Subjects were volunteers recruited from a patient population undergoing diagnostic and surgical procedures for medically intractable epilepsy. Protocols were approved by the University of Wisconsin - Madison and Middleton Veteran’s Affairs Hospital institutional review boards and all patients provided informed consent. Electrodes were placed intracranially according to the clinical need of each patient. Electrodes were implanted in the subdural space and leads were tunnelled transcutaneously. ECoG signals were split so clinical recording proceeded unimpeded. After implantation, the patient was returned to the Epilepsy Monitoring ward where they remained for a period of one to two weeks. Patients were fully awake during testing and able to follow complex instructions. Subjects (2 males and 2 females) ranged in age from 32–42 years (median = 38) and in composite IQ from 80–132 (median = 102.5). Prior to surgery, individual patients were on one or two of the following anti-convulstants: levetiracetam, phenytoin, carbamazepine, lamotrigine, phenobarbital, or topiramate. During monitoring, patients were weaned off their medications for clinical purposes. Seizure onset locations for the four patients were generally recorded from sites in the anterior lateral temporal lobe, inferior lateral temporal lobe, or ventral temporal lobe. Seizure activity was not predominantly located over regions of interest in the current experiment (i.e., postero-superior STG/SF).
Intracranial Electrodes and Data Acquisition
The subdural electrode array consisted of a grid of 64 platinum-iridium contacts encased in a silicone matrix. Intercontact distance was five millimeters (center-to-center) and contact diameter of each circular electrode was 2.3 millimeters. The electrodes were connected to an array of four 16-channel, electrically isolated pre-amplifiers (RA16PA, Tucker-Davis Technologies, Alachua, FL), which in turn were connected to four signal processing modules (RA-16). During data acquisition, ECoG signals were sampled at a rate of approximately 6kHz, bandpass filtered between 1.6 and 100Hz, notch-filtered at 60Hz, then down-sampled by a factor of 4 (P6–P11) or 6 (P15, P42) before saving to hard disk for later analysis. Recordings were made with reference to a ground tied to a single electrode placed either on the contralateral scalp (P6 & P9) or in the extra-cranial sub-galeal space over the ipsilateral parietal or frontal lobe (P15 & P42). Because reference electrodes were extracranial and significantly distant from the cortical electrodes, the recordings reported here are considered monopolar. This was confirmed by comparing ERPs directly averaged from each electrode recording with those averaged after sample-by-sample subtraction of a common-average reference formed from the grand mean across all remaining channels (Crone et al., 2001
). No significant differences were observed between the two types of averaged ERPs, or, more importantly, in the values and latencies of the N90stg and P170stg peaks. These results indicate that the references used here were inactive relative to the signal levels recorded intracranially.
The anatomic location of the implanted electrode grid was schematized by co-locating intraoperative photographs taken during initial surgical placement and/or resection onto a 3D, surface rendered magnetic resonance image (MRI) of each patient obtained pre-operatively. In two patients (P6 and P9), no pre-operative MRI was available, so photographs for these patients were localized onto a template MRI from a different patient (not analyzed here). For P15 and P42, electrode grid location was determined by co-registering the patient’s pre-implant MRI with a post-implant CT scan (Stealth Station, Medtronic Sofamor Danek, Memphis, TN, USA). Grids were found to cover at least a portion of the superior temporal gyrus in all patients.
Acoustic stimuli and behavioral task
A dichotic auditory oddball paradigm was used to investigate selective auditory attention in these experiments. A single experimental trial was defined as the dichotic presentation of a series of 200 tone bursts (30 msec duration; 5 msec rise/fall time; 100 in each ear) over ER-6 insert earphones (Etymotic, Elk Grove Village, IL) at nominal levels of 85 to 95 dB SPL. These levels were used to ensure clear audibility over the incidental noises of the patient’s hospital room. Depending upon experimental condition, mean ISI was set to either 400, 800 or 2000 msec, while actual onset times were randomly jittered from the mean ISI based on either a normal distribution with a standard deviation of 100 msec (P6 & P9) or a uniform distribution with a range of 300 msec (P15 & P42).
Attention was manipulated by requiring the subjects to attend to the tone series in a specified ear while ignoring the tone series presented concurrently in the other ear. Recordings made when subjects listened to the ear contralateral to the hemisphere of electrode location were labeled attend-toward
(AT) conditions; likewise, recordings made while listening to the ipsilateral ear were labeled attend-away
(AA). Directional terms were used to label these two attention conditions rather than “attend/ignore” in order to emphasize that any effects of attention direction were not due to general effects of arousal in auditory cortex, which could occur when comparing such conditions to passive listening or intermodal attention trials (Woods et al., 1992
). Electrode arrays were placed on the left hemisphere in all patients except for P15, who was implanted on the right hemisphere.
The subject was asked to detect rare deviations in the frequency of the attended tones by pressing a hand-held response button. The number of deviant tones was targeted to comprise 10% of the stimulus sequence on average, while “standard” tones constituted the remainder (actual number was randomly determined at the start of each session). Frequencies of standard tones were 1500 Hz (right ear) and 2300 Hz (left ear), and frequencies of deviants were set at the start of testing for each patient to 5% (P6, P9) or 10% (P15, P42) greater than the corresponding standard for each ear. P15 and P42 consistently achieved 100% performance at these starting levels, so deviants were adjusted in later sessions to 5% (P15) and 3% (P42) greater than the standard frequencies in an attempt to equate performance across subjects. For P15 and P42, only data from sessions with the smaller pitch difference deviants were analyzed in this report.
A single experimental block consisted of one AT trial and one AA trial, with the order of attention direction randomized within single testing blocks. Each trial presented 200 tones which were later segmented into individual recording epochs for averaging purposes (see below). Order of tone presentation was pseudo-randomized within a single trial, with the restriction that two deviants could not occur in a row; for a single experimental block, however, presentation order of stimuli was held constant across both attention direction trials. All patients were briefly exposed to the dichotic listening task prior to data collection. Multiple blocks of data were often recorded on separate days for each patient (range: 1–8 days after implant surgery). Data for the present analyses were taken from the following number of individual epochs recorded for each patient at each ISI (400/800/2000 msec): P6 – 148/316/235 epochs; P9 - 79/237/161; P15 – 101/200/50; and P42 – 160/160/39. Data averaging procedures weighted results equally across patients, rather than by number of epochs.
Performance in the attention task was assessed using percent correct (p(C)) and d-prime (Woldorff & Hillyard, 1991
; Teder et al, 1993
). A correct response (Hit) was any response occurring within 1.5 sec after a deviant was presented in the attended ear; responses occurring at any other time were labeled false alarms (FAs; multiple FAs between deviants were ignored to prevent negative d-prime estimates). The hit rate (equivalent to p(C)) was the number of correct responses divided by the number of deviants; the false alarm rate was the number of false alarms divided by the number of standard tones for the attended channel. For trials where either the hit rate was 100% or FA rate was 0%, these measures were adjusted in calculating d-prime to avoid infinite values (see Macmillian & Creelman, 1991, for details).
Intracranial ERP identification and nomenclature
Only ERPs formed from responses to standard tones presented contralateral to grid hemisphere were used in the following analyses (i.e., right ear standard stimuli for all patients but P15). Data recorded from all channels were considered valid and incorporated in the following results unless voltage values indicated amplifier blocking. Data from entire trials were also rejected if there was extreme noise in the ECoG or there were no responses to any deviants for an either an AT or AA trial (which may arise due to equipment failure, patient alertness, etc.). Due to inherent features of the TDT recording hardware used in the present experiment, all signals were high-pass filtered during recording at a cutoff frequency near 1.6 Hz (3-dB cutoff between 1.6 and 2.6 Hz). Entire ECoG data streams for accepted channels were bandpass filtered again off-line between 2–35Hz (-3dB cutoff points) using a phase-corrected FIR filter in order to alleviate potential delays normally caused by low- and band-pass filters. This added filtering may attenuate slow-wave attention-related negativities such as the PN; however, it should also help to isolate exogenous responses occurring in the alpha-band region, such as the N1 peak, which are of interest to the goals of this work. Individual ECoG epochs were isolated according to the recorded onset time of each tone burst, and began 50 msec prior to stimulus onset and continued for 300 msec post-stimulus. ERPs were then formed by averaging all epochs of the same stimulus/channel type.
Using a non-parametric bootstrap statistical test (Efron & Tibshirani, 1993
), the effects of attention direction were assessed for each subject and for each ISI separately by comparing the mean peak voltages from the AT and AA averaged ERPs taken in 20 msec time-windows centered on the maximal negative and positive peaks observed between 70–120 and 121–220 msec post-stimulus, respectively (these peaks were subsequently labeled N90stg and P170stg). For data recorded for each ISI separately, all epochs from the modal N90stg electrode from the two attention conditions (including those recorded over multiple days) were first combined into one master data matrix for each patient. 10,000 AT and AA bootstrap epochs (bAT, bAA) were formed by drawing randomly with replacement from this combined collection of AT and AA epochs. Selecting with equal likelihood from the combined set of recorded epochs approximated the null-hypothesis that no difference exists between the observed AT and AA conditions (Di Nocera & Ferlazzo, 2000
). Bootstrap ERPs were then formed by averaging across the bAT and bAA datasets, and the bootstrap peak means and latencies for the major negative and positive peaks during the aforementioned time-window were computed in the same manner as for the observed data. Finally, the p-values of the observed negative and positive peaks for the original AT and AA ERPs were computed by finding the number of bootstrap mean component values greater or less than (2-tailed statistical test) the observed mean component values, divided by the total number of bootstrap samples (10,000).
Bootstrap analysis of combined “grand average” data
The following procedure was performed separately on data recorded for each ISI. Data from all AA and AT epochs from the maximal N90stg electrodes of all patients were combined into two grand AT and AA datasets. The bootstrap procedure as described above would require drawing randomly with uniform probability from these combined epoch pools to form a null-hypothesis bootstrap N90/P170stg difference distribution. The observed grand average N1 difference, taken by simply averaging the total epochs in each pool and subtracting them from each other, would then be compared to this null-hypothesis distribution. However, this would over-represent data from those patients who were able to participate in many experimental sessions (e.g., P6 and P9 2000 msec ISI data), and result in weighting the bootstrap and observed grand average ERPs by epochs rather than by patient.
An alternative method was used here in which each null-hypothesis bootstrap grand average ERP was created by first forming a list equal in number to the total number of pooled epochs (for each ISI independently) of the 4 patient numbers selected with uniform probability; then, for each subject number in that list, individual epochs were drawn with uniform probability from that subject’s subset of AA and AT epochs. This process created 2 pools of null-hypothesis bootstrapped “AA” and “AT” epochs in which each subject’s epochs were equally represented. The procedure was repeated 10,000 times until the null-hypothesis bootstrap N90/P170stgdifference distribution was formed. Finally, the observed grand average ERP was found by averaging the ERPs from the maximal N90stg electrode separately averaged for each block for each patient. This method weighted the averaging and bootstrap formation processes equally across subjects rather than by number of epochs.
Tests were assessed at critical significance levels of pcrit ≤ 0.05 (single asterisk) and pcrit ≤ 0.01 (double asterisk). Significant effects of attention on the N90stg peak are reflected in observed p-values less than pcrit, which indicate that the negative peak of the AT waveform was more negative than the corresponding AA peak because very few bootstrapped component values were less than the observed component value (and vice versa for the p-value of the P170stg peak). The effect of attention on component latencies were also assessed in the same manner by forming bootstrap distributions of the latencies of the peak deflections in previously defined time windows.