Episodic memory refers to the human memory system that enables people to remember specific past events or information (Tulving et al., 1994). Its formation encompasses multiple processes, including encoding or the acquisition of learning material, consolidation or the retention of newly acquired information, and retrieval of the information already stored (Tromp et al., 2015). Aging is accompanied with episodic memory decline, which can lead to different degrees of damage in all associated memory processes (Koen and Yonelinas, 2014). It is therefore important to further investigate age-related decline of episodic memory and the underlying neural mechanisms of neurodegenerative diseases, and to use this information to formulate potential memory interventions.

Among the variety of standardized neuropsychological tests that measure episodic memory, list-learning tests are perhaps the most commonly used. Verbal list-learning tests are commendably employed to assess early cognitive decline in patients with mild cognitive impairment (MCI) and Alzheimer’s disease (AD), as well as to predict diagnostic conversion to AD (Arnaiz and Almkvist, 2003; Karrasch et al., 2005; Guo et al., 2009). These tests involve learning across multiple trials and typically produce several memory indices, such as immediate free recall, delayed free recall, and recognition (Ivnik et al., 1990; Bleecker et al., 2005). In terms of the different retention intervals, episodic memory is measured by considering immediate recall (IR) and delayed recall (DR), which require participants to retrieve memories immediately after learning and 30 min after learning, respectively.

Verbal learning and memory retrieval with different retention intervals may encompass different memory processes. For example, participants do not have enough time to rehearse or consolidate memory in the case of IR. However, during a delayed period in the case of DR, consolidation can occur; this stabilizes the newly encoded memory, and enhances and integrates the new information with pre-existing long-term memories (Marshall and Born, 2007). Memory consolidation relies on a dialog between the neocortex and hippocampus (Winocur et al., 2007). The newly acquired memory representations are stored in the hippocampus and are gradually redistributed and transferred to neocortical regions via the strengthening of cortico-cortical circuits (Rasch and Born, 2007; Takashima et al., 2009). Thus, we can speculate that DR depends on neocortex-subcortex (hippocampus) connections and cortico-cortical connections, while IR does not. Moreover, different memory deficits of IR and DR have been associated with different brain activity (Bosch et al., 2013). Taking all this evidence together, IR and DR may be based on dissociable neural substrates, or at least separate neural circuits.

However, no studies about neural correlates of IR and DR in traditional list-learning tests in the elderly have yet been conducted. While neuropsychological tests have been widely used to evaluate brain function, the underlying substrates of these test results have rarely been explored. Furthermore, different memory measures of IR and DR have been treated as a whole to evaluate general episodic memory function, making it impossible to detect the variability inherent in individual memory indices (Ystad et al., 2010; Wolk et al., 2011). Therefore, the potential differences in neural mechanisms underlying IR and DR in the elderly population need to be investigated.

Brain activation in resting state is highly similar to the functional activity in a cognitive task. It is increasingly believed that intrinsic activations shape patterns of cognitive task activations and predict cognitive ability (Smith et al., 2009; Cole et al., 2016; Schultz and Cole, 2016). Thus, resting state fMRI is a promising tool to explore intrinsic brain activity and its relationship with cognitive processes, with extensive use of regional homogeneity (ReHo) and functional connectivity (FC) analyses. ReHo supposes that voxels within a functional brain area are more temporally synchronous when this area is involved in a specific condition (Zang et al., 2004). It is applied in the prediction of variations in cognitive performance in healthy people by detecting regional signal changes (Tian et al., 2012). In addition, significant abnormalities in spontaneous low-frequency fluctuations coherence have been observed in patients with AD. ReHo has been speculated to be a biomarker of disease progression in patients with AD and MCI (He et al., 2007; Zhang et al., 2012). FC evaluates the interregional connectivity between spatially remote brain regions. Functional interactions between distinct regions also play a crucial role in cognitive processing and, similarly, altered patterns of FC is a proposed biomarker of AD and MCI (for a review, see van den Heuvel and Pol, 2010).

Therefore, in this study, we investigated the underlying brain systems of different episodic memory capacity in a cohort of normal elderly people using ReHo and FC of resting state fMRI signals. On the basis of previous work, we predicted that, as different memory retrieval indices, IR and DR depend on different neural correlates, and that DR may be supported by intercortical connections in memory-related brain regions to a greater extent.

We investigated the neural correlates of different verbal episodic memory indices, as defined by the AVLT, in a cohort of 66 healthy elderly participants. Using the ReHo method and FC analysis, we found that the ReHo of primary regions related to age-related decline of episodic memory capacity were all located within the DMN. Specifically, AVLT-TR was negatively correlated with regional coherence of local activity in the PHG and MFG, and positively correlated with that in the PCC and IPL. Furthermore, we demonstrated that IR and DR relied differentially on the DMN. ReHo of spontaneous activity in the DMN could significantly account for the variations in IR, while both ReHo and FC within DMN could account for variations in DR.

To our knowledge, this is the first study to report that ReHo of the DMN nodes is related to episodic memory in healthy older adults. The DMN is a particular brain system that shows greater activity during the resting state than during goal-oriented or attention-demanding tasks (Jeong et al., 2015). It contains a set of functionally connected brain regions, including the mPFC, PCC/PCu, IPL, lateral temporal cortex, and hippocampal formation (Raichle et al., 2001; Fox and Raichle, 2007; Buckner et al., 2008; Raichle, 2015). Multiple lines of research using resting-state FC analysis have revealed the contribution of the DMN to episodic memory, and abnormality in DMN connections has been suggested to be one basis for memory impairment in patients with neurodegenerative diseases. For example, greater functional connections within the DMN (e.g., MPF-LIP connectivity) (He et al., 2012), and between the DMN and the medial temporal lobe system (e.g., PCC-hippocampus connectivity) (McCormick et al., 2013), have been found to indicate better memory performance. Several resting state fMRI studies have shown that disrupted functional connections of the DMN is related to poor episodic memory performance (for a review, see Dickerson and Sperling, 2009), especially in patients with MCI and various stages of AD (Greicius et al., 2004; Mevel et al., 2011; Binnewijzend et al., 2012). Together, this previous work indicates that the DMN subserves episodic memory processing. In concordance with this idea, we found an association between memory function and ReHo of spontaneous activity in primary DMN brain areas in memory-unimpaired older adults. Furthermore, our findings extend previous work showing that, in addition to resting-state FC, ReHo in the crucial DMN nodes also predicts memory performance. Resting-state FC reflects inter-regional synchronization of time courses, whereas ReHo reflects intra-regional synchronization of time courses, which may be related to regional metabolic homogeneity (He et al., 2007).

An interesting finding in the present study was the different pattern of correlations between ReHo values in the DMN and memory function in normal older adults. In particular, AVLT-TR scores were positively correlated with regional coherence in the PCC/PCu and IPL, and negatively correlated with that in the PHG and MFG. Previous studies have also produced mixed results. For instance, He et al. (2007) found significant negative correlations between ReHo in the mPFC and Mini-Mental State Exam scores in patients with AD, but significant positive correlations between ReHo in the PCC/PCu and Mini-Mental State Exam scores. One explanation for these varied results is a functional dissociation of DMN components. Recently, it has been proposed that the DMN is not homogenous, as it has previously been considered (Uddin et al., 2009; Andrews-Hanna et al., 2010; Leech et al., 2011; Damoiseaux et al., 2012; Bado et al., 2014). Although the exact function and anatomy of distinct functional components is still unknown (e.g., anterior vs. posterior or ventral vs. dorsal), the notion of functional heterogeneity in the DMN and its dissociable roles during memory formation has been widely supported. For example, Maillet and Rajah (2014) found that activation in the left mPFC predicted encoding success; however, activation in the PCC was greater during unsuccessful encoding, which may reflect task-unrelated thoughts. Similarly, in another study, the PCC/precuneus and anterior DMN node in the mPFC also showed functional heterogeneity, with the former regions significantly activating during memory retrieval and the latter regions strongly deactivating (Sestieri et al., 2011). In line with these studies, our results showed that the mPFC and PCC/PCu played dissociable roles in memory formation in the healthy elderly adults, leading to different correlations with memory performance.

As another major discovery, our results revealed the discrepant neural circuits within the identical brain network between temporary memory store and the stores for long-term retention in older adults. Correlation analysis between different memory measures and resting state brain activity of the DMN demonstrated that age-related variety of immediate and delayed memory abilities relied on partially dissociable neural underpinnings. IR was related to the regional coherence of spontaneous activity in the DMN, and DR depended on both regional fluctuations and interregional integration within the DMN. Although the DMN is engaged in episodic memory processing, its different role in IR and DR in older adults was previously unclear. As far as we know, this is the first study to show that the DMN plays different roles in memory across different retention intervals in older adults.

Delayed recall may reflect memory consolidation in the elderly, which would explain its correlation with interregional connections within the DMN in the present study. On the one hand, there is gradual reorganization within long-term memory storage as time passes after learning. As a result, DR comprises memory consolidation. Our findings parallel the idea that IR trials may measure attention and working memory to a greater degree than DR trials, which predominantly tap retention, consolidation, and delayed retrieval effects (Shankle et al., 2005; Rabin et al., 2009). On the other hand, cortico-cortical pathways and hippocampal-neocortical interactions are generally considered as the neural correlates of consolidation, whereby consolidation relies on connections of brain regions. Meanwhile, patterns of neural activity when performing a task can be reproduced in resting state (Foster and Wilson, 2006; Albert et al., 2009), which is devoted to the reprocessing of past experiences. Miall and Robertson (2006) proposed that the DMN may support the off-line processing and consolidation of memories. Taken together, our findings that the DMN FC positively predicted DR performance provides further evidence for the role of the DMN in memory consolidation.

In addition, we also observed that ReHo of other regions were associated with episodic memory performance, mainly including the temporal cortex, the supramarginal gyrus (SMG), and bilateral cerebellum. The superior temporal cortex contains the primary auditory cortex and plays a crucial role in speech process and auditory information process (Binder et al., 2000; Zatorre et al., 2002; Leff et al., 2009; Rauschecker and Scott, 2009). Furthermore, the SMG is related to linguistic functions, such as phonological processing in the phonological-articulatory loop (Dehaene-Lambertz et al., 2005; Rauschecker and Scott, 2009). Therefore, it is reasonable to have found involvement of the temporal cortex and SMG because we used an auditory memory test. In addition, our results are consistent with the recent findings of cerebellum’s cognitive role. It is increasing recognized that the cerebellum is engaged in multiple domains of higher-order cognitive functions and emotional control (Schmahmann and Caplan, 2006), beyond the motor domain. Neuroimaging and neuropsychological evidence has shown that the cerebellum contributes to episodic memory processes, and patients with cerebellar lesions have been reported to suffer from episodic memory deficits (Gottwald et al., 2004; Mathiak et al., 2004; Weis et al., 2004). Thus, in the future, we may investigate the neural correlates of episodic memory in view of brain networks, combining the cerebellar network and the DMN.

Our study has some limitations that should be noted. First, we only recruited elderly adults with normal cognition, and our results cannot, therefore, be generalized to other populations. Future studies should recruit individuals with different cognitive statuses (i.e., patients with MCI or AD) to evaluate the changing trajectory of the roles of the DMN in episodic memory with aging and cognitive deterioration. Second, we did not consider the relationship between resting-state activity and task-related activity, though the latter could further support our results. For example, in a PET study, a significant correlation was found between the DR score and metabolism in the PCC (Brugnolo et al., 2014). A further study is required to combine the data during task and resting state and gather more conclusive evidence about the different contributions of the DMN to IR and DR. Third, we did not employ a younger control group in the present study. Thus, we cannot determine the effect of aging on the relationship of spontaneous activity in the DMN and recall performance. Further studies are required to investigate whether our observations reflect a normal phenomenon or an age-related alteration.

In summary, we explored episodic memory capacity of the healthy elderly adults using ReHo analysis of resting state fMRI data, providing vital insights into our understanding of brain activity of regional coherence and its cognitive implications. Our study indicates that ReHo, a feasible and easily applied approach, can be employed to characterize the neural correlations of traditional list-learning test results. Furthermore, the neural mechanism of memory consolidation has received increasing attention, and our findings reveal the possible relationship between the DMN and memory consolidation during wake for the first time. We also demonstrated that DMN activity measured with ReHo analysis can be used as a way to identify biomarkers for memory decline and neurodegenerative disease.

LH conceived the idea, designed the study, analyzed and interpreted data, and drafted the manuscript. RL, PW, and ZZ assisted the analysis and interpretation of data. JL participated in the writing and revision of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.