It is not just the category: behavioral effects of fMRI-guided electrical microstimulation result from a complex interplay of factors

Abstract

Functional imaging and electrophysiological studies in primates revealed the existence of patches selective for visual categories in the inferior temporal cortex. Understanding the contribution of these patches to perception requires causal techniques that assess the effect of neural activity manipulations on perception. We used electrical microstimulation (EM) to determine the role of body patch activity in visual categorization in macaques. We tested the hypothesis that EM in a body patch would affect the categorization of bodies versus objects but not of other visual categories. We employed low-current EM of an anterior body patch (ASB) in the superior temporal sulcus, which was defined by functional magnetic resonance imaging and verified with electrophysiological recordings in each session. EM of ASB affected body categorization, but the EM effects were more complex than the expected increase of body-related choices: EM affected the categorization of both body and inanimate images and showed interaction with the choice target location, but its effect was location-specific (tested in 1 subject) on a millimeter scale. Our findings suggest that the behavioral effects of EM in a category-selective patch are not merely a manifestation of the category selectivity of the underlying neuronal population but reflect a complex interplay of multiple factors.

Keywords: body patch; electrical microstimulation; inferior temporal cortex; macaque; visual categorization.

Fig. 1

Tasks, anterior STS body patch (ASB), and controls. A) Categorization tasks. Left: 2CC task with fixed saccade target locations. The tasks included categorization of animals versus objects, faces versus objects, and houses versus objects. The animals versus objects task is shown as an example. A red fixation dot appeared for 500 ms after which the cue stimulus was presented with or without EM (150 μA). After an ISI of 13.33 ms, the monkey was required to saccade (black arrow) to the correct image target indicating the category of the cue. Middle: 4CC task. Similar to the 2CC task, a cue appeared following the presentation of only the red fixation dot. The cue could be an image from 4 categories: animals, faces, houses, and objects. After an ISI of 13.33 ms, the monkey had to make a saccade to 1 of the 4 target images that matched the category of the cue. The targets on the horizontal axis were switched between hemifields for the 2 monkeys. Right: 2CC categorization tasks with different target locations. The task structure was identical to the 2CC described above except that the spatial locations of the targets for the category choices were identical to those of the 4CC task. In these 2CC tasks, monkeys had to categorize animals versus objects, houses versus objects (shown in the figure), faces versus houses, faces versus objects, faces versus animals, and houses versus animals. In 50% of the trials of all the tasks, EM (lightning bolt icon) of 150-ms duration with a delay of 50 ms was administered during the presentation of the cue. In all tasks, correct choices were followed by a fluid reward. B) Examples of stimuli with different SNRs and stimulus levels. The images were overlaid with noise (“salt-and-pepper”) by replacing a fixed percentage of image pixels with an equal number of black and white noise pixels. An SNR level of i.e. 20% corresponds to 80% of the image being masked by noise. The stimulus level is a signed SNR, with positive values corresponding to images of animals and negative values images of nonanimals in this example. C) Anterior STS body patch (ASB, red) on a coronal MRI section, which was defined using a threshold of t = 7.0. Colored lines indicate estimated electrode locations targeting ASB. The scale bar corresponds to 5 mm. D) Performance of monkeys in trials without EM (solid gray lines) and a linear decoder trained on AlexNet conv1 layer activations (black stippled line) for the 4CC task as a function of SNR. Error bars correspond to standard errors of the mean. Triangles and circles correspond to the categorization performance of MG (n = 6,460 trials) and MB (n = 1,263 trials), respectively. E) Performance in the 4CC task with novel and familiar stimuli compared; 60% familiar and 40% novel stimuli of each category (animal, face, house, and object) were presented with an SNR of 80% and 60% for MB and MG, respectively. The percentage of correct choices for the first presentation of each stimulus is plotted for the familiar (red bar) and novel (blue bar) stimuli for MB and MG separately. Error bars indicate 95% CIs, which were computed using the binomial distribution approximation. The chance level of 25% is indicated by the dotted horizontal lines. Abbreviation: n.s. : not significant.

Fig. 2

Effect of microstimulation on the performance in 2CC categorization tasks. The proportion of animal, face, or house choices in the animals versus objects A–C), faces versus objects D and E), and houses versus objects F) 2CC tasks, respectively, are plotted as a function of stimulus level. The red and blue dots represent the stimulated (μS) and nonstimulated (no μS) conditions, respectively. N = total number of trials. Significance of the difference between the stimulated and nonstimulated conditions are indicated by symbols: ^***  P < 0.001; n.s. P > 0.05. A, B) EM effect for the animals versus objects task in locations inside ASB. The negative stimulus levels represent the SNR levels of the nonanimal images, while the positive values represent the SNR levels of the animal images. A) Data of MB and B) data of MG. Left: Single session examples corresponding to the red circle in the inset distribution. The curves are a logistic fit (Equation (2) of Materials and methods). The top plots of the insets (shaded region) show the mean normalized high gamma LFP power for animals, faces, and objects measured at the sites in the same sessions preceding the EM task. Error bars represent standard errors. In conjunction with the LFP plots, the BSI calculated from the MUA is also shown. The bottom plots of the insets show the fitted shift values (𝜆/, fitted Equation (1) of Materials and methods), which were computed for all individual sessions. Positive shift values correspond to an increased proportion of animal choices with EM. Significant shift values are indicated by a dark gray color (P < 0.05). The mean shift is indicated by the black-filled circle. Horizontal back bars indicate the shift computed for psychometric functions that were fitted to the data pooled across all sessions. Vertical error bars represent 95% CIs, which were computed by bootstrap resampling. The symbols above the distributions indicate the statistical significance of the shift of the pooled psychometric functions, which was assessed with a randomization test (see Materials and methods). C) Performance of MG in the animals versus objects task with EM for a neighboring location in ASB. Same conventions as in A). D and E) EM effect for the faces versus objects task for locations inside ASB. Positive shift values correspond to an increased proportion of face choices with EM. Same conventions as in A). D) Data of MB and E) data of MG. F) EM effect for the houses versus objects task for a location inside ASB (MB). Same conventions as in A). Positive shift values correspond to an increased proportion of house choices with EM.

Fig. 3

Effect of microstimulation on the performance in the 4CC task. Left (first) column: The insets show the spatial location of the choice targets for 4 categories: animals (A), faces (F), houses (H), and objects (O). I and C indicate the hemifield ipsilateral and contralateral, respectively, to the stimulated hemisphere. The spatial locations of animal and house targets were switched between the monkeys along the horizontal axis. Second to fifth columns: The effect of EM on the animal (second column), face (third column), house (fourth column), and object (fifth column) choices, plotted as a function of SNR. Sixth column: Percent correct choices plotted as a function of SNR. N = number of trials. The full lines (and circles) and dotted lines (and triangles) represent the nonstimulated and stimulated conditions, respectively. Error bars represent 95% CIs for proportions. Right (seventh) column: Mean normalized high gamma LFP power for animals, faces, and objects before the corresponding EM sessions. Error bars represent standard errors. A) Top row: Effect of low current EM (50 μA) inside ASB for MG. Second row: Effect of low current EM (50 μA) outside ASB for MG. Third row: Effect of high current EM (150 μA) inside ASB for MG. B) Effect of low current EM (50 μA) inside ASB for MB. Same conventions of the symbols for statistical significance as given in Fig. 2.

Fig. 4

Effect of microstimulation for the 2CC tasks with the same target locations as for the 4CC task. First column: Spatial locations of the choice targets. Same conventions as in Fig. 3. Second to seventh columns: Top: Spatial locations of the targets used in the corresponding 2CC tasks for MG A) and MB B), which were preserved from the 4CC task. N = number of trials. Middle: Fitted shift values (and 95% CIs) for the data pooled across all sessions in the animals versus objects (second column), animals versus houses (third column), animals versus faces (fourth column), faces versus objects (fifth column), faces versus houses (sixth column), and houses versus objects (seventh column) categorization task in MG. Positive shift values correspond to an increased proportion of choices, with EM, of the image category shown below the abscissa (e.g. for the animals vs. objects task [second column], an increase of animal choices). Symbols indicate statistical significance: ^*  P < 0.025; ^**  P < 0.01; ^***  P < 0.001; n.s. P > 0.05. Bottom: Mean normalized high gamma LFP power for animals, faces, and objects measured at the sites in the same sessions preceding the EM. Same conventions as in Fig. 2. B) Effect of EM (50 μA) for the animals versus objects (second column), animals versus houses (third column), and houses versus objects (fourth column) tasks in MB. Same conventions as in A).

Fig. 5

Effect of microstimulation for the 2CC tasks with spatially flipped choice targets. A) MG and B) MB. Effect of EM in the 2CC task, with the spatial location of the animal and house targets being flipped along the horizontal axis (first column). Same conventions as in Fig. 4.

Fig. 6

Location dependency of the behavioral effect of EM. First column: Top: Spatial location of the choice targets for the animals versus objects 2CC task. Bottom: 5 different equidistant locations on the recording grid to examine the location specificity of the behavioral effect induced by EM (50 μA; MG). Lateral and posterior dimensions are in grid coordinates. Second to sixth columns: Effect of EM on behavioral performance in the 2CC task. The colored panel borders indicate the corresponding recording grid locations as shown in the bottom panel of the first column. Same conventions as given in Figs 4 and 5.

Fig. 7

Chronological order and experimental details of the 2CC EM experiments (50 μA; MG). The gray arrow indicates the temporal order of the experiments. The significance level of the behavioral effect of EM is indicated by the gray level of the circles in the arrow (see the legend on the right side). Each circle indicates a different condition and can correspond to multiple successive daily sessions. The filled and open horizontal bars below the arrow indicate that EM was inside or outside of ASB, respectively. First column: Top: Spatial locations of the choice targets (C1). Same conventions as in Fig. 5. Bottom: Spatial location of the animal and house targets being flipped along the horizontal axis (C2).