3.5 Análisis de las respuestas
3.5.3 Comparación de los resultados de ANVIS y HUF
In an attempt to distinguish perceptual judgments from motor outputs, we lastly analyzed whether the response of emotion-selective units was correlated with behavioral output. We analyzed the correlation between RT and spike timing or spike counts for the 41 emotion-selective neurons (Pearson correlation, false positive rate = 0.05, uncorrected for multiple comparisons). When correlating RT with peak firing time (center of the 250 ms bin which had the highest firing rate), we observed only two fear-selective neurons with a significant positive correlation (binomial test on the number of significant cells, p = 0.12; two neurons with a significant negative correlation, p = 0.12), while we observed no happy-selective neurons having a significant positive correlation (p = 0.58; one neuron had a significant negative correlation, p = 0.21). When correlating RT with the total number of spikes in a time window 500 ms prior to button press, we found only one
significant positive correlation for fear-selective neurons (p = 0.34) but no significant positive correlation for happy-selective neurons (p = 0.58; five fear-selective neurons (p < 0.001) and two happy-selective neurons (p = 0.050) had a significant negative correlation). Separate analyses for fearful trials and happy trials showed very similar results. We conclude that there was no significant correlation between firing rate and RT.
One confounding factor of our experimental setup is that the same motor action is always associated with the same emotion (the left button denoted fearful and the right button happy in our setup). To our knowledge there is no evidence that the amygdala encodes such specific motor actions. We further analyzed the distribution of emotion-selective neurons to investigate whether the button presses were associated with emotion coding in the amygdala. If the emotion neurons are associated with the button press, they should appear contralaterally to the pressed buttons. However, we did not observe such laterality. Of the total 210 cells, 92 cells were recorded from the left amygdala and 118 cells were recorded from the right amygdala (see Figure 4.20). Among the emotion-selective cells, 6 fear-selective cells were from the left amygdala, 18 fear-selective cells were from the right amygdala, 6 happy-selective cells were from the left amygdala and 11 happy- selective cells were from the right amygdala. The proportion of emotion-selective cell did not differ between left vs. right amygdala in any of these categories (Fisher’s exact test, p = 0.51; paired two-tailed t-test across patients on the percentages, all ps > 0.1), showing that the emotion neurons are not lateralized nor related to the output button response associated with the emotion. Further, the same results held when excluding neurons from the epileptic areas (all ps > 0.25).
!
A
B
0 50 100 Percentage Fear-Selective Neurons 0 50 100 Percentage Happy-Selective Neurons p17 p18 p19 p23 p25 p28-1 p28-2 p29-1 p29-2 p17 p18 p19 p23 p25 p28-1 p28-2 p29-1 p29-2 p17 p18 p19 p23 p25 p28-1 p28-2 p29-1 p29-2 0 20 40 # NeuronsTotal Number of Neurons L R 0 2 4 6 # Neurons Fear-Selective Neurons 0 2 4 6 # Neurons Happy-Selective Neurons p17 p18 p19 p23 p25 p28-1 p28-2 p29-1 p29-2 p17 p18 p19 p23 p25 p28-1 p28-2 p29-1 p29-2
Figure 4.20. Distribution of emotion-selective neurons.
(A) Number of emotion-selective neurons from each patient. (B) Percentage of emotion- selective neurons. Red: neurons recorded from the left amygdala. Blue: neurons recorded from the right amygdala. Patients 17 and 28 were diagnosed with ASD.
!
Our results suggest that the amygdala encodes the subjective judgment of emotional faces, but that it plays less of a role in helping to program behavioral responses.
!
4.6 Discussion
In this study, we found that a subset of amygdala neurons encode the subjective judgment of the emotion shown in faces. Behaviorally, our epilepsy patients did not differ from healthy controls in terms of learning performance on the task, and both epilepsy patients and control subjects primarily used the eye region of the stimuli to correctly judge fearful faces and primarily used the mouth region to correctly judge happy faces, findings consistent with prior studies (Smith et al., 2005, Scheller et al., 2012). 41 cells significantly differentiated the two emotions and subsequent analyses indicated that these cells encoded the patients’ subjective judgment regardless of whether it was correct or incorrect. Population permutation analysis with full independence between selection and prediction confirmed the robustness of this result when tested across the entire population. ROI analysis revealed that eyes but not mouth strongly modulated population neuronal responses to emotions. Lastly, when we carried out identical recordings, in the same patients, from neurons within the hippocampus, we found responses driven only by the objective emotion shown in the face stimulus, and no evidence for responses driven by subjective judgment.
It is notable that the population response metric for the correct trials was further away from the null distribution relative to the incorrect trials (25.0% vs. -4.63%). It is not
surprising that the strength of emotion coding in incorrect trials was weaker given fewer incorrect trials and thus potentially increased variability and decreased reliability. In addition, incorrect trials were likely a mixture of different types of error trials, such as true misidentifications of emotion, guesses, or accidental motor errors. Regardless, on average, the neural response during incorrect trials reliably indicated the subjectively perceived emotion. This suggests that a proportion of error trials were likely true misidentifications of the emotion rather than pure guesses.
Interestingly, there was a significant difference between the two types of happy subjective judgments (comparing happy correct and fearful incorrect, Figure 4.8E). This might reflect a different strategy used by subjects to compare the two emotions in our specific task. Future studies with a range of different tasks will be needed to understand how relative coding of emotion identity and task demands may interact in shaping neuronal responses.