Each day we are exposed to a plethora of complex visual scenes and events, yet only a minute portion of these can be successfully retrieved from memory after a long time. Various studies have demonstrated that emotionally arousing incidents and stimuli are much better remembered than those without emotional relevance (see McGaugh, 2004; Holland and Kensinger, 2013; Dolcos and Denkova, 2014 for a review). Importantly, these general findings have been replicated using various methodological approaches and experimental materials, mainly pictures and words (e.g., Bradley et al., 1992; Cahill and McGaugh, 1995; Ochsner, 2000; Kensinger and Corkin, 2003; Dolcos et al., 2004). Recent psychological and neuroscientific studies have also demonstrated that emotion influences not only the efficacy of memory, but various aspects of other cognitive processes, such as perception, attention, or decision-making (see Brosch et al., 2013; Dolcos et al., 2014 for a review). However, recognition memory may be significantly influenced by age-related differences in the processing of emotionally valenced information which have been recently reported (Kalpouzos et al., 2012; Reed et al., 2014; Ziaei and Fischer, 2016 for review).

It is interesting that only several long-term memory experiments have been conducted using a lengthy retention period. In one of the first studies investigating memory after around 1 year, Bradley et al. (1992) discovered a significantly better memory performance for both negative (low-valence), and positive (high-valence) highly arousing stimuli. In a functional resonance imaging (fMRI) study on female subjects after 1-year delay, Dolcos et al. (2005) found a significant improvement of recognition memory for emotional pictures in comparison to neutral ones. Analyses based on the regions of interest (ROI) revealed that successful retrieval of emotional pictures elicited higher activation in the amygdala, enthorinal cortex, and hippocampus. More recently, Weymar et al. (2011) have used behavioral and event-related potential (ERP) measures in passive viewing paradigm with a recognition test conducted 1 week and 1 year after encoding. The results obtained in a group of males showed a sustained enhanced recognition performance for emotionally unpleasant and pleasant pictures as compared to the neutral ones on both retention periods. On the neuronal level, boosted memory for unpleasant and pleasant stimuli after 1 week was related to an enhanced parietal ERP old/new effect. However, after 1 year, these neurophysiological effects were observed only for unpleasant, and not for pleasant, arousing pictures. In an fMRI study conducted on a group of males and females with intentional learning paradigm, valence was found to modulate brain responses after 3 days and 3 months (Harand et al., 2012). This study showed that hippocampal activations can decrease over time and the reorganization of memory traces takes place with additional activation in the ventromedial prefrontal and anterior cingulate cortex.

However, the effect of emotion on long-term recognition memory was only investigated in line with the dimensional models of affect, which describe emotions in a two-dimensional space consisting of valence (pleasantness or unpleasantness) and arousal (level of excitation (Osgood et al., 1957; Russell, 1980). These models are not the only way to conceptualize emotion. Another theoretical framework is Ekman's (1992) theory of basic emotions, according to which at least five different discrete emotion categories universally reflect facial and vocal expression, namely: happiness, sadness, anger, fear, and disgust. These two points of view are not mutually exclusive, having considerable explanatory value (Reisenzein, 1994) and at least two databases of emotionally charged images—International Affective Picture System (IAPS, Libkuman et al., 2007) and Nencki Affective Picture System (NAPS, Marchewka et al, 2014a; Riegel et al., 2016a,b) employ ratings in line with both models. Thus, together dimensional and categorical models of affect should be employed in the studies in order to understand the impact of emotion of cognitive processes (Mikels et al., 2005; Stevenson et al., 2007; Stevenson and James, 2008; Briesemeister et al., 2011).

Some insights into the differential influence of basic emotions on memory and attention can be gained from few existing behavioral and neuroimaging studies. It has been found that stimuli provoking disgust are later recalled better than stimuli provoking fear, and this applied both to words (Charash and McKay, 2002) and images (Croucher et al., 2011; Chapman et al., 2012). Moreover, disgust-evoking images, but not fear-evoking ones, keep hold of our attention for longer (Van Hooff et al., 2013). At the neuronal level, using fMRI and the directed forgetting paradigm, it was revealed that correctly recognized disgusting stimuli evoked the highest activity in the left amygdala compared to all other categories. This structure was also more activated for remembered vs. forgotten stimuli, but only in case of disgust or fear eliciting pictures (Marchewka et al., 2016).

To the best of our knowledge, there has been not a single study investigating the influence of emotion on long-term recognition memory using both dimensional and categorical models of affect.

The present study was aimed at extending the previous literature reporting long-term recognition memory effects for emotional stimuli (Bradley et al., 1992; Dolcos et al., 2005; Weymar et al., 2011; Harand et al., 2012). We were particularly interested in exploring the modulatory effect of emotions on long-term recognition memory by using normative behavioral ratings collected in line with dimensional and discrete models of affect. For this reason we developed a computerized rating procedure and conducted the memory test after a period of 6 months.

At the behavioral level, separate analyses examining the effect of emotion on memory performance according to both models of affect were computed. These analyses revealed that long-term recognition memory was significantly modulated by emotion conceptualized according to both theoretical models. Specifically, we found a significant effect of valence and arousal on recognition accuracy. Recognition accuracy was the highest for negative arousing stimuli, in line with previous studies where various types of material and methodological approaches were used (Bradley et al., 1992; Cahill and McGaugh, 1995; Ochsner, 2000; Kensinger et al., 2003; Dolcos et al., 2005). With respect to the ratings collected in line with discrete models of affect, we found emotional enhancement of memory effect only for negative emotions (sadness, fear, surprise, anger, and disgust). This effect was most pronounced in the case of disgust. Recent studies have already shown that stimuli eliciting disgust are able to capture and hold our attention much more effectively than fear-evoking images (Van Hooff et al., 2013; van Hooff et al., 2014) and are better remembered after a short retention period (Croucher et al., 2011; Marchewka et al., 2016). However, it is still debated whether disgust, since it is evidently a basic sensory affect (Rozin and Fallon, 1987) as well as a socially constructed moral emotion (Haidt, 2003), should be classified as a basic emotion (for discussion see Panksepp, 2007; Toronchuk and Ellis, 2007; Panksepp and Watt, 2011).

In the case of positive emotions (happiness) we found an opposite effect. A more positive rating of the stimuli corresponded with worse recognition success being observed. This result is in line with findings concerning the dimension of valence, reported also in the previous studies (Kensinger and Corkin, 2003; Kensinger et al., 2006, 2007). According to the proposed explanation, negative information may be encoded more automatically than positive information (Kern et al., 2005; Talmi et al., 2007), and negative items tend to be encoded with more perceptual processing than positive information (Mickley and Kensinger, 2008; Mickley Steinmetz and Kensinger, 2009). However, it is important to emphasize that the current sample was limited to young adults and observed behavioral effects could be different in older population. For example a large-scale meta-analysis showed that older adults demonstrate a significant information processing bias toward positive vs. negative information, whereas in younger adults an opposite effect was revealed (Reed et al., 2014). Nevertheless, age related effects on long-term memory have not been yet tested with long retention periods, such as the one in the current study.

At the neuronal level, using basic contrast-based fMRI analyses we found a number of brain areas associated with long-term recognition memory of emotional information, both cortical and subcortical. Cortical regions we found were previously indicated in the studies of various kinds of memory retrieval and recognition (Stock et al., 2009; Sitaram et al., 2014). Interestingly, some of these regions, i.e., the bilateral anterior insula and anterior cingulate cortex, are the core regions of the so-called salience network (Seeley et al., 2007). More and more evidence suggests that they play a critical role in mediating the interplay between brain areas involved in emotion perception and executive control (Menon and Uddin, 2010). A recent fMRI study on the influence of emotions on working memory found that the salience network plays a critical role in mediating interaction between emotion perception and executive control (Luo et al., 2014). Anatomically, core salience network regions (AI and ACC) have strong reciprocal connections with an emotional processing network including the amygdala and ventral sensory pathways (Krämer et al., 2014). It seems, based on the presented findings, that the regions of the salience network are also involved in successful recognition of stimuli after long retention periods.

In contrast to the previous study of Dolcos et al. (2005), we failed to observe any significant involvement of the hippocampus and amygdala in correct recognition of emotional items. This might be related to the fact that we computed fMRI analyses on the whole brain, whereas the results of the mentioned study were based on ROI analyses. Also, that study included only female participants, which might have influenced the pattern of brain activity. However, it seems that the involvement of the hippocampus and amygdala in long-term recognition memory may not be critical. It was first clinically demonstrated in HM patients that despite damage to the hippocampus resulting in profound amnesia, older memories, or information could still be consciously recollected and recognized. This fact led to a proposal that after a longer retention period information initially processed in the hippocampus is stored in a distributed cortical network. The ACC has recently been indicated in this process. Harand et al. (2012) found that hippocampal activity decreased over time for initially episodic, later semantic memories. For both types of memories, neocortical activations were observed with both delay periods (3 days and 3 months), specifically in the ventromedial prefrontal and anterior cingulate cortices. The authors argued that these cortical activations might reflect a gradual reorganization of memory traces within neural networks. In line with this result, studies on rodents have demonstrated that the ACC is necessary for recalling behaviors learned at least 1 month ago, but not for the same behaviors learned the preceding day (Weible et al., 2012).

The increased activity for correctly recognized compared to forgotten stimuli was also observed in the striatum, specifically in bilateral caudate and left pallidum as well as in the right thalamus. A recent study which explored the influence of genetic polymorphisms on long-term memory found that older people with the presence of an A1 allele of the DRD2/ANKK1-TaqIA polymorphism had worse memory performance and a weaker activation in left caudate nucleus (Persson et al., 2015). Further, a positive relationship between caudate activation and long-term memory updating performance among non-A1 carriers indicated that caudate activation was behaviorally relevant. Since this polymorphism is related to a reduced density of striatal dopamine D2 receptors (Jönsson et al., 1999), these results point to the critical role of dopamine in regulating long-term memory. In line with this hypothesis, emerging evidence from human and animal studies suggests that suboptimal dopamine modulation may be associated with increased forgetting of episodic information (Papenberg et al., 2013). What is more, the activation of the right thalamus found in the present study is consistent with neuropsychological findings in patients showing a double dissociation in material-specific long-term memory i.e., right thalamus lesion specifically impairs visual memory, while lesions of the left thalamus impairs verbal memory (Edelstyn et al., 2012).

Additional fMRI parametric analyses identified regions that were uniquely and increasingly recruited with a corresponding increase in subjective ratings of arousal. This modulation of brain activity by arousal was observed in the right caudate and the left anterior cingulate gyrus. Arousal was the only subjective rating significantly modulating brain activation. It seems therefore that these structures are not only significantly involved in long-term recognition memory but also in perceiving arousal per-se. Previous studies which showed brain correlates of arousal did not employ recognition memory paradigms but required participants to experience emotions or evaluate affective stimuli (Colibazzi et al., 2010; Viinikainen et al., 2010), thus it is difficult to directly compare them with the results from the present study. Nevertheless, they revealed the existence of distinctive brain networks managing emotions in line with the dimensional model of affect including the caudate and cingulate cortex. The activation patterns found in the present study probably reflect more complex brain networks serving both long-term recognition of emotional stimuli and experiencing of the feelings elicited by the material. Enhanced memory formation for unpleasant arousing photographs seems to be initiated already at stimulus encoding. In fact, increased stimulus processing is supposed to lead to the formation of inter-item associations and enhanced memory consolidation (Craik and Lockhart, 1972; Craik and Tulving, 1975; Cowan, 1995). Moreover, the exposure to unpleasant arousing stimuli is associated with increased retrieval of negative memories and unpleasant post-event recollection (Cuthbert et al., 2003). Superior memory formation for unpleasant arousing events seems to be adaptive, but it may also result in behavioral withdrawal and reinforce negative beliefs. These processes may impair our psychological functioning and, in extreme cases, assist in the development of mental disorders that may be difficult to alleviate because of poorer positive memory formation.

A strong limitation of the current study is the fact that we were unable to find significant differences in brain activations underlying long-term recognition memory in respect of subjective ratings collected in both the dimensional and discrete models of affect. These differences were visible only at the level of behavioral analyses.

We hypothesize that this lack of significant results in fMRI analyses could be a consequence of the small number of trials within each experimental condition and the relatively limited number of subjects in the recognition session with analyzable data. At the same time on the behavioral level we obtained moderate to high effect sizes. Therefore, we suggest that further experiments increase both the sample size and the number of stimuli. We also think that further studies in older adults should be conducted in order to understand the mechanisms of long-term recognition memory in humans in more details.

The most important conclusion that could be drawn from the current study is that only subjective ratings of arousal are directly associated with brain activation during long-term memory recognition. This findings may be interpreted against theories of basic emotions postulating that a discrete and independent neural system subserves every emotion (Posner et al., 2005; see also Clark-Polner et al., 2016).

In addition, the obtained results also support the hypothesis that the memorized material is firstly processed in the hippocampus and amygdala, yet over time it becomes stored in a distributed cortical network including the core salience network and basal ganglia.

AM was responsible for the study concept and design contributed to the acquisition of the data, performed the data analysis and wrote the article. MW performed the data analysis. AMo acquisition of the data. MR contributed to the writing of the article. JM study concept and design contributed to the writing of the article. KJ performed the data analysis and wrote the article. All authors critically reviewed content and approved final version for publication.