Perceptual shape sensitivity to upright and inverted faces is reflected in neuronal adaptation

Using an fMR-adaptation paradigm for different face morphing levels we have recently demonstrated a narrow neuronal tuning to faces even at the sub-exemplar level which was tightly related to perceptual discrimination (Gilaie-Dotan and Malach, 2007). However, it is unclear whether this relationship is unique to faces or is a general property of object representations including unfamiliar objects, and whether the adaptation tuning is due to physical changes in the stimulus or to changes in perceptual discrimination. Here we compared the same face-morph paradigm for upright and inverted faces, thus modulating familiarity and perceptual discrimination effects while equating all low-level features. We found, as expected, a perceptual “inversion effect”, i.e. a significant reduction in inverted face discrimination. Importantly, the fMR-adaptation tuning in the fusiform face area (FFA) changed in accordance with the different perceptual sensitivity both for upright and inverted faces. Additional object selective regions displayed differential tuning widths to the two categories. Our results are compatible with a model by which the ability of human observers to discriminate objects depends on the shape tuning properties of individual neurons.


Broad tuning of the adaptation effects to inverted faces
In general, across the different definitions of the FFA the pattern that emerges for inverted faces was that there were 2 significant steps in the recovery of the signal from adaptation. Critically, in all cases the signal was not completely recovered in the 1/3morph condition, indicating a broader tuning than that found for upright faces.

Alternative control definitions of the FFA and adaptation analysis
In order to expose neuronal populations that might be sensitive to inverted faces and that were "shaded" by the classical FFA definition favoring upright faces representations (faces vs. houses), we repeated the fMRI adaptation analysis for upright and inverted faces as sampled from two alternative definitions of the FFA that might expose those higher selectivities to inverted faces.
First alternative definition was based on preference to inverted faces over houses. The FFA was identified in 12 subjects using this definition (in 10: bilaterally, in 1: right only, in 1: left only). Data from this analysis are presented in Supp. Fig. 1A. Repeated measures ANOVA with condition (identical, 1/3morph, 2/3morph, and different) and hemisphere (right and left) on the time courses of 10 subjects (with bilateral foci) essentially replicated the results derived from the classically defined FFA: In the upright faces experiment: A significant effect for condition (F > 1, p < 0.0002), no effect for hemisphere (F < 1, p > 0.83), and no interaction (F = 1.583, p > 0.21). A further post-hoc analysis for condition revealed a significant identical-1/3morph difference (Bonferroni/Dunn: p < 0.0004), and no significant effects for 1/3morph-2/3morph (p > 0.96) or 1/3morph-different (p > 0.45).
Second alternative definition was based on preference to inverted faces over textures (Supp. Fig. 1B). Here also the alternative definition of the FFA did not alter the previously described adaptation profiles. Repeated measures ANOVA with condition (identical, 1/3morph, 2/3morph, and different) and hemisphere (right and left) on the experiment time courses of 11 subjects (with bilateral foci) revealed: In the upright faces experiment: A significant effect for condition (F > 1, p < 0.0001), no effect for hemisphere (F < 1, p > 0.28), and no interaction (F = 1.83, p > 0.16). A further post-hoc analysis for condition revealed a significant identical-1/3morph difference (Bonferroni/Dunn: p < 0.0001), and no significant effects for In the inverted faces experiment: A significant effect for condition (F > 1, p < 0.0001), no significant effect for hemisphere (F = 2.74, p > 0.12), and no interaction (F < 1, p > 0.49). A further post-hoc analysis for condition revealed a significant identical-1/3morph difference (Bonferroni/Dunn: p < 0.0008). 1/3morph condition was not significantly different from the 2/3morph (Bonferroni/Dunn: p > 0.26), but was significantly different from the different condition (Bonferroni/Dunn: p < 0.0002) as was the 2/3morph-different difference (Bonferroni/Dunn: p < 0.003).
Furthermore, we applied statistical analysis to the different FFA profiles in order to determine whether a statistical difference between these profiles could be detected.

FFA's response magnitude to upright and inverted faces in the category localizer experiment
When FFA was defined by the faces vs. houses contrast, a 1-tailed paired t-test across subjects showed that upright faces activated this region more than inverted faces (p < 0.002). However, note that this test was biased towards upright faces and not independent of the FFA's definition. When FFA was defined by inverted faces vs.
houses, no significant activation difference was found between upright and inverted faces (p > 0.34).

Adaptation profiles derived from peak responses
In our original analysis for each block we used the average PSC over two pre-defined time points (3 rd and 4 th TRs in morph experiments, 2 nd and 3 rd TRs in localizer) as the representative PSC for that block. Here we repeated the analysis using a single time point with the highest PSC (peak response) as the representative for the block response. Results presented in Supp. Fig. 2 replicated our original results. Detailed below are results from a repeated measures 2-way ANOVA with condition (identical, 1/3morph, 2/3morph, and different) and hemisphere (right and left) based on peak responses of each subject to each of the conditions: