Subjects

Subjects are epilepsy patients whose seizures are refractory to nonsurgical treatment and who undergo invasive seizure monitoring in an effort to determine whether they are suitable candidates for resection surgery.^5,6 Research protocols must be scrutinized and approved by the institutional review board where the research will be taking place according to the ethical guidelines of the institution’s governing bodies.⁷ In all cases, the plan for electrode placement is influenced exclusively by clinical criteria. Participation in this research does not change the risks associated with epilepsy surgery. The research plan is explained to research participants in detail, and informed consent for participation is obtained in advance. Particularly when chronic recordings are obtained over a period of days, this consent process is informally repeated before each experimental session. In most cases, patient-subjects wish to participate in these protocols but on certain days after surgery may elect to forgo testing. Because of this need for iterative informed consent and the practical considerations that accompany this requirement, most research of this type is carried out only with adult subjects.

The cognitive and behavioral functional status of each subject is evaluated extensively by a neuropsychologist before the electrode implantation surgery as part of the clinical diagnostic and treatment plan. It is desirable that the subject’s neuropsychological status fall within the normal limits of an age-matched control population. It is especially important to ensure that subjects do not have impairment in the cognitive functions of interest and that these functions be measured objectively and documented before surgery.

Electrodes

Several different types of clinical and combined clinical-research electrodes are available for invasive monitoring of seizure activity. Usually, the signals recorded from a given electrode contact can be split and used for both clinical monitoring and research purposes. This does not disrupt the clinical ECoG recording activity. There are two broad categories of intracranial electrodes: (1) subdural cortical surface electrodes in the form of either grids or strip electrodes and (2) depth electrodes (Fig. 59-1). The extent of coverage is decided solely by clinical necessity. Electrodes need to cover wide cortical surface areas and deep structures sufficiently to diagnose the seizure foci accurately. The specific implantation strategies used by different groups across the United States and elsewhere in the world to achieve this objective vary widely. Some programs use surface grids and strips almost exclusively, whereas other highly respected programs (e.g., Paris and Grenoble, France) use depth electrodes exclusively and implant more than 10 penetrating depth electrodes in a single hemisphere in many cases. There is no evidence proving the clinical superiority of any of these specific strategies, and in practice institutions have evolved to adopt a range of safe and effective approaches.

FIGURE 59-1 A, A photograph taken during electrode implantation surgery shows a standard clinical grid electrode with an intercontact distance of 1 cm (left upper) and a high-density custom grid electrode with an intercontact distance of 5 mm (right lower). B, Lateral skull radiograph of the same subject as in A. C, A photograph shows a standard clinical depth electrode (top) and a custom depth electrode with multiple high-impedance microcontacts on the electrode shaft (bottom). Arrows indicate positions of microcontacts.

(Courtesy of Ad-Tech Medical Instrument Corporation, Racine, WI.)

Many epilepsy patients who undergo invasive monitoring are suspected of having a temporal lobe focus. Standard preoperative evaluation typically includes video EEG monitoring with scalp electrodes, positron emission tomography, structural MRI, identification of clinical seizure semiology, and neuropsychological testing. Preoperative investigation may not lead to a conclusive site of seizure originfor example, the laterality of origination may be in question or the seizure focus may have been vaguely narrowed down to the temporal lobe of one hemisphere but the exact location of origin within that temporal lobe (i.e., medial structures versus lateral cortex) remains in question. In such cases, it is then necessary to extensively cover the surface and deep structures of the temporal lobe on the side that the seizure focus is likely to be located and also to have limited coverage of the contralateral temporal lobe. Contralateral coverage is usually achieved through a modified bur-hole exposure and placement of a small number of strip and depth electrodes. Such extensive recording from the temporal lobes provides an invaluable opportunity to investigate human brain functions that involve the temporal lobe and perisylvian brain regions.

In addition to standard clinical electrodes, a variety of specially modified electrodes are available that can collect research data in addition to clinical ECoG data. Some manufacturers make customized electrodes to suit each researcher’s need (e.g., Ad Tech Corporation, Racine WI). Most clinical grid or strip electrodes are constructed with a center-to-center intercontact distance of 1 cm. High-density electrodes with less than 5 mm of interelectrode distance provide better spatial resolution and can be fabricated without altering the clinical risk profile of the grid.⁸ Custom depth electrodes have several high-impedance microwire contacts in addition to clinical low-impedance contacts (see Fig. 59-1).^9–11 These high-impedance wires make it possible to record unit activity from the human brain. Some of the custom electrodes have more electrode contacts and lead cables attached to them than standard clinical electrodes do. The single-tailed electrode cables that some manufacturers provide can reduce the number of cables by combining multiple lead cables (up to 64 channels) into a single bundle, thus reducing the number of penetrations through the scalp.

Implantation Surgery

The surgical procedure to implant electrodes for a research participant is basically the same as that for a standard clinical epilepsy case. It is necessary to carefully plan placement of the electrodes in the optimal position so that the cables do not disturb each other or compress or displace the cortex. Displacement of cortex by grids or cables may occur because of the stiffness of the base plate of the electrodes and cables. Compression of the cortical surface can be minimized by making careful cuts on the base plate of the grid electrodes and meticulously looping cables to avoid undue torsion on the grid or strip electrodes. It is also important to pay careful attention to prevent leakage of cerebrospinal fluid (CSF) by placing a tight purse-string suture at each cable exit site on the scalp. It is not unusual to see CSF leakage around cable exit sites several days after the implantation surgery. This delayed leakage is probably due to subsidence of postoperative swelling of the scalp or breakdown of tissue around cable sutures that makes previously tight seals around cables loose enough to allow leakage of CSF. As soon as a CSF leak is noticed, the source of the leak should be identified and terminated by placing additional sutures at leak sites to reduce the risk for infection. In a series of approximately 200 patients who underwent implantation with chronic intracranial electrodes at the University of Iowa over a 15-year period, there was no significant difference in the infection rates of patients who were research participants and those who were not.

Accumulation of blood in the subdural space either beneath or above the grid electrodes sometimes occurs.^6,12–15 Although the exact mechanism of such blood accumulation is unknown, it is presumed that direct contact between the base plate of the grid electrodes and the dura mater may disturb normal hemostatic and resorptive processes. At our institution, we perform the following procedures in an effort to prevent accumulation of blood in the subdural space:

Verification of Electrode Placement

It is important to localize electrodes accurately in relation to surrounding brain structures to correctly interpret research data. Preimplantation and postimplantation computed tomography (CT), MRI, and photographs taken during both implantation and removal surgery are the three main tools used to localize the position of electrodes. Intracranially implanted electrodes create substantial artifact and distortion of images on CT and MRI, so extra caution is required when interpreting postimplantation imaging studies. The location of electrode contacts on a grid is best documented by photographs taken at the time of both implantation and explantation surgery. By matching the details of gyral and pial vessel anatomy, it is possible to localize surface contact locations with approximately millimeter accuracy. The position of electrodes is mapped onto a three-dimensional rendering of the brain surface drawn from each subject’s preoperative thin-slice MRI studies by referencing to a pattern of gyri and sulci on the cortical surface (Fig. 59-2).

FIGURE 59-2 Positions of subdural contacts are mapped on a three-dimensional surface rendering of a preoperative magnetic resonance image. The color on each contact represents the magnitude of electroencephalographic response in a high gamma band (approximately 70 to 150 Hz) to visual stimulation of static faces.

Localizing electrodes on the ventral surface of the brain is a difficult challenge because these electrodes cannot be viewed directly during surgery; as a consequence, electrodes in this location are not amenable to documentation by photography. MRI is also poorly suited for localizing ventral brain surface electrode contacts because of the susceptibility artifact created by brain-skull base bony structures interface. On thin-slice CT, metal contacts create such large amounts of artifact that it is impossible to observe brain parenchyma around a contact; however, each electrode contact and its relationship to the outline of the skull can be seen by adjusting the level and width of the display window. Therefore, it is possible to determine the position of contacts in relation to skull base bony structures. Images from CT and MRI are coregistered according to mutual voxel similarity. Finally, the position of electrodes can then be mapped onto the surface rendering of the preoperative brain image.

Electrode contacts on a depth electrode can be localized in relation to surrounding brain structures by postimplantation MRI. Only the larger, low-impedance contacts can be clearly delineated on postimplantation MRI, but with knowledge of the spacing of microwires positioned between these contacts, it is possible to depict accurately where these recording sites are within the brain, and these locations can be depicted on the preimplantation MRI study.^8,16,17

Recording of Electrical Activity

It is technically feasible to obtain massive amounts of ECoG recording data from hundreds of electrode contacts implanted in each surgical patient. Modern signal processing methods also enable investigators to use a wide range of analytic methods to discern what physiologic events are relevant to the cognitive functions being investigated. The practical challenge is to carefully plan and execute experimental protocols so that the results are interpretable and the limitations of the methods used are appropriately recognized. The key practical data collection issues are reviewed and discussed subsequently.

Modern clinical EEG recording equipment converts EEG potentials to digital signals and has the capability of recording more than 100 channels with high sampling rates. Although it is possible to use clinically recorded EEG signal for cognition research, it is better to have dedicated research recording equipment kept separately from the clinical EEG recording system because research recording can be performed more flexibly without disturbing the clinical EEG recording. In addition, the higher sampling rates used for research recordings enable investigators to study high-frequency brain activity that is not captured with standard clinical sampling rates. It is ideal to use battery-driven head stages and optical isolation of the EEG signal from the research amplifier-recording system to minimize the chance of injuring subjects by accidental leakage of current. Most institutional review boards and hospital biomedical engineers require this level of electrical isolation for the patient. This typically requires adaptation of research equipment designed for use in experimental animals, in which this level of isolation is not required. Almost all modern neurophysiologic recording systems have a digital recording design. Recorded data can be stored on digital recording media such as hard disk drives or optical recording media. Stored data can be analyzed offline with various commercially available or custom-made software. Because data are shared among many researchers, it is important to separate a subject’s identifiable information, such as name, initials, medical record number, or birthday, from the data recorded by replacing such information with unique research identifiers pursuant to regulations for the protection of personal health information.

At the University of Iowa we use custom-built connecting cables with a signal breakout box to split the ECoG signal picked up from the subject simultaneously to both a clinical ECoG recording device and a research recording device. These cables make it possible to conduct research recordings without disrupting clinical EEG monitoring. Researchers can use dedicated research recording equipment, and research cables can be disconnected for subjects’ convenience when research activity is not being performed. Depending on the specific research question being addressed, research recordings may require a wider frequency bandwidth.

Frequently, contamination by ambient electronic noise can become a problem, more so for the research recordings than for the clinical EEG recordings. Among various sources of noise, power line noise is typically the most disruptive and requires the greatest attention to eliminate. Although notch filtering may effectively reduce power line noise, such filtering distorts the EEG waveform and may affect the result of frequency analysis. Therefore, every possible effort must be taken to reduce noise contamination at its source. At our institution, research participants are housed in a specially constructed, electromagnetically shielded room in the National Institutes of Health–funded General Clinical Research Center. A significant amount of medical and nonmedical equipment is necessary for both the medical treatment and the convenience of the subjects, who spend up to 2 weeks in the room. It is useful to unplug as many power cords as possible when research recording is being performed. If any equipment can be run on battery power, it should be turned to battery mode. Shielding EEG connecting cables can reduce the extent of noise contamination. If some equipment has to be powered by alternating current, careful attention has to be paid to keep the power cords away from the ECoG recording equipment and connecting cables. Hospital-grade power cords must be used for all equipment, if possible, not only to reduce the noise level but also to reduce the chance of injuring the patient by leakage of current.

Cognitive Task

In properly selected patients with chronic intracranial electrodes, almost any test of cognitive function can be performed while ECoG recordings are under way. These tasks include those involving language functions and complex social behavior such as ethical decision making, as well as economic or financial decision making. Because these cognitive functions have been highly developed in humans, many of these functions cannot be studied in a substantive way in nonhuman animals. Emotion is another complex cognitive function that offers potential advantages for study in humans rather than other species because investigators can directly ask subjects how they are feeling and what kinds of emotions they are experiencing when they are exposed to experimental manipulations.

The following points must be considered carefully when a cognitive task is designed: (1) factors of interest and nuisance factors, (2) timing of stimulus delivery and response, (3) generation and recording of the timing signal, (4) order of stimulus presentation or response, and (5) number of trials.

The factors of interest must be defined clearly. After defining them, it should be determined how many levels each factor has and which nuisance, or confounding, factors need to be controlled. A randomized block design is often used to control contamination with nuisance factors. For example, the development of fatigue or fluctuation of attention level can be problematic in many cognitive tasks. When the effect of conditions A and B is to be compared, if experimental sessions are conducted sequentially by delivering condition A in the first session and condition B in the second session, it is likely that levels of fatigue and attention will not be the same between these sessions. If some difference is found in EEG measurement, it is difficult to assess whether these distinctions were due to differences in experimental conditions or differences in fatigue or attention level. In such cases, randomly interleaving the experimental conditions makes it possible to effectively equalize time-dependent nuisance factors (e.g., fatigue and attention level) between conditions A and B.

The timing of experiments must be planned carefully when the researcher wishes to detect correlations between the cognitive function of interest and EEG responses. Decisions on the interstimulus or intertrial interval should be based on considerations of what cognitive function is of interest and which region of the brain is being studied. For example, for investigation of lower level sensory cognition in a primary sensory cortex, the interstimulus interval can be shorter than that required for investigation of highly complicated cognitive function in the higher order cortex in the temporal or frontal lobe.

The sequence of stimulus presentations needs to be considered carefully to avoid unwanted effects on the EEG response that are related to stimulus order. When more than two stimuli are presented or more than two different responses are required, the order of their occurrence should be randomized, and the frequency of occurrence should also ideally be equalized.

It is necessary to conduct a sufficient number of trials to obtain robust statistical power. This issue is particularly important and challenging for experiments performed on epilepsy surgery patients. In most instances, these patients can maintain a high level of attention and fully participate in complex behavior tasks for less than 30 minutes per experimental block. This limitation has to be considered carefully when deciding how to balance the tradeoff between a large number of different stimuli or a large number of presentations of a smaller number of stimuli. The smaller the effect of experimental manipulation on the ECoG response, the larger the number of trials that are needed to detect and characterize the response induced.

Electrical Stimulation

Electrical stimulation of the brain has been used for the investigation of brain function for close to a century and is still the “gold standard” for neurosurgical functional mapping to minimize the risk for postoperative functional deficits.¹⁸ Electrical stimulation of the cortex is believed to either facilitate or disrupt local brain function in the vicinity of a stimulation site, depending on the stimulation parameters. Exactly how the frequency of electrical stimulation modulates neural activity within the stimulated structure is poorly understood and is the subject of many ongoing experimental investigations. With that caveat, it is generally believed that high-frequency stimulation well above 50 Hz exerts an inhibitory effect on the firing of neurons in the brain region exposed to suprathreshold currents. Stimulation in the 50-Hz range may activate sensory cortices and cause a subject to experience a sensory perception, but stimulation of language critical sites will disrupt function without generating a percept. This technique is useful for the investigation of local cortical and subcortical function not only in the clinical setting but also as a research tool.

Averaged evoked potentials elicited by electrical stimulation can be used for the investigation of connectivity between remote brain sites. By manipulating stimulus parameters, stimulation site, and recording site, it is possible to delineate functional connections in various brain regions.^19,20

One of the risks of electrical stimulation is the possibility of inducing a seizure. It is important that a qualified physician continuously monitor the ECoG recording during the electrical stimulation procedure to detect the occurrence of afterdischarges or local seizure activity induced by electrical stimulation. Once an afterdischarge is detected, electrical stimulation has to be halted temporarily until the afterdischarge disappears. On restarting the stimulation procedure, the stimulus intensity should be decreased. It is noteworthy that the afterdischarge threshold is not constant across various cortical locations. From our experience, the afterdischarge threshold in the primary sensory and primary motor cortices is lower than that of other brain regions. The threshold level may also be low in areas neighboring a seizure focus. Most afterdischarges stop spontaneously within a few minutes; however, they may occasionally progress to a generalized seizure.

Cooling

Cooling of brain tissue is known to modulate brain functions and has been used extensively in animal research,^21–24 as well as in human clinical settings.^25–27 Cortical cooling is a promising research tool to investigate local brain function.^28,29 Cooling brain tissue is known to block synaptic transmission reversibly within a highly localized region.³⁰ In contrast, electrical stimulation has the inherent problem of exerting effects well beyond the site of stimulation as a result of axonal conduction to functionally connected brain regions. The patterns of local spread of electrical stimulation currents within brain tissue are poorly understood. Neural activity induced by local stimuli may be transmitted to cortical and subcortical regions that are anatomically connected to the stimulation site, thereby disrupting or altering neuronal activity at these sites. Focal cooling can avoid the possibility of such remote effects. Brain blood flow exerts a robust effect on maintenance of tissue temperature. The area of low temperature is tightly restricted to within several millimeters of a brain cooling source, thus limiting the cooling effect to a local phenomenon.^29,31

Measurement of Physiologic Phenomena

While subjects are performing cognitive tasks, various physiologic measures can be recorded along with the ECoG recording. Such measurements include eye tracking, the electrocardiogram, the respiratory rate, oxygen saturation, and skin conductance. These simultaneous recordings are useful objective measures that complement the ECoG data set and allow investigators to examine underlying cognitive processing in the brain in a more sensitive and controlled manner. Contemporary eye-tracking methods allow the investigator to accurately measure and record eye movements, as well as alterations in pupillary size; in many experimental studies such data have been demonstrated to reflect underlying neuropsychological processes in human subjects.^32–36 Pupillary diameter readings are valuable measures of the status of the autonomic nervous system, along with skin conductance, heart rate, blood pressure, and respiratory rate.

Cognitive Studies in the Medial Temporal Lobe

The medial temporal lobe is often involved in the generation of seizures in surgically treatable epilepsy patients^37–39; thus, in many cases the ECoG recording electrodes are placed in this brain region. Neural structures in the medial temporal lobe play important roles in cognitive functions related to emotion, memory, and learning. For these reasons, the medial temporal lobe is one of the brain regions most extensively investigated by neurosurgeon-scientists who perform human cognitive research in epilepsy patients.⁴ Various recording techniques with modified electrodes permit finely detailed analysis of single-unit, multiunit, and local field potential activity in conjunction with sophisticated cognitive tasks. The combination of fine temporal and spatial resolution of these invasive recordings cannot be matched by any existing noninvasive research methods.

With this experimental approach, neurons in the medial temporal lobe, including the hippocampus, parahippocampal gyrus, and entorhinal cortex, were found to be activated by responding to visual stimuli of specific items.⁴⁰ The activity of a subset of these neurons was correlated with behavioral performance, in this case it was whether the visual stimulus was successfully memorized. The response patterns of memory performance–related neurons were different between the hippocampus and entorhinal cortex, thus suggesting that different medial temporal subregions may contribute differently to different aspects of the memory and learning processes.⁴¹ In addition to item-specific neurons, category-specific neurons were found.⁴² Some of these neurons responded not to what subjects were seeing but to what subjects perceived, imagined, or attended.^43,44 There were also neurons that were activated by spontaneous emergence of conscious recollection.⁴⁵

Involvement of the medial temporal lobe in spatial navigation has also been investigated extensively. Two different kinds of neurons were found, one representing “place” cells that respond to specific spatial locations and the other representing “view” cells that respond to specific views of task-relevant landmarks.⁴⁶ There was a difference in distribution of these cells between the hippocampus and entorhinal cortex. The activities of some place cells were modulated by specific combinations of goals of task and place; for example, certain cells were activated only when a particular house was found in a particular location. Another study revealed the existence of movement-related theta oscillations in the hippocampus and the neocortex, and these theta oscillations were significantly correlated with various cognitive functions.⁴⁷

Local field potentials in the medial temporal lobe were used for the investigation of mental activity as well. Gamma-band activity in the parahippocampal gyrus and ripple oscillations (100 to 200 Hz) in the hippocampus and entorhinal cortex were found to be correlated to awake and sleep states.⁴⁸ Ripples in the rhinal cortex were found to be correlated to memory performance.⁴⁹ Functional coupling between these regions, which can be measured by coherence in gamma-band activities, was found to be correlated to successful memorization.^50,51

The amygdala is known to be involved in defense mechanisms represented by the fight-or-flight response. A large volume of animal and human research has demonstrated that fearful emotion is modulated by activity of the amygdala.^52–56 It has been shown that gamma-band activity in the amygdala is enhanced by aversive pictures.¹⁷ The amygdala-specific research carried out in neurosurgical research subjects provides unique insight into the precise timing and electrophysiologic nature of amygdala activation.

Cognitive Studies in the Ventral and Medial Prefrontal Cortex

The ventral and medial sector of the frontal lobe has undergone extensive evolutionary development in humans in comparison to other primates, not to mention other animals. The function of this region is implicated in highly human behavior, such as social interaction, moral judgment, fairness, self-control, prediction of the future, and decision making in conflict situations.^57–59 A number of functional imaging studies have shown representation of these cognitive processes in this region of the brain.^60–70 Individuals who sustain damage to this region of the brain do not usually have deficits in intellect that can be measured by conventional intellectual tests; however, they often exhibit inappropriate social behavior and poor judgment with regard to their social and financial well-being.^59,71 Their choices are more influenced by immediate rewards even though the consequences of their choices are predicted to cause substantial disadvantage to them.^72,73

Anatomic studies have shown that the ventral and medial sectors of the prefrontal cortices have abundant connections with the sensory areas of multiple modalities and subcortical structures, including the hypothalamus, thalamus, amygdala, and brainstem.^74–76 The connectivity of this region supports the proposed function of this area, specifically, processing of various multimodal sensory inputs to modulate behavior, including visceral and autonomic function, to match behavior to fit appropriately to the situation in which an individual is placed.

The vast majority of this knowledge has been derived from lesion and functional imaging studies in humans and electrophysiologic studies in animals. Because the frontal lobe and functions of the frontal lobes are most developed in humans, there is a compelling rationale to investigate these functions in actual humans. Some of the functions in this region, such as ethics, morality, and emotional valence, are extremely difficult to study in other animal systems because it remains debatable whether and to what extent such functions exist in animals. Functional imaging methods provide an overview of wide brain regions with superior spatial resolution, up to a millimeter in scale. However, the greatest disadvantage of functional imaging methods, as well as lesion methods, is a lack of time resolution. In contrast, electrophysiologic studies in animals have the advantage of investigating finely detailed neural responses with precise time resolution, up to the millisecond scale, with simultaneous precision in spatial specificity.

Intracranial recording in epilepsy patients has the potential to fill the gap between lesion and functional imaging studies in humans and electrophysiologic studies in animals. Although the extent of the field of view is restricted to the vicinity of the area covered by electrodes, the electrophysiologic method in humans provides an incomparable level of time resolution and superb spatial resolution that can localize the neural activity of interest with unsurpassed precision.

Investigation of Emotion Representation

We have applied the single-unit recording technique to the investigation of responses of the ventromedial prefrontal cortex to pictures of complex emotional scenes.^16,77 All patients suffered medically intractable epilepsy, and the preoperative epilepsy evaluation suggested a seizure focus residing in the frontal lobes; therefore, chronic EEG monitoring with depth and subdural electrodes covering the frontal lobes was warranted. We implanted custom-made depth electrodes that had eight high-impedance microcontacts incorporated in an electrode shaft along with four low-impedance clinical EEG contacts. Selection of target locations was based on preoperative MRI, and implantation of the electrodes was performed by using either the CRW stereotactic implantation system (Integra, Plainsboro, NJ) in earlier cases or the Stealth Station frameless stereotactic system (Medtronic, Minneapolis, MN) in later cases. We waited more than 5 days after the implantation surgery until we started conducting research protocols that involved complicated cognitive tasks. By the time we started research recordings, doses of analgesic and antiseizure medications had been tapered so that they did not affect subjects’ cognitive functions. At the time the studies were conducted, the subjects had fully recovered from the effects of the implantation surgery. Because the primary purpose of chronic seizure monitoring is to record seizure activity, it is not uncommon for subjects to have partial or complex seizures that may affect local brain activity. To avoid unwanted influence of such seizure activity on subjects’ cognitive function, we conducted research recordings only when subjects had not suffered seizures within the 12 hours preceding the recording session. After the chronic seizure monitoring was completed, we confirmed that the regions from which the recordings were obtained were distant from the seizure foci and also that the recordings were not contaminated by frequent interictal discharges.

Unit recording picks up field potentials associated with action potentials of neurons that are located within 100 µm from a high-impedance contact. The locations of recording sites were determined with the methods described earlier in this chapter, and the location of each electrode contact was delineated on the preimplantation MRI studies (Fig. 59-3).

FIGURE 59-3 Neuroanatomic localization of recording sites (arrows) in individual slices (left column) and three-dimensional reconstructions (right column) of the brain of a research subject. A, A postimplantation scan shows the actual electrodes in situ (black signal artifact on the coronal scan). B, Recording sites are mapped on the preimplantation scans (small black squares). C and D, The location of the recording sites was also projected on three-dimensional reconstructions of the subject’s brain.

(From Kawasaki H, Adolphs R, Oya H, et al. Analysis of single-unit responses to emotional scenes in human ventromedial prefrontal cortex. J Cogn Neurosci. 2005;17:1509-1518.)

We investigated unit responses in the medial and ventral frontal cortex to visual stimuli that depicted pleasant, neutral, or aversive scenes. These scenes were part of the International Affective Picture System (IAPS), which has been used widely for studies of emotion.⁷⁸ The aversive stimuli include pictures of wartime, mutilated bodies, burn victims, and so on. These pictures are visually heterogeneous and do not have any low-level or simple visual features in common that might explain the different responses to different emotion categories that we observed. In one subject, we found unit responses that were specific to aversive stimuli and very rapid in onset (Fig. 59-4).¹⁶ The latency of responses spanned between 120 and 140 msec. Rapid responses of this latency are most likely explained by feed-forward responses from the occipitotemporal visual cortices to the medial and ventral frontal cortex without any intervening feedback modulation.⁷⁹ It is interesting to note that the unit responses to aversive pictures showed biphasic response patterns consisting of an initial rapid and brief decrease in firing and a later prolonged increase in activity. The later prolonged excitation is a possible reflection of recursive responses with multiple feed-forward and feedback interactions among multiple brain sites anatomically connected to the medial and ventral frontal cortex. These responses with different timing may reflect early coarse categorization of the emotional significance of visual scenes and later more detailed neural processing related to evaluation of the biologic value to an individual, prediction of consequences, triggering of autonomic responses, and planning of future action.

FIGURE 59-4 Composite peristimulus time histograms (top) and raster plots (bottom) from four individual neurons recorded medially in response to emotional scenes (stimulus onset, vertical line). Raster plots are shown at the bottom for presentations of each of 11 aversive stimuli. These units show a brief decrease in firing as early as 120 msec after the onset of aversive pictures, followed by a prolonged increase in firing.

(From Kawasaki H, Kaufman O, Damasio H, et al. Single-neuron responses to emotional visual stimuli recorded in human ventral prefrontal cortex. Nat Neurosci. 2001;4:15-16.)

In addition to the subject just described, we recorded unit responses to pictures from the IAPS battery in the bilateral medial and ventral frontal cortices of three more subjects. Figure 59-5 shows the distribution of emotion-selective neurons at four different sites in these regions from a total of four subjects.⁷⁷ Although aversive pictures recruited predominantly more neurons than did any other stimulus category in all sites, there is no clear topography of preferred emotional content. Because sampling of our unit recording is sparse (in other words, we recorded at only a few discrete brain sites in a small number of subjects), we might have missed an emotional topographic representation. It is premature to draw conclusions from our data about which hemisphere is predominantly involved in emotional processing, whether there is a preferred emotion that is processed in each hemisphere, or whether there is emotional topographic representation within a hemisphere. However, it is noteworthy that a minority of units responded to pleasant and neutral pictures at all locations. This provides evidence that information about the emotional content of visual scenes is intermingled within the medial and ventral frontal cortices.

FIGURE 59-5 Summary of recording sites at which significant selective responses to emotional stimuli were found. The upper middle panel shows a projection of the recording site of the right subgenual cingulate cortex (R sgCC) onto the sagittal surface, and the left upper panel indicates the number of neurons with particular response selectivities found in this location. The left lower, the right upper, and the right lower panels show bar graphs of the number of selective neurons found in the right medial orbitofrontal cortex (R mOFC), the left gyrus rectus (L GR), and the left medial orbitofrontal cortex (L mOFC), respectively. Projections of the recording sites onto the ventral surface of the brain are shown in the lower middle panel. Pink bars represent neurons that increased their firing rates in response to visual stimuli, and the tan bar represents a neuron whose response showed a decrease in firing rate. Each bar graph also indicates the total number of neurons that are isolated for each recording site. A, selective for aversive pictures; N, selective for neutral pictures; P, selective for pleasant pictures; R, nonspecific visual response.

(Adapted from Kawasaki H, Adolphs R, Oya H, et al. Analysis of single-unit responses to emotional scenes in human ventromedial prefrontal cortex. J Cogn Neurosci. 2005;17:1509-1518.)

Other investigative methods, including the blood oxygen level–dependent (BOLD) response, MEG, and ECoG, require many neurons to fire synchronously to be detected. Therefore, if a tiny fraction of neurons are active within a region, such activity cannot be detected by these methods. Conversely, it is possible to detect such underrepresented activities with unit recordings, as shown in the aforementioned study. Our findings are consistent with current knowledge about this region gained through a large amount of research in animal models and functional imaging studies in humans.^80,81 Neurons within the orbitofrontal cortex have been shown to respond selectively to stimuli of a variety of sensory modalities based on their emotional significance, rewarding or punishing valence, or motivational value.

Investigation of Expectation of Reward and Punishment

The Iowa Gambling Task is a well-validated test for identifying and quantifying functional deficits associated with damage to the ventromedial prefrontal cortex. Deficits in this region have otherwise been very difficult to detect with other cognitive tasks designed to investigate frontal lobe function.^58,82 In the Iowa Gambling Task, subjects choose cards from four decks in order to win or lose money. Each deck is assigned a different probability of gain and variance of win-loss value for each draw of a card. For example, one deck consists of a majority of a moderate amount of wins but occasional large losses, and in the long run the sum of draws would end up losing money. Another deck consists of many small wins and occasional moderate losses; however, the sum of draws from this deck would eventually end up winning money. Subjects do not know which deck is a winning deck and which one is a losing deck when they start the task. After drawing cards a number of times, subjects develop an expectation of what is likely to happen (win or lose) when they choose a card from a given deck on the basis of experience. Furthermore, subjects’ behavior comes to be influenced by the expectation. As such, subjects tend to choose more from decks where they feel that they will make more money in the long term and to avoid decks that may incur larger wins but lose money in the long term. Such expectation is probably unconscious in the early stages of its development. We applied a reinforcement-learning algorithm to model the process of development of expectation and choice selection behavior.⁸³ The model incorporated the probability of selecting from each deck and the degree of expectation. The model estimated the value of all choices, which parallels the expectation from each deck. A reward prediction error is an index for measuring the discrepancy between expectation and reality and is derived by subtracting the estimation from the result of a draw.

We presented the Iowa Gambling Task to an epilepsy patient who had multiple electrodes in his frontal lobes.⁸⁴ He performed the task as normal subjects would. He initially chose from all decks; however, with experience he learned to choose from good decks and to avoid choosing from bad decks. This provided evidence that the ventral and medial parts of the subject’s prefrontal cortex were functioning normally. While the subject was performing the task, we recorded local field potentials at three sites in the right frontal lobe with two custom depth electrodes: the granular paracingular cortex (area 10m), the middle frontal gyrus, and area 11l. The custom depth electrode had multiple high-impedance microwire contacts. To record local field potentials, we performed bipolar recording with a pair of closely positioned microcontacts at each recording site; as such, far field potentials would effectively cancel out.

We found that reward prediction error correlated with the alpha-band component of the event-related potential recorded in the ventral and medial prefrontal cortex. No such correlation was found in other parts of the frontal lobe or in other frequency bands. Further analysis revealed that this correlation was significant only when a subject chose the risky decks but punishment was not delivered (Fig. 59-6). In other words, the alpha-band component of the EEG in this region was positively correlated with the magnitude of the difference between what the subject obtained and expectation only when the subject expected punishment but was unexpectedly rewarded. These findings are consistent with other studies suggesting that the ventral and medial prefrontal cortex is involved in updating the expectations of punishment or reward and in reinforcement learning.^85–90

FIGURE 59-6 A, A scatter plot shows correlation between reward prediction error (PE, x-axis) and the magnitude of the alpha-band component (y-axis) of the event-related potential (ERP) recorded in the ventral and medial prefrontal cortex. B, A scatter plot shows the same correlation as in A, but restricted to trials from risky decks. Reward prediction error correlated with the alpha-band component of the ERP recorded in the ventral and medial prefrontal cortex. Further analysis revealed that this correlation was significant only when a subject chose the risky decks but punishment was not delivered. RMS, root mean square.

(From Oya H, Adolphs R, Kawasaki H, et al. Electrophysiological correlates of reward prediction error recorded in the human prefrontal cortex. Proc Natl Acad Sci U S A. 2005;102:8351-8356.)