Emotional influences on memory are clearly demonstrable; most people will have experienced a “flashbulb” memory of a dramatic or highly emotional event, such as the birth of their first child or the events of September 11th, 2001. A variety of studies have demonstrated that memories for emotional events are more persistent and vivid than their neutral counterparts (Ochsner, 2000; Sharot & Yonelinas, 2008; Todd, Talmi, Schmitz, Susskind, & Anderson, 2012), suggesting that emotional aspects of stimuli influence memory encoding, and hence subsequent recollection.

Emotion may be measured both in terms of its valence (i.e., happy/sad) and the degree of physiological arousal elicited. Both of these dimensions appear to influence memory processes. For example, the degree of arousal associated with the to-be-remembered items appears to be critical for feature binding in working memory (Mather, 2007) and for long-term memory (Judde & Rickard, 2010), whereas mood valence has been shown to influence associative memory (Isen, Johnson, Mertz, & Robinson, 1985). Corson and Verrier (2007) reported that high levels of arousal induced prior to a recognition test increased the false alarm rates for novel stimuli, though variations in mood valence had no effect. This evidence points to an effect of the emotional content and context of stimuli on the efficiency of memory encoding, and several imaging studies have examined the neural regions that mediate this process. For instance, the amygdala, medial temporal lobe, and prefrontal cortex have been shown to be involved in the successful encoding of emotional stimuli such as arousing pictures or scenes (Dolcos, LaBar, & Cabeza, 2004; Kalpouzos, Fischer, Rieckmann, Macdonald, & Backman, 2012; Kensinger & Corkin, 2004; Mickley Steinmetz & Kensinger, 2009), and inferior frontal cortex has been implicated in enhanced retrieval of emotionally valenced autobiographical memories (Denkova, Dolcos, & Dolcos, 2013a). Memory for emotionally valenced but nonarousing stimuli have been shown to elicit activation in a network encompassing the hippocampus and prefrontal cortex, whereas improved memory for highly arousing stimuli is dependent on response in the hippocampus and amygdala (Kensinger & Corkin, 2004). Response in the amygdala and hippocampus during retrieval has also been linked to the emotional valence of autobiographical memories (Denkova, Dolcos, & Dolcos, 2013b).

To understand the role of emotion in memory, it is fundamental to distinguish the effects operating during memory encoding from other emotional influences operating after stimulus encoding has taken place. In this vein, previous research has shown that postencoding manipulation of the individual’s emotional state may also influence memory for otherwise neutral events and stimuli (Anderson, Wais, & Gabrieli, 2006; Finn & Roediger, 2011; Greene, Bahri, & Soto, 2010; Kuhbandner & Pekrun, 2013; Liu, Graham, & Zorawski, 2008; Nielson & Arentsen, 2012; Nielson & Lorber, 2009). However, many of these studies have either conflated valence and arousal effects or focused on one dimension, to the exclusion of the other. Recent research has indicated that neither mood valence nor arousal alone is sufficient to explain emotional effects on memory performance; rather, their interaction appears to be most critical (Greene et al., 2010). In this study, increasing arousal after initial encoding but prior to memory testing was found to improve subsequent recognition of neutral stimuli when participants were in a positive mood, but to impair performance when combined with a negative mood state.

With the present study, we aimed to characterize the neurocognitive mechanisms that mediate the influence of mood valence and arousal on postencoding memory processes such as retrieval and recognition, following our prior behavioral work (Greene et al., 2010). One possibility is that emotional influences on recognition memory may be mediated by attention control mechanisms, of which the posterior parietal cortex is a key substrate (Ciavarro et al., 2013; Corbetta, Kincade, Ollinger, McAvoy, & Shulman, 2000; Han et al., 2004; Lane, Smith, Schenk, & Ellison, 2011; Yin et al., 2012). The posterior parietal cortex (PPC) is also one of the most commonly activated regions in fMRI studies of episodic memory and recognition (Kim, 2011; Wagner, Shannon, Kahn, & Buckner, 2005). Surprisingly, then, parietal lesions appear to cause only subtle impairments in recognition (Davidson et al., 2008; Haramati, Soroker, Dudai, & Levy, 2008; Rossi et al., 2006; though see Simons, Peers, Mazuz, Berryhill, & Olson, 2010). Despite its recurring role in memory retrieval studies, it has been suggested that the PPC’s role is not mnemonic per se, but may instead reflect attentional control processes in the service of recognition memory (Cabeza, Ciaramelli, Olson, & Moscovitch, 2008; Ciaramelli, Grady, Levine, Ween, & Moscovitch, 2010). To our knowledge, no studies to date have investigated the influence of emotional context on posterior parietal response during memory retrieval. Evidence that positive moods lead to greater incidence of global, rather than local, processing (Basso, Schefft, Ris, & Dember, 1996; Gasper & Clore, 2002) and to reduced attentional control (Jefferies, Smilek, Eich, & Enns, 2008) suggests that emotional state influences attentional selection processes. We hypothesized that if the attentional control aspects of recognition memory operate via the posterior parietal cortex, activity in the PPC during memory retrieval may be modulated by emotional state and by the interplay between mood valence and arousal.

Functional magnetic resonance imaging (fMRI) studies of emotional effects on retrieval are lacking in comparison to the number of studies assessing the mechanisms underlying emotional influences on memory encoding. To redress this balance, we manipulated emotional state following the presentation of neutral, abstract stimuli and used fMRI to assess recognition signals in a subsequent memory test. Music listening is an extremely effective method of inducing specific emotions in an experimental setting (Blood, Zatorre, Bermudez, & Evans, 1999; Greene et al., 2010; Rowe, Hirsh, & Anderson, 2007). Using the same method we had successfully employed in an earlier study (Greene et al., 2010), we carefully selected musical excerpts to vary postencoding emotional state in an orthogonal manner, and hence evoked different combinations of mood valence and arousal in our participants (positive mood/high arousal, positive mood/low arousal, negative mood/high arousal, negative mood/low arousal). This allowed us to assess recognition memory responses driven by mood valence, arousal, and their interaction, and critically, to determine whether any interaction effect reflected a quantitative change in responses in the same substrates that responded to mood or arousal, or instead represented a distinct effect in the operation of a different neural substrate. This was a critical aspect of our design, since, to the best of our knowledge, no previous fMRI studies, even those assessing the effects of emotion on memory encoding, have included independent manipulations of mood and arousal in a factorial design.

Method

Participants

A group of 24 healthy participants (14 male, ten female) between 20 and 24 years old were recruited for this study and were paid £30 for their involvement. All participants were right-handed and had normal or corrected-to-normal vision with no past history of head trauma or neurological problems. Each provided informed consent and ethics approval was granted by the West London Research Ethics Committee.

Music selection

Participants attended a training session one to two days prior to the scanning session, during which the pieces of music that would be used to induce the emotional states were selected. A series of short musical excerpts were played to the participants who were asked to indicate how each piece made them feel by placing a mark on a chart representing the dimensions of valence and arousal. The chart took the form of a large cross in which the vertical line denoted arousal, on a scale going from high (alert/energized) to low (relaxed), whereas the horizontal line represented mood valence on a continuum from positive mood to negative mood. The pieces of music were drawn from a wide variety of genres (e.g., classical, jazz, blues, rock, heavy metal, and electronica), with the sole restriction that they be instrumental pieces. The music selection process continued until participants had assigned at least one piece of music to each of the four quadrants, representing the four mood/arousal conditions.

During this session, participants also practiced the behavioral task described below to ensure their familiarity with it. The stimuli used during the training session were not presented during the scanning session.

Behavioral task

Participants underwent fMRI while performing a recognition memory experiment. The stimuli for this experiment were monochrome abstract shapes, generated from bitmap images by custom software programmed in MATLAB. Four blocks of the experiment, one for each emotional condition, were performed in counterbalanced order. Each block began with a study phase, in which twenty stimuli were presented sequentially for 3 s each, with a 500-ms intertrial interval. Participants were instructed to study these shapes and attempt to remember them for a subsequent recognition test. The study phase was followed by 3 min of music exposure. In order to enhance the emotional induction, participants were asked to focus on thoughts and memories consonant with the emotions evoked by the music. Participants rated their subjective mood (from very negative to very positive) and level of arousal (from very relaxed to very alert) before and after the presentation of each musical selection.Footnote 1 Visual analog scales (VASs), consisting of a horizontal line anchored at either end by the extremes of the dimension, were presented onscreen for 6 s each. Participants used the button box to move a cursor along the horizontal line to indicate their emotional state. The final position of the cursor was converted to a numerical score between 0 and 100. Following the second mood valence and arousal rating, participants performed a recognition memory test. Twenty old shapes (viewed during the study phase) and 20 completely new shapes were presented at center screen, one at a time. Each shape remained onscreen for 2 s before the presentation of the response cue, and remained onscreen throughout the response period. Participants were first asked to indicate whether the shape was old or new and then to rate their confidence in that decision (not sure, quite sure, or very sure). A two second window was provided for each response.

Image acquisition/scanning parameters

MRI scanning was conducted using a Siemens Avanto 1.5 Tesla MRI scanner with a 32-channel head coil. Following a brief localizer scan to determine the orientation of the participant’s head within the field, 176 high-resolution T1-weighted anatomical sagittal images were acquired with a field of view of 224 × 224 mm, TR of 2,000 ms, TE of 2.48 ms, and slice thickness of 1 mm, with a resulting voxel resolution of 1 × 1 × 1 mm. Four functional runs of 2*-weighted echoplanar imaging were then conducted to obtain 29 contiguous sagittal slices covering the whole brain. Each run contained 262 volumes that were acquired with a field of view of 224 × 224 mm, TR of 2,500 ms, TE of 44 ms, and slice thickness of 3.2 mm. Music was presented using Sensimetrics S14 MR Conditional headphones at a consistent volume across participants.

Imaging data analysis

The fMRI data processing was carried out using FEAT (fMRI Expert Analysis Tool), Version 5.98, which is part of FSL (FMRIB’s Software Library, www.fmrib.ox.ac.uk/fsl). The following prestatistics processing was applied: nonbrain removal using BET (Smith, 2002); motion correction using MCFLIRT (Jenkinson, Bannister, Brady, & Smith, 2002); grand-mean intensity normalization of the entire four-dimensional data set by a single multiplicative factor; high-pass temporal filtering above 100 s (Gaussian-weighted least-squares straight line fitting) and spatial smoothing using a Gaussian kernel of a full width at half maximum of 6.0 mm.

Time-series statistical analysis was carried out using FILM (FMRIB’s Improved Linear Model) with local autocorrelation correction for each run (Woolrich, Ripley, Brady, & Smith, 2001). Recognition test trials were modeled separately for old and new shapes. Each event had a duration of 2 s corresponding to the presentation of the memory probe and was modeled as a boxcar function from the onset of the stimulus convolved with the hemodynamic response function. Additional explanatory variables (EVs) included the onset of each stimulus within the study phase (duration = 3 s), the VAS rating sessions (duration = 6 s) and the music listening phase (duration = 180 s). Error trials and realignment parameters from the motion correction were added to the model as regressors of no interest and the temporal derivative of the hemodynamic response function was added for each explanatory variable in order to account for latency differences between slice acquisitions. Contrasts to test for simple effects of old and new trials, and for the difference between old and new trials were conducted.

Mean effects of trial type (old or new), irrespective of emotional condition, were assessed by means of cross-run individual analyses. Contrast parameter estimates from the four experimental runs were analyzed using fixed effects to derive Z statistic images for each participant. These were then added into a group-level analysis using FLAME (FMRIB’s Local Analysis of Mixed Effects) Stage 1 + 2 as implemented in FEAT (Beckmann, Jenkinson, & Smith, 2003; Woolrich, 2008; Woolrich, Behrens, Beckmann, Jenkinson, & Smith, 2004). In all cases, we report maps of BOLD responses thresholded using clusters determined by a voxelwise Z threshold of 2.3 and a corrected cluster significance threshold of p = .05. Each individual’s EPI scans were registered to high-resolution structural images using FLIRT (Jenkinson et al., 2002; Jenkinson & Smith, 2001), and were then co-registered and transformed to standard (Montreal Neurological Institute) space.

Since each emotional condition was assessed in a separate scanning run, effects of the emotional variables (valence and arousal) were computed via a quadrupled two-group difference (“quadrupled t test”) design, a natural extension of the paired t test and “tripled” t test procedures implemented in FSL (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FEAT/UserGuide. See Demeter, Hernandez-Garcia, Sarter, and Lustig (2011); Warbrick, Reske, and Shah (2013); and Zhu et al. (2010) for published examples of the tripled t test design). These tests were performed on the lower-level contrast of parameter estimates for the old-versus-new differences from each of the four emotional conditions. In this method, the four conditions (positive mood/high arousal; positive mood/low arousal; negative mood/high arousal; negative mood/low arousal) are represented by different combinations of three EVs. EV1 models the difference between Conditions 1 and 2 for each participant, EV2 models Conditions 1–3, and EV3 models Conditions 1–4. An additional 24 regressors were included to model the mean effect for each participant. The mean of Condition 1 is therefore given by EV1 + EV2 + EV3, the mean of Condition 2 by –EV1, the mean of Condition 3 by –EV2, and the mean of Condition 4 by –EV3. These contrast specifications were extended to test for main effects and interactions between the emotional variables (see Supplementary Data for further information). The resulting z-statistic images were then thresholded and co-registered as above.

Results

Emotional induction

VAS ratings of mood valence and arousal provided prior to the participants’ exposure to music were compared across the four experimental blocks using one-way repeated measures analyses of variance (ANOVAs). These revealed no significant differences in preliminary mood valence [F(3, 69) = 0.55, p > .05, η p 2 = .024] or arousal [F(3, 69) = 0.42, p > .05, η p 2 = .018] ratings across the conditions, indicating that participants’ emotional states returned to a baseline level between experimental blocks (see Fig. 1).

Fig. 1
figure 1

Subjective mood and arousal ratings before and after music exposure. a Ratings of mood increased following presentation of music rated as positive and decreased following music rated as negative. b Subjective arousal ratings increased following highly arousing music and decreased following non-arousing, or calming, music. Error bars represent standard errors of the mean

Full size image

A 2 × 2 × 2 repeated measures ANOVA was conducted to assess the effect of the mood dimension of the selected music (positive/negative), arousal dimension (high/low), and time (before music/after music) on the mean mood VAS ratings. A main effect of the mood dimension was noted [F(1, 23) = 66.44, p < .05, η p 2 = .74], indicating that mood was rated as more positive in the positive mood conditions (M = 67.88, SD = 14.95) than in the negative mood conditions (M = 46.6, SD = 18.45). A significant interaction between the mood dimension of the music and time was also observed [F(1, 23) = 60.31, p < .001, η p 2 = .72], such that ratings of mood valence increased after exposure to music rated as positive, and decreased after music rated as negative (Fig. 1a). Importantly, we observed no main effect of the arousal dimension of the music on mood valence ratings [high arousal, M = 57.08, SD = 17.65; low arousal, M = 57.4, SD = 15.76; F(1, 23) = 0.02, p > .05, η p 2 = .001].

Mean arousal scores were then entered into a separate 2 × 2 × 2 repeated measures ANOVA involving the same factors. High-arousal conditions led to increased ratings of arousal, whilst low-arousal conditions resulted in lower ratings of arousal, as was evidenced by a main effect of arousal [F(1, 23) = 78.64, p < .05, η p 2 = .77]. A significant interaction between the arousal dimension of the music and time [F(1, 23) = 68.86, p < .001, η p 2 = .75] was also noted: Subjective arousal ratings increased following music rated as highly arousing, but decreased following music rated as less arousing (Fig. 1b). We found no main effect of the mood dimension of the music on arousal ratings [F(1, 23) = 0.27, p > .05, η p 2 = .012]. These results indicate that music listening successfully induced different mood valence and arousal states.

Recognition test

Recognition test performance was assessed in terms of three variables: sensitivity to old/new differences, mean reaction time, and mean confidence rating. Sensitivity (d') was calculated using signal detection theory, with a hit defined as an “old” response to a previously viewed shape and a false alarm defined as an “old” response to a novel shape (Stanislaw & Todorov, 1999). A 2 (mood: positive/negative) × 2 (arousal: high/low) ANOVA was used to investigate the effects of the emotional variables on sensitivity. The confidence ratings and reaction time data were assessed using a 2 × 2 × 2 ANOVA in which Trial Type (i.e., old/new) was added as an additional factor. Descriptive statistics for each of these measures may be found in Table 1.

Table 1 Mean values for proportions of hits, proportions of false alarms, d', reaction times (RTs), and confidence ratings across all four emotional conditions (standard deviations in parentheses)
Full size table

Initial analyses indicated no significant effect of mood valence [F(1, 23) = 0.005, p > .05, η p 2 = 0] or arousal [F(1, 23) = 0.12, p > .05, η p 2 = .005] on the measure of sensitivity to old/new differences (d'). Similarly, no significant effects emerged of mood valence [F(1, 23) = 0.045, p > .05, η p 2 = .002] or arousal [F(1, 23) = 0.31, p > .05, η p 2 = .013] on confidence ratings, though participants rated their confidence as higher when responding to old (M = 2.27, SD = 0.33) rather than new (M = 1.84, SD = 0.31) shapes, a difference that was statistically significant [F(1, 23) = 169.07, p < .001, η p 2 = .88]. In accordance with the pattern observed in the confidence ratings, the mean reaction time was found to be shorter in old trials than in new trials [F(1, 23) = 27.37, p < .001, η p 2 = .54]; however, no main effects of mood or arousal condition were observed [respectively, F(1, 23) = 0.78, p > .05, η p 2 = .033, and F(1, 23) = 1.67, p > .05, η p 2 = .068].

A hypothesis that the emotional induction effect may not have persisted throughout the recognition test period led us to reexamine the measures of sensitivity, confidence, and reaction time during the first half of the recognition test only. The numerical means of d' in the first half of the trials were similar to those observed in our original behavioral experiment (Greene et al., 2010), such that sensitivity appeared to increase during positive mood/high arousal and negative mood/low arousal conditions (top panel of Fig. 2), but the interaction was not statistically significant [F(1, 23) = 0.44, p > .05, η p 2 = .02]. No effect of emotional condition on confidence ratings was observed. A significant three-way interaction between mood valence, arousal, and trial type was, however, observed in the reaction time data during the first half of the recognition test [F(1, 23) = 8.865, p < .05, η p 2 = .28].

Fig. 2
figure 2

Influence of mood and arousal on recognition memory during the first half of the recognition test: (Top) mean sensitivity; (Bottom left) mean reaction times (RTs) during new trials; and (Bottom right) mean RTs during old trials. Error bars represent standard errors of the means. *Significant difference between conditions (p < .05)

Full size image

Further investigation of the Mood Valence × Arousal × Trial Type interaction within the first half of the trials indicated that the effects of mood and arousal on mean reaction times differed in the presence of new and old shapes. In order to test these effects in more detail, we performed post-hoc analysis by splitting this three-way interaction into two separate two-way interactions. This allowed us to assess the effects of a Mood Valence × Arousal interaction on mean reaction times independently for old and new shapes. No main effects of valence or arousal were observed, but we found a significant interaction between valence and arousal in the presence of new shapes [F(1, 23) = 4.32, p < .05, η p 2 = .16; see Fig. 2, bottom left].Footnote 2 Post-hoc paired-sample t tests indicated that negative mood led to speeded reactions relative to positive mood under conditions of high arousal [t(23) = –2.662, p < .05] but not of low arousal [t(23) = 0.395, p > .05]. The interaction effect in the presence of old shapes did not reach statistical significance (p = .277, Fig. 2, bottom right). This finding suggests that emotional effects on recognition memory performance in the present study may have depended on the relative novelty of the stimuli.

fMRI results

The presentation of old versus new trials (i.e., a contrast of old > new) during the recognition test activated a network of regions typically observed in memory retrieval studies, including the posterior parietal cortex (PPC), precuneus, and hippocampus (see Fig. 3). Further analyses assessed the effects of emotional state on old/new responses. Here we analyzed the effects of mood valence and arousal conditions on the contrast parameter estimates for old-versus-new differences.

Fig. 3
figure 3

Regions showing increased BOLD response during presentation of old shapes relative to new shapes in recognition test

Full size image

A main effect of mood valence on old/new responses was observed, such that the recognition memory response (i.e., BOLD response detected using a contrast of old > new) increased during positive mood conditions in the left frontal cortex, including the inferior frontal gyrus and orbitofrontal cortex, as well as bilateral PPC. The bilateral PPC clusters encompassed the angular gyrus, inferior parietal lobule, and supramarginal gyrus, just encroaching on the anterior intraparietal sulcus (see Table 2 for peak coordinates). No clusters displayed the opposite pattern of response (negative > positive).

Table 2 Peak coordinates of significantly activated clusters from the whole-brain analysis
Full size table

A main effect of arousal on old/new responses was observed in the left superior and posterior parietal cortices, covering the superior parietal lobule and the intraparietal sulcus and in a large medial cluster covering the posterior cingulate, the precuneus, and the visual cortex. These regions displayed increased response to old relative to new trials under conditions of high arousal.

We next tested for interactions between the mood and arousal variables. A cluster in the left PPC—specifically, the inferior parietal lobule and angular gyrus—displayed increased response during the positive mood/high arousal and negative mood/low arousal conditions, but reduced response during the positive mood/low arousal and negative mood/high arousal conditions (see Fig. 4). Paired-sample t tests indicated a significant difference in BOLD signals in the activated voxels between the positive mood/high arousal and negative mood/high arousal conditions [t(23) = 2.75, p < .05], though not between the positive mood/low arousal and negative mood/low arousal conditions. We also observed a significant difference between the negative mood/high arousal and negative mood/low arousal conditions [t(23) = –2.37, p < .05]. Note that this is the same interaction effect between mood valence and arousal that we had previously shown to influence recognition memory performance in our behavioral work (Greene et al., 2010). Both main effects and the interaction described here resulted in heightened recognition (old > new) response in the left PPC. The interaction effect was observed in a section of the PPC inferior and lateral to the regions displaying main effects of mood valence and arousal, which also encompassed the superior parietal lobule (see Fig. 4).

Fig. 4
figure 4

Main effects of mood condition (red voxels in electronic version of the figure), arousal condition (blue voxels in electronic version), and the interaction between the two variables (green voxels in electronic version). The top right panel depicts the relative locations of activations in left posterior parietal cortex (PPC) from the three contrasts. Arousal effects are observed to be superior to mood effects, which are in turn superior to interaction effects. The bottom right panel depicts parameter estimates from the interaction contrast, indicating that BOLD response in the inferior PPC increased with arousal during positive mood, but decreased with arousal during negative mood. *Significant difference between conditions, p < .05

Full size image

To further investigate the effects in the posterior parietal cortex, we conducted region-of-interest (ROI) analyses in dorsal and ventral regions of the PPC. The coordinates for the ROIs were taken from Ciaramelli, Grady, and Moscovitch’s (2008) review of the attention-to-memory hypothesis. The dorsal ROI was centered on the coordinates –36 –57 42 in the intraparietal sulcus, whereas the ventral ROI was centered on the coordinates –50 –57 38 in the supramarginal gyrus. These regions have been shown to underlie top-down and bottom-up attention to memory, respectively. Spherical masks with a radius of 10 mm were drawn around these coordinates. Mood valence and arousal effects on recognition memory (old trials > new trials) were investigated within the masked regions, and a small-volume cluster-based correction was applied (voxelwise z = 2.3, p < .05, corrected).

Within the dorsal PPC ROI, main effects of mood (positive > negative) and arousal (high > low) was observed; there was no significant interaction effect. A main effect of mood (positive > negative) was also observed in the ventral PPC ROI, but no main effect of arousal. An interaction effect was, however, observed in the ventral PPC, such that BOLD signal increased during positive mood states relative to negative mood states, but only under conditions of high arousal. Paired t tests confirmed this interpretation, indicating a significant difference in BOLD signal within the voxels displaying the interaction effect between the positive mood/high arousal and negative mood/high arousal conditions [t(23) = 2.69, p < .05], but not between the positive mood/low arousal and negative mood/low arousal conditions (see Fig. 5 and Table 3).

Fig. 5
figure 5

Results of the region-of-interest (ROI) analysis. a ROIs in dorsal and ventral posterior parietal cortex (PPC). In the electronic version, red voxels = main effect of mood valence, blue voxels = main effect of arousal, and green voxels = interaction effect. b Contrast parameter estimates from the dorsal PPC ROI. c Contrast parameter estimates from the ventral PPC ROI. *Significant difference between conditions, p < .05

Full size image
Table 3 Peak coordinates of significantly activated clusters from region-of-interest analysis in dorsal and ventral posterior parietal cortex (PPC)
Full size table

In order to investigate the hypothesis that the emotional induction may have worn off during the recognition test, we repeated our analysis examining the first and second halves of the experiment separately. Whole-brain analyses revealed that the main effects of mood valence and arousal were stronger and more widespread during the first half of the recognition test trials than in the analysis of the full experiment, but that they covered similar regions (see Table 4). This interaction effect was not initially observed during the first half of trials; however, a significant interaction effect was observed in the ventral PPC ROI, following small-volume correction. This suggests that the interaction effect, which was somewhat weaker than the main effects, could not be detected with the reduced statistical power caused by discarding half of the data. Analysis of the second half of the experiment revealed a main effect of arousal in the left dorsal PPC, including the postcentral gyrus, supramarginal gyrus, superior parietal lobule, and inferior parietal lobule. No main effect of mood valence or significant interaction effect was observed in the second half of the experiment, even when the analysis was restricted to ROIs in the PPC. We interpret these results as indicating that the induced level of arousal persisted longer than the changes in valence, which appear to have worn off by the second half of the experiment.

Table 4 Peak coordinates of significantly activated clusters during the first and second halves of the recognition test
Full size table

Discussion

The two dimensions of emotional state—mood valence and arousal—were independently manipulated, and their effects on the neural substrates of recognition memory were assessed. Distinct effects of valence, arousal, and their interaction were observed in a number of regions, most interestingly in the posterior parietal cortex. We begin this section by summarizing our findings with respect to the neural correlates of recognition memory, and then will elaborate on the effects of the emotional manipulation.

The expected neural response to previously viewed shapes was seen in a network of cortical and subcortical areas, including the PPC, precuneus, and hippocampus—regions that are traditionally associated with recognition memory. Increased neural response to old/new differences was observed under conditions of positive mood (relative to negative mood) in bilateral PPC, and, interestingly, in the left orbitofrontal cortex. The left orbitofrontal cortex has been reported to be involved in reward and pleasure, memory formation, and retrieval of emotionally valenced information (Grabenhorst & Rolls, 2011; Petrides, 2007; Shigemune et al., 2010; Smith, Henson, Dolan, & Rugg, 2004; Tsukiura & Cabeza, 2011; Young & Shapiro, 2011). Our results indicate that the OFC (Brodmann’s area 47) may benefit processing of previously experienced neutral information that is subsequently reencountered under positive mood states. A main effect of arousal on old/new responses was also evident, with increased responses under conditions of high arousal in bilateral posterior cingulate gyrus, left superior and posterior parietal cortex, and the precuneus. The posterior cingulate cortex is tightly linked with arousal systems in the brainstem and has been posited to play a key role in the relationship between levels of alertness and levels of awareness (Vogt & Laureys, 2005). In addition, the posterior cingulate cortex and precuneus are well-established nodes within a broader parietal network for the control of memory retrieval (Wagner et al., 2005).

Interestingly, neural response in the PPC during memory retrieval was modulated by both mood valence and arousal, as well as by the interaction of those two factors. Although there was considerable overlap between the valence and arousal activations—specifically, in the inferior parietal lobule and around the intraparietal sulcus—the two main effects show a clear dorsal/ventral distinction. Arousal condition modulated recognition memory responses in regions of the dorsal PPC including the intraparietal sulcus and superior parietal lobe. In contrast, variation in mood valence (positive/negative) modulated recognition memory response in ventral PPC, in the angular gyrus and inferior parietal lobe. The interaction between valence and arousal condition activated a PPC cluster that was both ventral and lateral to the regions modulated by mood or arousal alone. These findings demonstrate that the neural activity underlying recognition memory retrieval is sensitive to relatively small fluctuations in both mood valence and arousal that also influence memory performance. Some previous studies have described an effect of stimulus valence on activity in the inferior parietal lobule (Kensinger & Corkin, 2004; Mickley Steinmetz & Kensinger, 2009), corresponding to the ventral PPC activations reported here, but these studies assessed the influence of the emotional characteristics of the items on memory encoding. To our knowledge, no previous studies have investigated the influence of arousal on parietal contributions to postencoding memory processes (i.e., retrieval) for otherwise neutral items.

Despite its frequent appearance in memory retrieval studies, it has been suggested that the PPC’s role is not mnemonic, but rather relates to the salience or “target value” of the recognized items (Vilberg & Rugg, 2008). According to this theory, the PPC may play a role in categorizing items according to their relevance to ongoing tasks. The attention-to-memory hypothesis (Cabeza et al., 2008; Ciaramelli et al., 2010) proposes a functional segregation of the PPC for this purpose, positing that dorsal PPC regions (including the superior parietal lobule and intraparietal sulcus) underlie the top-down, voluntary allocation of attention to memory, whereas ventral PPC (including the supramarginal gyrus and angular gyrus) is recruited during bottom-up attentional processes, such as those involved in the automatic guidance of attention to memory contents.

To investigate the parietal attention-to-memory hypothesis in the context of our emotional manipulation, we conducted ROI analyses in dorsal and ventral regions of the PPC that have been previously shown to underlie top-down and bottom-up attention to memory, respectively (Ciaramelli et al., 2008). These analyses confirmed that arousal effects were observed only in the dorsal PPC, whereas effects of mood valence were observed both dorsally and ventrally. The interaction between mood valence and arousal was only observed in the ventral ROI. Together, these findings demonstrate the parcellation of posterior parietal areas for recognition memory as a function of emotional state.

The finding that increased levels of arousal during recognition memory recruit dorsal posterior parietal regions suggests that arousal exerts top-down control over recognition memory, perhaps directing more attentional resources toward the memory task. This thesis is supported by research demonstrating that similar dorsal parietal regions are recruited when participants voluntarily shift their attention during performance of a memory task (Tamber-Rosenau, Esterman, Chiu, & Yantis, 2011), and by recent work showing that emotional arousal (irrespective of mood valence) enhances that top-down prioritization of goal-relevant memory traces (Sakaki, Fryer, & Mather, 2014). The effects of mood valence may be more passive; positive emotion has been associated with a broadening of spatial attention and increased susceptibility to peripheral stimuli and distractors (Basso et al., 1996; Gasper & Clore, 2002; Jefferies et al., 2008; Vanlessen, Rossi, De Raedt, & Pourtois, 2013). The ventral parietal regions that are reported here as being modulated by mood valence have previously been linked with bottom-up attentional processes during memory tasks (Cabeza et al., 2008; Ciaramelli et al., 2010), whereby attention is automatically captured by old stimuli, leading to retrieval of those items from memory. That recognition responses in ventral PPC were modulated by an interaction between mood valence and arousal is noteworthy, and indicates that response in ventral parietal regions is not merely modulated by changes in arousal, but is also dependent on the specific mood state experienced by the participants. In particular, the effect of mood on recognition responses in ventral PPC was enhanced under relatively high (vs. low) levels of arousal. On these grounds, future studies ought to avoid conflating levels of positive and negative arousal while controlling for the valence dimension.

The behavioral data reported here indicate an interaction between mood valence and arousal on recognition performance, such that the identification of novel items was faster during the negative mood/high arousal and positive mood/low arousal conditions. We also observed a nonsignificant trend toward increased sensitivity to old/new differences during the positive mood/high arousal and negative mood/low arousal conditions. The pattern of results described here is consonant with the emotional effects on recognition performance that we observed in our original behavioral study (Greene et al., 2010). The effects reported here were, however, weaker than those described in that study, and only manifested in the first half of the experiment, suggesting that the behavioral effects of the emotional induction wore off relatively quickly. The fMRI analyses supported this hypothesis, showing that the neural effects of emotion on recognition memory were stronger in the first half of the experiment. Despite the weak behavioral effects, the pattern of subjective ratings of mood valence and arousal indicated that the desired emotional states were successfully induced, and we observed clear effects of the emotional conditions on BOLD activation associated with memory recognition.

In sum, we describe for the first time a neural circuitry involving the PPC and orbitofrontal cortex that is sensitive to the modulation of memory responses by emotion. This novel delineation of distinct and overlapping networks mediating the effect of mood valence and arousal on the recognition of emotionally neutral stimuli indicates that the neural substrates of human memory can be fractionated on the basis of internal emotional state.