Activity of 98 single neurons in human lateral temporal cortex was measured during memory encoding for auditory words, text, or pictures and compared with identification of material of the same modality in extracellular recordings during awake neurosurgery for epilepsy. Frequency of activity was divided into early or late epochs or activity sustained throughout both; 44 neurons had significant changes in one or more categories. Polymodal and sustained changes lateralized to dominant hemisphere and late changes to nondominant. The majority of polymodal neurons shifted categories for different modalities. In dominant hemisphere, the timing and nature of changes in activity provide the basis for a model of the roles of temporal cortex in encoding. Superior temporal gyrus excitatory activity was related to the early epoch, when perception and processing occur, and middle gyrus to the late epoch, when semantic labeling occurs. The superior two-thirds of middle gyrus also demonstrated sustained inhibition. In a subset of lateral temporal neurons, memory-encoding activity reflected simultaneous convergence of sustained attentional and early perceptual inputs.
A role for temporal lobe in human recent memory is well established. Within temporal lobe, that role is usually related to medial structures, hippocampus, and parahippocampal gyrus (Scoville and Milner 1957). However, there is also evidence for participation of lateral temporal cortex in memory. Deficits in immediate and 30–40 min delay recall of prose passages have been reported with lateral temporal lesions sparing medial structures (Milner 1967; Ojemann and Dodrill 1987; Cheung and Chan 2003). Lateral temporal cortical electrical stimulation interferes with memory for explicit object names, particularly when applied during encoding or the 7–9 s of a storage phase filled with distracters (Ojemann 1978; Perrine et al. 1994). Functional imaging changes in lateral temporal cortex have been identified during the encoding stage of explicit verbal memory measures (Kirchhoff et al. 2000; Casasanto et al. 2002; Fletcher and Tyler 2002).
In addition, the activity of single neurons recorded in human lateral temporal cortex changes with memory measures. These changes were also particularly evident with encoding, when activity during encoding was compared with that during identification of similar verbal material, but without the instruction to retain the material in memory. We have now published recordings from a total of 236 neurons in human lateral temporal cortex obtained during this comparison in multiple series of subjects (Ojemann et al. 1988, 2002; Haglund et al. 1994; Weber and Ojemann 1995; Ojemann and Schoenfield-McNeill 1998, 1999). Activity was significantly altered by the instruction to retain the material in memory in 135 (57%) of those neurons, further establishing that lateral temporal cortex is part of the network that subserves encoding of explicit verbal material.
We previously reported anatomic subdivisions in temporal cortex for neurons encoding nameable object pictures, text, and auditory words, compared with those changing activity with the storage or retrieval stages of the memory measure, with implicit memory and with different types of retrieval, recall, or recognition (Ojemann et al. 2002). Here, we report the timing of the memory-encoding changes in these recordings, separating activity into an “early” epoch related to perception and processing, a “late” epoch related to verbal output, and activity “sustained” throughout both epochs. We found regional differences within lateral temporal cortex related to the timing and nature of this activity. Based on these data, we have developed the first model of the roles of different portions of this cortex in encoding.
The 29 subjects of the present study all had temporal resections for medically refractory epilepsy using a technique where the subject is awake for a portion of the operation under local anesthesia (0.5% lidocaine and 0.25% marcane) so that physiological guides unperturbed by anesthesia could be used to plan the resection (Ojemann 1995). Surgery was on the left in 21 subjects. Intracarotid amobarbital perfusion assessment of language lateralization (Wada and Rasmussen 1960) was available for 27 subjects: 22 had language changes predominately or exclusively with left perfusion and 5 predominately or exclusively with right perfusion. The remaining 2 subjects who were right handed and had right temporal operations for adult-acquired lesions without a postoperative language change were considered left dominant. One subject with right language dominance based on preoperative intracarotid amobarbital perfusion assessment had right temporal recordings and is included in the dominant group, and 4 subjects with left temporal recordings and exclusively or predominantly right language dominance with the same assessment are included in the nondominant group; otherwise, all dominant cases had left temporal recordings and all nondominant, right. Preoperative verbal IQ data were available for 25 subjects, with an average verbal IQ of 92.7 (range 72–115). Mean seizure onset was 16.7 years (range 4 months to 49 years). Mean age at operation was 34.6 years (range 18–61). One subject had a posterior temporal tumor, well behind the area of recording. No other subjects had imaging- or pathology-identifiable lesions in lateral temporal cortex, but 10 subjects had mesial temporal sclerosis and 5 mesial temporal tumors. Each subject gave informed consent to participate in this study, with the study and methods for obtaining consent reviewed annually by the University of Washington Institutional Review Board. Twenty-six of these subjects were included in our previous study of the anatomical distribution of temporal cortical neuronal activity related to various aspects of recent memory (Ojemann et al. 2002). The 3 additional subjects were excluded from that earlier study as the storage stage of the memory measure differed from that used in the other subjects. However, the encoding stage was identical with that used in the other subjects. As that is the focus of the present study, they were included here.
The microelectrode study was performed after completion of the physiological recording and stimulation needed to plan the resection. At that time, the subject was fully awake under local anesthesia, having awakened from the propofol intravenous anesthesia used for placement of the block and craniotomy at least an hour earlier. The sites of recording were in cortex that was subsequently resected as part of the surgical therapy of the subject's epilepsy. Two tungsten microelectrodes were back loaded through a translucent 1-cm diameter footplate into a hydraulic microdrive. The footplate was used to dampen cortical pulsations. Care was taken to avoid blanching pial vessels. Two microdrives were used in the majority of subjects; thus, activity at as many as 4 sites in 2 different regions could be simultaneously recorded. The sites of these recordings were identified by numbered tags and their location recorded photographically. Location of recording sites was established by the relation to the sulcal boundaries visible on those photographs (sylvian fissure, superior temporal sulcus, and middle-inferior temporal gyrus sulcus) and by measurements from the recording site to the tip of the temporal lobe. Other than avoiding recording from neurons with evidence of injury or epileptiform burst activity (Calvin et al. 1973), recordings were obtained from an unselected random sample of neurons at each recording site. Activity from each microelectrode, the subject's voice, and markers indicating changes in test items were recorded on frequency-modulated tape for later analysis. System frequency response for microelectrode channels was 100–6500 Hz.
Once stable recordings were obtained, the patient engaged in a series of tasks assessing various aspects of memory and identification. Tasks were presented in blocks, with the order of blocks varied between subjects. Stimuli consisted of nameable object pictures, text words, or auditory words. Items were presented by a video monitor controlled by a Mac 540 Powerbook computer running Psyscope software. Event codes distinguishing experimental trials were recorded on one channel of the analog tape. The behavioral measures analyzed in the present study were those assessing recent verbal memory and identification.
In the memory task, items were presented in sets of 5. The first item of each set was an identifiable item (encoding stage), followed by 3 different items that act as distracters (storage stage), and then a visual recall cue (retrieval stage). Distracters were of the same modality as the item presented on that trial. Items and cue were presented at the rate of one every 3 s. The subject's task was to identify each item aloud and then at the recall cue state aloud the item on the first slide of each set. Seven sets of items in each of the 3 different modalities—auditory words, text, or nameable object pictures—were randomly intermixed for all but the first 3 subjects, where 10–17 sets of each modality were presented.
The identification task included items and perceptual controls randomly presented at the rate of one every 4 s. Each identifiable item had a unique perceptual control of the same modality consisting of auditory words backward, scrambled pictures, or scrambled text. The patient was instructed to repeat auditory words, read text, and name pictures and not respond to any nonsense items. Twelve items of each modality and control were presented.
Each tape channel was digitized at 10 KHz using a program running on a Mac IIx computer. Activity of each microelectrode channel was divided into that of individual neurons with a window discriminator and visual separation of the resulting amplitude-frequency histograms. We have previously published examples of this separation (Schwartz et al. 1996). The time of presentation of each item was divided into 3 epochs. Epoch 2 began 300 ms before the patient's response and lasted 1500 ms. Thus, this epoch covers both any activity related to motor prepotentials (“bereitschaft” potentials; Libet 1985; Altenmuller et al. 2005) as well as overt speech output. Epoch 1 was the period from the appearance of the item on the video screen until the beginning of epoch 2 and thus contains activity related to perception or processing. Epoch 3 was the time from the end of epoch 2 until the next item appeared on the screen. Normalized frequency of activity was determined for each epoch by dividing the number of discharges in the epoch by its duration.
For each neuron, any relation to memory encoding was established by comparing activity in each epoch of the first (encoding) item of each memory set to the analogous epoch of identification in the same modality. Each of these comparisons was evaluated with the nonparametric Mann–Whitney U test. Statistical significance was set at P < 0.00625 for each epoch, or activity was summed over all epochs. Thus, a neuron could be related to identification or memory for one modality by a significant change in any one of the 3 epochs or overall activity, giving an overall chance of establishing this relationship on a random basis of 0.05, 2 tailed. For any epoch with significant changes, activity was further subdivided into 50-ms bins. These were examined to determine the time of appearance of peak change in activity in the direction of the significant change. For increased activity with encoding, this determination was made from activity during encoding alone, but for decreased activity with encoding, due to the low overall firing rates, the difference between encoding and identification activity was used. Statistical significance of regional differences in the distribution of neurons with different characteristics was determined with Fisher's exact test.
Recordings were obtained from 98 neurons at 38 sites in 29 patients. All recordings were obtained from lateral temporal cortex, in the area illustrated in Figure 1; 57 neurons at 20 sites in 15 patients were in the language-dominant hemisphere. Mean firing rate was 7.06 discharges/s (standard deviation = 5.53). Changes in frequency of activity for encoding compared with identification of material of the same modality that were significant at the 0.05 level, 2 tailed, and corrected for multiple tests were present in one or more epochs or overall activity for 51 neurons (52%). Changes were somewhat more frequent in dominant hemisphere recordings (33/57, 58% vs. 18/41, 44% nondominant, P = not significant). These changes were present in the first or second epochs or both in 44 of those neurons (45% of the total sample, 86% of those with significant encoding changes); changes were only in overall activity in the remainder.
Relation to Epochs
These 44 neurons represent the data set for this study of the timing of the changes in neuronal activity during memory encoding. They were recorded from 22 patients. Thirty of them were in the dominant hemisphere (51% of dominant neurons, 37% of nondominant). In 11 neurons (6 dominant), the only significant changes were confined to the first (early) epoch, the period from presentation of the item to 300 ms before the overt identification response that was required for both identification and memory encoding (mean duration of early epoch = 1451 ms, examples: Figs 2 and 3). Significant changes confined to the second (late) epoch, from 300 ms before the overt response to 1200 ms after, were the only changes present in 12 other neurons (5 dominant, examples: Figs 4 and 5). Significant changes sustained throughout both epochs were the only changes present for 9 neurons (8 dominant, examples: Figs 6 and 7). The remaining 12 neurons had significant changes in different epochs for different modalities (11 dominant, examples: Figs 5 and 7). These neurons were 57% of the neurons with significant changes for more than one modality; the other neurons with multiple modality changes had them confined to the same epoch. Of the neurons with changes in different epochs for different modalities, 10 had sustained activity for some modalities and early or late activity for others and the remainder early or late activity for different modalities. Neurons with sustained activity for some modality were significantly more likely in the dominant hemisphere (any sustained activity: 17/30 dominant, 2/14 nondominant, P = 0.01). Neurons with only late changes were lateralized to the nondominant hemisphere (7/14 nondominant, 5/30 dominant, P = 0.03). There was a trend for neurons with different relations for different modalities to be lateralized to the dominant hemisphere (11/30 dominant, 1/14 nondominant, P = 0.07). Increased activity characterized 80% of the changes confined to the early epoch, 68% of changes confined to the second epoch, and 56% of changes sustained throughout both epochs (Table 1). With one exception, the direction of change for an individual neuron was the same for all different epochs and modalities with significant changes.
|Location||Neurons recordeda||Only epoch 1||Only epoch 2||Both epochs 1 and 2|
|Location||Neurons recordeda||Only epoch 1||Only epoch 2||Both epochs 1 and 2|
Note: ST, superior temporal gyrus; SMT, superior third of middle temporal gyrus; MMT, middle third of middle temporal gyrus; IMT, inferior third of middle temporal gyrus. Significant differences, dominant hemisphere: only epoch 1 versus only epoch 2, ST versus all MT (P = 0.001); sustained inhibition versus excitation, SMT + MMT versus ST + IMT (P = 0.005); late epoch inhibition versus excitation, SMT versus MMT (P = 0.005).
Total sample/neurons with epoch 1, 2, or both changes.
Number with increased/decreased activity in those epochs during memory encoding compared with identification of items of the same modality. Three modalities sampled for each neuron.
Relation to Modalities
Of the 3 modalities assessed, auditory word repetition, text reading, and picture naming, changes were in a single modality for 23 neurons (10 dominant), in 2 modalities for 8 (7 dominant), and in all 3 modalities for 13, all in the dominant hemisphere. As previously reported for all neurons with significant encoding changes (Ojemann et al. 2002), neurons with encoding changes in early, late, or both epochs that involved multiple modalities were significantly more likely in the dominant hemisphere (P < 0.001). Only one of the 21 multimodality neurons was a combination of both visual modalities (text and picture naming); the remainder combined the auditory modality with one or both visual modalities. There were no significant differences between modalities in the overall proportion of neurons changing activity or in the distribution between early, late, or both epochs. Significant lateralization of neurons with changes to the dominant hemisphere was present for auditory (P < 0.001) and text (P = 0.02) but not picture modalities. Neurons with changes only during object picture naming were significantly lateralized to the nondominant hemisphere (P = 0.02).
Location of Changes Related to Different Epochs
Table 1 presents the location of recordings in relation to lateral temporal gyri, with the wide middle temporal gyrus divided into superior, middle, and inferior thirds. The proportion of neurons with changes in any epochs does not differ across the different regions. The number with significantly increased or decreased activity for the different epochs of memory encoding summed across modalities is also indicated for each region. For changes in only the first (early) or only the second (late) epochs, there was a significant difference between superior and middle temporal gyri in the dominant hemisphere, with superior temporal neurons more likely to show only early changes and middle only late (Table 1, P = 0.001). Within dominant middle temporal gyrus, late epoch excitation was significantly more likely in the superior third and inhibition in the middle third (Table 1, P = 0.005, example in Fig. 5). Sustained changes were present throughout lateral cortex. However, the superior and middle thirds of dominant middle temporal gyrus showed significantly more sustained inhibition than surrounding cortex, superior temporal gyrus, and inferior third of middle gyrus (Table 1, P = 0.005, example in Fig. 7). The location of the regions in dominant temporal cortex with inhibition sustained through both epochs and that confined to epoch 2 is indicated in Figure 1.
Timing of Changes in 50-ms Bins
To provide a more precise indication of when changes in neuronal activity occurred, activity for each epoch and modality with significant changes were divided into 50-ms bins (Figs 2–7). The bin with maximum excitatory activity was readily identified as illustrated in Figures 2–4 and 6. For early epochs with significant excitation, average peak activity was 18.3 discharges/s (range 7–55), peaking at 495 ms from presentation of the item to be encoded. Figure 8 illustrates the timing of these peaks. They are not randomly distributed. Although the average duration of the first epoch was 1451 ms, in only 3 of the 20 early epochs with significant changes was the peak later than 750 ms. Some of this peak activity was very early, within the first 150 ms for 7 of the 20 early epochs (Fig. 8, example in Fig. 3), one within the first 50 ms and 2 within 50–100 ms. Only one neuron showed this very early activity for more than one of the 3 modalities, although all modalities were represented in the very early sample (3 auditory, 2 pictures, 2 text). Of the remaining peaks, 4 occurred between 150 and 400 ms and 6 between 400 and 750 ms. Peak activity occurred earlier in the superior temporal gyrus (average 310 ms) than in the middle gyrus (680 ms, P = 0.05), and 5/7 recordings with activity within the first 150 ms were in superior gyrus. Differences in overall activity between modalities were not significant, although peak activity was earlier for auditory items (average 293 ms) than for picture (625 ms) or text (585 ms). For neurons with early epoch changes in more than one modality, the peaks for the different modalities differed by an average of 526 ms (range 50–950 ms).
For late epochs with significant excitation, average peak activity was 18.6 discharges/s, occurring 509 ms after beginning of this epoch. By definition, this epoch begins 300 ms before initiation of the overt response. Peak activity for 9 of the 17 epochs with significant changes was within that 300 ms (Fig. 8, example in Fig. 4). Five recordings had peaks 300–150 ms before overt response onset. Four had peaks within 50 ms of response onset. The remaining 5 peaks were 300–900 ms after. No overall significant differences were present for modality or anatomy, although peak activity occurred later for auditory items (average 508 ms after beginning of epoch) and pictures (560 ms) than text (320 ms). In contrast to early epoch changes, late epoch peaks were similar for different modalities with significant changes in the same neuron (mean difference 67 ms, range 50–100 ms).
For excitation sustained through both early and late epochs, the average peak activity was 20.8 discharges/s (range 8.5–40), occurring 1184 ms after initial presentation of the encoded item. Ten of the 16 peaks fell within the first epoch. As seen from Figure 8, those peaks had timing similar to peaks for only early epoch activity (example in Fig. 6). Peak sustained activity was significantly (P < 0.02) earlier for picture items (average 360 ms) than auditory (1475 ms) or text (1607 ms). Indeed, the peak for sustained excitation for all picture items was in the first epoch, in contrast to this peak for text or auditory items that were equally distributed between epochs and also in contrast to peak activity for picture items only in the first or only the second epochs, also equally distributed. There were no significant differences in peak activity related to anatomy. As with early changes, neurons with sustained changes in several modalities had peaks for the different modalities at substantially different times (mean 504 ms, range 50–1200 ms).
The timing of maximum “inhibition” was difficult to identify during memory encoding alone, due to the low average firing rate. This was readily identified on plots of the difference between memory encoding and identification (Figs 5 and 7). Values for maximum inhibition for all inhibitory neurons are shown in Figure 9. Maximum inhibition in the early epoch occurred at a mean of 670 ms (range 0–1200 ms) with a mean difference decrease of 12.2 discharges/s (range 7–17). Maximum inhibition in the late epoch occurred at a mean of 336 ms (range 50–1150) with a mean difference decrease of 13.9 discharges/s (3.5–24). Maximum inhibition sustained through epochs 1 and 2 occurred at a mean of 1358 ms after item presentation (range 0–2900 ms) with a mean difference decrease of 27 discharges/s (range 12–57). Ten of these fell in epoch 2 and 2 in epoch 1. Maximum inhibition was more randomly distributed than maximum excitation, with the exception of a cluster of maximum inhibition 250–300 ms before overt response.
Our earlier analysis of these recordings focused on changes in the frequency of temporal cortical neuronal firing across the encoding–storage–recall retrieval stages of the explicit verbal memory task. Changes modulated by the storage or retrieval stages of the task were overrepresented in recordings from basal temporal cortex, extending from inferior temporal gyrus to collateral fissure (Ojemann et al. 2002). However, encoding changes were much more widespread, throughout lateral temporal cortex. Here we have further analyzed that encoding activity in terms of timing and nature of the changes. The anatomical distribution of these changes provides the basis for a novel model of the role of superior and middle temporal gyri of dominant lateral temporal cortex in encoding. This model suggests a flow of activity during encoding, from the superior temporal gyrus, with early activation likely related to perception of the material to be encoded, to middle temporal gyrus with later activation, at a time when memory-dependent processes are thought to occur. Other neurons in middle gyrus have even later activation related to the overt verbal output of the encoded items. Focal regions of inhibition were also identified in middle gyrus. All these changes were independent of the modality of the encoded items.
However, lateralized differences in encoding of different verbal modalities were observed. Our earlier analyses of these recordings (Ojemann et al. 2002) indicated that neurons that are part of the networks for several different modalities were lateralized to the dominant hemisphere. In the present analysis, they constitute almost two-thirds of the neurons there changing activity with encoding. In nearly all neurons, this is a cross-modal interaction between auditory and at least one visual modality, text reading, or object naming. The proportion of neurons responding only to both visual modalities was substantially smaller than those with auditory and some visual modality. This is similar to findings in a previous series where few neurons showed memory responses to both text and objects (Ojemann and Schoenfield-McNeill 1999). Although neurons that were part of the networks for encoding auditory and text material were more frequent in dominant temporal cortex, neurons in the network for encoding object picture names were not significantly lateralized. Indeed, those showing only this effect were more frequently encountered in nondominant recordings. A role for nondominant hemisphere in object naming has been reported in some lesion studies (Luckhurst and Lloyd-Jones 2001). Nondominant recordings were also more likely to show activity confined to the late, output-related epoch. Later, output-related activity lateralized to nondominant hemisphere was a feature of 2 of our earlier studies of temporal neuronal activity during identification of nameable objects compared with spatial tasks (Schwartz et al. 1996, 2000). This suggests that portions of the network for identification of this type of material have a similar component, whether identification is part of encoding or not, but when part of encoding, additional neurons are recruited into the network, those neurons with a significant difference in activity when the same object names are identified as part of encoding compared with identification without the instruction to retain the item in memory.
Within dominant temporal cortex, certain timing patterns are more likely to be recorded in different anatomical regions. Early activity was more likely from superior gyrus recordings, whereas late activity was more likely from middle gyrus. Moreover, even during the early epoch, superior temporal gyrus increased activity peaks occurred significantly and substantially (360 ms) earlier than those in middle temporal gyrus. Together these findings relate the superior temporal gyrus activity to the early aspects of perception and processing involved in encoding and the middle gyrus to later aspects. Some of the increased activity in superior gyrus occurred very early, within the first 150 ms, even some within the first 50 ms. The earliest component of the visual evoked potential has an average peak at 55–60 ms in both scalp and intracranial recordings (Celesia and Peachey 2005; Farrell et al. 2007). Activity of the vertex electroencephalography (EEG) negative-evoked potential at 60–200 ms represents the earliest activity related to selective attention (Hillyard and Woods 1979). This potential is modulated by differences in simple physical cues. Thus, it is likely that the very early neuronal activity we recorded is related to early perceptual components of encoding, unexpectedly widely distributed beyond primary sensory cortices, even to dominant superior temporal cortex. The very early neuronal activity was present for both auditory and visual material rather than one modality stream. Moreover, activity of individual neurons differentiated not only between auditory and visual items but also within the 2 types of complex visual verbal material. At a neuronal level, then, very early excitation is highly selective for type of material, with different neurons recruited into the very early network for different material. The importance of this early activity for successful encoding is evident from the presence of neurons in the same region that have early encoding activity that discriminated correct from incorrect memory performance (Ojemann et al. 2004). Changes in activity related to recognition as opposed to recall retrieval and to implicit memory were also recorded from the same region, though generally not the same neurons (Ojemann et al. 2002), suggesting that perceptual processes may also have a role in these aspects of memory. Alternatively, the very early activity could represent a “top-down” attentional signal for encoding of items of a particular modality, although as illustrated in Figure 8, 7/10 recordings with very early (<150 ms) peak activation had no evidence for sustained activation that seems more likely to characterize attentional processes (Ojemann and Schoenfield-McNeill 1999).
Most of the remaining early neuronal activity peaks between 200 and 750 ms, a temporal interval similar to the “cognitive potentials” of the scalp EEG, at least a component of which has been thought to reflect memory-dependent processes (Hillyard and Woods 1979). This time interval is also similar to that for evoked potentials from rhinal cortex for remembered high frequency, but not low frequency, words, ascribed to “a semantically affected operation … supporting memory formation indirectly” (Fernandez et al. 2002). That a portion of the dominant middle temporal gyrus is part of this semantically related explicit memory-encoding network is suggested by our recording there of a predominance of peak excitatory activity in this 200–700 ms period. This is earlier than the hippocampal-evoked potential changes recorded in the Fernandez et al. study for remembered words regardless of frequency and considered “a direct correlate of declarative memory formation.” Based on our neuronal recordings and these evoked potential findings, the neurophysiological events during memory encoding in lateral temporal cortex precede those in hippocampus.
The late epoch reflects activity related to the overt verbal output labeling the material encoded in memory. Within dominant hemisphere, this change was more frequent in middle temporal gyrus recordings, further evidence for the role of this cortex in semantic labeling of encoded material independent of modality. Over half of the peak excitation occurred in the period from 300 ms before to just at the initiation of the overt response. This is a period that corresponds to the “bereitschaft” potential in frontal lobe, a potential that has been viewed as a motor prepotential perhaps reflecting unconscious intention to act (Libet 1985; Altenmuller et al. 2005). The uniformity of timing of peak activity for different modalities in the same neuron seems consistent with such a motor role, given the similar nature of the motor outputs (single words). Temporal cortical involvement related to motor aspects of speech is somewhat unexpected. However, middle temporal gyrus neuronal activity during overt speaking was previously identified by us in an auditory word repetition task (Creutzfeldt et al. 1989). In the present study, this middle temporal gyrus involvement is a feature of memory encoding after activity related to identification has been removed, another indication of the expansion of temporal cortical neural networks when motor output is part of explicit memory-encoding processes.
Sustained excitation was recorded throughout lateral temporal cortex. We have previously modeled this as reflecting attentional processes specific to memory (Ojemann and Schoenfield-McNeill 1999). Sustained activity also characterized human temporal cortical neuronal activity associated with learning of word pair associations (Ojemann and Schoenfield-McNeill 1998). The majority of the peaks of the sustained excitation during encoding have very similar timing to that of only early excitation, as though the output of these neurons represented a simultaneous convergence of a sustained tonic input increasing overall level of activity and the more phasic early changes of perception and processing aspects of encoding. The Hebbian model of encoding involves potentiating effects of simultaneous presynaptic inputs (Hebb 1949). Our finding extends that concept to simultaneous network inputs to individual neurons, from networks involved with task-specific attentional activation and from networks involved in perception or semantic attributes perhaps item specific. Neurons with activity reflecting such converging excitatory inputs for at least one modality represented 10/98 of our sample of lateral temporal cortical neurons and were predominately recorded from dominant hemisphere.
Although different timing patterns clustered in different anatomical regions, this does not seem to reflect intrinsic properties of the neurons or hard-wired circuits in those locations but rather depends on the network into which a neuron has been transiently recruited for encoding of one modality. Over half of the polymodal neurons sampled changed temporal patterns between early, late, and/or sustained categories with different modalities. Even when changes for different modalities were confined to the early epoch, polymodality neurons had 150 ms or greater separation of the peak increase for different modalities. Indeed, in individual neurons recorded from several subjects, this difference in peaks between modalities was as much as 900 ms. There was no evidence that neuronal activity was consistently earlier for one modality.
The average firing rate of the temporal cortical neurons recorded in this study was relatively low, about 7 discharges/s. Low firing rates have also characterized previous recordings from human and monkey temporal cortex (Fuster and Jervey 1982; Ojemann et al. 1988; Creutzfeldt et al. 1989; Haglund et al. 1994; Schwartz et al. 1996; Ojemann and Schoenfield-McNeill 1999). Although the average peak firing rates were only in the 18–21 discharges/s range, this represents a 2.5–3× increase in activity over baseline; moreover, in some neurons, the changes were larger (4.6× for neuron illustrated in Fig. 2). Parameters for models of neural networks related to human memory encoding should be based on these low firing rates and moderate dynamic range.
Relative inhibition of activity during encoding was more likely in recordings from the superior and middle thirds of the middle temporal gyrus of dominant hemisphere, effects that have not been previously documented at the neuronal level. Sustained inhibition was related to this entire area and inhibition during the late epoch only to the middle third. There is an additional evidence for the role of temporal lobe inhibition in verbal processes. Inhibition during overt speech characterized many of the changes in middle temporal gyrus activity recorded during auditory word repetition (Creutzfeldt et al. 1989). Inhibition was the feature lateralized to dominant temporal cortex in our earlier study of neuronal activity related to naming (Schwartz et al. 1996). Lateral temporal cortical networks for identification with or without encoding have similar neurophysiological features, but the network for identification with encoding is more extensive including neurons not in the network for encoding alone. Interestingly, inhibition also characterized human hippocampal neuronal activity during paired associate learning (Cameron et al. 2001). Both middle temporal gyrus and medial temporal sites of inhibition are in one of the regions where task-independent decreases in functional imaging activity have been identified (Gusnard and Raichle 2001). However, the inferior third of middle temporal gyrus, where we found no evidence of inhibition, is also part of that region. In contrast to the peaks of excitation, maximum inhibition values were not clustered at particular times, with one exception. The majority of late epoch inhibition maximums were 250–300 ms before the initiation of the overt output, possibly also reflecting a premotor role. Whether a neuron responds with increased or decreased activity with memory encoding seems to be a hard-wired property that did not change with different modalities. It remains to be determined if middle temporal gyrus inhibitory activity during encoding is part of an inhibitory surround of excitation for sustained changes in superior temporal gyrus and inferior third of middle gyrus and for late epoch, superior third of middle temporal gyrus. Alternatively, the relation of temporal cortex to subcortical structures may be similar to that of motor cortex to anterior horn cells, where subcortical structures with high levels of activity are modulated by relative changes in cortical activity.
Although obtained from patients with epilepsy, these recordings were from resected cortex that did not show epileptiform EEG activity and from a random sample of neurons that did not show epileptiform bursting activity (Calvin et al. 1973). Moreover, it seems unlikely that the patients' epilepsy would alter the distribution of neurons with different timing and excitation or inhibition patterns associated with the instruction to retain an item in recent explicit memory. These recordings suggest that superior temporal gyrus neurons are part of a perceptual and early processing network activated with memory encoding, with some of this activity occurring very early, at or not long after perceptual information reaches cortex. Middle gyrus was related to overt output during memory encoding and the superior two-thirds of it to inhibition. Individual neurons in dominant hemisphere were often polymodal but with shifting roles in the networks for the different modalities. A subset of lateral temporal neurons had memory-encoding activity reflecting convergence of inputs from attentional and more specialized neural networks.
National Institutes of Health (NS 36527 and EB 2663); Pew–McDonnell Cognitive Neuroscience grant and a grant from Shirley and Herb Bridge.
Ms L. Zamora, Mr E. Lettich, and Dr N. Oskin assisted in the recordings and analysis. Dr S. Hakimian provided advice on the manuscript. Conflict of Interest: None declared.