Abstract
Both, the hippocampal formation and the neocortex are contributing to declarative memory, but their functional specialization remains unclear. We investigated the differential contribution of both memory systems during free recall of word lists. In total, 21 women and 17 men studied the same list but with the help of different encoding associations. Participants associated the words either sequentially with the previous word on the list, with spatial locations on a well-known path, or with unique autobiographical events. After intensive rehearsal, subjects recalled the words during functional magnetic resonance imaging (fMRI). Common activity to all three types of encoding associations was identified in the posterior parietal cortex, in particular in the precuneus. Additionally, when associating spatial or autobiographical material, retrosplenial cortex activity was elicited during word list recall, while hippocampal activity emerged only for autobiographically associated words. These findings support a general, critical function of the precuneus in episodic memory storage and retrieval. The encoding-retrieval repetitions during learning seem to have accelerated hippocampus-independence and lead to direct neocortical integration in the sequentially associated and spatially associated word list tasks. During recall of words associated with autobiographical memories, the hippocampus might add spatiotemporal information supporting detailed scenic and contextual memories.
Introduction
The declarative memory system stores episodic memories, which represent information about events that have temporal, spatial and contextual details, and semantic memories, which hold facts and knowledge about the world (Renoult et al. 2019). Two systems have been investigated as locations of memory storage: the hippocampal formation and the neocortex, particularly prefrontal and parietal networks. The distinct functional contributions of these two systems have been a central, long-standing research question. Initially, the major distinctions between the medial temporal lobe (MTL)/hippocampus system and the neocortex were thought to be between conscious declarative memory and unconscious priming and perceptual learning (Squire et al. 2021), between recent and remote declarative memory (Gilboa and Moscovitch 2021), and between episodic memories including spatiotemporal details and more general semantic information (Moscovitch et al. 2016). Recent evidence puts these clear-cut distinctions into question. There is evidence that neocortical memory acquisition can occur within one training session (Brodt et al. 2016; Antony et al. 2017; Himmer et al. 2019), and even one-trial learning has been reported (Squire et al. 2021). Episodic memory representations have been found not only in the hippocampus, but also in the parietal cortex (Hutchinson et al. 2014; Gilmore et al. 2015; Sestieri et al. 2017). Especially, the precuneus has recently been identified as a central node in the episodic memory system (Mazzoni et al. 2019; Hebscher et al. 2019b; Ritchey and Cooper 2020). With both regions being implicated in episodic memory, it remains unclear which functional aspects of memory representation engage the hippocampal or the neocortical memory systems.
A number of properties have been closely associated with the hippocampus. It is involved in sequence learning (Buzsaki and Tingley 2018) and memory for temporal order (Lehn et al. 2009; Tubridy and Davachi 2011). Plus, it has been implicated in processing spatial sequences (Buzsaki and Moser 2013) and the sequential encoding of word lists by association with a spatial path (Fellner et al. 2016). Finally, its pivotal role for spatial and temporal context (Bird and Burgess 2008; Eichenbaum 2017; Banquet et al. 2021; Maurer and Nadel 2021) renders it of central importance to the (re)construction of detailed episodic and scenic information (Maguire et al. 2016; Thakral et al. 2020) and autobiographical memory (Cabeza and St Jacques 2007). The exact functional role of the precuneus in memory storage is less well known. It has been linked to attention-related functions (Cabeza et al. 2008), self-processing (Cavanna and Trimble 2006), and visual imagery (Fletcher et al. 1995; Wutte et al. 2012), and it is involved in memory retrieval rather than in encoding (Cavanna and Trimble 2006). Using functional and diffusion-weighted MRI, it was shown that there are enduring microstructural changes after learning in the precuneus, which indicate the actual location of an episodic memory engram (Brodt et al. 2018; Brodt and Gais 2020).
In the present study, our aim was to investigate the contribution of the neocortical and hippocampal memory subsystems to the recall of episodic memories, which are known to require the hippocampus during encoding. We were particularly interested in three types of episodic memory: sequences, spatial memories, and detailed autobiographical memories. We used a word list learning task in which concrete nouns had to be learned with the help of one of three types of encoding associations. The first experimental group associated each word sequentially with the next word on the list. The second group associated a location on a well-known spatial path with each word. In the third group, the task required subjects to associate each word with a detailed, autobiographical scene. After intensive rehearsal of the list and its associates through multiple encoding-recall repetitions, subjects recalled the words under identical conditions during fMRI.
Materials and methods
Subjects
Thirty-eight right-handed healthy volunteers with normal or corrected-to-normal vision gave written informed consent, in accordance with the Declaration of Helsinki, to participate in the study. The study was approved by the local ethics committee at the Department Psychology of the LMU. Subjects were monetarily compensated for their participation. Caffeinated or alcoholic beverages were not allowed on the day of scanning. One participant was excluded from the study prior to analysis due to an anatomical abnormality for a final cohort of 37 (21 female, mean age 24 ± 3 SD years). All participants had no history of neurological or psychiatric disorders and were not taking medication affecting the central nervous system.
Learning procedure
Forty mono or disyllabic German nouns were chosen from the Berlin Affective Word List for high concreteness and lack of emotional value (Vo et al. 2009). The same 40 words were learned by all subjects with the help of one of three encoding associations: (i) sequential, in which words were learned by repeated sequential association (n = 11), (ii) spatial, in which the learned words were associated with locations along a well-known path (n = 14), and (iii) autobiographical, in which each word was associated with a concrete, salient and personal memory (n = 12).
Learning took place in the presence of an experimenter on a computer using the Cogent 2000 toolbox (v1.33, Wellcome Trust Centre for Neuroimaging, London, UK) for Matlab (MathWorks, USA). Subjects were given written and verbal instructions about the learning condition and the task. In the sequential association group, participants viewed each word for 5 s and were instructed to form an association with the previously presented word. Visual imagery was explicitly discouraged, and subjects were asked to verbally repeat the word pair during learning over and over. Vocalization during learning was checked by the experimenter. After each repetition of the list, performance was tested by a verbal free-recall procedure. Learning was repeated until a criterion of 90% was reached. In the spatial association group, before learning, subjects were asked to choose a well-known path that they traveled regularly in the past and for which 40 salient locations could be clearly distinguished (e.g. their way to school or university). Subjects were then asked to name all the locations. This was repeated if the subject omitted one or more stations. When the locations were named without error, learning of the word list began. Words were presented for 7 s each in the first learning cycle to allow forming a concrete visual image and 5 s in all of the following cycles. Each cycle was followed by oral free recall of the list. Learning was repeated until 90% of the words could be recalled. The autobiographical association task followed the same procedure as the other two tasks. Here, subjects were given 20 s during the first learning-retrieval cycle in order to find a detailed, scenic autobiographical event to associate with the word. In all additional cycles, words were presented for 5 s. The order of the words in this task was different between the learning cycles in order to prevent sequential learning. In post-experimental questioning, subjects were able to describe the autobiographical events in detail and give an exact or approximate date for it. Only episodes that were concrete and detailed were considered in the analysis. Every-day events or events without a specific point in time were excluded. We use the term “autobiographical” in the context of our experiment to clarify that detailed episodic information from unique events in a person’s life was queried.
Recall procedure
fMRI of memory recall occurred approximately 30 min after the end of learning. Free verbal recall of the word list was used to test memory, using a procedure with appropriate inter-event timing for event-related scanning paradigms (Oztekin et al. 2010). The entire scanning session was divided into three segments: recollection, vocalization, and an odd-even number judgment control task. First, subjects were signaled to remember a word of the learned word list. Subjects were given at least 3 s or as much time as they needed to retrieve a word. When they remembered a word, they pressed a button, stated the word in a normal voice and then pressed the button again to start thinking about the next word. After every 10 words, subjects were given an odd-even digit task for 20 s. Here, numbers between 1 and 8 were presented and subjects were asked to press the left button of the keyboard for odd and the right button for even numbers. This control task was chosen as a baseline because it suppresses hippocampal activity (Stark and Squire 2001). When no more words could be remembered, the session was concluded. Three consecutive recall sessions were scanned.
MRI image acquisition
MR scanning was performed on a 3 T MRI Scanner (Signa HDx, General Electrics, Boston, MA) with a standard 8-channel head volume coil. Functional data were acquired using a gradient echo planar-imaging T2*-weighted sequence (TR 2.616 s., TE 40 ms, FOV 220 mm, matrix size 96 × 96, voxel size 2.29 × 2.29 × 3.5 mm3, 34 transversal slices). In addition, a T1-weighted structural image was acquired from each subject using a 3D fast spoiled gradient echo recalled (FSPGR) sequence (0.86 mm isotropic voxels).
Stimulus presentation and analysis
An LCD display projector (Christie Digital Systems, Germany) back-projected the visual stimulus via three front surface mirrors from a wave-guide through the Faraday cage onto a semi-transparent back-projection screen mounted behind the head coil. Subjects lay in a supine position with their head in the center of the magnet and viewed the stimulus via a mirror that reflected the images displayed onto the screen. The projection system resulted in a field-of-view of 25° horizontal and 19° vertical with a 1024 × 768 pixel resolution. Subjects responded via a 2-button, MR-compatible response device (Cedrus Corp., USA), and gave verbal responses over an MR-compatible optical microphone with preamplifier (MO2000, Sennheiser, Germany), connected to a USB audio interface (E-MU, Ireland), which were recorded in Matlab (MathWorks, USA) and analyzed offline. Recall events were separated into correct and incorrect, by two independent blind raters. ANOVAs and pair-wise Tukey HSD post hoc tests were used to test behavioral between group effects, including the number of repetitions required to achieve the threshold of 90% correct answers during the learning session, percent correct responses, and the average time required per word during the recall session.
fMRI data analysis
Image processing and data analysis were performed using the SPM8 toolbox for Matlab (Wellcome Trust Centre for Neuroimaging, University College London). All volumes were coregistered to the mean EPI image, and then spatially normalized to the MNI standard coordinate template echo planar image using the tissue segmentation parameters of the structural image. Functional images were spatially smoothed with an 8-mm full-width at half-maximum isotropic Gaussian kernel and temporally filtered with a 128-s high-pass filter. First-level analysis of fixed effects at the single subject level included regressors for key presses, vocalization, and correct word recall. Each regressor was modeled as an event convolved with the canonical hemodynamic response function. To account for motion artifacts, six motion regressors were added. Because of the increased variance in response time at the end of recall, the last two recalled words and their corresponding image volumes were not included in the analysis. The contrast images corresponding to the parameter estimates for correct word recall were used for the group-level random effects analysis, where a factorial design including the three learning associations was used. Group level effects were assessed with t-tests. The effect size Hedges’ g as well as corresponding 95% confidence intervals was calculated for each significant cluster at the peak voxel according to the procedure described in Gerchen et al. (2021). Areas of overlapping activity across groups were assessed using a conjunction for memory recall over all three groups using the conjunction null (Nichols et al. 2005). Percent signal change values were extracted from the parameter estimates of the first level model based upon the global mean activity over all brain voxels. Voxels exceeding a statistical threshold of P < 0.05 family-wise error (FWE) corrected for multiple comparisons in the full brain volume were used unless otherwise stated. Anatomical regions of interest were defined using the anatomical automatic labeling (AAL) toolbox (Tzourio-Mazoyer et al. 2002).
To examine the possible influence of the relative goodness of fit between the three experimental groups, we performed a similar analysis to the group-level mixed-effects analysis above, using the voxel-wise sum of squared residual error from each single-subject model.
Given that classical statistical methods in neuroimaging preclude inferences about the probability of no effect, we performed a supplementary analysis that used an empirical Bayesian approach to assess the functional segregation of learning conditions within the hippocampus by re-estimating the group level classical model (Friston and Penny 2003). Posterior probabilities (i.e. the probability of activity of a particular regressor given the data) were calculated based on a mean effect size of 2.2. All posterior probabilities above a threshold probability of P > 0.01 within the bilateral hippocampus were reported for this analysis.
Finally, to test the hypothesis that the hippocampal contribution declines with increasing age of a memory as derived from the assumptions of classical systems consolidation theory (Rekkas and Constable 2005), we correlated the age of the autobiographical events with the strength of the hemodynamic response in the hippocampus. The age of each autobiographical episode was recorded during post-scanning debriefing. Responses were converted into the number of days since the event and entered into an additional first-level model. Then, brain activity during autobiographical recall was parametrically modulated with the age of the episode in a linear model. The resulting contrast images for each subject were then entered into a second-level model to look for voxels with activity that correlated significantly with the age of each personal episode at an uncorrected level of P < 0.05. In addition, an average of all anatomically defined hippocampus voxels was correlated with memory age.
Results
We compared the neural correlates of memory retrieval of identical word lists that were encoded previously with three different types of associations. Healthy subjects had learned a list of concrete nouns to a criterion of 90% correct either using (i) sequential, (ii) spatial, or (iii) autobiographical associations (Fig. 1A). Approximately 30 min after learning, the subjects’ brain activity was measured during free verbal recall of the list (Fig. 1B). Recall settings were identical between groups.
Task and experimental design. (A) Learning: Subjects learned 40 concrete nouns with the help of one of three encoding associations. Sequential—subjects associated each list word with the previous word by vocal repetition. Spatial—subjects imagined the objects represented by the list words at locations along a well-known path. Autobiographical—subjects associated each word with a concrete, detailed autobiographical event. (B) Recall: Subjects were asked to recall all words of the list without external cueing during fMRI. For each word, they first received the instruction to remember the next word. Upon remembering, they pressed a button and verbalized the remembered word. After every 10th word, subjects performed a 20-second odd-even digit task, which served as baseline.
Brain activity during memory retrieval
In our first analysis, we investigated the areas that contribute to memory retrieval depending on how the memory content was learned. The sequential task showed significant activity in only a small number of regions (Fig. 2C): the strongest activity was found in the posterior precuneus (Brodmann Area [BA] 7), followed by the superior parietal lobule (SPL), and to a lesser extent in the middle occipital gyrus (BA 19) of both hemispheres (Table 1). Recall of words with spatial associates included the same regions as with sequential associates, but additionally showed significant activity in the retrosplenial cortex (BA 30). Finally, retrieval of words associated with autobiographical events showed more widespread activity. Again, as in the other two tasks, activation with the strongest peak and with the largest extent was the precuneus (BA 7). Additionally, again as for the previous two tasks, the middle occipital gyrus and the retrosplenial cortex showed activity. However, there was also a strong activity along the whole ventral visual pathway from lingual gyrus (BA 18) to fusiform gyrus (BA 37) to the MTL system including parahippocampal gyrus and hippocampus (Fig. 2D). Moreover, this task also induced activity in the caudate nucleus, the posterior cingulate (BA 23), and in frontal areas (middle frontal [BA 6], dorsolateral prefrontal [BA 9]).
Brain activity during word list retrieval after either sequential, spatial, or autobiographical association during learning. (A) Brain activity common to all three groups revealed by a conjunction analysis. (B) Average percent signal change during retrieval over the area of common activity in (A) after sequential (blue), spatial (red), and autobiographical (purple) association learning. No significant differences in signal strength were found between groups in this region (F2,34 = 1.3, P > 0.27). (C) Areas of significant retrieval activity in each of the groups separately seen in a glass brain view. A gradual increase in the areas recruited from sequential to spatial and autobiographical association learning can be seen. Color codes as in B. See also Table 1. (D) Location of the hippocampal retrieval activation for the word list associated with autobiographical information.
Regions of activity during recall in each group and in conjunction of all three groups.
| Region | Coordinates (x,y,z in mm) | Voxel level (z-score) | Cluster size (# of voxels) | Hedges’ g [CI] | |
|---|---|---|---|---|---|
| LH | RH | ||||
| Sequential association | |||||
| Precuneus (BA7) | −10, −74, 42 | 4.79 | 226 | 1.61 [0.77–2.69] | |
| 0, −70, 48 | 4.75 | 1.59 [0.75–2.65] | |||
| 12, −72, 48 | 4.34 | 84 | 1.41 [0.62–2.39] | ||
| SPL | 20, −74, 48 | 4.59 | 1.52 [0.70–2.55] | ||
| Middle occipital gyrus (BA19) | −30, −68, 38 | 4.56 | 41 | 1.51 [0.70–2.53] | |
| Spatial association | |||||
| Precuneus (BA7) | −6, −66, 46 | 5.17 | 365 | 1.62 [0.87–2.55] | |
| −10, −70, 52 | 4.87 | 1.49 [0.77–2.37] | |||
| 10, −66, 52 | 4.93 | 1.51 [0.79–2.40] | |||
| Middle occipital gyrus (BA19) | −34, −78, 32 | 5.37 | 208 | 1.71 [0.94–2.67] | |
| 40, −78, 28 | 5.17 | 95 | 1.62 [0.87–2.55] | ||
| Retrosplenial cortex (BA30) | −16, −58, 16 | 5.32 | 139 | 1.69 [0.92–2.64] | |
| 14, −54, 16 | 4.80 | 60 | 1.46 [0.75–2.32] | ||
| Autobiographical association | |||||
| Precuneus (BA7) | −32, −70, 38 | 5.75 | 687 | 2.03 [1.10–3.23] | |
| −34, −82, 24 | 5.22 | 1.75 [0.91–2.83] | |||
| Retrosplenial cortex Middle occipital gyrus (BA19) | −6, −68, 40 | 6.33 | 2620 | 2.38 [1.35–3.74] | |
| 10, −72, 44 | 5.79 | 2.05 [1.12–3.26] | |||
| 10, −54, 6 | 5.71 | 2.01 [1.09–3.20] | |||
| 34, −66, 34 | 5.09 | 395 | 1.69 [0.86–2.74] | ||
| 40, −76, 26 | 5.04 | 1.67 [0.85–2.71] | |||
| Caudate nucleus | −20, −24, 26 | 5.97 | 124 | 2.15 [1.19–3.41] | |
| Fusiform gyrus (BA37) | −34, −44, −8 | 5.49 | 269 | 1.89 [1.01–3.03] | |
| Lingual gyrus (BA18) | −2, −90, −10 | 4.59 | 1.47 [0.70–2.42] | ||
| Cerebellum | 8, −90, −10 | 4.90 | 329 | 1.60 [0.80–2.62] | |
| 18, −80, −16 | 4.98 | 1.64 [0.83–2.67] | |||
| Parahippocampal gyrus and Hippocampus | −8, −84, −16 | 5.01 | 100 | 1.65 [0.84–2.69] | |
| 34, −42, −2 | 4.86 | 28 | 1.58 [0.79–2.59] | ||
| Posterior cingulate (BA23) | 0, −30, 24 | 4.35 | 10 | 1.36 [0.62–2.28] | |
| Middle frontal gyrus (BA6) | −30, 18, 58 | 5.03 | 152 | 1.66 [0.84–2.70] | |
| Dorsolateral prefrontal cortex (BA9) | −50, 26, 30 | 4.62 | 14 | 1.48 [0.71–2.44] | |
| Conjunction | |||||
| Precuneus (BA7) SPL | −6, −68, 44 | 4.78 | 134 | ||
| 8, −70, 48 | 4.36 | ||||
| 12, −72, 48 | 4.34 | 5 | |||
| Middle occipital gyrus (BA19) | −30, −72, 48 | 4.37 | 3 | ||
| Region | Coordinates (x,y,z in mm) | Voxel level (z-score) | Cluster size (# of voxels) | Hedges’ g [CI] | |
|---|---|---|---|---|---|
| LH | RH | ||||
| Sequential association | |||||
| Precuneus (BA7) | −10, −74, 42 | 4.79 | 226 | 1.61 [0.77–2.69] | |
| 0, −70, 48 | 4.75 | 1.59 [0.75–2.65] | |||
| 12, −72, 48 | 4.34 | 84 | 1.41 [0.62–2.39] | ||
| SPL | 20, −74, 48 | 4.59 | 1.52 [0.70–2.55] | ||
| Middle occipital gyrus (BA19) | −30, −68, 38 | 4.56 | 41 | 1.51 [0.70–2.53] | |
| Spatial association | |||||
| Precuneus (BA7) | −6, −66, 46 | 5.17 | 365 | 1.62 [0.87–2.55] | |
| −10, −70, 52 | 4.87 | 1.49 [0.77–2.37] | |||
| 10, −66, 52 | 4.93 | 1.51 [0.79–2.40] | |||
| Middle occipital gyrus (BA19) | −34, −78, 32 | 5.37 | 208 | 1.71 [0.94–2.67] | |
| 40, −78, 28 | 5.17 | 95 | 1.62 [0.87–2.55] | ||
| Retrosplenial cortex (BA30) | −16, −58, 16 | 5.32 | 139 | 1.69 [0.92–2.64] | |
| 14, −54, 16 | 4.80 | 60 | 1.46 [0.75–2.32] | ||
| Autobiographical association | |||||
| Precuneus (BA7) | −32, −70, 38 | 5.75 | 687 | 2.03 [1.10–3.23] | |
| −34, −82, 24 | 5.22 | 1.75 [0.91–2.83] | |||
| Retrosplenial cortex Middle occipital gyrus (BA19) | −6, −68, 40 | 6.33 | 2620 | 2.38 [1.35–3.74] | |
| 10, −72, 44 | 5.79 | 2.05 [1.12–3.26] | |||
| 10, −54, 6 | 5.71 | 2.01 [1.09–3.20] | |||
| 34, −66, 34 | 5.09 | 395 | 1.69 [0.86–2.74] | ||
| 40, −76, 26 | 5.04 | 1.67 [0.85–2.71] | |||
| Caudate nucleus | −20, −24, 26 | 5.97 | 124 | 2.15 [1.19–3.41] | |
| Fusiform gyrus (BA37) | −34, −44, −8 | 5.49 | 269 | 1.89 [1.01–3.03] | |
| Lingual gyrus (BA18) | −2, −90, −10 | 4.59 | 1.47 [0.70–2.42] | ||
| Cerebellum | 8, −90, −10 | 4.90 | 329 | 1.60 [0.80–2.62] | |
| 18, −80, −16 | 4.98 | 1.64 [0.83–2.67] | |||
| Parahippocampal gyrus and Hippocampus | −8, −84, −16 | 5.01 | 100 | 1.65 [0.84–2.69] | |
| 34, −42, −2 | 4.86 | 28 | 1.58 [0.79–2.59] | ||
| Posterior cingulate (BA23) | 0, −30, 24 | 4.35 | 10 | 1.36 [0.62–2.28] | |
| Middle frontal gyrus (BA6) | −30, 18, 58 | 5.03 | 152 | 1.66 [0.84–2.70] | |
| Dorsolateral prefrontal cortex (BA9) | −50, 26, 30 | 4.62 | 14 | 1.48 [0.71–2.44] | |
| Conjunction | |||||
| Precuneus (BA7) SPL | −6, −68, 44 | 4.78 | 134 | ||
| 8, −70, 48 | 4.36 | ||||
| 12, −72, 48 | 4.34 | 5 | |||
| Middle occipital gyrus (BA19) | −30, −72, 48 | 4.37 | 3 | ||
All activations significant at PFWE < 0.05 (whole volume corrected for multiple comparisons). Coordinates are presented in Montreal Neurological Institute space. If no cluster size is reported then the coordinate belongs to the previous cluster. Regions were characterized based on subject anatomy, as well as the Harvard Oxford and Juelich atlases. Effect sizes are reported for peak voxel activation and with a confidence interval of 95%. LH: left hemisphere, RH: right hemisphere, CI: confidence interval
Regions of activity during recall in each group and in conjunction of all three groups.
| Region | Coordinates (x,y,z in mm) | Voxel level (z-score) | Cluster size (# of voxels) | Hedges’ g [CI] | |
|---|---|---|---|---|---|
| LH | RH | ||||
| Sequential association | |||||
| Precuneus (BA7) | −10, −74, 42 | 4.79 | 226 | 1.61 [0.77–2.69] | |
| 0, −70, 48 | 4.75 | 1.59 [0.75–2.65] | |||
| 12, −72, 48 | 4.34 | 84 | 1.41 [0.62–2.39] | ||
| SPL | 20, −74, 48 | 4.59 | 1.52 [0.70–2.55] | ||
| Middle occipital gyrus (BA19) | −30, −68, 38 | 4.56 | 41 | 1.51 [0.70–2.53] | |
| Spatial association | |||||
| Precuneus (BA7) | −6, −66, 46 | 5.17 | 365 | 1.62 [0.87–2.55] | |
| −10, −70, 52 | 4.87 | 1.49 [0.77–2.37] | |||
| 10, −66, 52 | 4.93 | 1.51 [0.79–2.40] | |||
| Middle occipital gyrus (BA19) | −34, −78, 32 | 5.37 | 208 | 1.71 [0.94–2.67] | |
| 40, −78, 28 | 5.17 | 95 | 1.62 [0.87–2.55] | ||
| Retrosplenial cortex (BA30) | −16, −58, 16 | 5.32 | 139 | 1.69 [0.92–2.64] | |
| 14, −54, 16 | 4.80 | 60 | 1.46 [0.75–2.32] | ||
| Autobiographical association | |||||
| Precuneus (BA7) | −32, −70, 38 | 5.75 | 687 | 2.03 [1.10–3.23] | |
| −34, −82, 24 | 5.22 | 1.75 [0.91–2.83] | |||
| Retrosplenial cortex Middle occipital gyrus (BA19) | −6, −68, 40 | 6.33 | 2620 | 2.38 [1.35–3.74] | |
| 10, −72, 44 | 5.79 | 2.05 [1.12–3.26] | |||
| 10, −54, 6 | 5.71 | 2.01 [1.09–3.20] | |||
| 34, −66, 34 | 5.09 | 395 | 1.69 [0.86–2.74] | ||
| 40, −76, 26 | 5.04 | 1.67 [0.85–2.71] | |||
| Caudate nucleus | −20, −24, 26 | 5.97 | 124 | 2.15 [1.19–3.41] | |
| Fusiform gyrus (BA37) | −34, −44, −8 | 5.49 | 269 | 1.89 [1.01–3.03] | |
| Lingual gyrus (BA18) | −2, −90, −10 | 4.59 | 1.47 [0.70–2.42] | ||
| Cerebellum | 8, −90, −10 | 4.90 | 329 | 1.60 [0.80–2.62] | |
| 18, −80, −16 | 4.98 | 1.64 [0.83–2.67] | |||
| Parahippocampal gyrus and Hippocampus | −8, −84, −16 | 5.01 | 100 | 1.65 [0.84–2.69] | |
| 34, −42, −2 | 4.86 | 28 | 1.58 [0.79–2.59] | ||
| Posterior cingulate (BA23) | 0, −30, 24 | 4.35 | 10 | 1.36 [0.62–2.28] | |
| Middle frontal gyrus (BA6) | −30, 18, 58 | 5.03 | 152 | 1.66 [0.84–2.70] | |
| Dorsolateral prefrontal cortex (BA9) | −50, 26, 30 | 4.62 | 14 | 1.48 [0.71–2.44] | |
| Conjunction | |||||
| Precuneus (BA7) SPL | −6, −68, 44 | 4.78 | 134 | ||
| 8, −70, 48 | 4.36 | ||||
| 12, −72, 48 | 4.34 | 5 | |||
| Middle occipital gyrus (BA19) | −30, −72, 48 | 4.37 | 3 | ||
| Region | Coordinates (x,y,z in mm) | Voxel level (z-score) | Cluster size (# of voxels) | Hedges’ g [CI] | |
|---|---|---|---|---|---|
| LH | RH | ||||
| Sequential association | |||||
| Precuneus (BA7) | −10, −74, 42 | 4.79 | 226 | 1.61 [0.77–2.69] | |
| 0, −70, 48 | 4.75 | 1.59 [0.75–2.65] | |||
| 12, −72, 48 | 4.34 | 84 | 1.41 [0.62–2.39] | ||
| SPL | 20, −74, 48 | 4.59 | 1.52 [0.70–2.55] | ||
| Middle occipital gyrus (BA19) | −30, −68, 38 | 4.56 | 41 | 1.51 [0.70–2.53] | |
| Spatial association | |||||
| Precuneus (BA7) | −6, −66, 46 | 5.17 | 365 | 1.62 [0.87–2.55] | |
| −10, −70, 52 | 4.87 | 1.49 [0.77–2.37] | |||
| 10, −66, 52 | 4.93 | 1.51 [0.79–2.40] | |||
| Middle occipital gyrus (BA19) | −34, −78, 32 | 5.37 | 208 | 1.71 [0.94–2.67] | |
| 40, −78, 28 | 5.17 | 95 | 1.62 [0.87–2.55] | ||
| Retrosplenial cortex (BA30) | −16, −58, 16 | 5.32 | 139 | 1.69 [0.92–2.64] | |
| 14, −54, 16 | 4.80 | 60 | 1.46 [0.75–2.32] | ||
| Autobiographical association | |||||
| Precuneus (BA7) | −32, −70, 38 | 5.75 | 687 | 2.03 [1.10–3.23] | |
| −34, −82, 24 | 5.22 | 1.75 [0.91–2.83] | |||
| Retrosplenial cortex Middle occipital gyrus (BA19) | −6, −68, 40 | 6.33 | 2620 | 2.38 [1.35–3.74] | |
| 10, −72, 44 | 5.79 | 2.05 [1.12–3.26] | |||
| 10, −54, 6 | 5.71 | 2.01 [1.09–3.20] | |||
| 34, −66, 34 | 5.09 | 395 | 1.69 [0.86–2.74] | ||
| 40, −76, 26 | 5.04 | 1.67 [0.85–2.71] | |||
| Caudate nucleus | −20, −24, 26 | 5.97 | 124 | 2.15 [1.19–3.41] | |
| Fusiform gyrus (BA37) | −34, −44, −8 | 5.49 | 269 | 1.89 [1.01–3.03] | |
| Lingual gyrus (BA18) | −2, −90, −10 | 4.59 | 1.47 [0.70–2.42] | ||
| Cerebellum | 8, −90, −10 | 4.90 | 329 | 1.60 [0.80–2.62] | |
| 18, −80, −16 | 4.98 | 1.64 [0.83–2.67] | |||
| Parahippocampal gyrus and Hippocampus | −8, −84, −16 | 5.01 | 100 | 1.65 [0.84–2.69] | |
| 34, −42, −2 | 4.86 | 28 | 1.58 [0.79–2.59] | ||
| Posterior cingulate (BA23) | 0, −30, 24 | 4.35 | 10 | 1.36 [0.62–2.28] | |
| Middle frontal gyrus (BA6) | −30, 18, 58 | 5.03 | 152 | 1.66 [0.84–2.70] | |
| Dorsolateral prefrontal cortex (BA9) | −50, 26, 30 | 4.62 | 14 | 1.48 [0.71–2.44] | |
| Conjunction | |||||
| Precuneus (BA7) SPL | −6, −68, 44 | 4.78 | 134 | ||
| 8, −70, 48 | 4.36 | ||||
| 12, −72, 48 | 4.34 | 5 | |||
| Middle occipital gyrus (BA19) | −30, −72, 48 | 4.37 | 3 | ||
All activations significant at PFWE < 0.05 (whole volume corrected for multiple comparisons). Coordinates are presented in Montreal Neurological Institute space. If no cluster size is reported then the coordinate belongs to the previous cluster. Regions were characterized based on subject anatomy, as well as the Harvard Oxford and Juelich atlases. Effect sizes are reported for peak voxel activation and with a confidence interval of 95%. LH: left hemisphere, RH: right hemisphere, CI: confidence interval
To identify regions commonly involved in word list recall regardless of encoding associations, we conducted a conjunction analysis (Fig. 2A). All three retrieval tasks showed significant activation in the posterior precuneus extending into the SPL, and to a lesser extent in the middle occipital gyrus of both hemispheres (Table 1, Fig. 2B). The three groups showed no difference in activity over all voxels active in the conjunction analysis even at an uncorrected threshold of P < 0.001, suggesting that responses in these parietal-occipital areas are comparable over groups and that the parietal cortex plays a central role in the recall of well-rehearsed word lists.
The hippocampus only supports recall of information encoded with autobiographical associations
The above analyses were done on a whole-brain level, and we chose a rather conservative FWE correction for multiple comparisons. Also with a more liberal uncorrected significance threshold of punc < 0.001, significant hippocampal activity was found only in the autobiographical group. These results were corroborated by a further region of interest (ROI) analysis on the anatomical hippocampus (Fig. 3A). Furthermore, a direct comparison between brain activity from the autobiographical and spatial association tasks revealed significant differences in activity only in the left posterior hippocampus and in the left lingual gyrus, with higher activity after autobiographically associated encoding (Fig. 3B). Similarly, average percent signal change over the entire bilateral hippocampus was elevated only during recall of words that had been linked with autobiographical events (Fig. 3C). Percent signal change was significantly higher than in the other two groups (F2,34 = 5.36, P = 0.009).
Hippocampal activity during the word list retrieval task after association with autobiographical memory during encoding. (A) An anatomically defined region of interest analysis (blue) showed bilateral posterior hippocampal activity (red-yellow) only during autobiographical recall. No other groups showed any significant activity in this search region, even at liberal significance thresholds. (B) Significantly higher left hippocampus activity after autobiographical encoding compared to after spatial encoding. (C) Percent signal change over the bilateral hippocampi was significantly higher during word recall after autobiographical encoding compared to after spatial and sequential encoding. Error bars represent SEM. *: P < 0.05, **: P < 0.01. (D) Maximum posterior probability for activity in the hippocampus. C and D use the same ROI as A.
In a supplementary analysis, the data were analyzed using an empirical Bayesian approach to examine activity in the hippocampus with respect to the probability of no effect. Using the global mean as the comparative effect size, we found a maximum posterior probability over the whole hippocampus of 0.01 and 0.1 for the sequential and spatial group, respectively, whereas the posterior probability in the autobiographical group was above 0.84 (Fig. 3D). This analysis also suggests that retrieval of words under sequential and spatial encoding conditions was indeed independent of the hippocampus.
In further analyses, we tested whether the activation differences between groups could be related to the goodness of fit of the first-level models. Timing differences, due to the subject-controlled nature of the experimental paradigm, could lead to a systematically different model. Then, additional variance not explained by the model would be found in the residuals of the single subject models. We therefore tested for group differences in the residuals by creating a group-level model of the residuals with a regressor for the encoding task. No voxel-wise differences in the model residuals from the single subject analysis were found between experimental groups, even at an uncorrected significance level of P < 0.01. Next, we looked at different time points within the retrieval phases. Here, the aim was to exclude the possibility that the lack of hippocampal activation after spatial and sequential encoding was due to differences in the duration of retrieval. We therefore created additional first-level models in which the time points of the recall events were delayed by 1, 2, or 3 seconds. In these models, we observed increasing levels of motor activity for later time points, suggesting motor preparation for speech. Apart from that, the patterns of brain activity remained similar to those without the delay factor. Together, all these analyses are in accordance with a specific role of the hippocampus for retrieving rehearsed word lists when they were associated with detailed, episodic memories of autobiographical events, but not with sequential or spatial associations.
Hippocampal activity is unrelated to the age of the autobiographical event
In a supplementary analysis testing for a negative relationship between hippocampal contribution and memory age, no correlation was found between activity in the hippocampus during word list recall and the age of the autobiographical episodes associated to the words during encoding. First, a whole brain analysis yielded no significant clusters for either positive or negative correlations, even at uncorrected significance levels (all P > 0.05, uncorrected). Second, the pattern of activity including the posterior hippocampus that was found during autobiographical memory retrieval remained the same after correcting for the age of the memory. Finally, the correlation of average activity in the hippocampus with individual memory events for individual subjects was not systematically related to memory age (Fig. 4). Taken together, these results suggest that hippocampal activity found during word recall was not influenced by the age of the associated autobiographical memory.
The age of autobiographic memories and hippocampal activity. Hippocampal activity shows no systematic relationship to the time since the episodic memories were formed (all P > 0.3). The within-subject correlation is shown between the age of the autobiographic episode and the average hippocampal activity during recall. Each symbol represents the data points from a single subject, and the regression lines are given in their corresponding shade.
Behavioral measures
The behavioral aspects of memory retrieval were similar between the three groups. To achieve comparable levels of recall performance in the learning session, the number of repetitions required to reach the 90% correct threshold was higher after sequential association (5.09 ± 2.8, M ± SD) than after autobiographical (3.42 ± 0.5) and spatial association (2.07 ± 0.62; ANOVA: F2,34 = 11.58, P < 0.001; Tukey HSD post hoc tests: P = 0.04, P < 0.001, respectively). Differences in the number of repetitions needed to reach the learning criterion (sequential>autobiographical>spatial) do not correspond with the extent of brain activity during recall in the respective group (autobiographical> spatial>sequential), and also do not explain the observed differences in brain activity (see Supplementary Table 1). During memory retrieval, the percentage of correctly recalled words remained above 90% in all three groups, but performance was higher when recalling spatially associated (98.70 ± 3.26) than sequentially (94.78 ± 6.01) or autobiographically (93.28 ± 5.32) associated lists (ANOVA: F2,34 = 4.30, P = 0.02; Tukey HSD post hoc tests: P = 0.11, P = 0.02, respectively). The average time to recall a word did not differ significantly between the three groups (ANOVA: F2,34 = 2.14, P = 0.13). An exploratory analysis using performance as a behavioral covariate in the group level analysis did not reveal any changes in the pattern of brain activity, indicating that the observed differences between the groups are not related to differences in retrieval difficulty.
Discussion
In this study, subjects learned a list of concrete nouns together with one of three types of associations, all of which are known to activate the hippocampus during encoding. The lists were encoded either in a sequential manner by associating each word with the previous one, or with locations on a well-known spatial path, or with unique autobiographical events. Because we were interested in the hippocampal involvement during retrieval of well-rehearsed memories, which have been shown to undergo rapid neocorticalization (Brodt et al. 2016; Brodt et al. 2018; Himmer et al. 2019), the lists were learned until a 90% criterion was achieved. The common denominator of brain activity during word list retrieval independent of association was activity in the posterior parietal cortex (PPC), in particular the precuneus. Depending on which kind of material had been associated with the words during learning, additional areas were involved in memory retrieval.
Precuneus activity during memory retrieval
There is now growing consensus that the precuneus fulfills a genuine mnemonic function in episodic memory retrieval (Gilmore et al. 2015; Mazzoni et al. 2019; Hebscher et al. 2019b; Ritchey and Cooper 2020), while its specific contribution is still under debate. A number of different classical declarative memory tasks elicit precuneus activity, such as autobiographical, source memory retrieval (Lundstrom et al. 2005; Hebscher et al. 2019a), and visuospatial memory (Brodt et al. 2016; Schott et al. 2019). It is also involved in memory for abstract content (Krause et al. 1999; Klostermann et al. 2008). Both familiarity and recollection invoke the precuneus (Frithsen and Miller 2014), it correlates with the vividness of a memory (Richter et al. 2016), and it has been linked to retrieval success (Lundstrom et al. 2005; Gilmore et al. 2015; Brodt et al. 2018; Tan et al. 2020). Earlier views of the specific roles of the precuneus during memory retrieval have emphasized auxiliary roles of the precuneus like attentional processes (Cabeza et al. 2008), the accumulation of memory-based evidence (Shannon and Buckner 2004; Wagner et al. 2005; Rutishauser et al. 2018), or visual imagery (Fletcher et al. 1995; Handy et al. 2004). In particular, world-list recall of the spatial and autobiographical associations likely comprises a strong visual component, as vivid memories and self-referential information were associated. However, visual imagery was explicitly discouraged in the experimental group that performed sequential learning and there was no difference in precuneus activity between these different types of associations, calling into question that visual imagery alone could explain the observed findings. Already in an older PET study, Krause and colleagues (Krause et al. 1999) report exclusive precuneus activation during word-pair retrieval independent of learning modality and imagery content, indicating a role of the precuneus in memory retrieval not solely related to visual imagery. More recent theories propose that the PPC might actually hold relations among entities in physical or in abstract conceptual space (Summerfield et al. 2020) and even represents semantic concepts (Fernandino et al. 2022). During our sequential learning task, relations between concrete nouns with existing semantic concepts were formed. As the only region being activated during retrieval of these sequentially associated words, the precuneus is a prime candidate to be involved in processing this information. As a high-order association area, it is perfectly suited to store this kind of relational information and evidence suggests it can actually hold engrams of learned information (Brodt and Gais 2020), which are manifest in persistent, learning-induced microstructural changes after repeated encoding of object-location associations (Brodt et al. 2018). The relevance of the precuneus for declarative memory retrieval is supplemented by recent clinical evidence indicating that parietal lobe atrophy can impair delayed autobiographical memory retrieval, especially in the recollection of episodic details of remote memories (Ahmed et al. 2018).
The retrosplenial cortex and spatial associations
In the second experimental group, sequential learning was aided by the addition of a highly familiar spatial structure. Encoding word lists in this manner have been reported previously to evoke hippocampal activity (Fellner et al. 2016). Word retrieval elicited activity in the same areas as sequential learning but additionally in the retrosplenial cortex (RSC). This region contributes to mental navigation and to the spatial aspects of episodic memory, as retrosplenial lesions can damage spatial memories (Wagner et al. 2005; Vann et al. 2009). The RSC has further been reported to be important for anchoring cognitive spatial maps to environmental landmarks (Epstein et al. 2017). As the RSC is highly interconnected with both subcortical and cortical areas and is densely connected to the medial temporal lobe (Vann et al. 2009), it constitutes a prime candidate for the integration of newly acquired episodic and semantic information into existing knowledge (Kaboodvand et al. 2018; de Sousa et al. 2019). We propose that the spatial information about the well-learned path is stored in the RSC (Milczarek et al. 2018) and adds an additional access vector to the word list information stored in the precuneus. Interestingly, adding spatial information per se seems to be insufficient to elicit a strong hippocampal response.
The hippocampus is preferentially involved in autobiographically associated word list retrieval
The third experimental group associated the words on the list with unique, scenic autobiographical events, which resulted in the most widespread neural activity during retrieval. Activity included all regions elicited in sequential and spatial recall but extended into the parahippocampal gyrus and hippocampus and included other regions like the dorsolateral prefrontal cortex. One obvious difference to the other two tasks is that during autobiographical encoding, no sequential structure was built, instead recall required a memory search strategy and working memory for already recalled words. This fits well with the additional activity in the DLPFC, which is central to executive function and working memory (Rottschy et al. 2012). Hippocampal activity might also be a reflection of the short-term memory processes required to remember the list words that have already been recalled (Manohar et al. 2017). However, we propose that the hippocampal activity during word list recall with associated autobiographical events also arises from a specific hippocampal contribution to episodic long-term memory. There are several main aspects of episodic memory that are suggested to evoke hippocampal activity: recency of the memory (Squire and Wixted 2011), the detailed scenic quality (Barry and Maguire 2019), item-context binding (Yonelinas et al. 2019), or the spatial and temporal detail (Eichenbaum 2017). Sequential and spatial associations, which place the words into a temporal or spatial context, did not evoke a hippocampal response. Only the combination of spatial and temporal context details of the associated autobiographical memories elicited hippocampal activity. Activation was found in old as well as recent autobiographical memories, and the strength of hippocampal activity did not diminish with the age of the autobiographical events linked to each word. Considering the idea that autobiographical memories stay hippocampus dependent as long as they retain their precise episodic qualities (Sekeres et al. 2018), we propose that hippocampal activity in the autobiographical association task mainly represents the detailed scenic reconstruction of the autobiographical events during word recall (however, see Gilmore et al. (2021)). Furthermore, the hippocampus-dependence in the autobiographical condition might indicate a freshly encoded memory trace holding an updated version of the existing memory including the association of the studied word (Moscovitch et al. 2016; Sekeres et al. 2018).
The lack of hippocampal activity in the sequential and the spatial association tasks is consistent with previous findings that also did not find any hippocampal activity during free recall of associated words (Krause et al. 1999) nor during sequential image learning (Konovalov and Krajbich 2018). However, our findings do not contradict a necessary contribution of the hippocampus during the encoding of new information (Duss et al. 2014) nor its role for the integration of information into existing knowledge networks (Shohamy and Wagner 2008; Zeithamova et al. 2012) as we concentrated on the retrieval process. The absence of hippocampal activity can be linked to the fact that the hippocampus displays a diminishing role in memory retrieval when the learning material is repeatedly rehearsed (Brodt et al. 2016; Himmer et al. 2019). Especially repeated retrieval of newly acquired information can accelerate integration into neocortical networks, reduce the time required for the formation of a neocortical memory trace, and prompt hippocampal independence (Antony et al. 2017; Brodt et al. 2018). Neocortical integration, particularly in the precuneus, might have been further facilitated by the fact that word lists consisted of concrete nouns that represent existing concepts in the semantic network and that only associations between these existing schemas had to be learned (Binder and Desai 2011).
Conclusion
In conclusion, our data demonstrate that retrieval of word lists centrally involves the PPC, most prominently the precuneus, regardless of the associated material. Depending on the association, additional regions are recruited. Most notably, the hippocampus comes into play when detailed autobiographical memories are associated. We suggest that the hippocampus most likely provides the scenic and contextual aspects of the memory and adds specific spatiotemporal information. Although this study cannot elucidate the function of the hippocampus in detail, what is most striking is the lack of hippocampal activity during recall in the sequential and spatial tasks. Finally, our findings support previous studies that show hippocampal independence without previous offline consolidation. Altogether, we suggest that the PPC is able to rapidly represent memories for established semantic concepts, while additional MTL regions are recruited for specific accompanying episodic features, providing an additional access route to the parietally stored concept memories. Future studies should investigate the existence of actual memory representations in the PPC, as this is beyond the scope of the current study looking only at retrieval-related activity.
Funding
This work was supported by the German Research Foundation (GA730/3–1).
Conflict of interest statement
The authors declare no competing financial interests.
