), V4 (blue arrow) and TEO (red arrows). Same as in d. Map thresholds were significant at P < 10−3 according to a one-sided t-test. All data are presented as the mean values ± s.e.m.

Source data

Foveolar mapping

To determine the visual representation of the foveal region we conducted visuotopic mapping using very narrow 0.15° lines and arcs during excellent visual fixation (Fig. 2). Maps of VM and HM marked the boundaries dividing cortical areas V1, V2, V3 and V4 (Fig. 2a,b). We also mapped isoeccentricities using arcs ranging from 1° to 6° isoeccentricity and, within the central 1°, several finely distinguished isoeccentricities (Fig. 2c,d) (summarized for each monkey in Fig. 2e,f).

Fig. 2: Mapping visual cortical borders by mapping HM and VM and isoeccentricity mapping.
figure 2

a,b, Stimuli for imaging VM and HM, in monkey E (a) and monkey J (b), comprised horizontal and vertical 0.15°-wide bands containing alternating saturated red and blue checkerboards (insets). All images were acquired within the same sessions but illustrated separately for each area (Methods). For each monkey, the four panels on the left were overlain in the rightmost images. Black and white dashed lines denote the VM and HM, respectively. LS, lunate sulcus; IOS, inferior occipital sulcus; STS, superior temporal sulcus. ECS, ectocalcarine sulcus. Monkey E, P < 10−3; monkey J, P < 10−9 (one-sided t-test). c, Isoeccentricity maps from monkey E. A paired stimulus paradigm (six sets of paired rings: 0.4° and 2°, 0.4° and 4°, 0.6° and 3°, 0.8° and 4°, 1° and 5° and 1.2° and 6°) reduced the number of stimulus conditions needed by half and further confirmed that the maps were stable across sessions (for example, similar 4° activations were obtained in 0° and 4°, 0.4° and 4° and 0.8° and 4° maps). White circle, intersection ___location between isoeccentricity and meridians. d, Isoeccentricity (0.8° and 4°, 1° and 5° and 1.4° and 7°) maps from monkey J. e,f, Summary of findings in c (e) and d (f).

We then used small stimuli placed at the fixation ___location to precisely identify foveolar locations. These comprised alternating saturated red and blue flashing spot stimuli (Methods); to examine the consistency of activation locations, three spot sizes were used (monkey E: 0.4°, 0.6° and 0.8°; monkey J: 0.6°, 0.8° and 1.0°). Given the distribution of eye positions during the fixation task, the most frequently activated locations are at the very center; thus, the foveolar center should be at the ___location of voxels with the highest statistical significance. Given the simple flashing visual spot stimuli used, the activation was strongest for early areas V1 and V2 and weaker at the more anterior V3, V4 and TEO locations; the maps were examined using thresholds (three P values per area) appropriate for the activation levels of each cortical area. Below, we describe how each foveolar ___location was identified in monkey E (Fig. 3). Findings for monkey J are presented in Extended Data Fig. 1. Division of voxels into V1/V2, V2/V3, V3/V4, V4/TEO and TEO/FST (fundus of the superior temporal visual area) regions were guided by locations of VM and HM (black and white dotted lines, respectively, shown in overlay in Fig. 3a, last column) and slice views (Extended Data Fig. 2). Below, we describe each foveolar ___location.

Fig. 3: Determining locations of foveolar representation.
figure 3

a, Activations to foveated small spot stimuli (monkey E: 0.4°, 0.6° and 0.8° flashing saturated red and blue squares alternating at 3 Hz, shown in the first, second and third row, respectively). The activation maps to each stimulus are shown in each column (TEO/FST, V4/TEO, V3/V4, V2/V3 and V1/V2). Each panel presents significant voxels at each of three thresholds (P values according to a one-sided t-test are color-coded: orange, lowest; green, middle; blue, highest). For each visual area, the center of highest significance is consistent across spot sizes (colored dashed lines). The last column consists of the set of all foveola (colored dashed circles, overlay from four leftmost columns). Black and white dotted lines denote the VM and HM, respectively. Gyri, light gray; sulci, dark gray. b, Foveolar activations in monkey E, right hemisphere (same as response to 0.6° in a). c, Foveolar activations in monkey E, left hemisphere. d,e, Foveolar activations in monkey J’s left hemisphere (d) and right hemisphere (e) (Extended Data Fig. 1). In be, small colored circles denote the foveolar activations in each cortical area. Lavender dotted oval, foveolar core.

V1/V2 (fourth column)

In monkey E, each of the three spot sizes activated a similar region (compare three rows in Fig. 3a). Examination of the activations at the lowest threshold (V1/V2 column, orange, 0.4° stimulus) revealed patchy activations across a centimeter-sized region; this activation region is anisotropic (more elongated along the VM axis), consistent with previously observed anisotropies in V1 and V2 because of functional rerepresentations21 and/or anisotropic representations along cardinal axes22. Although eye fixations away from the fixation ___location were relatively few (set by our criterion of <15% of fixations outside of a 1° radius), we cannot exclude the possibility that they contributed to some of the activation seen at lower thresholds. At higher thresholds, quite focal activations were revealed across smaller regions of the cortex (middle threshold, green: 3–8 mm; highest threshold, blue: 1–4 mm). We defined the ‘most foveal’ locus as the lateralmost (because the retinotopic center is located laterally on the macaque visual cortex) activation at the highest threshold (blue) that was consistent across the three spot size maps. The dorsal locus overlaid the V1/V2 border (red dotted lines) and another locus was slightly ventral to the HM in V1 (magenta dotted lines). Across the 0.4°, 0.6° and 0.8° spot sizes, the locations were similar (red dotted circles in overlay) and consistent with the overall topographic (HM and VM) organization.

V2/V3 (third column)

There were also two loci for V2/V3. The dorsal activation fell within the banks and depths of the lunate sulcus (LUS, yellow dotted lines) consistent with previous studies23,24, had a distributed patchy appearance and was topographically consistent with a foveolar ___location along the V2/V3 border (yellow dotted circle in overlay). A second foveolar locus (evident for the 0.6° and 0.8° spots but below the threshold for the 0.4° spot) fell in the inferior occipital sulcus (IOS, orange dotted lines; orange dotted circle in overlay).

V3/V4 (second column)

Two distinct foci were seen at the V3/V4 border, one on the anterior bank of the LUS (cyan dotted lines) and one at the tip of the IOS (green dotted lines). There were also a few patchy activations extending onto the cortical convexity (between the ends of the LUS and IOS and posterior to the superior temporal sulcus (STS)). The locations of these loci fell at the end of the V3/V4 border representing the VM (cyan dotted circle in overlay).

V4/TEO (second column) and TEO/FST (first column)

The TEO is a midtier area and likely to exhibit somewhat weaker responses to simple spots25. Despite this, a locus of patchy activation is visible at the tip of the IOS (white dotted lines, V4/TEO border) and at the posterior border of the STS (blue dotted lines, consistent with activation at the TEO/FST border16).

Bilateral representation

Because the fovea is known to be represented in both hemispheres, we also examined the foveolar representation in the contralateral hemisphere. The radiofrequency (RF) coil was typically placed slightly off-center biased toward one hemisphere, such that we could capture as much of the ventral pathway as possible. The SNR on the contralateral hemisphere was, therefore, slightly lower but substantial blood-oxygenation-level-dependent (BOLD) signals could still be captured. In addition to the activation shown above (Fig. 3b, same as that shown in Fig. 3a, middle row), foveolar activations in the contralateral hemisphere, obtained from the same sessions, are shown in Fig. 3c (slice views in Extended Data Fig. 2). Each of the foveolar locations identified in Fig. 3a are indicated by circles of the corresponding color. Activation in TEO on the right side was weak. Similar data for monkey J are shown in Extended Data Fig. 1 and summarized in Fig. 3d,e.

In sum, these methods revealed precise foveolar activation locations. Foveolar activation often comprises one to a few distinct patches. The locations of these activations were stable across multiple sizes of spot stimuli and were consistent across multiple thresholds. In addition, we show that these activations were obtained with both better and poorer eye fixation behavior (Extended Data Fig. 3), indicating that the results were not an artifact of eye movements and that they were not because of edges of the spot stimuli or subtraction of the fixation cross (Extended Data Fig. 4). These data suggest that, in contrast to what was previously viewed as a singularity, there are at least eight distinct nodes of foveolar representation, one near the end of each VM and HM for dorsal and ventral V1/V2, V2/V3, V3/V4, V4/TEO and perhaps TEO/FST. This mapping is simultaneously observed in both hemispheres. Thus, the foveolar representation is not a single point or confluence and is more than one per visual area, as previously suggested. Moreover, these findings suggest that there is cortical territory bounded by these foveolar representations that is not within the visuotopically mapped regions (Extended Data Fig. 4). We provide a schematic ring (dotted lavender ring in Fig. 3b–e) around this central region and refer to it as the foveolar core (or core). This ring is not meant to be a specific border but an indication of the region of cortex bounded by the foveolar loci.

Phase encoding reveals a nonhomogeneous region and distinct foveolar core responses

As phase encoding is the standard method for mapping retinotopy, we also conducted phase encoding for the central 3° in a fixating monkey (monkey J). Schira and colleagues15 studied the foveolar region in humans with phase encoding using very fine stimuli, high-resolution functional imaging (1.2-mm isotropic resolution) and sophisticated analysis. We attempted a similar study using a continuous sequence of very fine isoeccentric thin (0.15°) rings and narrow (5°) isopolar wedges. The resulting phase maps are shown in Fig. 4a,b (isoeccentricity and isopolarity, respectively); the foveolar core is indicated by a dashed lavender ring and foveolar loci are denoted by white circles. Consistent with phase-encoding studies10,14,15,16,26, in the isoeccentricity map, there was a systematic phase shift in V1, V2 and V3 on the lateral operculum, with the centermost (least phase-shifted) voxels (red) appearing most lateral on the cortex. This is further illustrated by the shifting time courses across regions of interest (ROIs) 1–6 along the V1/V2 border (Fig. 4d, left column; graphs 1–6, shifting red arrowheads); the regularity of the time courses and small degree of variability in these traces indicate that eye fixation was excellent.

Fig. 4: Phase encoding using fine isoeccentricity and isopolarity stimuli in visuotopic V1–V4 and in foveolar core.
figure 4

a, Isoeccentricity map. Inset, isoeccentricity stimulus: continuous increasing size of isoeccentric rings (each ring: 0.15° wide; one trial consisted of 50 rings over 3°, 0.06° per ring, lasting 50 s with a 10-s ISI; fixation point: 0.05° white dot, constant over a total of 21 trials). b, Isopolarity map. Inset, isopolarity stimulus: continuous rotating checkerboard wedge (5° wedge, rotated over 50 s from 90° to −90° in the contralateral visual field; size: 3°, over a total of 21 trials). Black circles in a and b denote the locations of voxel time courses shown in d,e. White circles in a and b denote the foveolar loci identified in Fig. 3d,e. c, Histogram showing that FFT amplitude at the stimulus frequency is weaker in ‘inside core’ compared to ‘outside core’ samples. FFT amplitude: inside core, mean = 1.2; outside core, mean = 3.0; P < 0.0001 (two-sided chi-square test). d, Isoeccentricity time courses (BOLD%). ROIs 1–6 are in V1 along the VM. Red arrowheads (center of mass of peak) indicate a peak shift as ROIs approach the center. ROIs 7–15 are in the foveolar core along an anteroposterior axis through the core. ROIs 16 and 17 are along a posterosuperior axis. ROIs 18 and 19 are along a posteroinferior axis. ROIs 20–22 are along an anteroinferior axis. e, Isopolarity time courses (BOLD%). ROIs 1–9 are along three isoeccentricity rings in V1. ROIs 10–12 are in foveolar core parallel to isoeccentricity rings in V1. Note: because the dorsal cortex represents ventral fields, the phase shift because of the wedge approaching the VM in the ventral field results in a shortening latency to response (red arrowheads). Red boxes represent time courses from ROIs in the core region. Numbers in the right corner of each graph denote FFT amplitude indices (Extended Data Fig. 6). All data in d,e are presented as mean values ± s.e.m.

Source data

We hypothesized that, if the foveolar center were within the core, we would observe continuously shifting phase from the several foveolar loci toward the center. We, therefore, sampled phase response at locations through the most central (red) regions of the foveolar core starting from (1) the dorsal V1/V2 foveolar locus (ROIs 7–15); (2) the dorsal V2/V3 to the ventral V2/V3 foveolar loci (ROIs 16–19); and (3) the ventral V3/V4 foveolar locus (ROI 20–22). Anterior V4/TEO locations were outside our imaging field of view (FOV) in this experiment. As shown in Fig. 4d (three rightmost columns), we found that these samples exhibited surprisingly noisy responses that were poorly phase-locked to the stimulus.

In the isopolarity map (Fig. 4b), outside the foveolar ring, the ROIs were well modulated and showed phase shifting (Fig. 4e, top three columns, ROIs 1–3, 4–6 and 7–9 along central isopolarities, shifting red arrowheads). Again, inside the core region, sampled from a parallel arc just inside the border of the core region (Fig. 4e, ROIs 10–12) and in the central zone of the core region (Fig. 4e, ROIs 13–15), the traces were noisy and not well modulated. This was supported by significant differences in the peak amplitude of the response frequency of ‘within core’ and ‘outside core’ samples (Fig. 4c and Extended Data Fig. 5).

Thus, whereas our methods could detect phase shifting outside the core, the responses in the core appeared noisy and poorly phase-locked. If the noisy signals were because of eye movements, the foveolar traces from outside core samples would also be noisy, inconsistent with the stable traces obtained. It may be possible that neurons in the core prefer even smaller stimuli (<0.15°); however, the within core traces did exhibit responsiveness to the stimuli (large magnitude fluctuations) but had poor phase correlation. While the nature of the noisy traces remains to be examined, these data suggest that responses in the core region are qualitatively distinct from those in retinotopic regions of visual cortex. We suggest that phase-encoding maps obtained in humans also show similar organization (Extended Data Fig. 6).

Are there functional domains within the foveolar cortex?

One of the defining features of visual cortical areas is their functional organization for different parameters of visual information. Well known are the stripes of V2 that range in size from 1 to 2 mm in width and the distinct sets of feature specific submillimeter-sized domains within the thin, pale and thick stripes21,23,27. While, imaging millimeter-scale or submillimeter-scale modular structures with fMRI in monkeys has been challenging17,23, we illustrate here that V2 color (thin) stripes were well mapped by color versus achromatic gratings but failed to evoke much activation in the foveolar core (Extended Data Fig. 7). We then proceeded to explore the foveolar core response using a small battery of visual stimuli, including high and low SFs, simple color and achromatic stimuli and motion dot stimuli.

High-SF domains

A defining feature of foveolar function is its high visual acuity, marked by neuronal responses to high SFs visible only to foveolar vision28. Neurons within foveolar locations of macaque V1 have extremely small receptive field sizes (predominantly 10–20 minutes of arc in size29) and exhibit peak SF tunings up to ~5–8 cycles per degree30. Peak SF tunings based on fMRI BOLD signals reach ~3 cycles per degree31. However, perceptual detection and discrimination can reach SFs well over 10 cycles per degree in humans28. Thus, a potential function of the core region may be to integrate lower-order SFs to achieve high spatial acuity28,32. In this section, we use the term ‘central’ to mean foveolar, not in the topographic sense but in the functional sense of central visual perception.

To capture central SF responses, we examined response to high-SF achromatic gratings (11, 15 and 18 cycles per degree) and, for comparison, to low-SF achromatic gratings (0.2 cycles per degree) and to small flashing color spots. Monkeys maintained excellent fixation behavior during presentation of these stimuli. As shown in Fig. 5a, activations to SF11 (yellow-green), SF15 (purple) and SF18 (orange) gratings revealed patchy activations scattered throughout the core (dashed oval). Moreover, SF11, SF15 and SF18 showed little overlap, suggesting that they are distinct functional domains (Fig. 5a,g; overlap index for all comparisons < 0.05). In sum, these high-SF stimuli appear to effectively activate the region within the core region and exhibit distinct spatial activations.

Fig. 5: Mesoscale functional domains within the foveolar core.
figure 5

Color-coded responses to different visual stimuli. a, Activations to achromatic high-SF gratings (in cycles per degree). Yellow, SF11; purple, SF15; orange, SF18. b, Overlay of high-SF (yellow, SF11 + SF15 + SF18) and low-SF (red, SF0.2) activations. c, Overlay of low-SF (red, SF0.2), motion (cyan, clockwise moving dots) and color (blue, 0.8° flashing spot) activations. d, Overlay of all domains. e, Four fields of view from d showing multiple types of functional domains within a small region. f, Averaged time course of each ___domain type selected from 3–5 clusters within the core. All data are presented as the mean values ± s.e.m. The numbers of voxels from top to bottom are 588, 420, 387 and 385, respectively. g, Overlay indices between different populations of domains show little overlap between domains (overlap indices: 0.09, range 0–0.25). h, Domain size distribution (total number: 111; Methods and Extended Data Fig. 8).

Source data

We then compared these maps to those obtained with a low-SF (0.2 cycles per degree) achromatic grating, an SF typically used for studying surface properties such as color and luminance. The SF0.2 responsive domains (Fig. 5b, red) were largely nonoverlapping with the high-SF domains (Fig. 5b, yellow: sum of SF11, SF15 and SF18; overlap indices < 0.1). Achromatic SF0.2 domains (Fig. 5c, red) also had low overlap with domains activated by a small (0.8°) color stimulus (Fig. 5c, blue; overlap index = 0.064). For comparison, response to small motion dot stimuli (2° patch of dots rotating clockwise) (Fig. 5c, cyan) elicited activations with low overlap with SF0.2 achromatic (red; overlap index = 0.25) and slightly lower overlap with color domains (blue; overlap index = 0.17). Domains of different types appeared within small regions within close proximity to one another and did not appear to coalesce into stripes or bands (examples in Fig. 5e). Averaged time courses of different domains within the foveolar core exhibited reasonable BOLD time courses (Fig. 5f), indicating that these activations were not artifactual.

Domain sizes

We also estimated the sizes of the activation domains. The area of each activation was calculated by measuring the number of nodes within the activation when overlaid onto the brain surface mesh33 (Extended Data Fig. 8). Given the 0.6-mm in-plane resolution, we conservatively classified domains as <1 mm, 1–2 mm and >2 mm in size. Domain diameters (the longest axis) revealed that they were largely <2 mm in size (Fig. 5h, top; 82% ≤2 mm in diameter) and held for different functional ___domain types (Fig. 5h, bottom).

Thus, remarkably, what originally appeared as a region of cortex with little response to foveolar spot stimuli was actually a region densely populated with spatially distinct functional ‘domains’ of different feature selectivity (___domain overlay in Fig. 5d; overlap indices < 0.15). Unlike the stripe and band organizations in early visual areas, each of the ___domain types appeared to be distributed broadly across this central zone, resulting in a mosaic-like functional architecture.

A view from optical imaging

To further examine the topographic and functional organization of the foveolar core, we conducted intrinsic signal optical imaging of this area in anesthetized monkeys. To functionally localize the foveal representation in anesthetized monkeys, determination of the foveolar ___location was determined by sequential, iterative imaging of large-to-small-interval vertical and horizontal bar stimuli to zero in on the precise x and y coordinates on the monitor (Methods). As shown in Fig. 6e, the lateral visual cortex was exposed, including lateral parts (central 0.5°) of V1, V2 and V4 and part of the TEO. Maps for ocular dominance revealed the V1/V2 border (Fig. 6a, fine dotted line). Mapping orientation preference (for example, horizontal versus vertical) revealed orientation maps in V1, orientation stripes in V2 (corresponding to the thick (‘disparity’) and pale (‘higher-order orientation’) stripes; regions between the yellow arrows) and larger-orientation (‘curvature’) responsive bands in V4 (Fig. 6b, outlined in green). Color versus achromatic response revealed blobs (‘color’) in V1 (pattern of dark dots), thin (‘hue and brightness’) stripes in V2 (yellow arrows) and larger bands of ‘contextual color and brightness’ response in V4 (outlined in red) (Fig. 6c). The color-responsive stripes in V2 aligned well with dark cytochrome oxidase histology (Fig. 6d, three stripes indicated by yellow arrows). Each area (V1, V2 and V4) was characterized by distinct and well-documented functional organizations34,35,36. These findings establish the quality of our optical imaging methods.

Fig. 6: Multiple foveolar representations revealed by optical imaging.
figure 6

a, Ocular dominance (OD) map. Scale bar, 1 mm. b, Orientation map (45° versus 135°). Green outlines, V4 orientation bands; red outlines, V4 color bands. c, Color map (color versus achromatic). Yellow arrows, color stripes in V2 (same locations as stripes shown in d). d, Cytochrome oxidase stained stripes in V2 align well with color stripes in c (yellow arrows). e, Blood vessel map over V1, V2, V4 and TEO. M, medial; L, lateral; A, anterior; P, posterior. f, Color-coded eccentricity map (red, 0.01°; yellow, 0.10°; green, 0.20°; cyan, 0.30°; blue, 0.40°; purple, 0.50°). Lavender dashed circle, foveolar core; colored arrows, shifted eccentricities. White arrow: V1/V2 border. g, Color-coded isopolarity map (red, 0°; yellow, 15°; light green, 30°; dark green, 45°; cyan, 60°; blue, 75°; purple, 90°). Numbered white dotted circles, foveolar locations at V1/V2 (1 and 2), V2v/V3v (4), V3v/V4v (6), V4v/TEO (7) and TEO/FST (8). Locations V2d/V3d (3) and V3d/V4d (5) are within the LUS and not visible in optical imaging. Colored arrows, shifted polar angles. h, For comparison, 7T MRI of all foveola in monkey E (left hemisphere, from Fig. 3b) numbered with the corresponding locations in g. The orientation of all optical imaging maps was rotated to correspond with the MRI results shown in h.

Central 0.5° visuotopy

To determine foveolar representations, we conducted visuotopic mapping using thin (two pixels wide) (1) isoeccentricity arcs at 0.01°, 0.1°, 0.2°, 0.3°, 0.4° and 0.5° eccentricities (Fig. 6f, inset; red, yellow, green, cyan, blue and purple, respectively) and (2) isopolarity lines, from VM to HM at 90°, 75°, 60°, 45°, 30°, 15° and 0° (Fig. 6g, inset; red, yellow, light green, dark green, cyan, blue and purple, respectively). These generated vector maps (the color of each pixel reflects the preference for one of the respective isoeccentric and isopolar conditions; single condition maps in Extended Data Fig. 9). In V1, isoeccentricity of the central 0.5° mapped mediolaterally (Fig. 6f, from purple to blue, green, yellow and red) and isopolarity mapped roughly posterior to anterior (Fig. 6g, from purple to blue, green, yellow and red). V2 (which only had a 2-mm-wide region visible on surface posterior to LUS) exhibited a similar mediolateral mapping of isoeccentricity (Fig. 6f) and shared the VM representation at the V1/V2 border (Fig. 6g, white arrow). The yellow dashed circles in Fig. 6f indicate corresponding locations of 0.1° eccentricity in V1, V2 and V4. Note that, in the isopolarity color code, the foveal center responds to all isopolar conditions; this led to no preference for any single condition and, therefore, resulted in black colored pixels. Thus, the corresponding foveolar representations (foveolar loci 1 (V1/V2d) and 2 (V1/V2v)) appeared red in the isoeccentricity map and black in the isopolarity map.

In V4, where receptive fields were larger than in V1 and V2, the map was weaker in response to the fine isoeccentricity and isopolarity stimuli but still consistent with known V4 topography. The isoeccentricity topography ran mediolaterally (Fig. 6f, colored arrows; the most medial regions were purple, blue and green, with yellow pixels further lateral). Even more lateral was a large dark region (likely indicating poor response to the central 0.01° stimulus) consistent with the foveal V4 ___location (dashed red arrow). In the isopolarity map, V4 exhibited weak responses but an anteroposterior isopolar topography could be discerned (in Fig. 6g, colored arrows; from purple to blue, green and red). Within TEO, the topography was less clear and also weak to central stimuli.

Foveolar centers and foveolar core

Two foveal centers at the V1/V2 border (1 (V1/V2d) and 2 (V1/V2v)) were seen as red in Fig. 6f and black in Fig. 6g. Foveolar loci 4 (V2/V3v, at dorsal lip of IOS), 6 (V3/V4v, more anterior on dorsal lip of IOS) and 7 (V4–TEOv, anterior on ventral lip of IOS) also appeared as red in Fig. 6f and black in Fig. 6g. Foveolar locus 8 (V4/TEOd) was visible on the surface and could extend into the STS (fewer red pixels in Fig. 6f because of weak response to tiny foveal stimuli). Foveolar loci 3 (V2/V3d) and 5 (V3/V4d) were not visible as they were within the LUS (Fig. 6h). The locations of foveolar loci 6, 7 and 8 were consistent with the corresponding loci shown in fMRI (Fig. 6h). In the core region (approximated by the lavender ring), there were intermingled patches of yellow, green, blue, red and purple, as well as dark regions.

In sum, the primary finding is that the optical imaging data were largely consistent with the fMRI findings. First, there were multiple foveolar loci (dark circled loci in Fig. 6g). Secondly, the visuotopic maps, most clear in V1 and V4, did not continue into the core region. Thirdly, there were patchy intermingled activations in the core region. Although these fMRI data are from different animals, these foveal centers provide support for those identified using fMRI (corresponding numbers in Fig. 6h). Following this reasoning, the lavender dashed oval region in the optical image in Fig. 6f,g indicates the approximate ___location of the foveolar core. Because of the folding of the cortex, the region corresponding to the foveolar core in the optical image may appear smaller than in fMRI. Thus, with a completely different imaging methodology and different spatial resolution, similar results were obtained, further strengthening these findings.

Foveolar cortical magnification

This finding introduces new questions regarding how much cortex is devoted to foveolar representation. The CMF, defined as the millimeters of cortical area devoted to processing a degree of visual angle, for each area has been well studied37,38,39,40; however, detailed data from the central 1° remain lacking41. CMFs calculated from retinotopic mapping (Fig. 2e,f) are shown from two monkeys in Fig. 7a,b, providing direct and precise fMRI measurements of cortical magnifications for the central 1°. CMF values at the very center reached up to around 24.6 mm per degree (red boxes), extending values from previous studies outside the central 1° (colored lines)38,41,42. Differences between the CMFs of V1 in the two monkeys were consistent with previously reported interindividual variability43 (previous values ranged from 4.5 mm per degree43 to 30 mm per degree20). From optical imaging data (Fig. 6), calculation of the CMF in V1 in the central 0.5° reached about 16 mm per degree (calculated from Fig. 7c). Furthermore, from the optical imaging data (Fig. 7f), the area of the central 0.1° (yellow pixels) was ~15 mm2, resulting in an areal CMF of 1,500 mm2 per degree, and that of the central 0.01° (for the V1 half of the red pixels in V1–V2d) was ~2.2 mm2, resulting in an areal CMF of 22,000 mm2 per degree. The greater precision of optical imaging provided better resolution for evaluating the relative increase in cortical magnification, illustrating that, rather than leveling off, the CMF was much higher at the most central visuotopic locations. Estimating the area measured from fMRI (Methods), a single quadrant of the central 1° occupied an area of ~70 mm2 for dorsal V1 or ~140 mm2 for a single hemisphere of V1. If we take a very crude approximation that V2, V3 and V4 have a similar area of representation (V2 and V3 may have slightly greater CMF15), then that number becomes four times larger (~560 mm2).

Fig. 7: The foveolar core has a large CMF.
figure 7

a,b, Linear CMF in central 1° of V1 as a function of eccentricity for monkey E (a) and monkey J (b) (Extended Data Fig. 1). Colored lines were taken from a previous study37. c, Linear CMF in central 0.5° of V1 (from optical imaging data in Extended Data Fig. 9). d,e, Foveolar core bounded by eight foveolar loci in monkey E (d) and monkey J (e). Yellow outlines denote the boundaries for area size calculation, approximated by a pink dotted oval. f, Optical image of central 0.01° and 0.1° (yellow circles). The CMF of the central 0.1° (yellow pixels) is 1,500 mm2 per degree and that of the central 0.01° (half of the red pixels in V1/V2d) is 22,000 mm2 per degree. g, A summary schematic of topographic (light gray) and nontopographic (dark gray) foveolar representation. Numbers denote the estimated area (mm2) per degree.

Source data

Size of foveolar core

To estimate the area in the foveolar core, we connected the eight foveolar loci identified in Fig. 2 and calculated the circumscribed area (Fig. 7d,e, yellow outlines). The approximate areas for this zone were 180 mm2 and 302 mm2 (for the left and right hemispheres of monkey E) and 291 mm2 and 271 mm2 (for those of monkey J), leading to rough estimates of the radius of the core regions of about 8–10 mm. To further check that our numbers were consistent with our optical imaging, the measured distance between the dorsal and ventral V1/V2 foveolar locations (loci 1 and 2 in Fig. 5g) was 7.2 mm (Extended Data Fig. 9h). This was comparable to the surface distance between the two V1–V2 foveolar locations from the fMRI map (Fig. 6h; loci 1–2, 7.2 mm; other example distances: loci 5–6, 5.5 mm; loci 7–8, 11 mm; loci 5–7, 6.3 mm; loci 5–8, 14.7 mm). Thus, both fMRI and optical imaging data provided areal estimates of hundreds of squared millimeters. Note that the centermost stimulus of 0.01° (red pixels in Fig. 6f and Extended Data Fig. 9h) covered a large area extending over 7 mm along the V1–V2 border, resulting in a centermost linear cortical magnification of ~700 mm per degree (Fig. 7e). In sum, the foveolar core region is large (roughly 200–300 mm2 in size). Adding in the approximated area of foveolar V1–V4 of 560 mm2, the total area devoted to foveolar vision is then upwards of 800 mm2 (Fig. 7g, dorsal foveolar V1 shown in light gray, foveolar core shown in dark gray).

Discussion

Summary

We report two primary findings on foveolar representation in this study. Previous visuotopic mapping studies characterized the centermost foveal region as a single foveal confluence13,16 (Fig. 8a, upper panel) or several foveolar loci arranged anteroposteriorly15,40,44,45,46,47,48,49,50,51 (Fig. 8a, lower panel). Our revised view (Fig. 8b, red stars) suggests that (1) there are four dorsal and four ventral foveolar loci per hemisphere, each at the centralmost point of topographic visual areal borders (V1/V2, V2/V3, V3/V4 and V4/TEO; eight red stars) and (2) these eight loci surround a newly identified cortical region, which we term the foveolar core.

Fig. 8: Revised view of visual foveolar representation.
figure 8

a, Previous classical view of topographic representation (light blue). The foveolar confluence (red stars) comprises a single foveolar locus (top)13 or one locus per area (bottom; in humans and macaques)15,16. b, Our study shows that (1) the foveolar center is represented eight times (eight red stars), one at each of the dorsal and ventral representations of V1/V2, V2/V3, V3/V4 and V4/TEO; (2) the foveolar core (orange) is an area within the ring of stars and outside the area of visuotopic representation; and (3) there are functional domains (colored dots) within the core that are responsive to large but not small foveolar stimuli (including color, orientation and motion stimuli). Red lines, HM; green lines, VM.

Foveolar maps

We used fMRI and optical imaging mapping approaches to arrive at these conclusions. Following on previous mapping studies of the foveolar visual cortex in humans and monkeys15,18,21,45, we further developed ultrahigh-field fMRI methods coupled with the use of very fine foveolar stimuli in well-trained fixating monkeys. To our knowledge, this is one of the few awake macaque monkey studies at 7T (refs. 45,52,53) and the second study obtained in a human 7T scanner45. We considered the possibility that these results are because of vascular artifacts, imaging artifacts, eye movement-related blurring issues, stimulation by edges of the stimuli and the subtraction of fixation stimulus but found these possibilities to be inconsistent with the data (Extended Data Figs. 37). We also mapped the central 1° using phase encoding in response to fine stimuli, which revealed systematic isoeccentricity and isopolarity maps for the visuotopic foveolar cortex but noisy responses with poor correlation to the phase of visuotopic stimuli in the foveolar core. This difference is not because of eye movements, as both foveolar loci and foveolar core responses were simultaneously obtained. Interestingly, the region within the core contained nonhomogeneous phase maps, also seen in previous studies11,15, raising potential links to described functional inhomogeneities in foveolar vision54. Lastly, we replicated previous extrafoveolar CMFs20,37,38,41,42,50,55,56 and reported new data on foveolar CMFs (V1 area, ~140 mm2, 16–25 mm per degree for the central 0.1° and approaching 700 mm per degree for the central 0.01° Fig. 7c,f). Combining these estimates of foveolar V1–V4 (~560 mm2) with the total area of the foveolar core (~200–300 mm2) comes to a total of roughly 800 mm2. As a point of comparison, the area of V1 (ref. 52) in one hemisphere ranges from 800 to 1,500 mm2, making the cortical area related to foveolar vision proportionally quite high.

Functional domains

Other features also distinguished this central area. The foveolar core was marked by interspersed millimeter-scale functional domains. Some domains preferred very high SFs consistent with high acuity in foveolar vision28,57,58,59 (11, 15 and 18 cycles per degree; higher SFs were not tested), much higher than previously mapped nonfoveolar SFs mapped with functional imaging30,54,60,61,62. Other domains were responsive to color and motion dot stimuli. Notably, these domains were not seen with small foveolar stimuli and yet were readily activated by larger stimuli (for example, high-SF, full-field and motion dot stimuli). It is possible that the populations of functional domains in the core contribute to the nonhomogeneous nature of phase-encoded maps. While further investigation is needed to understand the function and organization of these domains, we raise the possibility that populations of functional domains, distinct from those in areas V1–V4, may be specialized for handling the demands of foveolar vision.

An architectural specialization for foveolar rerepresentation

Our findings suggest that the foveolar core is a concept distinct from the foveolar confluence. Whereas the confluence refers to a singularity or multiple singularities where multiple cortical areas meet, the core reflects an independent area with distinct defining features. The foveolar core’s substantial size is an evolutionary statement of its behavioral importance. The architecture of a central zone nestled amongst multiple (eight) foveolar representations in each hemisphere encapsulates par excellence the concept of cortical specialization. While rerepresentation is itself a type of specialization (for example, ocular dominance columns in V1, color, shape and disparity stripes in V2, curvature and hue domains in V4 and motion and disparity domains in the middle temporal area), the foveolar core may have a role in coordinating multiareal rerepresentation and tying together the four quadrants of visual space.

Methods

Subjects

Animal care and experimental procedures were performed in accordance with the National Institute of Health’s Guide for the Care and Use of Laboratory Animal and approved by the Institutional Animal Care and Use Committee of Zhejiang University. Two rhesus monkeys (monkey E: Macaca mulatta, 8–10 kg, male, age 9; monkey J: M.mulatta, 4–6 kg, female, age 7) were used in this study. Monkeys were implanted with MRI-compatible headposts and trained to sit in a natural ‘sphinx’ position inside a plastic box. To motivate monkeys to learn the task, water was provided only during the task performance, 6 days per week. Water was provided ad libitum on the seventh day. A daily minimum of water was provided in the chair and supplemented with a variety of fruit. To ensure that monkeys remained healthy, food intake, urine and feces were monitored daily and body weight was monitored weekly.

Behavioral training

Motion training

The monkeys were trained to sit in a sphinx position in a customized plastic monkey chair20. The head was restrained with an implanted headpost secured to the chair. Although the head was restrained, body movements could still produce large motion-related imaging artifacts. We, therefore, spent some effort on training the monkeys to limit body movements. Our motion training system consisted of a three-dimensional (3D) motion sensor (placed on the monkey’s neck) that triggered both negative feedback during body movement (jingling sound and withholding of juice) and positive feedback when keeping still for a set period of time (low-frequency sound and juice reward). This training was largely conducted in the mock bore before onset of training in the scanner. Monkeys were trained to keep still for up to several minutes at a time, a period sufficient for scan acquisitions, and to adapt to the scanning noise.

Fixation training

Once the monkeys were trained to sit still in the monkey chair, they were placed in a mock scanner, directly facing a liquid crystal display screen, positioned 60 cm from the monkeys’ eyes. During initial training, they were required to maintain fixation within a 1° window centered on a red dot (0.3°) in the center of the screen. Eye position was monitored at 250 Hz through pupil position and corneal reflection (Eyelink 1000 plus, CR system). During the training phase, the monkeys were rewarded (fruit juice) for keeping fixation on the red dot within the fixation window for the required time. The fixation duration was gradually extended over time from 0.5 s to 3 s. After several months of training, both monkeys learned to maintain good fixation lasting 1–3 h. This resulted in stable activations in response to isoeccentricity and isopolarity stimuli. The stability of activation was assessed by calculating the right, anterior and superior coordinate distance between the center of mass across different trials (Fig. 1d) in the volume.

Scanning training

Even though the monkey had received training to become accustomed to scanning noise in the mock bore, scanning noises in the actual 7-T scanner were still much louder than the recorded ones. To help monkeys adjust gradually, a mild sedative (dexmedetomidine, gradually reduced from 20 μg kg−1 to 8 μg kg−1) was used over the initial 1–2 weeks. Monkeys were trained on the identical procedures learned in the mock bore. Monkeys quickly adjusted and were able to maintain good fixation in scan sessions lasting 1–4 hours. Only data acquired during good fixation (fixation maintained within a 1° radius >85% of the time) were used in this study, as plotted and calculated using Matlab version 9.13.0 (R2022b; MathWorks).

MRI

Custom RF coil

For the purpose of enhancing SNR, we built a custom 16-channel surface coil20 that greatly enhanced the SNR within a local cortical region. A special acquisition strategy with combined reduced-FOV imaging and high parallel imaging acceleration was used to maximize the spatial encoding capability of the MRI system. This is key to enabling submillimeter-resolution functional images (0.6-mm in-plane resolution). While this local transmit coil produces a nonuniform excitation profile, careful positioning of the coil provides a uniform field of foveolar cortex (defined as B1max/ B1min < 2). A spatial limitation of the coil is that it cannot reliably provide coverage for foveola of both hemispheres; therefore, we focused on a single hemisphere in each animal.

Anatomical

Scans were performed on a 7-T MRI system (Magnetom, Siemens Healthcare) for all experiments. Whole-brain structural images were acquired with a single-loop coil (RAPID Biomedical). To obtain T1-weighted structural images, we used a 3D magnetization-prepared rapid acquisition with gradient echo sequence with optimized parameters (repetition time, 2,590 ms; echo time, 2.73 ms; inversion time, 1,050 ms; matrix size, 192 × 192; FOV, 96 mm × 96 mm; slice thickness, 0.5 mm; flip angle, 7°; bandwidth, 250 Hz per pixel; three averages; scan time, 24 min and 52 s). Structural images of high quality were generated from averaging across multiple sessions.

fMRI

In awake monkeys performing a fixation task and a custom 16-channel dense multiarray coil, we imaged multiple visual cortical areas, including V1, V2, V3, V4 and TEO. Single-shot T2*-weighted gradient-recalled echo-planar imaging was used for functional scanning (repetition time, 2,000 ms; echo time, 35 ms; matrix size, 182 × 140; FOV, 110 mm × 85 mm; spatial resolution, 0.6 mm × 0.6 mm × 1 mm; flip angle, 70°; bandwidth, 784 Hz per pixel; echo spacing, 1.44 ms; parallel imaging acceleration factor, R = 3 according to generalized autocalibrating partial parallel acquisition). A block design was applied with a paradigm of 20 s of blank and 20 s of stimulus presented (each trial 40 s). Each run included seven trials, resulting in a scanning time of ~5 min with autocalibration reference scanning at the beginning. In total, 560 repetitions (four runs) were included for one functional mapping. Most of the brain images were acquired in a close-to-horizontal plane.

Monkey E provided spot imaging data from 32 runs (224 trials each, 7 trials per run, 4,480 trial repetitions) in four sessions and provided meridian and isoeccentricity data from 32 runs (224 trials, 7 trials per run, 4,480 trial repetitions) in six sessions. Monkey J provided spot imaging data from 21 runs (124 trials, 2,280 trial repetitions) in two sessions and meridian and isoeccentricity data from 12 runs (84 trials, 1,260 trial repetitions) in two sessions.

Visual stimuli

Stimulus presentation

Stimuli were generated by VPixx software (version 2.03, VPixx Technologies) and were projected by a projector (PROPixx, VPixx Technologies) onto a translucent screen placed in the magnetic bore 60 cm in front of the monkey’s eyes. The projector features a native resolution of 1,920 × 1,080, driven with a refresh rate of 60 Hz in red–green–blue (RGB) mode (1,440 Hz in grayscale) with deterministic timing. Eye movements were monitored by an MRI-compatible infrared eye-tracking system (Eyelink 1000 plus).

Meridian mapping

Our stimuli for direct meridian mapping consisted of a narrow band (vertical, 20° top to bottom; horizontal, 30° left to right) with a width of 0.15°; the band contained a red and blue checkerboard grid (0.25° × 0.75°) alternating at 3 Hz, a stimulus that evokes a robust visual response. The fixation point was a 0.2° dot presented during visual stimulation and blank.

Isoeccentricity mapping

Stimuli consisted of paired circular (ring) stimuli at two different eccentricities42. By presenting paired well-spaced rings, two distinct isoeccentricities could be mapped simultaneously without response ambiguity. In addition, in some experiments, maintaining the outer ring constant (for example, at 4°) across stimuli permitted confirmation of stability and more reliable comparisons of the maps across sessions. Inner ring and outer ring pair stimuli were as follows: monkey E, 0°–4°, 0.4°–4°, 0.6°–3°, 0.8°–4°, 1.2°–6°, 2°–4°, 1°–5° and 1.4°–7°; monkey J, 0.8°–4°, 1°–5° and 1.4°–7°. The inner ring consisted of 24 segments of red and blue checkerboard (red, 255,0,0; blue, 0,0,255; red CIE coordinate, 252.99, 120.39, 0.46; blue CIE coordinate, 59.33, 13.53, 330.62) and the outer ring had 32 segments, both alternately flashing at 1 Hz. The width of the rings was 0.6°.

Foveolar mapping

To directly map the foveal cortical representation, square patches of different sizes (0.4°–1.0° in diameter) were presented at the center of the visual field. For the monkeys to maintain excellent fixation during these scans, a fixation cross (0.3° × 0.1° lines) was included at the center of the square patches. The patches alternated between red and blue, a stimulus that is known to be effective for activation of color domains in visual cortex63 (red, 255, 0, 0; blue, 0, 0, 255; red CIE coordinate, 252.99, 120.39, 0.46; blue CIE coordinate, 59.33, 13.53, 330.62; 3 Hz).

Foveal functional ___domain mapping

To map different SF domains, full screens (horizontally extended 30°; vertically extended 20°) of isoluminant achromatic horizontal gratings with SFs of 0.2, 11, 15 and 18 cycles per degree were presented. The temporal frequency of these gratings was one cycle per second in all cases. A fixation cross (0.3° × 0.1° lines) was presented at the center of the square patches to help the monkeys maintain excellent fixation. For motion mapping, 20 random rotating clockwise dots (height, 5 pixels; width, 5 pixels) filling in a 2° visual field were used. The dots rotated coherently with an angle speed of 45° s−1.

Phase-encoding mapping

To further validate the multiple foveolar loci, very fine isoeccentricity and isopolarity angle mapping stimuli were used. Continuous expanding or contracting of isoeccentric rings was presented. Each ring was 0.15° wide and 50 rings in total covered 3°. Each block of expanding rings lasted 50 s with 10 s of a constant 0.05° white dot fixation point before the rings. A continuous clockwise rotating checkerboard wedge (5° wedge) rotated over 50 s from 90° to −90° in a contralateral visual field covering 3° of the visual field.

Data analysis

Anatomical image processing and surface reconstruction

To make the functional activations visible along the folding pattern of the cortex, we constructed a cortical surface mesh on an inflated view64. Programs implemented in FreeSurfer version 6.0 (https://surfer.nmr.mgh.harvard.edu/fswiki/FreeSurferWiki) were customized for the monkey cortex at ultrahigh-field MRI. First, the header orientation information was corrected from the supine position to sphinx position using mri_convert. Because of severe B1 inhomogeneity in ultrahigh-field fMRI images65, the whole-brain anatomical images from different sessions were corrected for intensity bias using N4BiasFieldCorrection66 in ANTs (version 2.2.0, https://github.com/ANTsX/ANTs). The bias-corrected anatomical images from different sessions were then aligned using rigid-body registration by mri_robust_register and were averaged for a high-quality anatomical image. The remaining procedures for surface reconstruction were similar to the recon-all work flow for high-resolution data67, except that the default human atlas was replaced with the D99 monkey atlas68.

Similar to de Hollander et al.69, the performance of surface reconstruction was carefully inspected and replicated for more than three iterations after manual correction for satisfactory surface reconstructions. Crucially, we checked whether white matter segmentation and the brain skull strip were correct. Error in white matter segmentation will lead to misposition of the white or gray matter surface. If the white matter mask extended into gray matter, the white matter mask was corrected using Freeview. If there was white matter lost, the ‘control point’ was added to give a prior intensity to the intensity bias correction. An error in the skull strip creates gray matter outside the brain (for example, dura). Therefore, the dura was manually corrected using Freeview where there was a misposition of gray matter or cerebrospinal fluid (CSF) surface.

Functional image preprocessing

The preprocessing was performed using AFNI (https://afni.nimh.nih.gov) and FreeSurfer version 6.0. DICOM files were transformed to the NIfTI data format with AFNI Dimon and, as with the anatomical images, the header orientation information was corrected from the supine position to sphinx position. EPI images were then preprocessed with slice-timing correction, distortion correction with a reverse phase-encoding image70 and motion correction. The first frame of the first run in each scan session (four runs) was used as the reference template frame for motion correction using rigid-body registration with cubic interpolation (using 3dvolreg). Then, the transformations from distortion correction and motion correction were concatenated and applied to the slice-timing-corrected data only once to reduce the spatial blur induced by interpolation of preprocessing71. No extra spatial smoothing was applied on either volume or surface data. For representing the volume data on the surface, the spatial transformation from functional image to whole-brain anatomical image was calculated by registering the reference frame to the whole-brain anatomical image through a tag-based registration method. A total of 9–12 tags distributed at different slices (dorsal, middle and ventral slices) were chosen and confirmed using the sagittal view, coronal view and axial view. Applying this spatial transformation, the activation volume resulting from general linear model (GLM) analysis was projected to the reconstructed surface at 50% of the cortical thickness (that is, middle surface between white matter and gray matter surface and between gray matter and CSF surface) using trilinear interpolation with the mri_vol2surf function in FreeSurfer.

GLM analysis

To detect the activated voxels, a GLM analysis was performed on the preprocessed data using the 3dDeconvolve function in AFNI to calculate t-score maps. One regressor representing the experimental design convolved with a canonical gamma hemodynamic response function, along with six motion regressors to remove residual motion effects, was used in the GLM model. Baseline shift was removed with a fifth-order polynomial function during GLM analysis. Activated voxels were extracted by 3dmaskdump in AFNI and the time courses were plotted using Matlab version 9.13.0 (R2022b, MathWorks).

Foveolar localization

We predicted that the loci of foveolar activation should remain stable across different sizes of foveolar stimuli and across multiple thresholds of significance. To examine this, each monkey was imaged with three spot sizes and each image was separated into V1–V2, V2–V3 and V3–V4 regions guided by VM and HM (Fig. 2). Given the simple flashing visual spot stimuli used, activations were strongest for early areas (V1 and V2) and weaker for areas with preferences for more complex stimuli (V3, V4 and TEO). Thus, each image was then analyzed at thresholds appropriate for each visual area (high, medium and low thresholds producing 1-mm-scale, 5-mm-scale and 1-cm-scale localization, respectively). The center of the foveolar region at each area was determined as the highest statistical significance, consistent across different spot sizes in the lateralmost regions of visual cortex.

Phase-encoding mapping

To improve SNR, three runs (21 trials) of each expanding rings, contracting rings and clockwise rotating wedges were averaged for phase-encoding mapping. The visual field maps were generated by 3dRetinophase in AFNI with the fast Fourier transform (FFT) function, which estimates phase on the basis of the fundamental frequency of stimuli. The time courses sampled from ROIs were transformed to the frequency ___domain using FFT.

CMF calculation

CMFs were quantified from topographic maps. The linear CMF (degrees per mm) was calculated on the basis of brain surface isoeccentricity maps (12 isoeccentricity rings) and the points of isoeccentricity line intersections with the VM and HM lines. Linear CMF was calculated by measuring the distance (mm) along the cortical surface (using mris_pmake from FreeSurfer function) between intersection points of two eccentricities along the meridians divided by the visual distance (degrees). Average linear CMFs along the polar axis were also calculated on the basis of the measured cortical distance between HM and VM at a given eccentricity and divided by the associated arc length for that eccentricity. Distances between HM and VM at a given eccentricity were generated by adding the distances between adjacent voxels activated along the eccentricity line to follow the surface curvature of the brain. Because the foveola loci were determined and these loci formed a ring-like zone, by connecting these identified foveolar loci, the central areas were outlined and labeled as a mask. The size of this connected area could be calculated with the mris_anatomical_stats function from FreeSurfer. The single quadrant 1° areal size in V1 was calculated in the same way by connecting a 1° eccentricity line with VM and HM.

Overlap calculation

To examine the degree of spatial overlap of functional domains, the Dice coefficient was used to quantify the overlap ratio. Specifically, the number of intersected vertices was divided by the total number of the activated vertices.

Domain size calculation

The area of the ___domain was calculated with mris_pmake. This function overlays the significantly activated voxels onto the surface mesh of the cortex33 and determines the number of vertices overlapping with the activation. It then provides the estimated length along the long axis of the ___domain (Extended Data Fig. 8).

Intrinsic signal optical imaging

Following craniotomy surgery, the brain was stabilized with agar and images were obtained through a cover glass. Images of reflectance change corresponding to local cortical activity were acquired (Imager 3001, Optical Imaging) with 630-nm illumination72. The SNR was enhanced by trial averaging (30–50 trials per stimulus condition). Frame sizes were 540 × 654 pixels representing ~16 × 19.4 mm of imaged area. Visual stimuli were presented in blocks. Each block contained all stimulus conditions (for example, different orientation gratings) and a blank condition, which was a gray screen at the same mean luminance level as the grating conditions.

The same gray screen was used for interstimulus intervals (ISIs), which were at least 8 s. For each condition, imaging started 0.5 s before the stimulus onset (imaging of the baseline), while the screen remained as a gray blank (the same as in the ISI). Then, a visual stimulus was presented for 3.5 s. All stimulus conditions were displayed in a randomized order.

Visual stimuli were created using ViSaGe (Cambridge Research Systems) and presented on a 20-inch cathode ray tube monitor (SONY CPD-G520). The stimulus screen was gamma-corrected and positioned 57 cm from the eyes. For ocular dominance in orientation maps, full-screen drifting square-wave gratings were used (SF = 1.5 cycles per degree; temporal frequency = 4 Hz). For color maps, responses to red and green isoluminant sinewave gratings and black and white sinewave gratings (SF = 1 cycle per degree; temporal frequency = 4 Hz) were compared. Isoeccentric arcs (width, 0.02°; radius, 0.01°, 0.1°, 0.2°, 0.3°, 0.4° or 0.5°) and isopolar bars (width, 0.02°; angular rotation, 0°, 15°, 30°, 45°, 60°, 75° or 90°) were presented on a uniform gray background at a refresh rate of 8 Hz.

For each stimulus condition, we constructed a ‘single-condition map’. The gray value of each pixel in the single-condition map represents the percentage change of the light reflectance signal after stimulus onset. These single-condition maps were used for calculating difference maps (ocular dominance map, left eye versus right eye; orientation map, horizontal versus vertical grating; color map, red–green versus white–black grating) and vector maps (color-coded angularity and eccentricity maps).

Color-coded eccentricity and angularity maps (vector summation of conditions) were drawn on the basis of the results of each condition. Each isoeccentric arc map or isopolar bar map was obtained first. The responses to different eccentricities or angularities at each pixel were then vector-summed to obtain a polar map73. These data analyses were performed using custom scripts in Matlab (R2020b).

Mapping foveolar ___location in anesthetized monkey with optical imaging

The position of the most foveal representation in the image maps was determined by topographic mapping of bars in horizontal and vertical orientation. Following an estimate of the foveal ___location on the monitor by backprojection of the retinal foveolar density visualized through an ophthalmoscope, we began by mapping with a sequence of five equally spaced bars placed at sequential neighboring positions on the screen (the range of the changes in bar positions spanning a total of 4°). Of the five cortical images to these bars, the one that produced the strongest response was identified. This ___location was then used as a center for testing five new bars spanning a total of 2°. Again, we chose the ___location of the strongest response as the one including the foveolar ___location and reduced the spanning size by half. This was repeated iteratively to identify the centermost foveolar ___location. By imaging sets of horizontal and vertical bars in this fashion, a central foveal ___location (precision within 0.1°) on the cortex was identified.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.