All studies conformed to the guidelines of the Medical Institutional Review Board at the University of California, Los Angeles. Electrode locations were based exclusively on clinical criteria and were verified by magnetic resonance imaging (MRI) or by computer tomography coregistered to preoperative MRI. Each electrode probe had nine microwires protruding from its tip, eight high-impedance recording channels (typically 200–400 kΩ), and one low-impedance reference with stripped insulation. The differential signal from the microwires was amplified using a 64-channel Neuralynx system, filtered between 1 and 9000 Hz, and sampled at 28 kHz. Spike detection and sorting was performed after bandpass filtering the signals between 300 and 3000 Hz (Quiroga et al., 2004 (link)).
Each recording session lasted ~30 min. Subjects were sitting in bed, facing a laptop computer on which pictures of famous individuals, landmarks, animals, or objects were shown. A median number of 97 (range, 60–202) different images were shown per session, centered on a laptop screen and covering ~1.5°, and displayed six times each for 1 s in pseudorandom order (Quiroga et al., 2005 (link)). After image offset, subjects had to indicate whether the picture contained a human face or something else by pressing the “Y” and “N” keys, respectively. This simple task, on which performance was virtually flawless, required them to attend to the pictures. Every stimulus presentation was preceded by a fixation cross for 500 ms to assess baseline firing activity. In a slightly different variant of the paradigm (23 of96 sessions), images were presented for 500 ms (20 sessions) or 750 ms (3 sessions), and the attention task was omitted. Absence of a significant influence of the presentation time on the observed response latencies was confirmed post hoc by nonparametric one-way ANOVA (Kruskal–Wallis; p = 0.18).
To determine whether a unit responded selectively to one or more of the stimuli presented, we divided the 1000 ms after stimulus onset into 19 overlapping 100 ms bins, and for each bin we compared the spike rates for the six presentations of each stimulus to the baseline intervals of 500 ms before all of the stimulus onsets in a session (~100 × 6) by means of a two-tailed Mann–Whitney U test, using the Simes procedure (Rodland, 2006 ) to correct for multiple comparisons and applying a conservative significance threshold of p = 0.001 to reduce false-positive detections. Only responsive units were included in the subsequent latency and selectivity analyses.
Onset latencies for responsive units were determined by Poisson spike train analysis (Hanes et al., 1995 (link)). For this procedure, the interspike intervals (ISIs) of a given unit are processed continuously over the entire recording session, and the onset of a spike train is detected based on its deviation from a baseline Poisson, i.e., exponential, distribution of ISIs (regardless of the experimental paradigm). For each response-eliciting stimulus, we determined the time between stimulus onset and the onset of the first spike train in all six presentations. Only spike train onsets within the first 1000 ms after stimulus onset were considered. The median length of these six time intervals was taken as response latency. For sparsely firing units with mean baseline firing activity of <2 Hz, Poisson spike train analysis generally failed to pick up any onset spike, thus we used the median latency of the first spike during stimulus presentation instead. To minimize spurious latency values, we excluded responses for which the onsets of the three trials closest to the calculated response latency were >200 ms apart. For a neuron responding to more than one stimulus, the median of the different stimulus latencies was taken.
For the nonparametric correlation analysis, selectivity of each unit was operationally defined as the reciprocal value of the relative number of response-eliciting stimuli.
Baseline firing rates of the responsive cells were calculated from the 500 ms before stimulus onset and quantified as the median across six presentations. For a neuron responding to more than one stimulus, the median of the baseline rates for different stimuli was taken.
Each recording session lasted ~30 min. Subjects were sitting in bed, facing a laptop computer on which pictures of famous individuals, landmarks, animals, or objects were shown. A median number of 97 (range, 60–202) different images were shown per session, centered on a laptop screen and covering ~1.5°, and displayed six times each for 1 s in pseudorandom order (Quiroga et al., 2005 (link)). After image offset, subjects had to indicate whether the picture contained a human face or something else by pressing the “Y” and “N” keys, respectively. This simple task, on which performance was virtually flawless, required them to attend to the pictures. Every stimulus presentation was preceded by a fixation cross for 500 ms to assess baseline firing activity. In a slightly different variant of the paradigm (23 of96 sessions), images were presented for 500 ms (20 sessions) or 750 ms (3 sessions), and the attention task was omitted. Absence of a significant influence of the presentation time on the observed response latencies was confirmed post hoc by nonparametric one-way ANOVA (Kruskal–Wallis; p = 0.18).
To determine whether a unit responded selectively to one or more of the stimuli presented, we divided the 1000 ms after stimulus onset into 19 overlapping 100 ms bins, and for each bin we compared the spike rates for the six presentations of each stimulus to the baseline intervals of 500 ms before all of the stimulus onsets in a session (~100 × 6) by means of a two-tailed Mann–Whitney U test, using the Simes procedure (Rodland, 2006 ) to correct for multiple comparisons and applying a conservative significance threshold of p = 0.001 to reduce false-positive detections. Only responsive units were included in the subsequent latency and selectivity analyses.
Onset latencies for responsive units were determined by Poisson spike train analysis (Hanes et al., 1995 (link)). For this procedure, the interspike intervals (ISIs) of a given unit are processed continuously over the entire recording session, and the onset of a spike train is detected based on its deviation from a baseline Poisson, i.e., exponential, distribution of ISIs (regardless of the experimental paradigm). For each response-eliciting stimulus, we determined the time between stimulus onset and the onset of the first spike train in all six presentations. Only spike train onsets within the first 1000 ms after stimulus onset were considered. The median length of these six time intervals was taken as response latency. For sparsely firing units with mean baseline firing activity of <2 Hz, Poisson spike train analysis generally failed to pick up any onset spike, thus we used the median latency of the first spike during stimulus presentation instead. To minimize spurious latency values, we excluded responses for which the onsets of the three trials closest to the calculated response latency were >200 ms apart. For a neuron responding to more than one stimulus, the median of the different stimulus latencies was taken.
For the nonparametric correlation analysis, selectivity of each unit was operationally defined as the reciprocal value of the relative number of response-eliciting stimuli.
Baseline firing rates of the responsive cells were calculated from the 500 ms before stimulus onset and quantified as the median across six presentations. For a neuron responding to more than one stimulus, the median of the baseline rates for different stimuli was taken.
