Alzheimer's disease (AD) is the most common type of dementia in senile individuals, characterized by episodic memory impairment in the incipient stage, and a subsequent progressive and irreversible cognitive decline at later stages. Mild cognitive impairment (MCI) is a subjective cognitive impairment which initially, does not affect social functioning or activities associated with daily living. Impairment is usually first recognized by patients or their relatives, and confirmed with neuropsychological testing (Ewers et al., 2011). MCI is an umbrella term for a heterogeneous group of cognitive impairments; however amnestic MCI (aMCI) in particular, is thought to be a transition stage between normal aging and AD (Jicha et al., 2006; Whitwell et al., 2008; Mitchell and Shiri-Feshki, 2009).

Literature regarding structural and metabolic imaging in AD has highlighted several abnormalities in cortical and subcortical gray matter (GM), including areas in the hippocampus, parahippocampus gyrus, amygdala, entorhinal cortex, posterior cingulate cortex, and thalamus (Nestor et al., 2003; Whitwell et al., 2008; Ewers et al., 2011; Liu et al., 2014; Delli Pizzi et al., 2016; Pini et al., 2016). Many of these areas or subdivisions of these areas, such as the medial temporal lobe (MTL), anterior thalamus, mammillary bodies and posterior cingulate are thought to be strongly anatomically inter-connected via a limbic-diencephalic network termed the Papez circuit (Acosta-Cabronero and Nestor, 2014). The Papez circuit constitutes a group of neural projections involved in memory and emotional expression that originates in the cortex, and subsquently communicates with the hippocampal formation and mammillary bodies which in turn send projections to the anterior thalamic nuclei where information is relayed the cortex of cingulate gyrus (Vann and Nelson, 2015). Besides the circuit of Papez, Yakovlev proposed another limbic network which was consisted of the orbitofrontal cortex, insula, amygdala, and anterior temporal lobe, as an underlying pathway for emotion and motivation (Yakovlev, 1948). Maclean concluded that the limbic cortex, together with the limbic subcortical structures, is a functionally integrated system interconnected by short- and long-range fiber bundles (Maclean, 1949, 1952). Recently, a revised model was proposed by Catani to give a better and more comprehensive understanding of the limbic system. In Catani's model, there were three distinct but partially overlapping networks of the limbic system: the hippocampal-diencephalic and parahippocampal-retrosplenial network, the temporo-amygdala-orbitofrontal network, and the medial default-mode network, each responsed for different cognitive behaviors (Catani et al., 2013). In a recent study, higher mean diffusivity (MD) values were found bilaterally in the thalamic sub-regions projecting to temporal cortex, showing that the anterior nuclei projecting to temporal lobe were damaged in AD patients (Delli Pizzi et al., 2014). In fact, it has been proposed that AD was associated with the degeneration of the limbic cortices and their white matter (WM) connections (Acosta-Cabronero et al., 2010).

Because episodic memory loss is the hallmark of AD, it is necessary to understand different dimensions of that memory loss, and whether any episodic memory components, such as emotional memory, are relatively preserved in the disease progression. Normally, emotional stimuli are more easily remembered than neutral stimuli, a phenomena known as emotional enhancement of memory (EEM) (Hamann, 2001). Emotional stimuli attract more attention and are processed more fully than neutral stimuli, which in turn, enhances memory by increasing the distinctiveness with which they are encoded (Ochsner, 2000). The EEM effect exists in normal aging (Denburg et al., 2003; Murphy and Isaacowitz, 2008). On the contrary, studies have indicated that individuals with AD have impaired EEM following emotional stimuli. Hamann et al. (2000) showed that AD patients had impaired emotional memory in recall tests after viewing negative pictures only, but impairment of EEM was evident for both positive and negative pictures in the recognition task. Using recognition tests, other researchers have also found that AD patients do not retain EEM following emotional pictorial stimuli (Abrisqueta-Gomez et al., 2002; Landre et al., 2013). Specifically, Chainay et al. (2014) found impaired EEM of positive and negative pictures in AD patients after both intentional and incidental encoding. Additionally, AD patients also show disrupted EEM in recall tasks for emotional words, neutral words embedded in emotional context, and negative verbal information (Kensinger et al., 2002, 2004). Nevertheless, there are discrepancies in regard to whether the EEM effect is retained in patients with MCI and early AD. Although, some researchers have found that aMCI patients exhibit normal EEM in immediate recall tests (Nieuwenhuis-Mark et al., 2009), others have found that patients with aMCI do not retain EEM following visualization of emotional pictures (Parra et al., 2013; Wang P. et al., 2013). Sava et al. (2015) demonstrated that AD patients had a positive memory bias when the stimulus was sufficiently rich and deep encoded to the individual, and when support was provided the time of retrieval such as cued recall or recognition tasks. Another free-recall study highlighted the impaired EEM for verbal stimuli but preserved EEM for non-verbal declarative memory in patients with mild AD (Baran et al., 2014). These discrepancies may be a result of different materials used to elicit the emotional response, different arousal of stimuli, and also different encoding and retrieval methods used in each specific study. Alternatively the inconsistency may be due to the composition of patients with MCI and mild AD, which constitute a heterogeneous population with various possible levels of EEM preservation, depending on the state of their EEM-related brain structures.

Since the limbic system has been considered to play important roles in memory, emotion, spatial orientation, visceral sensation and behavior and associated with pathophysiology of many diseases (Catani et al., 2013; Guo et al., 2014; Liu et al., 2015). There might be some associations between the limbic system and emotional memory. Previous studies have focused on task related functional magnetic resonance imaging (MRI) or GM changes relating to EEM of AD patients (Dolcos et al., 2005; Perrin et al., 2012; Kumfor et al., 2013; Landre et al., 2013; Parra et al., 2013; Philippi et al., 2015), however, little is known about emotional memory related changes in WM. Thus, the purpose of the present study was to investigate whether the EEM is absent in patients with aMCI and AD, and whether emotional memory is related to WM connections of the limbic system. Our primary hypothesis was that the WM tracks of the limbic system would be impaired in AD patients, and this impairment would have a detrimental effect on their emotional memory and EEM. In regards to patients with aMCI, we hypothesized that any damage to the WM tracks would be more localized and less severe and therefore limit the extent of loss of EEM in this cohort.

Based on our findings, using pictures as stimuli, incidental encoding and recognition tasks, NC participants recognized previously seen stimuli better than aMCI and AD patients. Compared with NC group, aMCI patients showed decreased Hits but normal FAs, which indicated that the impaired episodic memory in amnesia appeared to affect the correct recognition rate firstly. Furthermore, for the index d′, only the NC group benefited from the emotional stimuli of both negative and positive images. aMCI patients benefited only from positive stimuli. AD patients didn't differ in terms of discrimination of old stimuli according to emotional valence. The results were agreed with our hypothesis in regards to emotional memory in AD and aMCI patients. The EEM effect is impaired in AD patients but preserved in aMCI patients to some extent. However, aMCI didn't discriminate old negative pictures better than that of neutral ones. One probable explanation is that aMCI patients may spontaneously focus their attention on positive information and process it in a more self-referential way because of their limited cognitive resources. Another explanation is that the atrophied amygdala volumes in aMCI patients may preferentially affect negative emotion processing. However, in this study, for the index Hits, there were no significant valence × group difference and no EEM effect, even in the NC group. Since the definition of Hits is the correct recognition rate of previously seen stimuli. It is well accepted that recognition relies on both familiarity and recollection (Yonelinas, 2001). In fact, there would be three kinds of psychological state, “remember,” “know,” or “guess,” when a picture was judged by someone as old. So, only Hits alone could not reflect the real discrimination, and this might influence the detection of EEM effect. What's more, here in this study, we conducted an incidental encoding procedure, in which less cognition resource was involved, leading a participant more likely to depend his/her answer on a sense of familiarity. Several behavioral studies have shown that the memory-enhancing effect of emotion specifically modulates recollection rather than familiarity (Dolcos et al., 2005; Wang P. et al., 2013). Thus, the EEM effect of hit rate might be contaminated due to greater proportion of familiarity-based response for all stimuli. Our results were partly consistent with Chainay et al. who did not observe EEM for Hits after incidental encoding either (Chainay et al., 2014). This may also suggest that the index d′ is more suitable and sensitive than the index Hits to reflect EEM effect.

According to the TBSS results, AD patients showed widespread WM impairments, including association fibers, commissural fibers and projection fibers, which indicated impaired WM connections both between and within hemispheres. In addition, the WM connections of the limbic system networks, including the column and body of fornix, bilateral crus of fornix and stria terminalis, cingulum bundles, anterior thalamic radiation, and uncinate fasciculus were impaired in AD patients. Consistent with the proposed functionally integrated viewpoint, VBM results from the present cohort confirmed that AD patients had GM atrophy of the limbic system, such as the MTL structures (i.e., amygdala, hippocampus, parahippocampal gyrus), thalamus, posterior cingulate gyrus, orbitofrontal and medial prefrontal cortex, and precuneus.

Firstly, the hippocampal-diencephalic and parahippocampal-retrosplenial network was impaired in AD patients. The hippocampal-diencephalic limbic circuit is connected through the fornix and mammillo-thalamic tract, and the parahippocampal-retrosplenial circuit is connected through the ventral cingulum (Catani et al., 2013). The hippocampus has long been considered as a central structure for episodic memory encoding, consolidation and retrieval. However, a role for the hipocampus in emotional control has become more widely appreciated recently. The fornix has played a fundamental role in linking MTL and diencephalon (Acosta-Cabronero and Nestor, 2014). Using tractography, it has been shown that the integrity of the fornix is linked to memory (Metzler-Baddeley et al., 2012). Nestor et al. found λ1 abnormalities in the fornix, parahippocampal WM, and anterior thalamus of an AD cohort (Nestor et al., 2003). Specific research targeting the fornix, cingulum bundle and corpus callosum, found significant cross-sectional and longitudinal tensor differences for all three structures (Keihaninejad et al., 2013). These major WM connections of the limbic system have been highlighted in the “early hypometabolic landscape” of AD. Secondly, the temporo-amygdala-orbitofrontal network was impaired in AD patients. The temporo-amygdala-orbitofrontal network is connected through the uncinate fasciculus, and is dedicated to the integration of visceral and emotional states with cognition and behavior (Catani et al., 2013). Previous studies have indicated that emotional memory is dependent on the integrity of the amygdala, and the interaction between the amygdala and other brain regions, such as the hippocampal formation, prefrontal cortex, sensory neocortex, striatum, and hypothalamic–pituitary–adrenal axis (Labar and Cabeza, 2006). The amygdala receives information related to the external environment from the sensory thalamus and sensory cortices, and is reciprocally connected with cortical regions, especially the hippocampus, and the midline and orbital prefrontal cortices (Janak and Tye, 2015). It has been proposed that EEM may be based on two different mechanisms related to the amygdala; one based on the influence of the amygdala on attention processes (attention-mediation hypothesis) and the other related to modulation of consolidation processes (consolidation-mediation hypothesis) (Hamann, 2001; Talmi et al., 2008; Hermans et al., 2014). In fact, the uncinate fasciculus links the amygdala and temporal lobe with the prefrontal lobe. It is critical for episodic memory. Some researches have indicated that the uncinate fasciculus was impaired in AD and MCI patients (Fujie et al., 2008; Yasmin et al., 2008; Kiuchi et al., 2009), and might be associated with impaired facial emotional recognition in aMCI (Fujie et al., 2008). Our results confirmed that the integrity of uncinate fasciculus was aslo associated with the discrimination of previous seen emotional pictures in AD patients. Thirdly, the medial default-mode network was impaired in AD patients. Actually, the medial regions of the default-mode network are inter-connected through the dorsal cingulum. It is well known that the cingulum is an important component of the limbic system, and can be divided into an anterior-dorsal component, which constitutes most of the white matter of the cingulate gyrus, and a posterior-ventral component running within the parahippocampal gyrus, retrosplenial cingulate gyrus, and posterior precuneus (Catani et al., 2013). The anterior cingulate/medial prefrontal cortex and the posterior cingulate/precuneus form the medial default-mode network. The DMN network is involved in introspection and active episodic memory (Barkhof et al., 2014). The majority of resting-state functional MR studies found reduced DMN connectivity in patients with AD but also in the earlier phase of MCI, even in unaffected carriers of familial AD and subjects with cognitive complaints (Greicius et al., 2004; Sheline and Raichle, 2013; Wang Y. et al., 2013). Consistent with previous DTI studies (Liu et al., 2011; Shu et al., 2011; Wai et al., 2014), our results indicated structures outside the limbic system, including the sagittal stratum (include inferior longitudinal fasciculus and inferior fronto-occipital fasciculus), superior longitudinal fasciculus, superior fronto-occipital fasciculus, corona radiata, posterior thalamic radiation and external capsule, were also affected in AD patients.

Moreover, by combining these microstructure impairments of the limbic system networks with emotional memory scores, we found that the uncinate fasciculus, corpus callosum, left cingulum bundles, left parahippocampal WM, left crus of fornix and stria terminalis had correlations with the index d′ of emotional valence pictures or EEMd′ in AD patients, thus supporting our previous hypothesis that WM connections of the limbic system were associated with emotional memory and EEM impairment. It tends to indicate that the few AD patients still benefiting from the effect of emotional valence on memory are the ones who have more intact WM fiber connections, even though most patients do not benefit from this effect at all. Additionally, the GMV of limbic system networks such as the amygdala, insula, medial frontal lobe, and cingulum gyrus, were also associated with emotional memory in AD patients, which exactly reveals the WM and GM functional integrity theory of the limbic system networks.

By contrast, the atrophied brain areas of aMCI were more limited, nevertheless, structures of the MTL, precuneus and medial prefrontal cortex were already affected. Although, several WM tracks associated with the limbic system were also affected in aMCI patients, these impairments were more localized and less severe, and no significant correlations except the positive correlations between the FA value of right uncinate fasciculus and the d′(positive), and between the volume of right amygdala and d′(positive) were found in aMCI patients. One probable explanation is that the WM impairments in aMCI patients were still too subtle to cause disfunctions of those fibers. Another possible explanation is the higher brain reserve of aMCI patients. Here in our study, aMCI patients had a higher educational level and more localized impaired brain areas than AD patients. It has been hypothesized that “brain reserve” may account for inter-individual differences in the recruitment of neural networks and cognitive processes, and may thus compensate for age-related brain dysfunction (Freret et al., 2015). In fact, one study has revealed that the preserved executive functioning in MCI patients was associated with a higher education time (Andrejeva et al., 2016). Another 12-month follow-up study has demonstrated that higher cognitive reserve contributed to protecting against cognitive decline, and higher brain reserve was a strong indicator for reversion and conversion in patients with MCI (Osone et al., 2016). Thus, the potential compensational effects due to a higher brain reserve in the aMCI corhort may limit the detection of correlations between the limbic system and EEM effect.

Using a combination of VBM and TBSS analyses, we found that both the emotional memory-related GM regions and WM connections that are anatomically inter-connected via the limbic networks were damaged in AD and aMCI patients. Besides primary degeneration of WM, such as demyelination and axonal degeneration, WM impairment can be the secondary result of primary neuronal cell body degeneration, a process known as Wallerian degeneration. We speculate that primary neuron loss of interacting and correlated emotional memory-related brain structures may cause secondary lesions to their interconnecting WM fibers, which exacerbate brain atrophy reciprocally, thus accelerating the progress of neuropathological damage of the limbic system networks and the impairment of emotional memory.

Nevertheless, there are still some limitations to the present study. Firstly, the slice thickness of raw DTI data was 4 mm, which may be too thick to view intricate structures. Secondly, as mild to moderate WMHs are common during aging, we did not consider them as exclusion criteria in the current study, and the education level of AD patients was lower than the NCs and aMCI group. However, on one hand, we have regarded these possible confounding factors as covariates, and therefore nullified them in our statistics models. On the other hand, large well-matched samples will be collected in the future to verify those findings and for further studies. Thirdly, the NC group was lack of a deep neuropsychological evaluation, and this might bring in some bias of group entrance. Fourthly, given the potential role of inflammation in psychopathology, it has been thought that chronically activated inflammatory signals in aging may contribute to increased vulnerability to neuropsychiatric and neurodegenerative disorders (Popa-Wagner et al., 2014; Sandu et al., 2015). Thus, inflammation might also be associated with the impairment of emotional memory. However, in the present study, we didn't check for systemic inflammatory markers, thus a further research is needed to focus on this potential interference. Finally, we did not distinguish the role of collection and familiarity in the present study, which may have limited the detection of EEM to some extent, and further investigation is warranted regarding these two factors.

In conclusion, our results give support that there are impairments of the limbic system networks in AD and aMCI patients, and this impairment is associated with emotional memory and the loss of EEM in AD patients. Furthermore, neuronal loss of interacting and correlated emotional memory-related brain structures may cause secondary lesions of WM fibers interconnecting these regions which exacerbate brain atrophy reciprocally, thus accelerating the progress of neuropathological damage of the limbic networks, associated with the impairment of emotional memory and loss of EEM in AD patients.

YY designed the study and revised it critically for important intellectual content. XSL performed the research and drafted the manuscript. HW helped in data analyses. XHL helped draft the work. KW,YT and SZ participated in the clinical evaluation of patients. All authors read and approved the final manuscript.