We determined the sample size based on previous studies that assessed the effects of aging on emotional memory encoding, with n = 30–37 participants (Kensinger and Schacter, 2008; St Jacques et al., 2009; Addis et al., 2010). We screened 70 volunteers from the greater New Haven, Connecticut area. The inclusion/exclusion criteria included (1) all participants needed to be physically healthy, with no major medical conditions and score ≥ 27 in Mini Mental State Examination (MMSE); (2) those with current use of prescription medications or with a history of head injury or neurological illness were excluded; (3) those who met criteria of current or history of Axis I disorders according to the Structured Clinical Interview for DSM-IV (First et al., 1996) or reported current use of illicit substances or tested positive for cocaine, methamphetamine, opioids, marijuana, barbiturates, or benzodiazepines were excluded. Post-screening, we invited 62 volunteers for study participation. Among the 62, 2 were excluded because of MMSE score < 27, suggesting cognitive impairment (Salis et al., 2023), 4 could not complete the MR scan, and 5 with insufficient number of events in each fMRI conditions (see Section 2.4). Thus, we presented the data from 51 healthy adults (23 women) 22 to 80 years of age (Table 1). These participants were assessed with the State–Trait Anxiety Inventory (STAI) (Spielberger, 1989). The STAI state score ranged from 20 to 63 with a mean ± SD of 32.80 ± 10.54 in the current sample. Men and women did not differ in STAI state score (t = 1.33, p = 0.189), and differed marginally in age (t = 2.00, p = 0.050). The Human Investigation Committee at Yale School of Medicine approved the study procedures. All participants signed an informed consent prior to the study.

We scanned participants during encoding of negative emotional and neutral pictures selected from the International Affective Picture System (IAPS) image set (Figure 1). Prior to MRI and again at the beginning of task during MRI, participants were instructed to remember the pictures and that they would be tested for how well they did after the scan. Forty negative emotional (normative valence = 2.72 ± 0.83; normative arousal = 5.67 ± 0.73) and 40 neutral (normative valence = 5.26 ± 0.89; normative arousal = 3.83 ± 0.83) pictures were presented in pseudorandomized order each for 4 s, with an intertrial interval (blank screen) 8 to 16 s in duration to provide jittering of the stimuli. The total task duration was approximately 8 min. We presented the task in two sessions, each with a different set of pictures.

The task was presented using a Dell Desktop (Windows 10; NVIDIA GeForce GT 1030, resolution: 1024 × 768; active signal resolution: 1024 × 768; refresh rate: 60 Hz) with Presentation Software (version 22.1 Build 04.30.21 by Neurobehavioral Systems, Inc.). The stimuli were projected onto an Optoma projector positioned at the front of the magnet bore opening, approximately 150 cm from the subject’s eyes. Subjects viewed this display through an inverted mirror placed on the head coil, positioned directly above their eyes.

Following the MR scan, participants were presented a picture one at a time from the entire set of pictures shown, interleaved with an equal number of new pictures (new negative picture normative valence = 2.72 ± 0.83, new negative picture normative arousal = 5.67 ± 0.73; new neutral picture normative valence = 5.34 ± 1.11, new neutral picture normative arousal = 3.86 ± 0.82) never shown during the MR scan, and queried whether the pictures were presented before. Participants answered yes or no to each picture.

According to signal detection theory (SDT) (Yu et al., 2018), the four possible outcomes for each trial/picture were: hits, misses, false alarm (FA), and correct rejection (CR). A hit and miss each represent an accurate and failed recognition that the picture was presented during the MR scan. An FA is false recognition of a picture never presented as presented and CR is correct recognition that such a picture was not presented. We computed the probability of hit rate (HR) and false alarm rate (FAR) as follows: P(HR) = n(hit)/[n(hit) + n(miss)]; P(FAR) = n(FA)/[n(FA) + n(CR)]. The sensitivity (d’) and bias were calculated as d’ = Z_HR − Z_FAR; bias = Z_FAR, where Z_HR and Z_FAR are the Z-score transforms, based on the normal distribution, of the P(HR) and P(FAR). We computed the Z-scores on the basis of negative and neutral trials combined, so the d’ and bias (Z_FAR) could be contrasted between negative and neutral trials in statistical analyses.

Brain images were collected using multiband imaging with a 3-Tesla MR scanner (Siemens Trio, Erlangen, Germany). Conventional T1-weighted spin echo sagittal anatomical images were acquired for slice localization. Anatomical 3D MPRAGE image were next obtained with spin echo imaging in the axial plane parallel to the AC–PC line with TR = 1900 ms, TE = 2.52 ms, bandwidth = 170 Hz/pixel, field of view = 250 × 250 mm, matrix = 256 × 256, 176 slices with slice thickness = 1 mm and no gap. Functional, blood oxygen level-dependent (BOLD) signals were acquired with a single-shot gradient echoplanar imaging (EPI) sequence in two-sessions. Fifty-one axial slices parallel to the AC–PC line covering the whole brain were acquired with TR = 1,000 ms, TE = 30 ms, bandwidth = 2,290 Hz/pixel, flip angle = 62°, field of view = 210 × 210 mm, matrix = 84 × 84, 51 slices with slice thickness = 2.5 mm and no gap, 480 volumes, and multiband acceleration factor = 3. Images from the first ten TRs at the beginning of each scan were discarded to ensure that only BOLD signals in steady-state equilibrium between RF pulsing and relaxation were included in data analyses.

Data were analyzed with Statistical Parametric Mapping (SPM12, Welcome Department of Imaging Neuroscience, University College London, United States), following our published routines (Chaudhary et al., 2023b). Images of each individual subject were first realigned (motion corrected) and corrected for slice timing. A mean functional image volume was constructed for each subject per run from the realigned image volumes. These mean images were co-registered with the high-resolution structural image and segmented for normalization with affine registration followed by nonlinear transformation. The normalization parameters determined for the structure volume were then applied to the corresponding functional image volumes for each subject. The resampled voxel size is 2.5 × 2.5 × 2.5 mm³. Finally, the images were smoothed with a Gaussian kernel of 8 mm at Full Width at Half Maximum.

We separated the negative and neutral trials according to the answers – “yes, I have seen the image” or “no, I have not seen the image” – to post-scan queries and categorized trials as negative-correct, negative-incorrect, neutral-correct, and neutral-incorrect. A statistical analytical design was constructed for each individual subject, using a general linear model (GLM) with the onsets of the event for each trial convolved with a canonical hemodynamic response function (HRF) and with the temporal derivatives of the canonical HRF and entered as regressors in the model (Friston et al., 1994). Realignment parameters in all six dimensions were also entered in the model. Serial autocorrelation caused by aliased cardiovascular and respiratory effects was corrected by a first-degree autoregressive or AR (1) model. The GLM estimated the component of variance explained by each of the regressors.

Following previous studies of memory encoding (Kensinger and Schacter, 2008; Spaniol et al., 2009; St Jacques et al., 2009; Fischer et al., 2010; Clemens et al., 2015; Dennis and Turney, 2018), we contrasted (negative-correct – negative-incorrect) versus (neutral-correct – neutral-incorrect) to identify the neural correlates specific to successful encoding of negative versus neutral emotional images. This contrast allows assessment of a direct association between brain-activity and performance specific to negative emotion memory (Dolcos et al., 2012). All subjects had 7 or more events in each of the 4 conditions (negative-correct, negative-incorrect, neutral-correct – neutral-incorrect), ensuring a sufficient number of events in each condition to observe a significant effect (Fischer et al., 2010).

To identify age-related and anxiety-related correlates, we performed a whole-brain linear regression of the contrast “(negative-correct – negative-incorrect) – (neutral-correct – neutral-incorrect)” on age and STAI state anxiety score, respectively. We evaluated the results at voxel p < 0.005, uncorrected in combination with cluster p < 0.05, FWE-corrected. For the clusters that showed a significant correlation with age, we checked for the correlation of their activities (β’s) with STAI state score. Likewise, for the clusters that showed a significant correlation with STAI state score, we checked for the correlation of their β’s with age.

For the clusters with activities of “(negative-correct – negative-incorrect) – (neutral-correct – neutral-incorrect)” showing a significant correlation with both age and STAI state score, we performed mediation analyses to characterize the inter-relationships of these variables (See Results). We performed mediation analyses as with our previous studies (Chaudhary et al., 2023a,b) and described in the Supplemental methods. The analysis was performed with package ‘Medsem’ (Mehmetoglu, 2018) in Stata: Statistical software for data science (StataCorp LLC).

The mean ± standard deviation (SD) of hit rate (HR), false alarm rate (FAR), sensitivity d’, and bias Z(FAR) are shown in Figure 2A. Participants showed higher bias or Z(FAR) (t = 2.43, p = 0.019, paired t-test) and indistinguishable sensitivity d’ (t = 1.05, p = 0.296, paired t test) in recognizing negative versus neutral emotional images.

Across subjects, the differences in Z(FAR) [negative Z(FAR) – neutral Z(FAR); r = 0.31, p = 0.025] and in d’ (negative d’ – neutral d’; r = −0.44, p = 0.001) in recalling negative versus neutral images were both significantly correlated with age, with sex as a covariate (Figure 2B). Thus, age was associated with higher bias but lower sensitivity in recognizing negative relative to neutral images.

Participants showed a STAI state score of 32.8 ± 10.5 (mean ± SD). STAI state score was significantly and negatively correlated with age, with sex as a covariate (r = −0.52, p < 0.001). STAI state score was not significantly correlated with the difference in d’ (negative d’ – neutral d’; r = 0.15, p = 0.309) or with the difference in Z(FAR) [negative Z(FAR) − negative Z(FAR); r = −0.11, p = 0.452] in a linear regression with sex as a covariate (Figure 2B). Thus, individual variation in state anxiety was not significantly associated with the bias or sensitivity in recognizing negative emotional relative to neutral images.

To identify the correlates specific to negative emotional memory, we contrasted “negative-correct – negative-incorrect” with “neutral-correct – neutral-incorrect” and showed the regional activities in Figure 3A. Two clusters each of the left insula/inferior precentral gyrus/superior temporal gyrus (INS/iPrCG/STG; x = −43, y = −1, z = 18, Z = 3.95, 335 voxels) and right superior temporal gyrus/temporo-parietal junction (STG/TPJ; x = 43, y = −34, z = 6, Z = 3.86, 424 voxels) showed lower activation during correctly versus incorrectly recognized emotional images, as compared to the same of neutral images. At the same threshold, no regional activities were significant for the reverse contrast. The mean brain activations for all relevant contrasts are presented in the Supplementary Figures S1, S2 and Supplementary Tables S1, S2.

To identify age-related correlates of negative emotional memory, we performed a whole brain regression of the contrast “(negative_correct - negative_incorrect) – (neutral_correct - neutral_incorrect)” on age, with sex as a covariate, and showed the regional activities in Figure 3B. Only the cluster that included the superior parietal cortex and precuneus (SPC/PCu) met cluster p < 0.05, FWE-corrected (x = 3, y = −59, z = 53, Z = 3.72, 437 voxels). We extracted the β value of “(negative_correct - negative_incorrect) – (neutral_correct - neutral_incorrect)” of the SPC/PCu cluster for testing in mediation analysis, and correlation with clinical variables. As expected, the β value of right SPC/PCu was significantly and negatively correlated with age, with sex as a covariate (r = −0.49, p < 0.001; Figure 4A). The β value of the SPC/PCu cluster was also significantly and positively correlated with anxiety score (r = 0.33, p = 0.018; Figure 4B), and with d’(negative-neutral) (r = 0.35, p = 0.013), with sex as a covariate. We did not test for correlation with the reporting bias Z(FAR) as these did not represent neural activities during memory encoding.

To identify anxiety-related correlates of negative emotional memory, we performed a whole brain regression of the difference in sensitivity “(negative_correct - negative_incorrect) – (neutral_correct - neutral_incorrect)” on STAI score, with sex as a covariate, and showed the regional activities in Figure 3C. Four clusters met cluster p < 0.05, FWE-corrected: the left parahippocampal gyrus and part of the cerebellar cortex (x = −23, y = −56, z = −12, Z = 4.15, 1,541 voxels), right middle/superior temporal gyri and temporo-parietal junction (MTG/STG/TPJ; x = 58, y = −44, z = 1, Z = 4.19, 626 voxels), cuneus and precuneus (x = −3, y = −79, z = 48, Z = 3.83, 472 voxels), and the supplementary and pre-supplementary motor areas (x = −13, y = 12, z = 51, Z = 4.00, 418 voxels). Of these four clusters, only the right MTG/STG/TPJ showed a β contrast in significant correlation with age, with sex as a covariate (r = −0.33, p = 0.018; p’s > 0.07 for other clusters) (Figure 4C). Right MTG/STG/TPJ β (r = 0.32, p = 0.022) also correlated positively with d’ (negative-neutral) with sex as a covariate.

As presented in the previous section, we observed 3-way pair-wise correlations among age, STAI score, and the β “(negative_correct - neutral_correct) – (negative_incorrect - neutral_incorrect)” of the SPC/PCu identified from age regression (Figures 4A,B) and of the right MTG/STG/TPJ identified from STAI score regression (Figures 4C,D). We ran mediation analyses with sex as a covariate to further investigate the inter-relationship amongst these clinical and neural measures. For each cluster, we tested all six models and evaluated the results at a corrected p < 0.05/6 = 0.0083. It’s important to note that, while all six mediation models were tested, our focus was specifically on models with age as the independent variable. For SPC/PCu, a significant mediation was noted in model β → age → anxiety (Figure 4E), whereas for the right MTG/STG/TPJ, mediation analyses showed a significant model with age → anxiety → β value (Figure 4F), both with sex as a covariate. The statistics of all other models are shown in Supplementary Tables S3A,B. If we considered solely models where age served as an independent variable, only the model age → anxiety → MTG/STG/TPJ β was significant.

Negative emotions can enhance both the accuracy (Kensinger, 2007; Mather and Sutherland, 2011) and/or subjective feelings of vividness or confidence for retrieved memory (Dougal and Rotello, 2007; Phelps and Sharot, 2008). Previous studies observed higher sensitivity in recalling negative versus neutral images (Anderson et al., 2006; Rimmele et al., 2011; Xie and Zhang, 2017; Yeh et al., 2021). Here, we did not observe a significant difference in sensitivity in recalling negative versus neutral images. On the other hand, we observed higher reporting bias in recalling negative versus neutral images. Indeed, while recollection was more accurate for negative than neutral scenes (Anderson et al., 2006; Yeh et al., 2021), Rimmele et al. noted enhanced subjective feeling of remembering negative versus neutral scenes without actually remembering the scene details (Rimmele et al., 2011). The other study noted item memory as enhanced under negative versus neutral conditions was also associated with elevated subjective feelings of remembering the conditions (Xie and Zhang, 2017). Together, these findings suggest the critical importance of accounting for reporting bias in evaluating the accuracy of recall.

We noted age-associated reduction in sensitivity d’ in recognizing negative emotions, consistent with some (Charles et al., 2003; Leigland et al., 2004; Kensinger and Schacter, 2008; Leal et al., 2015; Stam et al., 2022) but not other (Kensinger et al., 2002, 2007a,b; Fischer et al., 2010) studies. Notably, Fischer et al., 2010 did not correct for bias when evaluating response accuracy. With bias accounted for, the memory performance would trend similarly to our findings (Hit-false alarm rate for negative emotional (fearful face): young = 0.35 ± 0.15, old = 0.24 ± 0.19; and for neutral face: young = 0.26 ± 0.19, old = 0.17 ± 0.14) (Fischer et al., 2010), suggesting an increase in false alarms with age (as we discuss next). In addition to reporting bias, a number of other factors may influence age-related differences in encoding accuracy. For instance, Kensinger et al., noted that in studies where participants passively viewed but were not instructed to memorize the pictures, young and older adults performed equally in memory (Kensinger et al., 2002, 2007a). However, in studies that required encoding, younger adults performed better than older adults (Kensinger et al., 2007b). Thus, the differences in study design and the role of active versus passive engagement in encoding may contribute to the observed discrepancies in age-related memory performance across studies. Further, although age was significantly associated with lower levels of anxiety, STAI scores were not significantly correlated with d’. This finding suggests that age-related reduction in negative emotional memory is unlikely to account for age-related reduction in anxiety. Together with the earlier reports, the present findings are consistent with age-related positivity in emotional memory, which, however, cannot explain the changes in state anxiety in this sample.

We found that age was associated with higher reporting bias during recall of negative versus neutral images. Compared to young adults, older adults have objectively less accurate event memory, but over-estimated event vividness (Folville et al., 2020, 2022). Age-related cognitive changes may have imposed on older people to rely more on familiarity-based monitoring, which along with reduced inhibition of irrelevant information, induces false memories (Devitt and Schacter, 2016). However, how emotion affects such an age-related tendency toward false memories is still not clear. Gallo et al., noted false memory to be higher in response to emotional than to neutral pictures, with older versus young adults showing higher frequency of false memory of emotional pictures (Gallo et al., 2009). In another study with emotional/neutral words presented during encoding and new emotional/neutral lure words during retrieval, both young and older adults performed equally and showed less false memory of emotional versus neutral words (Kensinger and Corkin, 2004). A possible explanation for the discrepant findings concerns the stimuli, with studies using images but not those using words showing an age effect, although depending on the context in which lure words were presented, word stimuli could elicit false memories (Garry and Wade, 2005; Smith and Hunt, 2020).

We did not observe a significant association between sensitivity or reporting bias with anxiety, in-line with earlier reports that did not find any differences in recall in high versus low anxiety individuals (Wolfensberger et al., 2008; van Tol et al., 2012), and in contrast with other reports of better recall of negative versus neutral emotional stimuli in high anxiety individuals (Reidy and Richards, 1997; Lee and Fernandes, 2018; Yu et al., 2018; Hakamata et al., 2022; note that only Lee and Fearnandes and Yu et al. controlled for reporting bias in evaluating accuracy in recall). Exposure to stress prior to encoding enhances encoding of emotional versus neutral stimuli (Payne et al., 2007; Hoscheidt et al., 2014). On the other hand, negative emotional states can disrupt memories; a recent meta-analysis showed that negative emotions elevated recall failures in a working memory task and that the effects were moderated by participants’ self-reported severity of negative life experiences (Xie et al., 2022). Other studies with image pairs-negative and neutral images presented in pairs of both neutral, both negative, or mixed; noted varied impacts of context on memory (Kensinger and Schacter, 2006; Bisby et al., 2016; Ritchey et al., 2019). While we cannot currently explain the absence of a behavioral association between anxiety and memory, future studies may explore these explanations. Further, despite the lack of a behavioral association, we observed a correlation between regional encoding of negative emotion and anxiety, consistent with previous studies (Wolfensberger et al., 2008; van Tol et al., 2012). This is probably because encoding may be more sensitive to anxiety, while retrieval may share a complex relationship with anxiety (Gagnon and Wagner, 2016).

Age and anxiety correlated negatively and positively, respectively, with right MTG/STG/TPJ activity during “(negative correct > incorrect) – (neutral correct > incorrect)” trials, and anxiety mediated the association between age and MTG/STG/TPJ activity. The MTG has been implicated in successful encoding of emotional memory (Murty et al., 2010; Dahlgren et al., 2020; Riegel et al., 2022). The STG is involved in social cognition (Allison et al., 2000) and emotion processing (Tseng et al., 2016; Liu et al., 2022), and the TPJ in differentiating old and new memory in memory updating (Simon et al., 2017). Previous studies have noted either no age-related changes (Kensinger and Schacter, 2008) or reduced MTG (hippocampal) activity (St Jacques et al., 2009; Fischer et al., 2010) during correct versus incorrect negative versus neutral encoding in older adults. Our findings are consistent with the study that used scenes (St Jacques et al., 2009), but not those that used objects (Kensinger and Schacter, 2008; Addis et al., 2010). This discrepancy may be due to different stimuli engaging different brain regions during processing (Preckel et al., 2019). In anxiety disorders, previous literature supported structural/functional deficits of the temporal lobe. For instance, relative to healthy participants, individuals (15–55 years age) with anxiety disorders demonstrated lower left MTG/STG volumes (van Tol et al., 2010). Another study of young adolescents noted greater right MTG volume in individuals scoring higher in social anxiety (Wang et al., 2021). A meta-analysis of structural and resting-state functional connectivity identified altered amygdala-MTG/STG connectivity in anxiety disorders (Cao et al., 2021). Exposure to stressful versus neutral stimuli led to higher magnetoencephalographic activity in the medial temporal lobe, especially during memory encoding (Heinbockel et al., 2021). Collectively, diminished anxiety with age may dampen negative emotion encoding in MTG.

Age and anxiety also correlated negatively and positively, respectively, with SPC/PCu activity in the contrast of “(negative correct > incorrect) – (neutral correct > incorrect)” trials. Along with the MTG, the parietal cortex is implicated in threat detection and encoding as well as hypervigilance and excessive monitoring of environment in the pathophysiology of anxiety (Carlsson et al., 2006). Higher SPC activity during anticipation of an unpredictable aversive event was associated with subjective anxiety scores in healthy participants (Hasler et al., 2007). Individuals with social anxiety disorder demonstrated higher SPC activity in processing threatening versus neutral visual scenes (Goldin et al., 2009; Heitmann et al., 2016). As the hub of default mode network and in support of self-reflection and self-focused thinking (Cavanna and Trimble, 2006), the precuneus is also implicated in anxiety. The precuneus showed elevated amplitude of low-frequency fluctuations in individuals with social anxiety as compared to controls (Yuan et al., 2018). An earlier meta-analysis noted precuneus hyperactivity during processing of negative versus neutral (and positive) emotional stimuli in individuals with post-traumatic stress disorder (Etkin and Wager, 2007). Importantly, a broad literature highlighted age-related changes in precuneus activity in a wide range of behavioral tasks. For instance, while watching a video and responding to each meaningful event, which may be important for episodic memory retrieval, the PCu amongst other posterior medial brain regions showed higher activity and this increase in activity became less prominent with age (Reagh et al., 2020). Our findings are thus consistent with this literature. Further, the SPC/Pcu activity was positively correlated with both anxiety score and negative versus neutral d’ (sensitivity), though mediation models failed to substantiate the relationship amongst [age, SPC/PCu β, and d’] (data not shown) or amongst [age, SPC/PCu, and anxiety] (model in support of SPC/PCu β → age → anxiety; Figure 4E). A larger sample size is needed to revisit this issue.

Averaging across all participants, we observed lower left insula/iPrCG/STG and right STG/TPJ responses to successfully versus unsuccessfully remembered negative versus neutral images. The insula supports visceral/interoceptive, sensorimotor, and homeostatic processing as well as emotional awareness and response selection (Nieuwenhuys, 2012). In the context of memory encoding, a recent animal study noted enhanced consolidation of object memory by the basolateral amygdala through suppression of insular cortical activity, suggesting a possible role of the insula in allocating resources for memory (Chen et al., 2022). The STG is involved in social cognition (Allison et al., 2000) and emotion processing (Tseng et al., 2016; Liu et al., 2022), and the TPJ in differentiating old and new memory in memory updating (Simon et al., 2017). A previous study noted greater STG activation during successful versus unsuccessful encoding of emotional (negative + positive) versus neutral images (Mickley Steinmetz and Kensinger, 2009). A meta-analysis observed enhanced STG activity during the encoding and PrCG activity during the retrieval of emotional items or neutral items encoded in an emotional context (Dahlgren et al., 2020). In our study, we observed an overall reduction in activity, which may be attributed to the inclusion of subjects across the lifespan. Notably, a previous study that similarly encompassed subjects across the lifespan reported reduced activity in the PrCG and STG during the processing of emotional compared to neutral items (Daley et al., 2020). Nevertheless, further investigation of this issue is warranted.

A number of limitations need to be considered. First, we focused here only on negative and neutral emotional memory. Investigating the neural processes of positive emotion processing will better our understanding of the mechanisms of “positivity effect” and age- related reduction in anxiety. It is also important to note that, whereas the great majority of studies employed the contrast “negative correct - neutral correct” to identify the correlates of memory, we considered incorrect trials in the contrast to control for the perceptual processes, as with several earlier studies (Kensinger and Schacter, 2008; Spaniol et al., 2009; St Jacques et al., 2009; Fischer et al., 2010; Clemens et al., 2015; Dennis and Turney, 2018). The specific contrasts used would dictate the results and may account for the discrepancy between the current and some of the previous findings. Second, we did not scan subjects during the retrieval phase. Previous studies have shown common and distinct regions involved in encoding and retrieval (Prince et al., 2005; Dahlgren et al., 2020), both of which may be altered by age (Srokova et al., 2022) and anxiety (Wolfensberger et al., 2008). Imaging data collected during recall would help in elucidating the neural correlates of reporting bias and how neural processes of encoding and recall may be related. Third, the STAI state score ranged from 20 to 63 in the current sample. Participants with higher state score are needed to better understand the effects of anxiety as well as the impact of age on anxiety, with the caveat that most people with the highest state scores have a clinical diagnosis of anxiety and depressive disorders and typically receive mediations for their conditions. Our sample size is moderate; studies of a larger sample may allow us to investigate the mediation models with an adequate power and sex differences in the roles of emotional memory in supporting age-related reduction in anxiety. Finally, not all older adults display reduced anxiety. Increased anxiety is evident in older adults with subjective cognitive decline (SCD) (Liew, 2020), mild cognitive impairment (MCI) (Devier et al., 2009; Li et al., 2014; Klimova et al., 2015), and Alzheimer’s Disease (AD) (Devier et al., 2009). Along with anxiety, these cohorts also display emotion processing dysfunction. For instance, individuals with AD display impaired identification of negative facial emotions, inconsistent emotional memory benefit (Brueckner and Moritz, 2009; Klein-Koerkamp et al., 2012; Mistridis et al., 2013; Cárdenas et al., 2021), and inconsistent positivity effect (Goodkind et al., 2015; Zhang et al., 2015). People with MCI exhibit impaired negative emotion recognition in people with MCI (Elferink et al., 2015; Morellini et al., 2022); and those with SCD show altered facial emotion processing(Perez et al., 2022) and emotion dysregulation (Mah et al., 2021). Further, anxiety represents a risk factor of cognitive decline in the elderly (Pietrzak et al., 2012; Santabárbara et al., 2020; Gracia-García et al., 2023). Thus, impaired emotion processing and anxiety may be linked to early cognitive changes. Our study cohort includes typically healthy, cognitively intact participants without severe anxiety. Thus, the findings represent healthy aging. Future studies should target older adults with high anxiety and cognitive impairments.

In conclusion, age is associated with lower levels of anxiety and reduced sensitivity but higher reporting bias in recognizing negative emotions. The right middle temporal gyrus/temporal parietal junction/superior temporal gyrus (MTG/TPJ/STG) show activity (β) during “negative correct – incorrect” – “neutral correct – incorrect” trials each in positive and negative correlation with anxiety and with age. Mediation analyses supported a complete mediation model of age → less anxiety → less MTG/TPJ/STG. Together, the findings demonstrate age-related changes in negative emotional memory and how age-related reduction in anxiety is reflected in diminished temporal–parietal cortical activities during encoding of negative emotional memory.