Main

Studies with humans1,2 and experimental animals3,4 have demonstrated that memory is better for emotionally arousing stimuli than for emotionally neutral stimuli. From an evolutionary perspective, it is clearly adaptive for memory for emotional stimuli to be enhanced, because emotional stimuli, whether pleasant (for example, sexual) or aversive (for example, frightening) are generally more important than neutral stimuli for reproductive success. Considerable evidence from animal studies indicates that the amygdala has a crucial role in enhancing the strength of long-term memory for emotional stimuli3,5,6 through the interaction of peripheral adrenergic systems with cholinergic, opioid-peptidergic and GABAergic systems in the amygdala3. Two recent single-case studies of amygdala-damaged human subjects7,8 suggest that the amygdala also enhances episodic memory (conscious memory for events) for aversive stimuli. However, there are no conclusive neuroimaging data linking amygdala activity to enhanced episodic memory for specifically emotional stimuli in humans9. The role of the amygdala in memory for pleasant stimuli is of particular interest, because there are to date no data linking the amygdala to memory for pleasant, rewarding stimuli in humans. Here, using positron emission tomography (PET), we examined the relationship between amygdala activity during memory encoding of pleasant and aversive stimuli and enhanced long-term episodic memory (recall and recognition) for these stimuli. We tested the hypothesis that amygdala activity is related to enhanced episodic memory for pleasant and aversive emotionally arousing stimuli but not for nonemotional stimuli. Our primary hypothesis is consistent with the 'memory modulation' theoretical framework of amygdala function3,5,6 from experimental animal data, which predicts that the amygdala should enhance long-term memory formation through modulation of hippocampal activity for emotionally arousing stimuli but not for emotionally neutral stimuli.

We measured brain activity with PET H215O studies of regional cerebral blood flow (rCBF) while ten healthy male volunteers viewed four types of pictures. Representative pleasant pictures depicted sexually arousing scenes, appealing animals or appetizing food. Representative aversive pictures depicted mutilated and diseased bodies, frightening animals or lethal violence. These conditions were contrasted with two nonemotional control conditions. Representative neutral pictures depicted chess players, plants and animals or household scenes. Interesting pictures, such as a chrome rhinoceros, a scene from a surrealistic film or an exotic parade, were designed to attract interest and attention and to be highly memorable as a control for these factors in the emotional pictures, yet to be emotionally unarousing. Twelve scans were done, three each for the four picture types. Skin-conductance and heart-rate data collected during scanning confirmed substantial physiological arousal elicited by emotional pictures, corroborating subjects' emotion ratings taken after each scan ( Table 1 ). PET data revealed an association of emotional-stimulus pictures with significantly greater amygdala activation than the neutral pictures, consistent with prior findings10 (pairwise comparisons with PET data will be reported separately; S.H, T.E., J.H. & C.K.). As expected, interesting pictures were rated as similar in interest and attention to the emotional pictures but were rated relatively unemotional ( Table 1 ). After scanning had ended, a surprise memory-recall test was given for the pictures. Four weeks later, subjects were given another unannounced recall test followed by a recognition-memory test, in which they were asked to discriminate between pictures seen during the PET session and similar but previously unseen pictures.

Table 1 Behavioral data.
Full size table

RESULTS

We examined correlations between rCBF in the amygdala during encoding of pleasant, aversive or interesting pictures and the corresponding memory enhancement for each picture type relative to performance in the neutral control condition. The resulting difference scores (for example, pleasant-picture recognition minus neutral-picture recognition) isolated the memory-enhancement effect from individual subject differences in general, nonemotional (neutral) memory ability. Isolation of the enhancement effect was crucial, as a given subject might exhibit better memory for pleasant or aversive pictures than another subject because of superior general memory ability unrelated to the amygdala or to greater emotional-memory enhancement. For each subject, the rCBF and memory data were averaged across the three scans for each picture type.

Correlational analyses were done for recognition data and immediate- and delayed-recall test data. Long-term episodic recognition memory was substantially enhanced for the pleasant, aversive and interesting pictures relative to the neutral pictures (data for the three memory tests are presented in Table 1 ). Significant correlations with amygdala rCBF were observed only for the recognition data; therefore only those results are presented here. The absence of significant amygdala correlations for the four-week-delayed-recall test data may have resulted from a statistical artifact (restriction of range) associated with the low level of memory performance and consequently, small variance for enhancement scores in this condition. In contrast to the four-week, delayed-recall data, there was substantial individual subject variance in the four-week recognition and ten-minute recall memory-enhancement scores. Absence of significant amygdala correlations for the ten-minute recall data is consistent with the memory-modulation theory, which postulates that a major function of the amygdala is to modulate long-term memory consolidation3,5,6. At short delays, the amygdala's role may not be detectable because little consolidation has taken place.

Significant correlation was found between individual subjects' recognition memory enhancement for pleasant pictures and bilateral amygdala activity measured during the pleasant-picture scans ( Fig. 1a ). For aversive pictures, recognition-memory enhancement was also correlated with bilateral amygdala activity measured during the aversive picture scans, though at a slightly more inferior level and more right-lateralized ( Fig. 1b ). Three-dimensional coordinates11 of representative maximally correlated pixels for these regions are presented in Table 2. In contrast, recognition-memory enhancement for interesting pictures was not correlated with activity in the amygdala or in the periamygdaloid cortex, even using liberal statistical thresholds, indicating that the amygdala's relation to enhanced recognition memory was specific to emotional stimuli.

Figure 1: Brain activity correlated with memory enhancement.
figure 1

Maps of pixels in which individual subject rCBF was significantly correlated (yellow, p < 0.05, two-tailed test for significant correlations) with individual-subject episodic-memory enhancement superimposed on an axial magnetic-resonance reference image averaged from a separate group of subjects. (a) Correlation map for pleasant stimuli at z = –10.5. (b) Correlation map for aversive stimuli at z = –16.5.

Full size image
Table 2 Loci of maximal correlation.
Full size table

This conclusion was corroborated by an analysis that compared strength of correlations between amygdala rCBF at encoding and memory enhancement for pleasant and aversive pictures ( Table 2 ) with corresponding correlations for interesting pictures at the same loci. In each case, amygdala rCBF correlations were significantly greater for the emotional pictures than for the interesting, emotionally neutral pictures, all t (7) > 1.77, p < 0.05, for the directional t-test of significant difference between two correlations. Indeed, the correlations for interesting pictures at the left and right amygdala correlation loci for pleasant pictures were negligible or negative (–0.19 and 0.01, respectively), as were the correlations for the aversive picture loci (–0.21 and 0.11, respectively).

The outcome of the correlational analyses suggests a specific relationship between enhancement of episodic memory for pleasant and unpleasant emotional stimuli and activity in the amygdala. Subjects who showed the greatest episodic-memory enhancement associated with emotional stimuli also tended to have higher levels of activity in the amygdala than subjects who showed the least enhancement.

Because the amygdala has been hypothesized to influence episodic memory indirectly, by modulating3,5,6 the activity of a hippocampal/medial–temporal lobe memory system12,13 (according to the memory-modulation theory), we did the same correlational analysis for a second region of interest, the hippocampal/parahippocampal region, which had been defined a priori. Consistent with this hypothesis, recognition-memory enhancement for both pleasant and aversive pictures was correlated with bilateral activity in the hippocampal region ( Table 2 ), although a causal relationship between amygdala and hippocampal activity cannot be inferred from these correlations. Consistent with the hippocampus's role in general episodic memory encoding for novel stimuli14, recognition-memory enhancement for interesting pictures was also correlated with hippocampal activity in the right hemisphere ( Table 2 ).

If the amygdala influences memory for emotional stimuli by modulating activity in the hippocampal/parahippocampal region as specified by the memory-modulation theory, then activity levels in the amygdala and the hippocampal/parahippocampal region should be highly correlated across individual subjects. Conversely, uncorrelated activity in these two regions would suggest that, although activity in each region is independently related to enhanced memory, no modulatory relationship exists between them. Importantly, the memory-modulation theory does not predict that activity at all loci in the amygdala should be correlated with all loci in the hippocampal/parahippocampal region. Instead, it makes the more specific prediction that particular loci within these regions whose activation is significantly related to memory performance should show correlated activity. Therefore, the critical loci to test for intercorrelation are loci in the amygdala and hippocampal/parahippocampal regions of interest at which memory-correlated activity was maximal, specifically the loci listed in Table 2.

The results of this analysis were generally consistent with the memory-modulation theory. For both pleasant and aversive picture conditions, activity (rCBF) at critical amygdala loci was significantly correlated with activity at the corresponding hippocampal or parahippocampal loci, both ipsilaterally and contralaterally (r > 0.68, p < 0.05) with two exceptions: nonsignificant positive correlations between the left amygdala and left hippocampus for pleasant stimuli (r = 0.64), and between the right amygdala and right hippocampus for aversive stimuli (r = 0.57). The finding of significant contralateral correlations is important because the close proximity of some of the amygdala and hippocampal/parahippocampal loci within the same hemisphere raises the possible concern of partial overlap of smoothed rCBF data between loci, which would tend to inflate ipsilateral correlations; the contralateral correlations are not subject to this concern.

It is probable that the amygdala is part of a larger network of brain regions involved in encoding conscious memory for emotional stimuli. For example, for pleasant stimuli, the regions of correlated rCBF in the amygdala ( Fig. 1a ) were spatially contiguous with a larger, superior region of correlated rCBF including the right nucleus accumbens/subcallosal cortex (9, 9, –7 mm11, r = 0.93, p < 0.001) and right anterior cingulate cortex (7, 21, –7 mm11, r = 0.82, p < 0.01). The nucleus accumbens has been linked to pleasure and reward mechanisms15,16 and receives prominent input from the amygdala17. These interconnected regions may compose a functional network preferentially involved in encoding pleasant episodic memories. However, as these regions were not included in the a-priori regions of interest, their involvement and functional interpretation should be viewed as tentative, pending replication.

The reliability of these correlational data is supported by several lines of evidence. First, it is unlikely that correlated activity in the regions of a-priori interest is a chance artifact because activity correlated with memory enhancement for pleasant and aversive pictures falls within discrete, contiguous regions, and the memory-correlated activity is largely confined to the a-priori regions of interest ( Fig. 1 ). For example, within a large region of the brain extending from the horizontal level corresponding to the inferior limit of the most inferior region of correlated activity in an a-priori region of interest (z = –22, Table 2 ) up to and including the horizontal slice at z = 0, memory-correlated brain activity was largely confined to the a-priori regions of interest. Exceptions included the putative network for encoding pleasant episodic memories noted above, two small regions in visual cortex and one in right frontal lobe showing correlated activity for pleasant pictures, and one small region of visual cortex and a small right medial-temporal-lobe region showing correlated activity for aversive pictures. Inspection of correlation plots relating rCBF to memory enhancement also suggests that the observed correlations are not due to extreme scores skewing the observed correlations (see Fig. 2 for correlation plots for the pleasant-picture correlation data; removal from the correlational analysis of the apparent outlying score on the far right in scatterplots a, b and c reduced the observed correlations slightly but not significantly).

Figure 2: Relationship between pleasant-picture memory and brain activity for individual subjects.
figure 2

Correlation scatterplots for the pleasant-picture memory correlations with rCBF, n = 10. Memory enhancement is defined as recognition for pleasant pictures minus recognition for neutral pictures, in d ´ accuracy units24. rCBF is in units of ml per min per 100 g of brain tissue, normalized across subjects and scans by proportionate scaling of global cerebral blood flow. (a) Correlation plot for left amygdala. (b) Correlation plot for right amygdala. (c) Correlation plot for left hippocampal region. (d) Correlation plot for right hippocampal region.

Full size image

DISCUSSION

Only one previous neuroimaging study has examined the relationship between amygdala activity at encoding and episodic memory for emotional stimuli9. Using a different PET methodology with [18F]fluoro-2-deoxyglucose (FDG-PET), that study examined recall for aversive and neutral films. It found a significant positive correlation between brain glucose metabolic rate in the right amygdala during encoding and the number of emotional but not neutral films recalled three weeks later, demonstrating a relation between amygdala activity and episodic memory for aversive stimuli. Although significant correlations with neutral-film recall were observed in other brain regions outside the amygdala, the specificity of the relation of amygdala activity to aversive emotional memory could not be established as conclusively as in the current study, because memory of neutral films was poor, reducing likelihood of significant correlation. To establish that amygdala activity is not related to memory for nonemotional stimuli, it is necessary to rule out the possibility that the amygdala has a nonspecific memory-enhancing role for all stimuli. The use of interesting and memorable yet nonemotional pictures as control stimuli in the current study allowed the selectivity of the amygdala's memory enhancement for emotional stimuli to be established more conclusively. The specificity of the amygdala's mnemonic role for emotional stimuli is also supported by a recent neuroimaging study that reported inactivity of the amygdala during formation of nonemotional declarative memory for verbal stimuli18. Although the FDG-PET study9 used total recall rather than an enhancement measure as the behavioral correlate of amygdala activity, reanalysis of the FDG-PET study data using enhancement scores (number of emotional films recalled – number of neutral films recalled) does not significantly change the conclusions of that correlational analysis (L. Cahill, personal communication, 1998).

Our results confirm and extend these previous FDG-PET findings9 by demonstrating that activity in the left as well as right amygdala is related to enhanced episodic memory for aversive stimuli, and that this relationship is specific to emotional stimuli. In addition, the current study is the first demonstration that the amygdala is involved in encoding memory for emotionally arousing, pleasant stimuli in humans. The role of the amygdala in memory for pleasant, rewarding stimuli in nonhuman animals has received some attention5,15,16,19, but relatively little in comparison to the intense scrutiny afforded the amygdala's role in aversive memory phenomena such as fear conditioning20,21,22 and fear-potentiated startle23. The current findings suggest that the amygdala's role in encoding memory for emotionally arousing, pleasant stimuli in humans is significantly greater than has been appreciated previously. In summary, the present findings suggest that the amygdala in humans is important in modulating memory for events according to their emotional importance, regardless of whether the nature of the emotion is pleasant or aversive. The hippocampal data are consistent with the notion that the amygdala may exert its influence on episodic memory for emotionally arousing stimuli in part through modulation of the hippocampal/medial-temporal-lobe memory system3,5,6.

METHODS

Subjects.

The ten subjects were right-handed male university students with a mean age of 22.8 years (range 20–32) who had been screened for neurological and psychiatric conditions. To ensure effectiveness of the sexually arousing, pleasant pictures, only self-identified heterosexual subjects were included in the study. The study was approved by the Emory University School of Medicine Human Investigations Committee; written informed consent was obtained from the subjects.

Behavioral tasks.

During each PET scan, subjects viewed pictures on a computer monitor that spanned their field of view, and were instructed, "pay attention and experience whatever thoughts or feelings the pictures may elicit in you." No mention of the later memory tests was made. Following each scan, subjects rated the set of pictures they had just viewed (13 pictures) on 3 dimensions using scales of 1–5: emotional arousal (1, lowest arousal; 5, highest), emotional valence (1, most unpleasant; 3, neutral; 5, most pleasant) and degree of interest (1, lowest interest; 5, highest interest). Mean ratings for the pleasant, aversive, interesting and neutral stimuli are shown in Table 1. Each picture was presented for 7.5 s, beginning 10 s before and during each scan. Different sets of stimuli were used for each picture-presentation condition, and the order of conditions was counterbalanced across subjects.

Skin conductance and heart rate were measured at 100 Hz beginning 1 min before each scan. Table 1 shows the mean skin-conductance response to each stimulus type averaged across the 10 subjects, in microsiemens (μS, units of electrical conductance). The mean skin-conductance response (SCR) for each subject was calculated by first determining the SCR to each picture stimulus, defined as the maximal positive deflection in skin-conductance level with onset occuring 0.5–4.0 s after stimulus presentation, and then averaging the four largest SCRs for each scan. (The SCRs corresponding to the first three stimuli of each scan were excluded from this analysis because their SCR values were affected by the SCR generated by the injection of the radioactive tracer.) Mean SCRs for pleasant, aversive or interesting conditions differed significantly from the mean SCRs in the neutral condition (all t (8) > 3.65, p < 0.01, paired t-test), but did not differ significantly from each other (p > 0.10). (SCRs were elevated in the interesting condition because skinconductance was affected by the orienting response to interesting stimuli as well as by emotional arousal.) Heart-rate change was assessed by subtracting the lowest heart rate observed during stimulus presentation (in beats per minute; bpm) from the baseline heart rate measured 20 s before the injection of the radioactive tracer. Heart-rate deceleration to emotional and attention-provoking stimuli is typically observed in situations where such stimuli are presented for passive viewing. Heart rate decelerated significantly more during viewing of the pleasant, aversive, or interesting pictures than for the neutral pictures (all t (8) > 2.37, p < 0.05, paired t-test).

Memory measures.

The immediate-recall test was given 10 min after the end of the final scan. Subjects were instructed to try to recollect the pictures they had seen during the scans in any order they chose and to write down words or phrases that would describe the pictures so that they could be identified from the set they had seen. Two independent judges determined which picture (if any) each verbal response described. Four weeks later, subjects were given another unannounced recall test using the same procedure as the 10-min recall test, followed by a recognition test in which all of the pictures that had been viewed during the PET session (except for the first picture of each set, which was a buffer item not included in the analysis) were presented in a pseudo-random order together with an equal number of similar but previously unseen pictures. Subjects were asked to indicate which pictures they had seen before and which were new. Memory-enhancement scores for each subject were calculated by converting individual-subject recognition scores from each picture condition to d´ recognition-accuracy statistics24 (to control for the effects of differing response criteria across subjects) and then subtracting the d´ scores for the neutral condition from each of the corresponding d´ scores from the other conditions. All pictures are from a standard set of affective pictures25 except for the interesting pictures, which were collected from commercial sources and standardized on a separate group of subjects.

PET image acquisition.

PET images of task-related rCBF were acquired in two-dimensional mode with an ECAT 951 or 921 scanner (Siemens, Knoxville, Tennessee) following the intravenous administration of the blood-flow tracer26 H215O. Acquisition of emission data was initiated 10 seconds after tracer injection and continued for 90 s, with 10 min between each tracer injection. Head movement was minimized by use of the TruScan restraint system. Images were reconstructed using a calculated attenuation correction. PET images for each subject were aligned and resliced27. The PET scans were coregistered to a population-average PET target28 centered in Talairach coordinates11, derived from 20 subjects who also had structural MRI. The average MRI from this population was used as the anatomical reference. PET images were smoothed to a final isotropic resolution of 9 mm and normalized to each other by proportionate scaling of global activity. The PET data for one subject was not available for part of the right amygdala and right anterior regions corresponding to inferior horizontal slices at z = – 18 and below11. Therefore, all correlational analyses in this study were done with the other nine subjects with complete PET data. The correlational analyses did not change significantly when recalculated with all 10 subjects for loci at which PET data was available for all subjects. For reliability checks on the correlations and the correlation scattergrams shown in Fig. 2 that assessed the effects of extreme scores, all available data from all 10 subjects were used because the 10th subject had the potential to reveal instability in the correlations if his data were highly discrepant from those of the other subjects.

Region of interest selection.

We defined a priori our two regions of interest, the amygdala and the hippocampal/parahippocampal region, by tracing the neuroanatomical boundaries of these brain regions on 1.5 mm-thick axial slices of the averaged-subject magnetic-resonance reference image that was used for PET coregistration. We also traced the boundaries of these brain regions onto a second reference, the horizontal slices from the standard atlas of Talairach and Tournoux11. The correlational analysis examined rCBF, estimated from radioactivity relative to the global mean, computed for every 1.5 mm × 1.5 mm × 1.5 mm pixel in the entire brain that fell within the field of view of the PET scanner. For the region-of-interest analysis, a strict criterion was used for neuroanatomical localization. Specifically, a correlated pixel was accepted as being localized within the amygdala or hippocampal/parahippocampal region only if it fell both within the traced ROI on the averaged subject structural magnetic resonance image and within the Talairach-based ROI as determined by the plotted Talairach coordinates of the correlated pixel. This same procedure was also used to determine the neuroanatomical localization of correlated regions that were not defined a priori (for instance, nucleus accumbens).