Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Neuropsychologia 39 (2001) 556– 573 www.elsevier.com/locate/neuropsychologia Differential amygdala activation during emotional decision and recognition memory tasks using unpleasant words: an fMRI study Matthias H. Tabert a,b,*, Joan C. Borod a,b, Cheuk Y. Tang c,d,e, Gudrun Lange f, Tsechung C. Wei c, Ray Johnson b, Annette O. Nusbaum d, Monte S. Buchsbaum c b a Department of Neurology, Mount Sinai Medical Center, New York, NY, USA Department of Psychology, Queens College and The Graduate Center, City Uni6ersity of New York (CUNY), New York, NY, USA c Department of Psychiatry, Mount Sinai Medical Center, New York, NY, USA d Department of Radiology, Mount Sinai Medical Center, New York, NY, USA e Department of Radiological Sciences, UC Ir6ine, Ir6ine, CA, USA f Departments of Psychiatry and Radiology, UMDNJ – New Jersey Medical School, Newark, NJ, USA Received 14 October 1999; received in revised form 26 October 2000; accepted 17 November 2000 Abstract This study used fMRI to examine the response of the amygdala in the evaluation and short-term recognition memory of unpleasant vs. neutral words in nine right-handed healthy adult women. To establish specificity of the amygdala response, we examined the fMRI BOLD signal in one control region (visual cortex). Alternating blocks of unpleasant and neutral trials were presented. During the emotional decision task, subjects viewed sets of three unpleasant or three neutral words while selecting the most unpleasant or neutral word, respectively. During the memory task, subjects identified words that were presented during the emotional decision task (0.50 probability). Images were detrended, filtered, and coregistered to standard brain coordinates. The Talairach coordinates for the center of the amygdala were chosen before analysis. The BOLD signal at this location in the right hemisphere revealed a greater amplitude signal for the unpleasant relative to the neutral words during the emotional decision but not the memory task, confirmed by Time Course× Word Condition ANOVAs. These results are consistent with the memory modulatory view of amygdala function, which suggests that the amygdala facilitates long-term, but not short-term, memory consolidation of emotionally significant material. The control area showed only an effect for Time Course for both the emotional decision and memory tasks, indicating the specificity of the amygdala response to the evaluation of unpleasant words. Moreover, the right-sided amygdala activation during the unpleasant word condition was strongly correlated with the BOLD response in the occipital cortex. These findings corroborate those by other researchers that the amygdala can modulate early processing of visual information in the occipital cortex. Finally, an increase in subject’s state anxiety (evaluated by questionnaire) while in the scanner correlated with amygdala activation under some conditions. © 2001 Elsevier Science Ltd. All rights reserved. Keywords: fMRI; Emotion; State anxiety; Amygdala; Occipital cortex; Unpleasant words; Neutral words; Emotional decision; Emotional memory 1. Introduction Studies of patients with bilateral amygdala lesions have provided evidence that the human amygdala plays a role in the evaluation of emotional facial expressions, particularly those related to fear and anger [6– * Corresponding author. Address: New York State Psychiatric Institute, 722 West 168th Street, Unit 126, Kolb Annex, Rm 417, New York, NY 10032, USA. Tel.: + 1-212-543-5046; fax: + 1-212369-8764. E-mail address: [email protected] (M.H. Tabert). 8,27,35,90,91]. Human bilateral lesion studies have also implicated the amygdala in the recognition of non-verbal threat-related sounds (e.g., screams and growling) [80] and in the evaluation of words [80] and sentences denoting negative emotions [4]. These findings are consistent with a large body of animal literature that implicates the amygdala in the evaluation of cues that predict danger to the organism [10,60]. These data also suggest that bilateral amygdala lesions in humans can result in a modality-independent impairment in the recognition of threat-related emotional expressions. Moreover, since the amygdala receives inputs from 0028-3932/01/$ - see front matter © 2001 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 8 - 3 9 3 2 ( 0 0 ) 0 0 1 5 7 - 3 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 multiple sensory modalities [12] and is capable of multimodal processing [11], these data argue that the amygdala is part of an underlying processing mechanism geared to interpret emotional signals of threat or danger regardless of their source. It should be noted, however, that not all patients with bilateral amygdala damage demonstrate selective emotional processing impairments. A number of studies have demonstrated that bilateral amygdala damage does not necessarily impair the ability to evaluate facial [9,48,49,51] or prosodic [5,13] emotional expressions. Moreover, amygdala lesions do not appear to block normal autonomic [19,84] or self-reported reactions [2,31,49] to salient emotional stimuli. While reasons for this discrepancy in the bilateral amygdala lesion literature are not fully understood (for possible explanations, see Refs. [1,5,6,27,48,51]), there is overall convincing evidence from animal and human studies that the amygdala does play a role in selectively processing emotional, particularly threatening, information [1,9,11,86]. PET and fMRI studies investigating the perception and evaluation of prototypical emotional facial expressions (e.g., fear, anger, disgust, and happiness) in normal individuals have provided further evidence that the amygdala plays a role in the evaluation of emotional information. In a PET-O15 study, Morris and colleagues [68] observed increasing amygdala activation as a function of the intensity of fearful facial expressions. Interestingly, in a subsequent analysis of these data, Morris et al. [67] found that amygdala activation in response to fearful expressions was strongly correlated with activity in the occipital cortex, suggesting that the amygdala can modulate early processing of visual information. Using fMRI, Breiter et al. [26] also observed amygdala activation to fearful facial expressions. In the studies of Morris et al. [67] and Breiter et al. [26], subjects passively viewed facial expressions or made judgments as to the gender of the posers. Whalen et al. [88] presented fearful faces to subjects below the threshold of conscious awareness, using a backward masking procedure, and found that the amygdala revealed a significant increase in activation to masked fearful expressions. Phillips and coworkers [73,74] and Baird et al. [17] have also reported amygdala activation to facial expressions of fear. Imaging studies have investigated brain activation in response to emotionally intoned vocal expressions and to the emotional content of negatively valenced words. Phillips et al. [73] reported right amygdala activation in response to vocal expressions of fear (e.g., screams) in normal subjects. Kiehl et al. [55] reported right amygdala activation in response to viewing, rehearsing, and recalling negatively valenced emotional words. These studies provide further evidence that the amygdala is involved in processing negatively valenced stimuli, regardless of presentation modality. 557 A large body of behavioral, pharmacological, lesion, and imaging data also suggests that the amygdala plays an important role in memory processes related to emotion [1,72]. Normal subjects generally show superior memory for emotionally arousing stimuli relative to emotionally neutral stimuli (for review, see Ref. [44]). Lesion and functional neuroimaging findings have illuminated the importance of the amygdala in facilitating the acquisition of emotional memories (reviewed by Phelps and Anderson [72]). Adolphs et al. [2], Babinsky et al. [16], Cahill et al. [31] and Markowitsch et al. [64] have all reported selective long-term memory impairment for verbal and non-verbal emotional materials following bilateral amygdala lesions. Cahill et al. [32] (using FDG-PET) and Hamann et al. [50] (using PET-O15) have also found that increases in amygdala activity in subjects viewing emotional and neutral stimuli are strongly correlated with performance on longterm but not short-term free recall and recognition memory tasks for emotionally arousing relative to neutral information. Together, these findings support the memory modulatory theoretical framework of amygdala function, developed by McGaugh and colleagues [30,33,65,66] in experimental animal research. This view suggests that when the human amygdala becomes active in the presence of emotionally arousing stimuli (e.g., during encoding), it weighs conscious memory for the triggering stimuli in proportion to their salience by influencing long-term memory storage and consolidation via interactions with neurotransmitter systems, particularly the adrenergic system, and via anatomical connections to the hippocampus, striatum, and other brain regions [66]. An important aspect of this view is that the amygdala’s role in declarative memory is unrelated to memory for non-emotional material. Moreover, with respect to emotional material, this view emphasizes that the amygdala’s involvement is time dependent in that its memory-enhancing effects only become apparent after enough time (i.e., several hours to days) has passed to allow for memory consolidation in remote brain regions to occur [30,65,66]. The amygdala itself is, therefore, not thought to be the site where actual emotional memories are stored and is not thought to participate directly in the retrieval or recall of emotional information [70]. In this vein, a recent study by Bianchin et al. [20] demonstrated that the administration of several drugs into the rat amygdala at the time of training in a one-trial, step-down, inhibitory avoidance paradigm had no effect on either working memory (tested at 3 s post-training) or short-term memory (tested at 1.5 h post-training). However, all drugs had strong modulatory effects on long-term memory (tested at 24 h post-training), some enhancing (e.g., norepinephrine) and others impairing performance. 558 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 In summary, neuroimaging studies corroborate human bilateral amygdala lesion and animal studies, implicating the amygdala in the processing of emotional stimuli. Further, the amygdala appears to have a special role in the perception of emotional facial expressions. Facial expressions of anger and fear are especially salient in that they provide potent, possibly hardwired cues to the presence and source of personal danger [39,86]. Also, there is evidence from both lesion and imaging data that vocal expressions of fear [73,80] and the evaluation of emotionally arousing negative lexical stimuli (i.e., words and sentences) [4,55,80] involve the amygdala. Finally, there is evidence that amygdala activity while viewing (encoding) stimuli is correlated with subsequently enhanced long-term recall of emotional as compared to neutral information [32,50]. The purpose of the current study was to further examine the extent to which the amygdala is recruited in response to the evaluation, or presumably encoding, and the immediate recall (via recognition memory) of highly unpleasant and neutral words. In the current study, two separate fMRI scans were conducted. During the first scan, subjects were required to explicitly evaluate the relative emotional significance of highly unpleasant vs. neutral word sets (i.e., degree to which the words were perceived as threatening). Subjects were instructed to base their evaluations of the word sets on their personal experience with the concepts and connotations conveyed by the words. During the second scan, which immediately followed the first scan, memory for the words presented during the first scan was assessed. Together, these scans were used to examine the differential involvement of the amygdala during the explicit evaluation vs. the short-term recognition of highly unpleasant as well as neutral words. We hypothesized that the amygdala would show greater time-locked activation in response to the evaluation of highly unpleasant than neutral words [55]. Because normal subjects generally show superior immediate and long-term memory for emotional as compared to neutral lexical stimuli (for review, see Ref. [44]), we hypothesized that subjects would recall more emotional than neutral words immediately following the first scan (i.e., evaluation and encoding). However, on the basis of the memory modulatory view of amygdala function (see above), we predicted that this shortterm memory enhancement would not be related to amygdala activity. To establish the specificity of the amygdala response to unpleasant words, we examined the BOLD signal waveform in one control region, the occipital cortex (primary and secondary visual areas). A general response to stimuli (unpleasant and neutral) was expected in this control region for both tasks (emotional decision and recognition memory) since the visual cortex re- sponds selectively to visually-presented stimuli [68]. Moreover, a recent study by Morris et al. [67] found that while amygdala and occipital activations to fearful expressions were positively correlated (r= 0.62), amygdala and occipital activations to happy expressions were inversely correlated (r= − 0.55). These findings suggest that amygdala activation in response to emotional stimuli modulates the early processing of visually presented information in the occipital cortex in a category-specific way. Hence, in the current study, we further hypothesized that the response in the occipital cortex to words in general (i.e., unpleasant and neutral) would be positively correlated to increases in amygdala activation during the unpleasant word condition. A further goal of this study was to investigate the influence of anxiety created by the scanning procedure on amygdala activation. Based on our own pilot work and reports from previous studies [46,47,78], we predicted a significant increase in self-reported state anxiety [81] from subjects’ scores immediately before entering the scanner to scores reflecting their anxiety level during the scanning session (measured immediately after the scan). We also conducted exploratory correlational analyses to assess the relationship between changes in state anxiety as a result of the scanning procedure and amygdala activation. Finally, the a priori focus of this study was on the amygdala. Hence, only results pertaining to this region of interest and one control area (occipital cortex) are discussed at length. To further establish the specificity of amygdala findings, however, whole brain activation is presented on an exploratory basis. 2. Methods 2.1. Subjects Nine women (mean age, 28.2 yr; S.D., 4.7 yr; range, 23–38 yr) participated in this study; all were righthanded and native English-speakers and held a bachelor’s or doctoral degree (Mill Hill vocabulary test [76]: mean verbal IQ estimate [71], 111; S.D., 7.5). Handedness was determined using the Coren et al. [37] inventory. Subjects had no history of psychiatric, neurological, or other major medical illness, and had never been treated with psychotropic medication. Written informed consent was obtained from all subjects, and each subject was paid for participation. To eliminate sex differences and improve statistical power, only females were scanned in the current study. 2.2. Procedures 2.2.1. Measure of state anxiety To assess anxiety related to the scanning procedure, M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 we measured state anxiety with the State Trait Anxiety Inventory (STAI [81]), both immediately before and immediately after scanning. During the second administration, subjects were instructed to complete the STAI in terms of their experience inside the scanner. 2.2.2. Emotional decision scan The words used in this study came from the database of Toglia and Battig [83] (Table 1) which includes 2854 words that have been rated on concreteness [from 1 (low concreteness) to 7 (high concreteness)], imagery [from 1 (low imagery) to 7 (high imagery)], and pleasantness [from 1 (unpleasant) to 7 (pleasant), with 4 as the neutral midpoint]. For the current study, 30 highly unpleasant words and 30 neutral words (mean pleasantness ratings= 2.1 vs. 4.0) that did not differ significantly on word length (mean word length= 5.4 vs. 5.7 letters, P=0.509), word frequency [43] (mean frequency of occurrence= 62.1 vs. 61.8 per million, P= 0.989), concreteness (mean concreteness ratings= 4.4 vs. 4.7; P= 0.289), and imagery (mean imagery ratings=5.0 vs. 4.7; P = 0.152) were selected. Although words were not specifically matched for part of speech, Table 1 Word stimuli used in the current study, selected from Toglia and Battig [83] Neutral Unpleasant following step referee rows academy cable size meant therefore loop chart pots glasses empire tray differences sequel doorman suit age cents garment boot angle mile rapid wind tradition root chestnut cancer murder slavery morgue greedy kill gun polio measles misery war cruel hate crime poor failing pollution dead bomb prejudice ugly death agony coffin tragedy sewer trouble hell rape vulgar 559 both lists contained nouns, verbs, adjectives, and adverbs. The selected words were used to generate unpleasant and neutral word sets, where each word set comprised a unique combination of three words. Word sets were presented in eight alternating unpleasant and neutral blocks (4 cycles; Fig. 1(a) and (b)). A 2- to 3-min break occurred halfway through the scan to allow acquired data to be transferred to a remote computer workstation for storage and analysis. Each block began with a 2-s cue indicating whether the block consisted of unpleasant or neutral trials (Fig. 1b). Five word set trials were presented during each block, each appearing for 4 s with a 2-s inter-stimulus interval (ISI). Thus, the BOLD response was recorded to four blocks of unpleasant words and to four blocks of neutral words. During each ISI, a centrally placed fixation-cross replaced the word set. A 20-s resting baseline preceded the first block of trials, and a 24-s resting period separated all subsequent blocks of trials, during which subjects also viewed a central fixation cross. During each trial, a word set was projected to the center of the subject’s field of view via an SVGA computer-controlled projection system that presented stimuli to a rear-projection screen located at the entrance of the magnet bore. Subjects viewed stimuli projected to the screen via a 1.5× 3 in.2 mirror attached to the head coil and positioned : 6 in. from and directly above the subject’s eyes. Each word appeared twice over the course of the scan (i.e., first presentation occurred during the first two blocks, and the second presentation during the last two blocks of each stimulus type) within unique word sets. Word sets were presented in a fixed pseudo-randomized order across the blocks with the condition that a word did not reappear until all words had been presented once. The position of a word within a word set was counterbalanced across presentations of that word. The sequence in which blocks of unpleasant and neutral trials were presented was counterbalanced across subjects by reversing the trial order for four of the nine subjects. Immediately before scanning began, subjects were given 10 practice trials (5 unpleasant and 5 neutral). The negative and neutral words presented during practice did not overlap with the experimental words and were selected from the list of Brown and Ure [28], which includes 650 words that were also rated for pleasantness using a 7-point scale. During unpleasant blocks, subjects were instructed to select the most unpleasant or most threatening word from the three negatively valenced words (word set) presented on each trial. Subjects were instructed to base their decision on their personal knowledge of and experience with the concepts and connotations conveyed by the words. Similarly, during neutral blocks, subjects 560 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 Fig. 1. (a) Overview of blocked stimulus presentation paradigm for the Emotional Decision Scan. Eight alternating blocks of Neutral (N) and Unpleasant (U) trials were separated by Rest Periods ( + ). Total scan time was 488 s (8 min and 8 s), yielding 244 images of 14 axial slices (3416 slices). Each slice contains a maximum of 64 × 64 pixels, where each pixel is 3.59 mm2 (FOV= 23 cm2). (b) Summary of the stimulus presentation parameters within a block of unpleasant and neutral trials for the Emotional Decision Task (one complete cycle of unpleasant and neutral trials). (i) Block of unpleasant trials and preceding baseline. (ii) Block of neutral trials preceded by a rest period. Each block began with a cue indicating ‘‘unpleasant’’ or ‘‘neutral.’’ Subjects selected a word judged to be the most unpleasant or neutral for each word set by pressing one of the three buttons. M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 561 Fig. 2. (a) Overview of blocked stimulus presentation paradigm for the Recognition Memory Scan. Twelve alternating blocks of Neutral (N) and Unpleasant (U) trials were separated by Rest Periods ( +). Total scan time was 520 s (8 min 36 s), yielding 260 images of 14 axial slices (3640 slices). Each slice contains a maximum of 64 × 64 pixels where each pixel is 3.59 mm2 (FOV= 23 cm2). (b) Summary of the stimulus presentation parameters within a block of unpleasant and neutral trials for the Recognition Memory Task (one complete cycle of unpleasant and neutral trials). (i) Block of unpleasant trials and preceding baseline. (ii) Block of neutral trials preceded by a rest period. The 30 unpleasant and 30 neutral words presented during scan 1 (targets) were randomly mixed with 30 foils of each word type (0.5 probability of target appearing). Subjects selected a word judged to be the most unpleasant or neutral for each word set by pressing one of the three buttons. selected the word they deemed to be the most neutral (i.e., non-emotional or non-threatening) from each word set. Subjects indicated their response by pressing one of the three buttons on a response pad. 2.2.3. Recognition memory scan Immediately following the first scan (described above), a second scan was performed to assess recognition memory for the words presented (targets) during the emotional decision task. Sixty additional new words were selected from Toglia and Battig [83] to serve as foils or distractors in the memory scan (30 highly unpleasant and 30 neutral). Given the limited number of words available, particularly for the unpleasant category, the foils were matched to the targets on word length only. Together, 120 words (60 previously seen and 60 not previously seen) were presented one at a time in the center of the subject’s field of view in 12 alternating unpleasant and neutral blocks (6 cycles; Fig. 2(a)). Each block consisted of a fixed random sequence of 10 words, with a 0.50 probability of being a previously seen word (Fig. 2(b)). Each word was presented 562 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 for 2 s and was immediately followed by the next word. A 20-s resting period preceded the first block of trials and separated successive blocks, during which time subjects viewed a central fixation cross. The sequence in which blocks of unpleasant and neutral trials was presented was counterbalanced across subjects by reversing the trial order for four of the nine subjects. Subjects were instructed to press a key with their index finger whenever they recognized a word from the previous scan. This method produces symmetrical brain images around the midline so that hemispheric contrasts can be examined. The fMRI data of consecutive images (224 for Scan 1 and 260 for Scan 2) were arranged as a pixel time-series with each pixel in the image having entries representing the 2-s time points. The anatomical center and extent of the amygdala region were obtained a priori from Talairach and Tournoux [82] atlas coordinates (center: 23, −6, −12; with : 10-mm diameter corresponding to 3 pixels; 3.6 mm2/pixel). 2.3. Image acquisition and post-processing 2.4. Statistical analyses An initial T1-weighted sagittal localizer (spoiled gradient recall acquisition in a steady state; SPGR) sequence was acquired using a 1.5 T GE Horizon LX scanner (General Electric, Milwaukee, WI). Subsequently, 14 axial spin-echo T1-weighted images were acquired, encompassing the whole brain (TE/TR = 18/ 600 ms, FOV =23 mm2, slice thickness= 5 mm, skip = 2.5). Echoplanar magnetic resonance brain images were acquired with a multi-slice 2D EPI sequence (64×64 matrix, FOV= 23 cm2, TE =40 ms, TR =2000 ms, thickness 5-mm, 2.5-mm gap). For the first (emotional decision) scan, 244 T2*-weighted images depicting BOLD contrast [69] were acquired over 8 min and 8 s. For the second (recognition memory) scan, 260 images were acquired. Each image consisted of 14 near-axial, non-contiguous slices, providing whole-brain coverage. Preprocessing of the raw fMRI signal for each scan involved the removal of the mean, detrending, and digital filtering to remove low and high frequencies outside the range of the primary blood flow response. The mean and linear trends were removed using a least squares line-fit from each continuous data set. A pixelby-pixel fast Fourier transform (FFT) was then performed, and all frequency components below 0.018 Hz and above 0.071 Hz were removed. These cut-off frequencies were selected to reduce signal contributions slower than 1 cycle in 40 s (greater than the block length) and faster than 1 cycle in 15 s (about 14 of the block length), based on the on – off cycle period of 56 s. A new filtered fMRI image dataset was obtained by an inverse FFT. The 14 reconstructed anatomical and functional images were visually inspected and matched to standard axial brain slices from the Talairach and Tournoux [82] atlas. The edges of the anatomical slices for each subject were traced and midline structures identified and marked using our own image-processing system program (multi-image processing-system; MIPS). The traced edges and identified landmarks were used for coregistration of each subject’s functional and anatomical data and to morph the images into a standardized stereotactic space derived from 70 normal adults [52]. Each pixel in the images was subjected to a two-way repeated-measures analysis of variance (ANOVA) involving Word Condition (unpleasant vs. neutral) and Time Course (28 images for Scan 1 and 20 images for Scan 2) for a 3-pixel diameter region of interest (ROI) centered on the left and right amygdala (Talairach coordinates: 23, − 6, − 12, chosen in advance from the atlas). A three-way Condition (unpleasant vs. neutral)× Time Course (28 images for Scan 1 and 20 images for Scan 2)× Hemisphere (right vs. left) ANOVA was conducted to assess hemispheric asymmetries. Habituation of the amygdalar response to the stimuli across blocks of trials was also examined [threeway Condition (2)× Time Course (28)× blocks (4 blocks of each word type) ANOVA]. It should be noted that without very high-resolution MRI templates for each subject, differentiation of the amygdala from the end of the hippocampus is difficult. However, coordinates from the Talairach atlas were used to identify activation in this study and thus allow for the replication of our findings by other imaging centers. To establish the specificity of our results, we also examined the BOLD response to stimuli in the occipital cortex. Since the exact location of this control area (occipital cortex) was not specified a priori, a resampling procedure was used to protect against Type I error [14,15,53]. A random starting point within the time-series, subjects, and word type conditions was selected. From the current sample, each word stimulus trial was then drawn until the set of two word type conditions for all nine subjects was completed. Each cluster containing contiguous pixels with P-values below the 0.05 threshold level was identified: emotional decision scan, F(27,216)= 1.54; recognition memory scan, F(19,152)=1.66. Within each cluster so identified, the sum across all pixels of the F-ratio height above the critical threshold level (1.54 for the emotional decision scan and 1.66 for the recognition memory scan) was calculated. For example, the volume for a cluster with two contiguous pixels with the F-values of 2.54 and 3.54 would be 1+ 2= 3. These pixel cluster volumes were tabulated across 5000 iterations (from M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 563 Table 2 Whole brain activation sites in Talairach space for the Condition×Time interaction during the emotional decision taska,b Condition×Time interaction Location x y z Volume F (a) Unpleasant\neutral Amygdala Middle temporal gyrus (BA 21) Frontal lobe Parahippocampal gyrus Temporal lobe Midbrain Transverse temporal gyrus (BA 42) Superior temporal gyrus (BA 39) Precentral gyrus (BA 6) Medial frontal gyrus (BA 9) Superior frontal gyrus (BA 10) Cuneus (BA 7) Precentral gyrus (BA 6) 26 63 −7 −24 −44 −4 29 54 58 −2 13 2 −50 −6 −0.5 2 −48 −31 −24 −31 −56 1 47 55 −67 −4 −12 −12 −12 −4 −4 −4 12 20 20 24 24 32 40 8.33 2.74 6.16 3.44 14.14 1.95 2.23 4.43 6.27 7.32 0.73 4.39 1.81 5.81 4.43 6.58 3.90 5.46 3.32 3.76 4.84 4.57 6.54 2.98 4.80 4.57 (b) Neutral\unpleasant Parahippocampal gyrus (BA 36) Fusiform gyrus (BA 37) Superior temporal gyrus (BA 38) Inferior frontal gyrus (BA 47)** Parahippocampal gyrus Internal capsule Claustrum Lingual gyrus Claustrum Temporal lobe Inferior frontal gyrus (BA 44) Inferior frontal gyrus (BA 44) Frontal lobe Middle frontal gyrus (BA 46) Middle frontal gyrus (BA 8) 35 −40 60 −49 21 −24 −33 23 −27 −48 −44 58 −31 −46 35 −33 −56 11 23 −45 −22 0 −79 20 −63 13 12 22 25 38 −12 −12 −12 −12 −4 −4 −4 4 4 12 12 20 24 24 40 4.78 1.93 2.46 3.11 6.12 7.80 4.15 6.47 2.95 6.85 0.25 3.96 4.96 1.54 4.64 5.49 3.02 3.86 4.63 4.29 6.61 3.73 4.47 3.41 4.63 2.55 3.68 3.80 3.56 4.97 a Sites where the BOLD response during stimulus presentation was greater to the unpleasant than neutral Word Condition. Sites where the BOLD response during stimulus presentation was greater to the neutral than unpleasant Word Condition. Significance thresholds were set at PB0.001 for the maximum pixel and a minimum of at least 4 contiguous pixels. The volume index (used to enter the resampling table for the resampling test) is the number of contiguous pixels in the patch multiplied by the difference between the average F value across the patch and the PB0.001 F value (see text). b nine independent subject and time-interval starting point samples), and cluster volumes in the fifth percentile or larger anywhere on the brain slice (Talairach coordinate: z= + 12) were considered statistically significant. Note that this step also adjusts for any inflation of the degrees of freedom in repeated-measures ANOVA designs because the cluster volume appearing less than one time in 20 is identified from the empirical table. The resampling method avoids potentially unwarranted assumptions about image smoothness and other data features needed to estimate the F-level of significance with a purely theoretical approach. Resampling F-tables can be relatively easily calculated via the same computer code used to initially organize the data and to conduct the pixel-by-pixel F-tests. The first stage, randomizing the trials and time starting points, can be conveniently developed from programs which select the trial data points (in our case, 2-s epochs) from the longer series of blocks with only minor alterations. The second stage, the ANOVA mapping, supplies the clusters to the third stage (i.e., cluster identification and volume computation). A fourth sim- ple operation forms the table of cluster volumes across the 5000 iterations. To further establish the specificity of the amygdala finding, we present in tablular form the whole brain activation as it relates to the Word Condition by Time Course interaction (Tables 2 and 3). For these whole brain exploratory analyses, we set significance thresholds at P B 0.001 for the maximum pixel within the cluster and only listed clusters with at least 4 contiguous pixels. Although a significance threshold of PB 0.001 and the 4-pixel cluster-size constraint are thought to be adequate corrections for multiple comparisons [42], our resampling criterion was more conservative. None of the activations listed in Tables 2 and 3 met the resampling criterion for statistical significance. Exploratory analyses examined the correlation between amygdala activation during the emotional decision task and measures of state anxiety. Similarly, the correlations between performance on the recognition memory task and amygdala activation during the evaluation (Scan 1) and recall (Scan 2) tasks were assessed. 564 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 Table 3 Whole brain activation sites for the Condition×Time interaction during the recognition memory taska,b Condition×Time interaction Location x y z Volume F (a) Unpleasant\neutral Midbrain Temporal lobe Precentral gyrus (BA 6) Middle frontal gyrus (BA 8) Superior frontal gyrus 2 48 −53 −24 −10 −29 −40 0 23 16 −4 −4 40 45 55 6.15 2.45 6.73 1.24 3.51 5.52 4.24 5.83 3.57 4.89 (b) Neutral\unpleasant Superior temporal gyrus (BA 38) Caudate Frontal lobe Precuneus Medial frontal gyrus (BA 8) 35 13 31 16 3 5 5 36 −60 51 −12 20 24 40 40 3.06 1.40 2.01 5.46 4.08 4.48 2.78 3.91 6.44 5.63 a Sites where the BOLD response during stimulus presentation was greater to the unpleasant than neutral Word Condition. Sites where the BOLD response during stimulus presentation was greater to the neutral than unpleasant Word Condition. Significance thresholds were set at PB0.001 for the maximum pixel and a minimum of at least four contiguous pixels. The volume index (used to enter the resampling table for the resampling test) is the number of contiguous pixels in the patch multiplied by the difference between the average F value across the patch and the PB0.001 F value (see text) b Finally, in order to replicate findings by Morris et al. [67] (see above), correlations between amygdala and occipital (control) activation were computed. 3. Results 3.1. Beha6ioral data During the emotional decision task scan, all subjects made an appropriate response (chose one word from each word set) on 100% of the trials. This demonstrates that subjects attended to the stimuli and responded as instructed for the entire duration of the scan.1 During the recall task scan, subjects recognized more emotionally negative words as compared to neutral words. Overall, 91% of the unpleasant and 82% of the neutral targets were correctly identified (paired t(8) = 3.59, P= 0.007), demonstrating a significant memory enhancement for the unpleasant words. 1 Given that the task instructions for the emotional decision task required subjects to make a subjective choice that was based on their own personal experiences and knowledge of the words involved in each word set, we did not include the behavioral data for the first scan here. As a manipulation check, we questioned subjects immediately following the experiment about their experience in the scanner. Debriefing revealed that all subjects found the unpleasant word sets to be highly unpleasant, particularly when compared to the neutral word sets. This agreed with findings from our pilot work in which a number of individuals were asked to perform and assess the validity of the task. In the current study, subjects were not asked to rate each word separately in terms of emotional valence. Future studies using a similar lexical paradigm may find such rating data useful as an additional manipulation check in order to assess the degree to which subjects agree with the normative data used to select the lexical stimuli. 3.2. Emotional decision scan 3.2.1. Amygdala Acti6ation A two-way ANOVA (Word Condition×Time Course) for the 3-pixel diameter ROI chosen a priori from the Talairach atlas (ROI: left and right amygdala) revealed a significant Word Condition× Time Course interaction (F(27,216)= 5.95, PB0.0001) for the right amygdala, demonstrating greater time-locked activation for unpleasant than neutral words. Pixel-by-pixel ANOVAs surrounding the center of the right amygdala also revealed a significant Word Condition×Time Course interaction (F(27,216)=5.81, PB 0.001) for a total of 9 contiguous pixels (Fig. 3). The pixel with the maximal response was located approximately at the center of the amygdala (Talairach coordinates: 26, − 6, −12) and was highly significant for the Word Condition× Time Course interaction (Fig. 3). A main effect of Time Course was also observed at this pixel (Fig. 3). Activation in the left amygdala did not reach significance (F(27,216)= 1.54, P=0.05, for the main effect of Time Course; F(27,216)=1.40, P=0.10, for the Word Condition× Time Course interaction). A three-way ANOVA (Word Condition×Time Course× Hemisphere) revealed a significant Time Course× Hemisphere interaction (F(27,216)=2.84, PB 0.0001) for the ROI analysis. However, the Word Condition× Time Course ×Hemisphere interaction was not significant. To assess habituation effects of the BOLD signal waveform in the right amygdala across successive blocks of trials, a two-way ANOVA (Block×Time Course) was conducted for the emotional condition at the pixel of maximum activation. No significant effects were observed with the blocking factor. These data suggest a sustained amygdala response across the blocks of trials. M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 3.2.2. Acti6ation in control region To establish the specificity of the amygdala response to unpleasant words, we also examined the BOLD signal waveform in a control region, the bilateral occipital cortex (primary and secondary visual areas). As predicted, this area showed significant task activation but no different pattern with neutral and unpleasant words. Pixel-by-pixel two-way ANOVAs (Word Condition × Time Course) in visual cortex (Brodmann areas 17, 18, and 19; Fig. 4) revealed widespread activation confirmed for Time Course, but without the Word Condition ×Time Course interaction observed for the amygdala. A three-way ANOVA (Word Condition× Time Course ×Hemisphere) did not reveal any interactions with Hemisphere, suggesting bilateral activation of visual cortex to both unpleasant and neutral words. This bilateral time-locked activation in primary and secondary visual cortices to both unpleasant and neu- 565 tral words is consistent with the well-known role of these areas in processing visually presented information. 3.2.3. Correlation between amygdala and occipital acti6ation The peak BOLD response from the pixel of maximal activation (Talairach coordinates: 26, − 6, − 12) in the right amygdala during Scan 1 was correlated with the peak BOLD response from the pixel of maximal activation (Talairach coordinates: − 21, − 85, 12) in the occipital cortex. The peak BOLD response for the neutral and unpleasant blocks was derived by subtracting the maximal pixel intensity value during the stimulus-on period (averaged across the four blocks of each trial type) from the average value of the baseline preceding the stimulus-on period (Figs. 3 and 4). The correlation analysis revealed that the peak BOLD re- Fig. 3. Emotional Decision Task: (Top left) The Condition by Time Course interaction (F(27,216) = 5.25; PB 0.0001) in the pixel with maximum activation (26, − 6, − 12) within a patch of 9 contiguous pixels surrounding the center of the right amygdala. (Bottom left) The main effect of Time Course (F(27,216) =2.29, P= 0.0006) for the pixel with the maximal response (26, − 6, − 12) within the same patch of 9 contiguous pixels. (Right) This 9-pixel patch was the largest area/volume of activation (maximum patch) on this axial slice. Other areas that met the resampling criteria for significance were omitted for clarity. 566 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 Fig. 4. Emotional Decision Task: (Left) The main effect of Time Course (F(27,216) =13.57; P B0.000) for the pixel of maximum activation (Talairach coordinates: − 21, −85, 12). (Right) The patch of 138 contiguous pixels containing the maximum activation pixel (resampling volume=342, resampling PB 0.025) was located in visual cortex (Brodmann areas 17, 18, 19). This 138-pixel patch is the largest area/volume of activation (maximum patch) on this axial slice. Other areas that met the resampling criteria for significance were omitted for clarity. sponse in the right amygdala to the unpleasant words was highly correlated with the overall peak BOLD response (i.e., averaged across Word Condition) in the occipital cortex (r= 0.87, P =0.003). On the other hand, the peak BOLD response in the right amygdala to the neutral words did not correlate with the response in the occipital cortex (r = −0.04, P = 0.912; Fig. 5). Correlation analyses between the left amygdala and the occipital cortex were not performed since significant amygdala activation was not observed on the left side. 3.2.4. Correlation between measures of anxiety and amygdala acti6ation Raw scores for state anxiety [81] were determined. To assess the effect of anxiety due to the scanning procedure on the statistically significant BOLD response in the right amygdala, the difference score for the two time points of the state anxiety measure (during scanning as assessed retrospectively and before scanning) was correlated with the peak BOLD response from the pixel of maximal activation (Talairach coordinates: 26, − 6, −12). State anxiety ratings went from a mean score of 33.3 99.1 before scanning to 44.29 8.8 after scanning (paired t(8)= 2.69, P =0.027), demonstrating an overall significant increase in anxiety presumably due to the scanning procedure. Difference scores for each subject were also calculated (i.e., score during scan− pre-scan score). These scores ranged from −8 (relative decrease in anxiety during scan) to 24 (relative increase in anxiety during scan), with a mean of 8.99 9.9 and a median of 10. State anxiety difference scores (scan report−prescan report), reflecting the change in anxiety from the pre-scan period to the report about the actual scanning period, were significantly correlated with the peak BOLD response of the right amygdala to the neutral (r=0.70, P= 0.035) but not to the unpleasant word condition (r= 0.33, P= 0.380). Together, these findings suggest that state anxiety levels do increase as a result of the scanning procedure. However, while this increase in state anxiety was related to the amygdala response during the neutral word condition, it was not related to the amygdala response during the emotional word condition. 3.3. Recognition memory scan 3.3.1. Amygdala acti6ation Analyses similar to those described above for the emotional decision task were conducted to examine the amygdala response during the recognition memory test for words presented during the first scan. Pixel-by-pixel ANOVAs around the center of the right and left amygdala revealed no significant Word Condition×Time Course interaction or significant increases to word recognition in general. These findings suggest that, unlike the emotional decision task scan (see above), amyg- M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 dala activation was not elicited during the recognition of unpleasant or neutral words that were presented earlier. 3.3.2. Acti6ation in control region Activation in the occipital cortex, on the other hand, demonstrated a somewhat similar activation pattern to that observed in the emotional decision scan (Fig. 6). Pixel-by-pixel two-way ANOVAs (Time Course× Word Condition) in visual cortex (Brodmann areas 17, 18, 19) revealed widespread activation confirmed for Time Course, but not for the Time Course × Word Condition interaction. Also, a three-way ANOVA (Word Condition×Time Course ×Hemisphere) did 567 not reveal higher order interactions with Hemisphere, again, suggesting bilateral activation of visual cortex to visually presented information irrespective of unpleasantness. 3.3.3. Correlation between recognition memory and amygdala acti6ation A correlational analysis was conducted to assess the possibility that the enhanced behavioral performance on the recognition memory task for the unpleasant words was related to the amygdala activation observed during Scan 1. The peak BOLD response of the pixel of maximal activation (Talairach coordinates: 26, −6, − 12) in the right amygdala during the presentation of unpleasant word sets was not significantly correlated with the number of correct responses (Hits) on the memory test (unpleasant words: r= − 0.11, P=0.782; neutral words: r= 0.08, P= 0.840). Similarly, the peak BOLD response of the pixel of maximal activation (Talairach coordinates: 26, − 6, − 12) in the right amygdala during the presentation of neutral word sets was not significantly correlated with the number of correct responses (Hits) on the memory test (unpleasant words: r= −0.05, P= 0.899; neutral words: r=0.48, P= 0.196). These findings suggest that the differential enhancement in recognition memory for the unpleasant words, as described above (Section 3.1), is not related to the increase in amygdala activation observed during Scan 1. 4. Discussion Fig. 5. Scatterplots illustrating the correlation between the peak BOLD response in the pixel of maximum activation in the right amygdala (26, − 6, −12) and occipital cortex ( − 21, − 85, 12). (a) Amygdala response to unpleasant words (Wds) vs. the overall occipital response to words (i.e., unpleasant and neutral). (b) Amygdala response to neutral words vs. overall occipital response to words. Regression lines have been fitted to the data (slopes: + 0.989 and −0.038, respectively; correlation coefficients (r): + 0.866 and − 0.043, respectively). To date, lesion and imaging studies have demonstrated that the amygdala plays a role in both the evaluation of emotional information and in the modulation of emotional memories. The amygdala has also been implicated in the modulation of attentional systems related to the early processing of visually presented emotional information. Evidence for such roles has come largely from studies using emotional facial and vocal expressions. The current fMRI activation paradigm was designed to assess the extent to which the amygdala is recruited in response to the evaluation and short-term recognition of highly unpleasant words. A further goal of this study was to investigate the influence of anxiety created by the scanning procedure on amygdala activation. In summary, our results revealed a Time Course× Word Condition interaction in the right amygdala for the emotional decision but not the recognition memory task, demonstrating selective time-locked activation for the evaluation of unpleasant but not neutral words. In contrast, a main effect of Time Course was observed in the occipital cortex for both the emotional decision and recognition tasks, demonstrating a general time-locked 568 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 Fig. 6. Recognition Memory Task: (Left) The main effect of Time Course (F(19,152) =12.88; P B0.0001) for the pixel of maximum activation (Talairach coordinates: − 29, −81, 12). (Right) The patch of 103 contiguous pixels containing the maximum activation pixel (resampling volume=197, resampling PB 0.025) was located in visual cortex (Brodmann areas 17, 18, 19). This 103-pixel patch is the largest area/volume of activation (maximum patch) on this axial slice. Other areas that met the resampling criteria for significance were omitted for clarity. response to visually presented words irrespective of valence and task manipulation. Interestingly, the general response observed in the occipital cortex was strongly correlated with right amygdala activation to the unpleasant but not neutral words. Although subjects showed significantly enhanced short-term memory for the emotional as compared to neutral words, this memory advantage was not correlated with the observed increase in amygdala activation during encoding (Scan 1). Finally, the increase in state anxiety experienced by subjects while in the scanner was significantly correlated with amygdala activation during the neutral but not unpleasant Word Condition. A number of previous imaging studies presenting words [18,62,87] have not reported specific amygdala activation in response to the presentation of unpleasant words. However, consistent with the findings of Kiehl et al. [55], the present study found a robust BOLD signal in the right amygdala that was greater for unpleasant than neutral word sets. The discrepancy between the findings from the current study and those from previous studies, which did not observe amygdala activation in response to the presentation of unpleasant words, may be due to methodological differences. While previous studies employed more implicit processing tasks, such as passive viewing of emotional words or counting the number of negative vs. neutral words on a screen, in the present study, subjects were required to evaluate a group of highly unpleasant or neutral words in relation to their own personal experience (i.e., the degree to which the words were perceived as unpleasant or threatening). Careful analysis of the amygdala response in this study revealed a number of interesting results. First, consistent with the Kiehl et al. [55] study, amygdala activation to unpleasant words was statistically confirmed only in the right hemisphere. However, an analysis specifically testing for hemispheric asymmetries [39,40] (Time Course× Word Condition× Hemisphere ANOVA) did not reveal significant Word Condition× Hemisphere or Word Condition× Time Course × Hemisphere interactions for the 3-pixel diameter ROI centered on the amygdala. Asymmetries in the perception, expression, and experience of emotion have been well-documented in patients with focal unilateral cortical lesions [3,21 –23,45,54,79]. In general, lesion studies support the notion that the right cerebral hemisphere is specialized for the processing of emotional information (i.e., right hemisphere hypothesis) [24]. As the literature focusing specifically on the amygdala reveals (Section 1), however, selective emotional processing deficits have primarily been confirmed in patients with bilateral amygdala damage. Also, no clear consensus about the laterality of the amygdala response to emotional stimuli can yet be reached on the basis of imaging studies, as different imaging studies employing the same experimental conditions have reported left-sided, right-sided, or bilateral activation. Importantly, Davidson and Ir- M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 win [39,40] have pointed out that although many imaging studies have reported asymmetric changes associated with emotion, the data by-and-large have not been properly analyzed with specific right – left, laterality-focused contrasts. Although our findings are suggestive of right greater than left amygdala activation, they underscore the need to perform appropriate laterality tests before claiming hemispheric asymmetries. Higher resolution BOLD images, larger samples, and anatomical coregistration may be necessary to confirm lateralization effects. This is particularly true since the right and left amygdalae may not be centered on the same 5-mm thick slice. Second, unlike previous studies [26,29,56,75,88], the current study did not find habituation of the amygdala response to emotional stimuli over the course of the scan. Rather, a sustained amygdala response was observed across the blocks of trials to the unpleasant as compared with the neutral words. Methodological differences between the current and previous studies may account for this difference. For example, previous studies that have observed habituation of the amygdala response presented threat-related facial expressions [26,75,88] or aversive conditioned stimuli [29,56] for very brief durations (B 200 ms). In fact, Whalen et al. [88] presented stimuli below the threshold of conscious awareness (30 ms) using a backward masking procedure. Further, in these other studies, subjects were kept naı̈ve as to the emotional manipulations employed in the experiments (i.e., implicit processing) and were exposed to the same stimuli more than once over the course of the scan. In contrast, the current study exposed subjects to stimuli for extended periods of time (6 s). Further, subjects were also explicitly made aware of the word type manipulation used and were instructed to carefully evaluate the relative emotional significance of word sets in terms of their own knowledge and experience. Finally, each word set was composed of a unique combination of words to prevent habituation due to familiarity with the stimuli. The selective amygdala response to unpleasant words observed in the current study is consistent with a large body of animal and human literature suggesting that the amygdala plays a role during the evaluation of emotionally significant and/or biologically relevant cues, particularly those that predict possible threat to the organism. However, the exact nature of this role remains controversial [9]. While a number of studies have demonstrated selective emotional processing deficits following bilateral amygdala lesions in humans [6 – 8,27,36,90,91], others have not [5,9,13,48,59,51]. Moreover, bilateral lesions in humans generally do not prevent normal autonomic [19,84] and self-reported emotional reactions [2,31,49] to emotionally significant stimuli. Similarly, while studies have implicated the amygdala in the production of conditioned emotional 569 responses and as the location where conditioned memories are indelibly stored [60,61,63], other studies (for review of this literature, see Ref. [34]) have demonstrated that the amygdala is not necessary for fearbased autonomic responses [84] or for fear-conditioned learning [85]. The sustained response of the amygdala to the evaluation of unpleasant words in this study may, on the one hand, reflect retrieval of past emotional experiences for the purpose of appraising the emotional significance of current stimuli in the context of the emotional decision task (i.e., choosing the most unpleasant word from a set of three highly unpleasant words based on subjective experience) and guiding related behavioral responses. Such emotional memories may be stored in the amygdala per se or in widespread associative neural networks of the cortex that are reciprocally connected to the amygdala through the hippocampal system [38]. This interpretation is consistent with some imaging studies that have found right amygdala activation during retrieval of highly personal emotional memories [75] and during the retrieval of affect-laden autobiographical memories with personal emotional significance [41]. On the other hand, a recent PET study by Reiman et al. [77] found that although viewing emotionally arousing films activated the amydala, recall of previously experienced emotional events did not. Further, a number of imaging studies [32,50] have reported that amygdala activity while viewing emotional stimuli is strongly related to long-term memory performance for emotionally significant, as opposed to, neutral material. These findings are consistent with the memory modulatory framework [30,33,65] discussed above which suggests that amygdala activation in response to emotionally significant stimuli regulates the long-term storage of this information in other brain areas in a time-dependent manner. Hence, according to this view, the increase in amygdala activity observed here, while subjects evaluated the unpleasant vs. neutral word sets, may reflect the encoding and ultimate long-term storage of the lexical stimuli in general proportion to their emotional significance [30]. The lack of an amygdala response during the shortterm memory task used in this study (Scan 2) is also consistent with the memory modulatory view [20,30,32,50]. During the memory scan, which occurred immediately after the first (encoding) scan, subjects were simply instructed to rapidly identify any word that occurred in the previous scan. Interestingly, although subjects were able to more accurately recognize the emotional than non-emotional words, their enhanced performance on the emotional task was not related to an increase in amygdala activation. This finding is consistent with the memory modulatory view, which posits that the amygdala is not involved in memory retrieval processes but is involved in memory consolida- 570 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 tion. Moreover, similar to the Hamann et al. study [50], amygdala activation during encoding (Scan 1) was not correlated with short-term memory performance. This finding is again consistent with the memory modulatory view that predicts amygdala activity during encoding to be correlated with long-term but not short-term memory [20]. In the current study, however, we did not administer a test of long-term memory for the lexical stimuli. Of note, both Cahill et al. [32] and Hamann et al. [50] did find a significant correlation between amygdala activation at encoding and long-term memory performance (i.e., after 1 month). To establish the specificity of amygdala activation to the emotional condition, we also analyzed activation in a control region – the occipital cortex. For the emotional decision and recognition memory tasks, widespread bilateral activation was observed in the visual cortex for both unpleasant and neutral words. The time-locked activation in this region to both unpleasant and neutral words suggests that visually presented unpleasant and neutral words are processed in primary and secondary visual cortices independent of the behavioral task used. By contrast, amygdala activation was specific to the unpleasant word condition and seems to be more dependent on the behavioral task used, such that making emotional decisions about highly unpleasant words, and presumably encoding them into longterm memory, activates the right amygdala whereas identifying unpleasant words stored in short-term memory does not. In the current study, similar to the findings by Morris et al. [67], there was a strong correlation between the overall occipital response to words (i.e., the averaged response across unpleasant and neutral trials) and the amygdala response to the unpleasant words (Fig. 5). This finding is consistent with a number of studies [26,58,59,67,77,89] that provide support for the notion that the amygdala modulates early visual processing in the occipital cortex via reciprocal projections [12]. In this vein, Whalen [86] has recently proposed that the amygdala is part of a larger vigilance system that selectively responds to aversive stimuli that are ambiguous or pose an uncertain threat to the organism. When activated by the amygdala, this vigilance system potentiates neuronal responsiveness or cortical processing to gather additional information about the predictive biological significance of the stimulus. A final goal of the current study was to assess the effects of anxiety during the scanning procedure on amygdala activation. Not surprisingly, state anxiety scores increased significantly due to the scanning procedure. Moreover, this increase in anxiety (i.e., difference between anxiety level as measured before vs. during the scan) predicted the response of the right amygdala to neutral but not unpleasant words. This finding has several important implications for imaging studies. First, the fMRI scanning procedure can induce elevated levels of anxiety in non-claustrophobic, normal adult subjects [46,47,78]. Similar and possibly more severe stress reactions can be expected from clinical populations who, by definition, are more sensitive to environmental stressors. Second, it calls into question the assumption that the anxiety effects of scanning are similar across the experimental conditions of a particular study. In the present study, the amygdala response to the neutral words varied as a function of subjects’ anxiety levels, whereas the amygdala response to the unpleasant words was related to the presentation of emotional words and not to state-anxiety levels. Direct comparison of conditions that were differentially affected by extraneous factors, such as anxiety, may result in confounded results. Hence, it would seem prudent for imaging studies to pay close attention to the effect of factors such as anxiety when interpreting activation patterns in normal and patient populations. 5. Summary and future directions In summary, the current study has demonstrated that the amygdala is involved in the explicit evaluation of unpleasant words. Findings from our study and those from previous studies support the view that the amygdala is involved in the processing of emotional information independent of the modality of presentation. While the exact nature of the amygdala’s role in emotional processing remains controversial, findings from the current study fit best with the memory modulatory view of amygdala functioning. Although the emotional decision and short-term recognition memory tasks in this study elicited the same response from control areas (i.e., visual cortex), the latter task did not result in amygdala activation. Moreover, amygdala activity during encoding was not correlated with short-term memory performance. Together, these findings are consistent with the notion that the amygdala facilitates the acquisition of long-term declarative knowledge about emotionally significant material. Given that the current study did not include a long-term memory condition, we are unable, however, to directly assess the degree to which amygdala activity during encoding selectively modulates memory for highly unpleasant words at longer delays. To more fully assess the memory modulatory view with respect to lexical stimuli, future studies should include long-term memory recognition and free recall tasks. Future studies should also obtain physiological (e.g., galvanic skin response) and self-report measures of emotional arousal from subjects while they engage in similar lexical emotional decision tasks. Such manipulations would make it possible to assess the degree to which decisions about highly unpleasant words elicit emotional arousal and, furthermore, the degree to M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 which emotional arousal is related to amygdala activity during encoding and to the regulation of memory consolidation processes [57]. In addition, including highly positive words as a third word condition would allow one to directly assess brain lateralization hypotheses as a function of emotional valence [25,36,39,40]. Finally, the findings of this study corroborate findings by other research groups that the amygdala’s response to visually presented emotional information modulates early processing in the occipital cortex. The current findings also highlight the importance of evaluating future imaging data in terms of changes in state anxiety due to the scanning procedure. In the current study, such changes were shown to have differential effects in the emotional vs. neutral condition. Acknowledgements This study is based on a doctoral thesis conducted by the first author at Queens College and The Graduate Center of the City University of New York (CUNY) and in affiliation with the Departments of Neurology and Radiology and the Neuroscience Brain Imaging Laboratory at Mount Sinai Medical Center. This research was supported, in part, by NIMH Grant no. MH42172 and Professional Staff Congress – CUNY Research Award no. 69257-0029 to the second author at Queens College. We are grateful to Drs. Scott W. Atlas and Ronald Bloom for their useful advice during the initial stages of this project, to Karen Metroka and Adam Brickman for their assistance, and to the anonymous reviewers for their helpful suggestions. A portion of this study was presented at the International Conference on Functional Mapping of the Human Brain in Düsseldorf, Germany, in June of 1999. References [1] Adolphs R. The human amygdala and emotion. The Neuroscientist 1999;5:125– 37. [2] Adolphs R, Cahill L, Schul R, Babinsky R. Impaired declarative memory for emotional material following bilateral amygdala damage in humans. Learning and Memory 1997;4:291–300. [3] Adolphs R, Damasio H, Tranel D, Damasio AR. Cortical systems for the recognition of emotion in facial expressions. Journal of Neuroscience 1996;16:7678–87. [4] Adolphs R, Russell JA, Tranel D. A role for the human amygdala in recognizing emotional arousal. Psychological Science 1999;10:167– 71. [5] Adolphs R, Tranel D. Intact recognition of emotional prosody following amygdala damage. Neuropsychologia 1999;37:1285– 92. [6] Adolphs R, Tranel D, Damasio AR. The human amygdala in social judgement. Nature 1998;393:470–4. [7] Adolphs R, Tranel D, Damasio H, Damasio AR. Impaired recognition of emotion in facial expressions following bilateral damage to the human amygdala. Nature 1994;372:669–72. 571 [8] Adolphs R, Tranel D, Damasio H, Damasio AR. Fear and the human amygdala. Journal of Neuroscience 1995;15:5879–91. [9] Adolphs R, Tranel D, Hamann S, Young AW, Calder AJ, Phelps EA, et al. Recognition of facial emotion in nine individuals with bilateral amygdala damage. Neuropsychologia 1999;37:1111– 7. [10] Aggleton JP. The functional effects of amygdala lesions in humans: a comparison with findings from monkeys. In: Aggleton JP, editor. The amygdala: neurobiological aspects of emotion, memory, and mental dysfunction. New York: Wiley/Liss, 1992:485– 504. [11] Allman J, Brothers L. Faces, fear and the amygdala. Nature 1994;372:613– 4. [12] Amaral DG, Price JL, Pitkanen A, Carmichael ST. Anatomical organization of the primate amygdaloid complex. In: Aggleton JP, editor. The amygdala: neurobiological aspects of emotion, memory, and mental dysfunction. New York: Wiley/Liss, 1992:1– 66. [13] Anderson AK, Phelps EA. Intact recognition of vocal expressions of fear following bilateral lesions of the human amygdala. Neuroreport 1998;9:3607– 13. [14] Arndt S, Cizadlo T, Andreasen NC, Heckel D, Gold S, O’Leary DS. Tests for comparing images based on randomization and permutation methods. Journal of Cerebral Blood Flow and Metabolism 1996;16:1271– 9. [15] Arndt S, Gold S, Cizadlo T, Zheng J, Ehrhardt J, Flaum M. A method to determine activation thresholds in fMRI paradigms. Psychiatry Research: Neuroimaging 1997;75:15– 22. [16] Babinsky R, Calabrese P, Durwen HF, Markowitsch HJ, Brechtelsbauer D, Heuser L, et al. The possible contribution of the amygdala to memory. Journal of Neurology 1993;6:167–70. [17] Baird AA, Gruber BM, Fein DA, Mass LC, Steingard DA, Renshaw RJ, et al. Functional magnetic resonance imaging of facial affect recognition in children and adolescents. Journal of the American Academy of Child and Adolescent Psychiatry 1999;38:195– 9. [18] Beauregard M, Chertkow H, Bub D, Murtha S, Dixon R, Evans A. The neural substrate for concrete, abstract, and emotional word lexica: a positron emission tomography study. Journal of Cognitive Neuroscience 1997;9:441– 61. [19] Bechara A, Tranel D, Damasio H, Adolphs R, Rockland C, Damasio AR. Double dissociation of conditioning and declarative knowledge relative to the amygdala and hippocampus in humans. 1995; 269: 1115– 8. [20] Bianchin M, Mello e Souza T, Medina JH, Izquierdo I. The amygdala is involved in the modulation of long-term memory, but not in working or short-term memory. Neurobiology of Learning and Memory 1999;71:127– 31. [21] Borod JC. Interhemispheric and intrahemispheric control of emotion: a focus on unilateral brain damage. Journal of Consulting and Clinical Psychology 1992;60:339– 48. [22] Borod JC. Cerebral mechanisms underlying facial, prosodic, and lexical emotional expression: a review of neuropsychological studies and methodological issues. Neuropsychology 1993;7:445– 63. [23] Borod JC. Emotional disorders (emotion). In: Beaumont JG, Kenealy PM, Rogers MJC, editors. The Blackwell dictionary of neuropsychology. Cambridge: Blackwell, 1996:312– 20. [24] Borod JC, Bloom RL, Santschi-Haywood C. Verbal aspects of emotional communication. In: Beeman M, Chiarello C, editors. Right hemisphere language comprehension: perspectives from cognitive neuroscience. Mahwah, NJ: Lawrence Erlbaum, 1998:285– 307. [25] Borod JC, Madigan NK. Neuropsychology of emotion and emotional disorders: an overview and research directions. In: Borod JC, editor. The neuropsychology of emotion. New York: Oxford University Press, 2000:3– 28. 572 M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 [26] Breiter HC, Etcoff NL, Whalen J, Kennedy WA, Rauch SL, Buckner RL, et al. Response and habituation of the human amygdala during visual processing of facial expression. Neuron 1996;17:875– 87. [27] Broks P, Young AW, Maratos EJ, Coffey PJ, Calder AJ, Isaac CL, et al. Face processing impairments after encephalitis: amygdala damage and recognition of fear. Neuropsychologia 1998;36:59– 70. [28] Brown WP, Ure DMJ. Five rated characteristics of 650 word association stimuli. British Journal of Psychology 1969;60:233– 49. [29] Buchel C, Morris J, Dolan RJ, Friston KJ. Brain systems mediating aversive conditioning: an event-related fMRI study. Neuron 1998;20:947–57. [30] Cahill L. A neurobiological perspective on emotionally influenced, long-term memory. Seminars in Clinical Neuropsychiatry 1999;4:266– 73. [31] Cahill L, Babinsky R, Markowitsch HJ, McGaugh JL. The amygdala and emotional memory. Nature 1995;377:295– 6. [32] Cahill L, Haier RJ, Fallon J, Alkire MT, Tang C, Keator D, et al. Amygdala activity at encoding correlated with long-term, free recall of emotional information. Proceedings of the National Academy of Sciences USA 1996;93:8016–21. [33] Cahill L, McGaugh JL. Mechanisms of emotional arousal and lasting declarative memory. Trends in Neuroscience 1998;21:294– 9. [34] Cahill L, Weinberger NM, Roozendaal B, McGaugh JL. Is the amygdala a locus of ‘‘conditioned fear’’? Some questions and caveats. Neuron 1999;23:227–8. [35] Calder A, Young AW, Rowland D, Perret DI, Hodges JR, Etcoff NL. Facial emotion recognition after bilateral amygdala damage: differentially severe impairment of fear. Cognitive Neuropsychology 1996;13:699–745. [36] Canli T, Desmond JE, Zhao Z, Glover G, Gabrieli JDE. Hemispheric asymmetry for emotional stimuli detected with fMRI. NeuroReport 1998;9:3233–9. [37] Coren S, Porac C, Duncan P. A behaviorally validated self-report inventory to assess four types of lateral preference. Journal of Clinical Neuropsychology 1979;1:55–64. [38] Damasio AR. Toward a neurobiology of emotion and feeling: operational concepts and hypotheses. The Neuroscientist 1995;1:19– 25. [39] Davidson RJ, Irwin W. Functional MRI in the study of emotion. In: Moonen C, Bandettini PA, editors. Medical radiology – diagnostic imaging and radiation oncology: functional MRI. New York: Springer, 1999:487–99. [40] Davidson RJ, Irwin W. The functional neuroanatomy of emotion and affective style. Trends in Cognitive Neuroscience 1999;3:11– 21. [41] Fink GR, Markowitsch HJ, Reinkemeier M, Bruckbauer T, Kesseler J, Heiss W-D. Cerebral representation of one’s own past: neural networks involved in autobiographical memory. The Journal of Neuroscience 1996;16:4275–82. [42] Forman SD, Cohen JD, Fitzgerald M, Eddy WF, Mintun MA, Noll DC. Improved assessment of significant activation in functional magnetic resonance imaging (fMRI): use of a cluster-size threshold. Magnetic Resonance in Medicine 1995;33:636– 47. [43] Francis WN, Kucera H. Frequency analysis of English usage: Lexicon and grammar. Boston: Houghton Miffin Company, 1986. [44] Gabrieli JDE. Cognitive neuroscience of human memory. Annual Review of Psychology 1998;49:87–115. [45] Gainotti G, Caltagirone C, Zoccolotti P. Left/right and cortical/ subcortical dichotomies in the neuropsychological study of human emotions. Cognition and Emotion 1993;7:71–93. [46] Gur RC, Gur RE, Resnick SM, Skolnick BE, Alavi A, Reivich M. The effect of anxiety on cortical cerebral blood flow and [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] metabolism. Journal of Cerebral Blood Flow and Metabolism 1987;7:173– 7. Gur RC, Gur RE, Skolnick BE, Resnick SM, Silver FL, Chawluk J, et al. Effects of task difficulty on regional cerebral blood flow: relationships with anxiety and performance. Psychophysiology 1988;25:392– 9. Hamann SB, Adolphs R. Normal recognition of emotional similarity between facial expression following bilateral amygdala damage. Neuropsychologia 1999;37:1135– 41. Hamman SB, Cahill L, Squire LR. Emotional perception and memory in amnesia. Neuropsychology 1997;11:104– 13. Hamann S, Ely T, Grafton S, Kilts CD. Amygdala activity related to enhanced memory for pleasant and aversive stimuli. Nature Neuroscience 1999;2:289– 93. Hamann SB, Stefanacci L, Squire LR, Adolphs R, Tranel D, Damasio H, et al. Recognizing facial emotion. Nature 1996;379:497. Hazlett EA, Buchsbaum MS, Haznedar MM, Singer MB, Germans MK, Schnur DB, et al. Prefrontal cortex glucose metabolism and startle eyeblink modification abnormalities in unmedicated schizophrenia patients. Psychophysiology 1998;35:186– 98. Haznedar D, Buchsbaum MS, Metzger M, Solimando A, Spiegel-Cohen J, Hollander E. Anterior cingulate gyrus volume and glucose metabolism in autistic disorder. American Journal of Psychiatry 1997;154:1047– 50. Heilman KM, Blonder LX, Bowers D, Crucian GP. Neurological disorders and emotional dysfunction. In: Borod JC, editor. The neuropsychology of emotion. New York: Oxford University Press, 2000:367– 412. Kiehl KA, Smith AM, Mendrek A, Forster BB, Hare RD, Liddle PF. Activation of the amygdala during an affective memory task. NeuroImage 1998;7:S902. LaBar KS, Gatenby C, Gore JC, LeDoux JE, Phelps EA. Human amygdala activation during conditioned fear acquistion and extinction: a mixed-trial fMRI study. Neuron 1998;20:937– 45. LaBar KS, Phelps EA. Arousal-mediated memory consolidation: role of the medial temporal lobe in humans. Psychological Science 1998;9:490– 3. Lane RD, Chua PM-L, Dolan RJ. Common effects of emotional valence, arousal and attention on neural activation during visual processing of pictures. Neuropsychologia 1999;37:989–97. Lang PJ, Bradley MM, Fitzsimmons JR, Cuthbert BN, Scott JD, Moulder B, Nangia V. Emotional arousal and activation of the visual cortex: an fMRI analysis. Psychophysiology 35:199– 210. LeDoux JE. Emotion: clues from the brain. Annual Review of Psychology 1995;46:209– 35. LeDoux J. The emotional brain: the mysterious underpinnings of emotional life. New York: Simon & Schuster, 1996. Maddock RJ, Buonocore MH. Activation of left posterior cingulate gyrus by the auditory presentation of threat-related words: an fMRI study. Psychiatry Research: Neuroimaging 1997;75:1– 14. Maren S, Franselow MS. The amygdala and fear conditioning: has the nut been cracked? 1996; 16: 237– 40. Markowitsch HJ, Calabrese P, Wurker M, Durwen HF, Kessler J, Babinsky R, et al. The amygdala’s contribution to memory – a study on two patients with Urbach-Wiethe disease. NeuroReport 1994;5:1349– 52. McGaugh JL. Memory – a century of consolidation. Science 2000;287:248– 51. McGaugh JL, Cahill L, Roozendaal B. Involvement of the amygdala in memory storage: interaction with other brain systems. Proceedings of the National Academy of Science USA 1996;93:13508– 14. M.H. Tabert et al. / Neuropsychologia 39 (2001) 556–573 [67] Morris JS, Friston KJ, Buchel C, Frith CD, Young AW, Calder AJ, et al. A neuromodulatory role for the human amygdala in processing emotional facial expressions. Brain 1998;121:47– 57. [68] Morris JS, Frith CD, Perrett DI, Rowland D, Young AW, Calder AJ, et al. A differential neural response in the human amygdala to fearful and happy facial expressions. Nature 1996;383:812– 5. [69] Ogawa S, Lee T-M, Nyak AS, Glynn P. Oxygenation-sensitive contrast in magnetic resonance image of rodent brain at high magnetic field. Magnetic Resonance in Medicine 1990;14:67– 78. [70] Packard MG, Cahill L, McGaugh JL. Amygdala modulation of hippocampal-dependent and caudate nucleus-dependent memory processes. Proceedings of the National Academy of Science USA 1994;91:8477– 81. [71] Peck DF. The conversion of progressive matrices and Mill Hill vocabulary raw scores into deviation IQs. Journal of Clinical Psychology 1970;26:67–70. [72] Phelps A, Anderson AK. Emotional memory: what does the amygdala do? Current Biology 1997;7:R311–4. [73] Phillips ML, Young AW, Cott SK, Calder AJ, Andrew C, Giampietro V, et al. Neural responses to facial and vocal expressions of fear and disgust. Proceedings of the Royal Society: B 1998;265:1809– 17. [74] Phillips ML, Young AW, Senior C, Brammer M, Andrews C, Calder AJ, et al. A specific neural substrate for perceiving facial expressions of disgust. Nature 1997;389:495–8. [75] Rauch SL, van der Kolk BA, Fisler RE, Alpert NM, Orr SP, Savage CR, et al. A symptom provocation study of posttraumatic stress disorder using positron emission tomography and script-driven imagery. Archives of General Psychiatry 1996;53:380– 7. [76] Raven JC. Revised manual for Raven’s progressive matrices and vocabulary scale. Windsor, UK: NFER Nelson, 1982. [77] Reiman EM, Lane RD, Ahern GL, Schwartz GE, Davidson RJ, Friston KJ, et al. Neuroanatomical correlates of externally and internally generated human emotion. American Journal of Psychiatry 1997;154:918–25. [78] Reivich M, Gur R, Alavi A. Positron emission tomographic studies of sensory stimuli, cognitive processes and anxiety. Human Neurobiology 1983;2:25–33. [79] Ross ED. Right hemishpere syndromes and the neurology of emotion. In: Schacter SC, Devinsky O, editors. Behavioral neu- [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] 573 rology and the legacy of Norman Geschwind. Philadelphia: Lippincott-Raven, 1997:183– 91. Scott SK, Young AW, Calder AJ, Hellawell DJ, Aggelton JP, Johnson M. Impaired auditory recognition of fear and anger following bilateral amygdala lesions. Nature 1997;385:254–7. Spielberger CD, Gorsuch RL, Lushene RE. Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologists Press, 1970. Talairach J, Tournoux P. Co-planar sterotaxic atlas of the human brain: 3-dimensional proportional system; an approach to cerebral imaging. Stuttgart: George Thieme Verlag, 1988. Toglia MP, Battig WF. Handbook of semantic word norms. Hillsdale, NJ: Lawrence Erlbaum, 1978. Tranel D, Damasio AR. The covert learning of affective valence does not require structures in hippocampal systems or amygdala. Journal of Cognitive Neuroscience 1993;5:79– 88. Vazdarjanova A, McGaugh JL. Basolateral amygdala is not critical for cognitive memory of contextual fear conditioning. Proceedings of the National Academy of Science USA 1998;95:15003– 7. Whalen PJ. Fear, vigilance, and ambiguity: initial neuroimaging studies of the human amygdala. Current Directions in Psychological Science 1998;7:177– 88. Whalen PJ, Bush G, McNally RJ, Wilhelm S, McInerney SC, Jenike SC, et al. The emotional counting Stroop paradigm: a functional magnetic resonance imaging probe of the anterior cingulate affective division. Biological Psychiatry 1998;44:1219– 28. Whalen PJ, Rauch SL, Etcoff NL, McInerney SC, Lee MB, Jenike MA. Masked presentations of emotional facial expressions modulate amygdala activity without explicit knowledge. The Journal of Neuroscience 1998;18:411– 8. Wik G, Fredrikson M, Eriksson L, Stone-Elander S, Greitz T. A functional cerebral response to frightening visual stimulation. Psychiatric Research: Neuroimaging 1993;50:15– 24. Young AY, Aggleton JP, Hellawell DJ, Johnson M, Broks P, Hanley R. Face processing impairments after amygdalotomy. Brain 1995;118:15– 24. Young AW, Hellawell DJ, Van de Wal C, Johnson M. Facial expression processing after amygdalotomy. Neuropsychologia 1996;34:31– 9. . .