Development of the macaque face-patch system

Abstract

Face recognition is highly proficient in humans and other social primates; it emerges in infancy, but the development of the neural mechanisms supporting this behaviour is largely unknown. We use blood-volume functional MRI to monitor longitudinally the responsiveness to faces, scrambled faces, and objects in macaque inferotemporal cortex (IT) from 1 month to 2 years of age. During this time selective responsiveness to monkey faces emerges. Some functional organization is present at 1 month; face-selective patches emerge over the first year of development, and are remarkably stable once they emerge. Face selectivity is refined by a decreasing responsiveness to non-face stimuli.

Figure 1. Visual stimuli used for longitudinal scanning.

The first three rows show a subset of the static images that were shown in blocks; the bottom row shows examples of the videos that were shown to the monkeys younger than 2 months. (Top row) Monkey faces. (Row 2) Scrambled monkey faces, scrambled using the method of Portilla and Simoncelli. (Row 3) Familiar rectilinear objects. (Row 4) The three categories of videos shown to monkeys <2 months old: L to R: monkeys grooming each other with prominent faces; same movies with the faces pixelated; traffic or time-lapse nature.

Figure 2. Development of face selectivity.

Maps of activations (beta coefficients) for the contrast faces-minus-objects for each static-image scanning session for each of four young monkeys projected onto the standard F99 macaque brain flat map. Each monkey's age in days is indicated beside each map. Where the age is indicated in black, the beta coefficients were thresholded at P≤0.05, FDR corrected; where the age is indicated in blue, no faces>object activations were present in the STS at P<0.05, so the beta coefficients were thresholded at P≤0.25, FDR corrected. In the top centre are enlarged maps for monkeys B3 and B4 for the dates we used for defining the ROIs for quantitative ROI analysis; M indicates the region outlined in white corresponding to the middle face-patch cluster, and A indicates the region outlined in white corresponding to the anterior face-patch cluster; the object patches are outlined in black. Corresponding visual areas are mapped in Supplementary Fig. 1 and non-overlapping maps are shown in Supplementary Fig. 2.

Figure 3. Very early organization revealed using movies.

(a) Representative activations for the contrast faces-minus-scenes movies in one monkey at two ages, once before static-image face patches appeared at 30 days old (above) and once after the static-image face patches appeared, 836 days (below), both thresholded at P<0.05 FDR corrected. Dotted white outlines indicate faces>object static-image activations for the same monkey at 276 days old. Additional examples of movie data for other young monkeys are shown in Supplementary Fig. 4. (b) Maps of the contrast pixelated faces>scenes movies for the same monkey. (c) ROI analysis of three infant (top) and two juvenile (bottom) monkeys (two hemispheres each) in response to movies. Activations in anterior (triangles) and middle (circles) face patches in response to movie stimuli for (top) three monkeys at the youngest ages at which they showed significant faces>scenes movie activations and (bottom) two monkeys at ages after the static-image face patches had appeared.

Figure 4. Average raw response time courses.

Each panel shows the normalized MR signal time course in the ROI indicated on the left, averaged over blocks for each stimulus category; shading indicates 95% confidence limits. Face and object ROIs were identified using data from monkey B3 at 284 days old and from monkey B4 at 276 days old. The central V1 ROI (central 6–7° of visual field) was identified by retinotopic mapping in each animal at >2 years of age. Numerals above each column indicate the monkey's age in days; outlined age indicates movie stimuli; others were blocks of static images.

Figure 5. Development of category-selective responses in face and object ROIs in two individual monkeys for all the static-image sessions.

Responses to static-image blocks of faces (red), scrambled faces (green) and objects (blue) in the anterior-face ROI, the middle-face ROI, and in the object-selective ROI for monkey B3 (top row) and monkey B4 (middle row). Shading indicates ±s.e.m. Red asterisks at the top of each graph indicate sessions in which that ROI showed a significantly greater response to faces than to objects at P<0.05; blue asterisks indicate sessions in which that ROI showed a significantly greater response to objects than to faces at P<0.05; black dots indicate sessions in which there was no significant difference between faces and objects. Data from the scan sessions from which the ROIs were calculated are not connected by lines to the rest of the dates, and are indicated by open symbols for B3 and filled symbols for B4. (Bottom row) Responses by age in each ROI to faces (red) and objects (blue) divided by the responses in central V1 for that day to that image category in the same two monkeys. Lines indicate linear fit to each category data. B3 filled symbols, solid lines; B4 open symbols, dashed lines. Asterisks indicate slopes that were significantly less than zero at P<0.01. Anterior face patch, slope of object responses (relative to y intercept at birth): B3, m=−33%/year, P=5 × 10−5; B4, m=−57%/year, P=6 × 10−4. Middle-face patch, slope of object responses: B3, m=−21%/year, P=2 × 10−5; B4, m=−37%/year, P=1 × 10−4. The slopes for face responses in the face patches were not significantly different from zero (P>0.05). For both monkeys the slopes for the face responses in the face patches were significantly different from the slopes for object responses at P<0.01 (assessed by a 3-way ANOVA with age, category, and face patch; interaction between age and category was significant at F>7.09, P<0.01; no other significant interactions, P>0.10).

Figure 6. Development of category selectivity in face and object ROIs.

Plots show D-primes for distinguishing faces from objects (left), faces from scrambled faces (middle) and scrambled faces from objects (right) for the anterior face-patch cluster, middle face-patch cluster and object patches, as indicated. Data from the scan sessions from which the ROIs were calculated (B3 284 days; B4 276 days) are indicated by desaturated symbols in the first panel; we did not present scrambled faces in that session.

Figure 7. ROI-independent analysis of IT organization.

Signal change in response to faces/face movies (top row), scrambled faces/pixelated movies (middle row) and objects/scene movies (bottom row) as a function of age (vertical dimension) and distance along the STS from anterior to posterior (horizontal dimension). Black lines separate early movie data from later static-image data.