Functional Specialization and Coordination in the Amygdala-Hippocampus Circuit Support Working Memory

doi:10.21203/rs.3.rs-1500621/v1

Download PDF

Article

Functional Specialization and Coordination in the Amygdala-Hippocampus Circuit Support Working Memory

https://doi.org/10.21203/rs.3.rs-1500621/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this older preprint version

Read the latest preprint version →

Both the hippocampus and amygdala are involved in working memory (WM) processing. However, it is still an open question how the two structures interact to represent and maintain WM content. Here, we simultaneously recorded intracranial EEG from the amygdala and hippocampus of epilepsy patients while performing a WM task. Using a series of univariate, multivariate, and connectivity analyses, our results revealed a functional specialization of the amygdala-hippocampus circuit: The mnemonic representations in the amygdala were highly distinct but decreased from encoding to maintenance. The hippocampal representations, however, were less specific but remained stable in the absence of stimulus. Furthermore, the amygdala and hippocampus coordinated their activity during WM: The hippocampal representation was reinstated in the amygdala during maintenance, with task induced low-frequency information flow from the hippocampus to the amygdala. Importantly, the functional specificity and coordination were correlated only when they appeared in correct trials. Our study provides new evidence that successful WM processing in humans relies on specialized and coordinated functions within the amygdala-hippocampus circuit.

Human working memory (WM) processing involves the hippocampus and amygdala, as evidenced by persistent single neuron firing. However, it remains unclear what role the two regions play and how they interact during WM. By combining intracranial EEG recordings from the two regions and representational analysis, we show that the mnemonic representations during encoding in the amygdala are more specific but decrease during maintenance. Hippocampal representations are less specific but remain stable during maintenance. We also reveal coordinated representational reinstatement and information flow between the amygdala and hippocampus. These results suggest that functional specialization and coordination are fundamental to the formation and retention of WM.

Working memory (WM) is fundamental for human cognition and allows storing information in an active and readily available state for a short period of time (1). Studies have reported widely distributed brain regions, including the prefrontal, parietal, and sensory cortices, that are essential for forming and maintaining working memory content (2). However, whether the amygdala-hippocampus circuit participates in working memory processes has remained unclear. The amygdala is classically associated with emotional processing (3), while recent studies also showed that the amygdala has multidimensional response properties (4) and plays a role even in memorizing non-emotional stimulus material (5). On the other hand, beyond its well-known role in forming new long-term memories (6), the hippocampus is also implicated in short-term storage of information under high memory load or interference from distractors (7). Leveraging the high spatiotemporal resolution (8), recent intracranial EEG (iEEG) recordings in the amygdala and the hippocampus have revealed persistent firing as a mechanism for WM at the cellular level (9–11). However, is still unknown how the hippocampus and amygdala represent WM content at the neural population level.

Multivariate analysis methods such as representational similarity analysis have been used to inform the representation of specific components (see ref. (12) for a review). Research has identified two properties related to memory performance, the representational dissimilarity between distinct pieces of information (13) and the reinstatement of memory representations (14). The amygdala is known as a detector of goal-related stimuli (15) and receives major projections from the anterior temporal lobe (16) that convey highly processed object information (17). On the other hand, the hippocampus is crucial for memory consolidation (18). For instance, recent studies suggested notable overlap in representational patterns between the encoding and the post-encoding period in the hippocampus (14, 19). However, no study has yet simultaneously tracked amygdala and hippocampal representations in humans and addressed their interaction specifically in WM.

Tract tracing studies have uncovered structural connections between the amygdala and the hippocampus (20). This anatomical connectivity suggests a functional communication between these structures. This is supported by rodent studies (21, 22) showing that activation by direct electrical stimulation of one region can improve synaptic plasticity in the other region and by human studies (5) indicating that stimulation of the amygdala led to increased power in the hippocampus and improved subsequent memory even during non-emotional events. Further evidence came from human iEEG studies demonstrating functional interactions between these two structures during episodic memory processing (14, 23). However, it remains to be established whether the engagement of the two structures is coordinated or independent.

To address these issues, we recorded iEEG simultaneously from the amygdala and the hippocampus in 14 pre-surgical epilepsy patients while they performed a WM task. Here we utilized three analytical approaches: univariate activation analysis (Fig. 1A), multivariate representational similarity analysis (RSA) (Fig. 1(B-D)), and connectivity analysis (Fig. 1(E)). We found that the amygdala contributed directly to WM by constructing highly distinct mnemonic representations during encoding while the hippocampus participated in WM by spontaneously reinstating stimulus-driven activity patterns during maintenance. Next, the information in the hippocampus during encoding was reinstated in the amygdala during maintenance, which indicates coordinated reinstatement. This was accompanied by unidirectional information flow from the hippocampus to the amygdala during encoding and maintenance. Finally, this functional specialization and coordination pattern was predictive of later memory performance. Together, these findings provide evidence for functional specialization and coordinated memory reinstatement between the amygdala and the hippocampus as a mechanism that supports WM processing.

Task, behavior and recording channels

Fourteen patients with drug resistant epilepsy (7 females) (Table 1) performed a modified Sternberg WM task (65 total sessions from 14 subjects) during an invasive presurgical evaluation. In this task, the items were presented simultaneously rather than sequentially, thus separating the encoding period from the maintenance period. In each trial, the subject was asked to memorize a set of 4, 6, or 8 letters presented for 2 s (encoding). After a delay (maintenance) period of 3 s, a probe letter was presented and the subject responded whether the probe letter was identical to one of the letters held in memory (retrieval) (Fig. 2(A)). The average accuracy was 91.9%±3.2% (range 86.1%-97.6%). The mean response time was faster for correct than incorrect trials (1.44 ± 0.36 versus 1.95 ± 0.66 seconds, paired t test: t (13) = -4.15, p = .0011). Hence, the subjects performed well in the task. Unless stated otherwise, we report results for correct trials only.

Table 1

Subject characteristics
Subject	Age	Gender	Recording sites (amygdala)	Recording sites (hippocampus)	Retrieval Accuracy	RT for correct trials (mean ± std)
1	24	Female	2	6	0.925	1.26 ± 0.63
2	39	Male	4	8	0.864	1.28 ± 0.65
3	18	Female	2	6	0.932	1.10 ± 0.34
4	28	Male	4	8	0.949	1.48 ± 0.63
5	20	Female	2	4	0.913	1.41 ± 0.90
6	31	Male	4	8	0.943	1.47 ± 0.59
7	47	Male	4	8	0.949	1.44 ± 0.57
8	56	Female	4	7	0.900	1.41 ± 0.51
9	19	Female	4	8	0.928	1.24 ± 0.33
10	35	Male	4	6	0.930	1.64 ± 0.73
11	51	Female	4	6	0.914	1.30 ± 0.58
12	30	Male	4	7	0.895	1.49 ± 0.77
13	29	Female	4	4	0.976	1.05 ± 0.28
14	56	Male	4	8	0.861	2.58 ± 1.22

Local field potentials (LFPs) were recorded simultaneously from depth electrodes implanted in the amygdala and hippocampus (Fig. 2(B)). In total from all subjects, we recorded from 94 channels in the hippocampus and 50 channels in the amygdala (see the details in Methods).

Increased theta-alpha power in the amygdala and hippocampus

We calculated the time-frequency power for each channel within the amygdala and the hippocampus, and the power outputs were z-scored against pretrial (500 ms) baseline distributions to assess the significance of the task-induced power effects per trial. Both the amygdala and the hippocampus showed sustained activity in the frequency range (1–40 Hz), especially during the encoding and the maintenance phase (Fig. 3(A)), suggesting that both the amygdala and the hippocampus are involved in WM. The elevated activity in low-frequency oscillation in the present study matches the results reported in our previous study in the hippocampus during WM processing (24). We then averaged the spectral power during encoding and maintenance and compared the two spectra. The z-scored power in the hippocampus during maintenance was higher than during encoding in the theta-alpha frequency band (3–13 Hz, gray shaded window in Fig. 3(B) right, p < .05, cluster-based permutation test). No significant difference between task periods was found in the amygdala (p > .05, cluster-based permutation test, Fig. 3(B) left).

We next asked whether the engagement of the amygdala and the hippocampus was preferentially associated with a particular task period. We extracted the z-scored theta-alpha power (3–13 Hz) during encoding and maintenance for each channel in the amygdala and hippocampus. We then performed a repeated measure ANOVA (RMANOVA) with the extracted z-power as the dependent variable and two within-subject factors, area (amygdala/hippocampus) and task period (encoding/maintenance), and their interaction as independent variables. As shown in Fig. 3(C), we found a significant interaction (p = .001, F (1, 13) = 19.043). A simple effect analysis further indicated that the z-scored power in the amygdala was higher than in the hippocampus during encoding (p = .017). The reverse pattern appeared during maintenance but this did not reach statistical significance (p = .054).

As a further measure, we computed the count of responsive channels for each task period based on a magnitude threshold of the theta-alpha power (see Methods for details). In the amygdala, 35/50 channels (70.0%) responded above threshold during encoding (Fig. 3(D)), and this percentage did not increase greatly during maintenance (40/50; 80.0%, permutation test, p = .037). Interestingly, in the hippocampus only a few channels responded during encoding (35/94; 37.2%), but a much higher percentage responded during maintenance (66/94; 70.2%, permutation test, p < .0001).

Together, these three univariate analyses indicate that encoding preferentially engages the amygdala and maintenance preferentially engages the hippocampus.

Functional specialization: Representation in the amygdala is specific during encoding

We next applied multivariate analysis to investigate how information is represented across multiple channels and different spectral powers at different times. We performed a series of representational similarity analyses (RSA) to investigate two representational properties crucial for memory performance, i.e., the distinctiveness and the stability of neural representations (25).

We first quantified representational dissimilarity separately for the amygdala and hippocampus. Given that both the amygdala and hippocampus showed sustained activity in the low-frequency range (1–40 Hz, Fig. 3(A)), we used the spectral power in this frequency band as the feature for the RSA. We correlated the z-scored power from two trials across channels and frequencies (1 to 40 Hz in steps of 1 Hz) in consecutive overlapping time windows of 100 ms (step width 10 ms Fig. 2(C)). Then the dissimilarity (1 - similarity) of the representational patterns was averaged across the pairs of trials, resulting in a temporal map of encoding-encoding dissimilarity (EED) across all subjects for the amygdala (Fig. 4(A) left) and the hippocampus (Fig. 4(A) right) separately. Next, we compared the EED map between the amygdala and the hippocampus using cluster-based permutation tests. Across all encoding time windows, a cluster with higher EED values in the amygdala than in the hippocampus appeared (outlined in black in Fig. 4(B), p = .007, cluster-based permutation); there was no cluster with higher EED values in the hippocampus than in the amygdala. To elucidate whether differences in EED were congruent across subjects, we extracted the average EED values in the significant cluster and then compared the average EED values for the amygdala and the hippocampus. This revealed higher EED values in the amygdala than in the hippocampus (p = .027, t(13) = 2.49, paired t test, Fig. 4(C)). As the letter strings were different across trials, these findings indicate that the amygdala represented distinct WM information in a more specific neural pattern during encoding.

Functional specialization: Hippocampal representation is stable from encoding to maintenance

We next examined whether and how stable representational structures were maintained in the absence of stimuli in the amygdala and the hippocampus. Representational stability was indexed by memory reinstatement, an approach borrowed from the long-term memory literature (26). Memory reinstatement was quantified as the correlation between patterns of oscillatory power across channels and frequencies (1 to 40 Hz in steps of 1 Hz) within consecutive overlapping time windows of 100 ms (step width 10 ms) for each combination of the encoding-maintenance time bins in the same trial. The correlation matrix was then averaged across trials, resulting in a temporal map of encoding-maintenance similarity (EMS) for the amygdala and the hippocampus (Fig. 2(D)). The EMS values were higher in the hippocampus than in the amygdala for every encoding-maintenance time pair in the EMS map (Fig. 4D). For each subject, the extracted EMS value in the hippocampus was higher than that in the amygdala (p = .0049, t(13) = 3.38, paired t test, Fig. 4(E)). These findings indicated that the hippocampus retained WM information in a more stable representation during maintenance than the amygdala.

Was the functional specialization of the amygdala and the hippocampus correlated? To answer this question, we carried out a correlation analysis between the EED_Amy−Hipp (EED_Amy - EED_Hipp) and the EMS_Hipp−Amy (EMS_Hipp - EMS_Amy) for each subject. We found a significantly positive correlation (Pearson’s correlation, p = .0004, r = 0.811, Fig. 4(F), which remained significant after removing an outlier (p = .025)). This suggests that in subjects in which the amygdala represented more WM information, the hippocampus maintained the representation more stably.

Functional coordination: Coordination of the amygdala and hippocampus in representational reinstatement

In addition to functional specialization, functional coordination between brain regions is considered to be another important mechanism in WM (27). To test whether the amygdala and hippocampus work independently or interactively in WM, we investigated whether the representational structure during encoding in one structure was reinstated in the other structure during maintenance and vice versa.

We calculated cross-region EMS on each channel pair (one from the amygdala, one from the hippocampus in the same hemisphere in the same subject). The cross-region EMS maps were then averaged across channel pairs and trials, resulting in two cross-region EMS maps, one between the hippocampal encoding-amygdala maintenance combination (Fig. 5(A), left), the other between the amygdala encoding-hippocampal maintenance combination (Fig. 5(A), right). We found substantial cross-region EMS between the amygdala and the hippocampus (Fig. 5(A), $\rho$ (rho) > 0). Next, we compared the two cross-region EMS maps using cluster-based permutation tests and found a significant cluster (outlined in black in Fig. 5(B), p = .015, cluster-based permutation test), showing that stronger hippocampal representational structures during encoding were reinstated by the amygdala during maintenance. The extracted averaged EMS value within the significant cluster was higher in the hippocampal encoding-amygdala maintenance direction than in the opposite direction across subjects (p = .0015, t(13) = 4.006, paired t test, Fig. 5(C)). These results provide evidence that the amygdala and the hippocampus coordinate their activity to maintain the representation and that the hippocampus conveys information to the amygdala.

Since we found both functional specialization and coordination between the amygdala and the hippocampus in WM, we then tested for a possible correlation between the functional differentiation and collaboration. Pearson’s correlation was separately carried out between the coordinated EMS and the EED_Amyg−Hipp and between the coordinated EMS and the EMS_Hipp−Amyg. Intriguingly, we found a positive, albeit not significant correlation between the coordinated EMS and the EED_Amyg−Hipp (Pearson’s correlation, p = .092, r = 0.467, Fig. 5(D)) and a significant positive correlation between the coordinated EMS and the EMS_Hipp−Amy (Pearson’s correlation, p = .021, r = 0.607, Fig. 5(E)). These findings suggest that in subjects with more specific engagement of the amygdala or hippocampus in WM the activity between the two areas was more coordinated.

Functional coordination: Directional information transfer from the hippocampus to the amygdala tracks WM processing

A possible mechanism that could underlie the coordination of reinstatement structure across brain regions is information flow. We examined the functional synchronization between the amygdala and the hippocampus and the direction of information transfer. Here we applied two complementary measures, the phase locking value (PLV) and the spectral Granger causality (GC) index.

First, the phase synchrony of the PLV between the amygdala and the hippocampus was computed to measure the consistency of the phase relationship for each amygdala-hippocampus channel pair. As shown in Fig. 6(A), there was significant phase synchronization (p < .05, permutation test) between the amygdala and the hippocampus in the 1–40 Hz range throughout the entire WM processing. To further examine the directionality of the amygdala-hippocampus synchronization, we analyzed the effective connectivity indexed in GC within this frequency band. We compared the GC from the hippocampus to the amygdala with the GC in the reverse direction during encoding and during maintenance. The GC for both directions was significantly above the threshold (see Methods for details, grey lines denote the thresholds in Fig. 6(B)). Moreover, the GC from in the hippocampus to the amygdala was higher than the GC in the reverse direction in the 2–40 Hz band both during encoding (Fig. 6(B) top, black line, p = .066, cluster-based permutation test) and during maintenance (Fig. 6(B) bottom, black line, p = .017, cluster-based permutation test). Next, we calculated the difference in GC between the two directions and compared the directional difference between encoding and maintenance across the 1–40 Hz band. We found more directional information transfer from the hippocampus to the amygdala during maintenance than that during encoding across 5.5–36.5 Hz (Fig. 6(C), black line, p = .030, cluster-based permutation test). These findings match the coordinated reinstatement and provide converging evidence of a hippocampus-driven directional hippocampal-amygdala communication during WM maintenance.

Functional specialization and coordination in the amygdala-hippocampal circuit support successful WM outcomes

Functional specialization and coordination were found in the amygdala-hippocampal circuit during WM processing for the correct trials. However, it was still unclear whether the two properties were specific to successful WM processing. As a control, we repeated the above analyses for the incorrect trials.

We calculated the z-scored power within the amygdala and the hippocampus across 3–13 Hz during encoding and maintenance for the incorrect trials and compared them using RMANOVA with two within factors: area (amygdala/hippocampus) × phase (encoding/maintenance). No interaction effect was found for the incorrect trials (p = .11, Fig. 7(A)). We also calculated the EED map during encoding within the amygdala and the hippocampus for the incorrect trials. The EED map of the amygdala was compared with that of the hippocampus using cluster-based permutation, and no significant clusters were found (Fig. 7(B), p > .05, cluster-based permutation test). Next, we computed the EMS for the incorrect trials within the amygdala and the hippocampus separately. Contrasting the within-region EMS between the two regions revealed no difference (Fig. 7(C) left, p = .086, t(13) = 1.861, paired t-test). In addition, we did not find any difference in the cross-region EMS based on the error trials (Fig. 7(C) right, p = .17, t(13) = 1.47, paired t-test). This suggests that functional specification and coordination support successful WM processing.

To exclude possible bias from having an imbalanced number of correct and incorrect trials, we performed control analyses by bootstrapping 10 times. Specifically, we recomputed the EMS within the amygdala and the hippocampus on a subset of trials that contained the same number of correct trials and incorrect trials. These subsets of trials were randomly selected from all the correct trials, and the process was repeated 10 times. As shown in Fig. S1(A), Pearson’s correlation was computed and revealed high correlations between the EMS value for each subset and those from all the trials both in the amygdala (r = 0.975 ± .013) or the hippocampus (r = 0.976 ± .025). In each subset, the EMS within the hippocampus was higher than that within the amygdala (Fig. S1(B)). We thereby replicated our findings using a small subset of trials. Next, we recomputed the EMS across regions based on 10 subsets. Again, we found significant association between the EMS from each subset and those from all the trials, both in the combination of hippocampal encoding and amygdala maintenance (r = 0.965 ± .015, Pearson’s correlation) and in the amygdala encoding and hippocampal maintenance pair (r = 0.961 ± .014, Pearson’s correlation) (as shown in Fig. S1(C)). Again, for each subset, the EMS between the hippocampal encoding and the amygdala maintenance was higher than the EMS in the opposite direction (Fig. S1(D)). Taken together, our findings ruled out bias from having an imbalanced number of correct and incorrect trials.

To further validate synchronization between the amygdala and the hippocampus for the incorrect trials, we computed the PLV across the time-frequency domain up to 40 Hz. The amygdala-hippocampus PLV values for the incorrect trials that survived significance testing (p < .05) showed low phase synchronization (Fig. 7(D)). We also computed the GC values between the amygdala and the hippocampus during encoding and maintenance for the incorrect trials. We did not find any difference in directional information transfer (subtraction of GC values between the two directions) between the encoding and the maintenance (Fig. 7(E), p > .05, cluster-based permutation test).

To conclude, successful WM performance was promoted both by functional differentiation, including preferred stages of increased power, more distinct representation during encoding in the amygdala, and more stable representation reinstatement during maintenance in the hippocampus and by functional coordination, including biased cross-region reinstatement and directional information flow from the hippocampus to the amygdala.

Recordings in the epileptic seizure onset zone

Our analyses included recordings from channels in the seizure onset zone (SOZ). To assess possible effects of the SOZ on our findings, we compared the channels within and outside the SOZ for the amygdala and hippocampus, respectively. Since the representational structure is based on the distribution of power, we first compared the z-scored power across 1–40 Hz between the SOZ and the non-SOZ channels. The results indicated that no significant clusters were found for either the amygdala or the hippocampus (p > .05, cluster-based permutation test). That means the epileptogenic hippocampus/amygdala retained the normal activation pattern during WM. For the Granger causality analysis, we separated the channel pairs into SOZ (at least one SOZ channel) and non-SOZ (non-SOZ to non-SOZ channel) pairs and compared them in two directions across 1-40Hz separately for encoding and maintenance. There was no significant difference either in the direction from the hippocampus to the amygdala or from the amygdala to the hippocampus between the SOZ and non-SOZ channel pairs for either encoding or maintenance (all clusters: p > .05). Our analysis indicated that there was residual function in the epileptogenic hippocampus. This is consistent with our previous findings of single neurons in the SOZ that responded to this task (11), of LFP in the anterior/posterior hippocampal SOZ (28), and of a more general finding suggesting normal physiological responses to cognitive stimuli in epileptogenic brain areas (29).

Because previous research focused mainly on the role of the neocortex in WM, ranging from the sensory to the parietal and prefrontal cortices (for a review, see (2)), the contribution of the amygdala-hippocampal circuit in WM had only begun to be elucidated. Here we addressed the question of how the amygdala and hippocampus contribute to WM by analyzing oscillatory activity, neural representations, and inter-regional information flow. We found that 1) the amygdala was involved in memory specificity during encoding; 2) the hippocampus retained the representation of WM information in the absence of a stimulus on the screen; 3) the representational reinstatement in the amygdala and the hippocampus was coordinated; 4) WM enhanced low-frequency coupling between the amygdala and the hippocampus; and 5) functional specificity and functional coordination within the amygdala-hippocampus circuit occurred during correct trials but was absent during incorrect trials.

The amygdala has long been known to be related to emotion, thus, a role of the amygdala in memory representational specificity during encoding might appear surprising. However, a significant body of evidence has indirectly implicated a role of the amygdala in memory specificity. First, a recent human study reported that the amygdala has a higher proportion of concept cells with specific responses to preferred stimulus than the hippocampus (10). Other studies suggested that the amygdala plays a stimulus specific role in novelty detection (30, 31) and encodes state-dependent exploratory behavior (32). Both functions require specific memory representations. Our finding is also consistent with recent literature highlighting a broader function for the amygdala, including processing of sensory, memory, valence, etc. (4). Taken together, our study provided new evidence that the amygdala might contribute to WM, in particular by encoding specific memories with distinct representations.

The hippocampal representations, on the other hand, were less specific but could be better retained than those in the amygdala. The lower specificity of hippocampus representation is consistent with the recent idea that the hippocampus supported less-specific representations that hold across experiences (33). The higher EMS in the hippocampus is in line with our previous work revealing a higher proportion of maintenance cells in the hippocampus than in the amygdala (11). Studies in long term memory also suggested that memory consolidation in the hippocampus may start early, even at the end of encoding (14), suggesting that during this post-encoding period, memory representations are maintained in the hippocampus as memory engrams, so that they can be used to recover information later. Taken together, we infer that the hippocampus may contribute to WM maintenance by keeping representations stable in the absence of the stimulus.

In addition to the above-mentioned functional specificity, we found functional coordination between the amygdala and the hippocampus during WM processing. This coordination included coordinated representational reinstatement, phase synchronization, and inter-regional informational flow between the two structures. The interregional communication is consistent with rodent (34) and monkey (35) studies, which reported anatomical connection by using tracing techniques, and with rodent studies, which found synaptic plasticity in the amygdala induced by electrical stimulation to the hippocampus (22), and vice versa (21). Human iEEG studies found inter-regional functional connectivity during emotional information (36) and emotional memory (23) processing. In the current study we extended the functional coordination in the amygdala-hippocampal circuit to WM processing even in the absence of explicit emotional content.

Both the coordinated representational reinstatement analysis and the GC analysis revealed a directional coordination driven by the hippocampus. This indicated a modulatory effect of the hippocampus on the amygdala in WM processing. This direction is opposite to that during emotion processing, which showed an amygdala influence on the hippocampus (36). This raises the novel possibility that the directional information transfer from the amygdala could contribute to emotion, while that from the hippocampus could have a role in supporting WM. This inference fits with studies of fear learning in which hippocampal neurons convey contextual representations through projections to the amygdala (37, 38). Further evidence came from a rodent study demonstrating that a subset of hippocampal neurons, whose activity was necessary for memory encoding, was enriched in the hippocampal projection to the amygdala (39). Together, our data support a model in which the representation in the amygdala during encoding is more specific than that in the hippocampus but cannot be well retained. To at least partially solve this dilemma, the hippocampus would convey WM information to the amygdala through coordinated representational reinstatement and information transfer to help the amygdala better retain the WM information, which in turn would facilitate subsequent WM processing.

There are two major theories about the brain's cognitive function, the first is the theory that modularity supports functional specialization (40). Our results support this theory by showing that more specific representation structures occur in the amygdala during encoding and that a higher encoding-maintenance similarity in representation structure exists in the hippocampus. The second theory, distributive processing, proposes that the brain is more interactive and its regions are functionally interconnected (41). Our findings support this theory by showing cross-regional EMS and inter-regional information flow. Interestingly, we also observed a correlation between functional coordination and functional specialization, but functional specialization and coordination only existed in correct WM trials. Thus, these two properties of the amygdala-hippocampus circuit might be essential mechanisms supporting successful WM processing.

In summary, our results demonstrated functional specialization between the amygdala and hippocampus in working memory. In addition, we showed both coordinated reinstatement and information flow from the hippocampus to the amygdala. Taken together, we provide a novel and mechanistic explanation of how the oscillatory signatures of neuronal activity in the two structures especially contribute and orchestrate to support working memory.

Subject characteristics

Data were obtained from epilepsy patients undergoing intracranial EEG monitoring at the Swiss Epilepsy Clinic, Klinik Lengg, Switzerland, to localize epileptic foci for potential surgical resection. Intracranial depth electrodes (1.3 mm diameter, 8 contacts of 1.6 mm length, 5 mm spacing; Ad-Tech, Racine, WI, www.adtechmedical.com) were stereotactically implanted. The electrode placements were guided exclusively by clinical needs. Subjects having channels in both the amygdala and the hippocampus in the same hemisphere were included in the current study. Before testing, all the subjects provided written informed consent for the study, which had been approved by the relevant institutional ethics review board (Kantonale Ethikkommission Zürich, PB 2016–02055). In total, 14 patients (mean ± SD [range]: 34.5 ± 12.6 [18–56]; 7 females) participated in this study. There were no seizures recorded during any of the epochs, and any epochs with interictal epileptiform activity were excluded from analysis.

Task

We used a modified Sternberg task in which the encoding of memory content, maintenance, and recall were temporally separated (Fig. 2(A)). Each trial started with a fixation period (1 s) followed by the presentation of the stimulus in the encoding period (2 s). The stimulus consisted of a set of letters at the center of the screen. After the encoding period, the stimulus was replaced by a fixation square during the maintenance period (3 s). Finally, a probe letter appeared and the subjects responded with a button press (“IN” or “OUT”) to indicate whether the probe was part of the stimulus letter set. The subjects performed 50 trials per session, which lasted approximately 10 min. Some subjects performed up to seven sessions of the task.

Channel localization

The channels were localized using postimplantation computed tomography (CT) scans and postimplantation structural T1-weighted MRI scans. For each patient, the CT scan was co-registered to the postimplantation scan, as implemented in FieldTrip (42). The channels were visually marked on the coregistered CT-MR images. Channel positions were verified by the neurosurgeon (L.S.) after merging pre-operative MRI with postimplantation CT images of each individual patient in the plane along the electrode (iPlan Stereotaxy 3.0, Brainlab, München, Germany). Channel locations in native space for each patient were projected to MNI space and are shown in Fig. 3(B). The final dataset contained 94 channels in the hippocampus and 50 channels in the amygdala across all patients. There were 6.7 ± 1.4 (range 4–8) channels per patient in the hippocampus and 3.6 ± 0.9 (range 2–4) channels per subject in the amygdala.

Channel selection

Each subject had 0–1 electrode targeting the anterior and posterior hippocampus and the amygdala per hemisphere. Targeted regions and hemispheres varied across subjects for clinical reasons and included the hippocampus in the left (n = 13) and right (n = 14) hemispheres and the amygdala in the left (n = 13) and right (n = 11) hemispheres. We selected the two most medial channels on each electrode targeting the hippocampus or the amygdala, as was done in previous studies (26, 43). This procedure was used to minimize inter-individual variability, which would be higher if different numbers of channels would have been selected across subjects. The final number of selected channels in each region for each subject is listed in Table 1. We included only ipsilateral channel pairs in the analysis.

Data acquisition and preprocessing

Intracranial data were acquired against a common intracranial reference using a Neuralynx ATLAS recording system, sampled at 4 kHz, and analog-filtered above 0.5 Hz. After data acquisition, neural recordings were down sampled to 1 kHz and band-pass filtered between 1 to 200 Hz using the zero-phase delay finite impulse response (FIR) filter with Hamming window. Line noise harmonics were removed using a discrete Fourier transform. The filtered data were manually inspected to mark any channels containing epileptiform activity or artifacts for exclusion. The data were then re-referenced to the average of the signal over all the clean channels (24). We then segmented the preprocessed data into event-related epochs; 1 s fixation period, 2 s encoding period, 3 s maintenance period, and 2 s retrieval period. We rejected trials with artifacts by visual inspection (53/3250 or 1.6% of all trials). We performed preprocessing routines with the FieldTrip (42), EEGLAB (44) toolboxes and custom scripts in MATLAB.

Time-frequency analysis

Time-frequency power was separately computed for correct and incorrect trials. For each trial for each channel, we convolved the signal with complex-valued Morlet wavelets (6 cycles) to obtain power information at each frequency from 1 to 100 Hz in 1 Hz steps with a time resolution of 1 ms (19). The task-induced power was analyzed per trial using a statistical bootstrapping procedure as was done in previous studies (24, 45). Briefly, for each channel and frequency, a null distribution was created by randomly selecting and averaging several data points in the baseline power (500 ms pretrial) 1,000 times, the raw power for each time point during task was then z-scored by comparing it to the null distribution to generate the z-scored power.

For each subject, the z-scored spectral power was separately averaged across the encoding and maintenance periods within the amygdala and the hippocampus. Next, we compared the z-scored spectral power between encoding and maintenance across subjects separately for the amygdala and the hippocampus using a cluster-based permutation test which computes statistics at the cluster level and corrects for multiple comparisons (46). Differences were quantified by t values. The t values corresponding to uncorrected p values of .05 or less were clustered. The differences were considered significant when the maximum of the cluster-level summed t values in the true data exceeded the threshold of a Monte Carlo distribution, which was created by randomly shuffling the labels of the regions 1000 times, and p < .05 was considered as significant.

We found significantly higher z-scored spectral power during maintenance than encoding period across 3–13 Hz (theta-alpha band) in the hippocampus via a cluster-based permutation test and found no significant difference in any frequency band period in the amygdala. Then we performed two additional analyses on this frequency band. First, we extracted the theta-alpha band z-scored power during encoding and maintenance within the amygdala and the hippocampus separately. We performed RMANOVA with the extracted z-power as the dependent variable and two within-subject factors and their interaction as independent variables: area (amygdala/hippocampus) and period (encoding/maintenance). A simple effect analysis was made if the interaction effect was significant (p < .05). Next, we calculated the proportion of responsive channels within this frequency range separately for the amygdala and the hippocampus, as in our previous study (47). Specifically, for each channel and each trial, we averaged the theta-alpha z-scored power, resulting in two time series during the encoding and the maintenance periods. Then, two windows extending from 200 ms prior the stimulus onset to the end of the encoding (-200 ms to 2000 ms) and from 200 ms prior the start of the maintenance to the end of the maintenance (1800 ms to 5000 ms) were used to identify consecutive responsive time windows (> 100 ms) of the theta-alpha power above the threshold (z > 1.96). This trial was considered as responsive. A channel was defined as responsive to a WM period if at least 70% of trials of the channels were found to respond. Finally, we computed the proportion of responsive channels for encoding or maintenance in the amygdala and the hippocampus by dividing the number of responsive channels by the number of recording channels.

Representational dissimilarity analysis

A sliding time window approach was applied to calculate the representational dissimilarity in a 100 ms sliding time window (step width 10 ms). The z-scored power was first averaged across the time points within each sliding window for each trial. The generated z-scored power of all channels and frequencies (1–40 Hz) within each time window were then vectorized. The Spearman’s correlation between the features of the two trials was calculated and Fisher z-transformed. The generated values were subtracted from 1 and then averaged across trial pairs to index the dissimilarity among trials in the given time window pair. After these steps, we got the encoding-encoding dissimilarity (EED) map across all time windows. Notably, for each subject, the number of trial pairs computed in the correct trials was the same as those in the incorrect trials, to exclude the impact of different number of trials for the correct and incorrect outcomes. The analysis procedure is presented in Fig. 2(C). We then compared the EED map in the amygdala with the hippocampus at the group level by using a cluster-based permutation test (46). We also extracted the average EED values in the significant cluster in the two regions for each subject and then contrasted them via paired t tests at the group level.

Representational similarity between encoding and maintenance

Next, we calculated the representation similarity between encoding and maintenance (EMS) periods within the same trials for the correct and the incorrect trials separately. First, we built representational patterns based on distributed oscillatory power (1–40 Hz) across all channels available for each subject between the time windows pairs (one from encoding and the other from maintenance), resulting in EMS maps within the same trials between all encoding-maintenance time window pairs, which were then averaged across trials. Figure 2(D) describes the detailed analysis procedure (red and blue boxes). Next, we contrasted the EMS maps for the amygdala and the hippocampus by using a cluster-based permutation test. Since we found significant contrast in every encoding-maintenance time pair in the EMS map, we then extracted the average EMS values in the whole map for each subject from the two regions and contrasted them via paired t tests at the group level.

To rule out the confounding effect of different numbers between the correct and incorrect trials in the EMS calculation, for each subject we selected subsets with the same number of trials from the correct trials as from the incorrect trials, and recalculated the EMS based on the extracted trials. This procedure was repeated 10 times. For each subset, EMS_amygdala and EMS_hippocampus were generated to see if the finding of higher EMS_hippocampus value found from all correct trials could be replicated with a reduced number of trials.

Coordinated reinstatement analyses between the amygdala and the hippocampus

To assess whether there was a coordinated representational reinstatement between the amygdala and the hippocampus, we also computed the cross-regional representation reinstatement. The z-scored power spectral values (1–40 Hz) of one channel within each time window were vectorized. The cross-region representational reinstatement was then obtained by calculating the Spearman’s rho between the features of one channel from the amygdala during encoding and of one channel from the hippocampus in the same hemisphere during maintenance and vice versa. We then Fisher z-transformed the rho scores and averaged the coordinated reinstatement maps across all trials. For each subject, we obtained two coordinated reinstatement maps: one was the representational reinstatement between hippocampus encoding and amygdala maintenance (EMS_{hippenc_amymaint}) and the other was between amygdala encoding and hippocampus maintenance (EMS_{amyenc_hippmaint}). We also presented the coordinated reinstatement maps in the bottom of Fig. 2(D) (purple and orange squares). Then, we compared the EMS_{hippenc_amymaint} with the EMS_{amyenc_hippmaint} at the group level using a cluster-based permutation test (46). Then we extracted the values within significant clusters for each subject and contrasted them via paired t tests at the group level.

Again, to avoid the confounding effect of unbalanced trial numbers between correct and incorrect trials, we performed a control analyses. Specifically, we re-performed the same analysis 10 times by randomly selecting the same number of correct trials as the incorrect trials. For each subset, we computed the average EMS_{hippenc_amymaint} and EMS_{amyenc_hippmaint} for each subject and then averaged these across subjects to see if the result from all correct trials could be replicated in a reduced number of trials.

Inter-regional phase synchronization

The strength of the inter-regional neural synchrony was quantified by the phase locking value (PLV), which calculates the average phase φ difference between two channels (a,b) across trials for a given time point t and frequency f:

$${PLV}_{a,b}\left(t,f\right)=\frac{1}{Ntrials}\left|\sum _{n=1}^{n=Ntrials}\text{exp}\left(i\left[{\phi }_{n,a}\left(t,f\right)-{\phi }_{n,b}\left(t,f\right)\right]\right)\right|$$

The PLV indexes the degree of consistency for each pair of channels’ phase relationship independent of the absolute phases and amplitudes among repeated trials and ranges from 0 to 1 with values approaching 1 if the two signals show a constant phase relationship. We calculated the PLV for each channel pair (one from the amygdala and one from the ipsilateral hippocampus) from 1–40 Hz (1 Hz steps) through the WM periods across the correct and the incorrect trials separately. The trial-wise mean was subtracted from each trial to minimize contamination from simultaneous voltage changes of the phase estimates before calculating the PLV (48). To test the statistical significance of the PLV values for the correct and the incorrect trials, we permuted the trials 200 times to create a null distribution and computed the 95th percentile threshold for each pair. The thresholds were then averaged across all channel pairs and only the time-frequency PLV values above the corresponding thresholds were kept as significant PLVs. The values below the thresholds were set to 0.

Granger causality analysis

We examined the directionality of the amygdala-hippocampus synchronization across 1–40 Hz using spectral Granger causality (GC), which quantifies the prediction error of the signal in the frequency domain by introducing another time series. For each channel pair, the trial-wise mean was subtracted from each trial before being fit to an autoregressive model and computing the spectral GC. We then applied the Multivariate Granger Causality Matlab Toolbox (49) based on the Akaike information criterion to define the model order for each pair. The GC index was computed for both directions (from the amygdala to the hippocampus and the reverse direction) during encoding and maintenance for the correct and incorrect trials separately. Then we created a null distribution by randomly swapping the signal between channels 200 times. A GC value with a channel pair above the 95th percentile of the null distribution was considered as significant. The spectral GC values from two directions across 1–40 Hz were then averaged across all pairs for each subject.

We then used a cluster-based permutation test (46) to further examine whether the spectral GC values differed between directions at frequency points at the group level. As described above, we created a Monte Carlo distribution by randomly shuffling the labels of the directions 1000 times. The differences between directions were considered significant when the maximum of the cluster-level summed t values in the true data exceeded the 95th percentile of the null distribution (p < .05). We found more directional information transfer from the hippocampus to the amygdala in both encoding and maintenance. Next, we calculated the net information transfer (GC_{hipp to amyg} – GC_{amyg to hipp}) in each subject in the encoding and maintenance period separately and then compared the net information transfer value between the two periods at the group level via paired t tests.

The dataset is freely available for download at https://doi.gin.g-node.org/10.12751/g-node.d76994/. The task is freely available for download at http://www.neurobs.com/ex_files/expt_view?id=266. Links to updates and further data sets can be found at https://hfozuri.ch.

Funding and Disclosures

This work received support from the following sources: Science and Technology Innovation 2030 - Brain Science and Brain-Inspired Intelligence Project (Grant No. 2021ZD0200200), Science Frontier Program of the Chinese Academy of Sciences (grant No. XDBS01030200), National Key R&D Program of China (grant No. 2017YFA0105203), National Natural Science Foundation of China (grant Nos. 31300934, 82151307), Open Research Fund of the State Key Laboratory of Cognitive Neuroscience and Learning (CNLYB2004), Open Research Fund of the CAS Key Laboratory of Behavioral Science, Institute of Psychology, and the Swiss National Science Foundation (SNSF 176222 to J.S.).

Acknowledgments

The authors also thank Rhoda E. Perozzi and Edmund F. Perozzi, PhDs, for English and content editing assistance.

Conflict of Interest statement

The authors declare no competing financial interests.

Baddeley A (2012) Working memory: theories, models, and controversies. Annu. Rev. Psychol. 63:1–29.
Christophel TB, Klink PC, Spitzer B, Roelfsema PR, & Haynes JD (2017) The Distributed Nature of Working Memory. Trends Cogn Sci 21(2):111–124.
LaBar KS & Cabeza R (2006) Cognitive neuroscience of emotional memory. Nat. Rev. Neurosci. 7(1):54–64.
Gothard KM (2020) Multidimensional processing in the amygdala. Nat. Rev. Neurosci. 21(10):565–575.
Inman CS, et al. (2018) Direct electrical stimulation of the amygdala enhances declarative memory in humans. Proc. Natl. Acad. Sci. U. S. A. 115(1):98–103.
Eichenbaum H (1993) Memory, amnesia, and the hippocampal system (MIT press).
Jeneson A & Squire LR (2012) Working memory, long-term memory, and medial temporal lobe function. Learn. Mem. 19(1):15–25.
Parvizi J & Kastner S (2018) Promises and limitations of human intracranial electroencephalography. Nat. Neurosci. 21(4):474–483.
Kornblith S, Quian Quiroga R, Koch C, Fried I, & Mormann F (2017) Persistent Single-Neuron Activity during Working Memory in the Human Medial Temporal Lobe. Curr. Biol. 27(7):1026–1032.
Kaminski J, et al. (2017) Persistently active neurons in human medial frontal and medial temporal lobe support working memory. Nat. Neurosci. 20(4):590–601.
Boran E, et al. (2019) Persistent hippocampal neural firing and hippocampal-cortical coupling predict verbal working memory load. Sci Adv 5(3):eaav3687.
Freund MC, Etzel JA, & Braver TS (2021) Neural Coding of Cognitive Control: The Representational Similarity Analysis Approach. Trends Cogn Sci 25(7):622–638.
Favila SE, Chanales AJ, & Kuhl BA (2016) Experience-dependent hippocampal pattern differentiation prevents interference during subsequent learning. Nat Commun 7:11066.
Zhang H, et al. (2021) Awake ripples enhance emotional memory encoding in the human brain.2021.2011.2017.469047.
Sander D, Grafman J, & Zalla T (2003) The human amygdala: an evolved system for relevance detection. Rev. Neurosci. 14(4):303–316.
Aggleton JP (1992) The amygdala: neurobiological aspects of emotion, memory, and mental dysfunction (Wiley-Liss).
Ranganath C & Ritchey M (2012) Two cortical systems for memory-guided behaviour. Nat. Rev. Neurosci. 13(10):713–726.
Vanz F, et al. (2018) Role of dorsal hippocampus kappa opioid receptors in contextual aversive memory consolidation in rats. Neuropharmacology 135:253–267.
Liu J, et al. (2020) Stable maintenance of multiple representational formats in human visual short-term memory. Proc. Natl. Acad. Sci. U. S. A. 117(51):32329–32339.
McDonald AJ & Mott DD (2017) Functional neuroanatomy of amygdalohippocampal interconnections and their role in learning and memory. J. Neurosci. Res. 95(3):797–820.
Abe K, Niikura Y, & Misawa M (2003) The induction of long-term potentiation at amygdalo-hippocampal synapses in vivo. Biol. Pharm. Bull. 26(11):1560–1562.
Maren S & Fanselow MS (1995) Synaptic plasticity in the basolateral amygdala induced by hippocampal formation stimulation in vivo. J. Neurosci. 15(11):7548–7564.
Zheng J, et al. (2019) Multiplexing of Theta and Alpha Rhythms in the Amygdala-Hippocampal Circuit Supports Pattern Separation of Emotional Information. Neuron 102(4):887–898 e885.
Li J, et al. (2021) Anterior-posterior hippocampal dynamics support working memory processing. J. Neurosci.
Sommer VR & Sander MC (2021) Contributions of representational distinctiveness and stability to memory performance and age differences. Neuropsychol. Dev. Cogn. B Aging Neuropsychol. Cogn.:1–20.
Pacheco Estefan D, et al. (2019) Coordinated representational reinstatement in the human hippocampus and lateral temporal cortex during episodic memory retrieval. Nat Commun 10(1):2255.
Gazzaley A, Rissman J, & D'Esposito M (2004) Functional connectivity during working memory maintenance. Cogn. Affect. Behav. Neurosci. 4(4):580–599.
Li J, et al. (2022) Anterior-Posterior Hippocampal Dynamics Support Working Memory Processing. J. Neurosci. 42(3):443–453.
Liu S & Parvizi J (2019) Cognitive refractory state caused by spontaneous epileptic high-frequency oscillations in the human brain. Sci. Transl. Med. 11(514).
Balderston NL, Schultz DH, & Helmstetter FJ (2013) The effect of threat on novelty evoked amygdala responses. PLoS One 8(5):e63220.
Camalier CR, Scarim K, Mishkin M, & Averbeck BB (2019) A Comparison of Auditory Oddball Responses in Dorsolateral Prefrontal Cortex, Basolateral Amygdala, and Auditory Cortex of Macaque. J. Cogn. Neurosci. 31(7):1054–1064.
Fustinana MS, Eichlisberger T, Bouwmeester T, Bitterman Y, & Luthi A (2021) State-dependent encoding of exploratory behaviour in the amygdala. Nature 592(7853):267–271.
Schapiro AC, Turk-Browne NB, Botvinick MM, & Norman KA (2017) Complementary learning systems within the hippocampus: a neural network modelling approach to reconciling episodic memory with statistical learning. Philos. Trans. R. Soc. Lond. B Biol. Sci. 372(1711).
Kishi T, Tsumori T, Yokota S, & Yasui Y (2006) Topographical projection from the hippocampal formation to the amygdala: a combined anterograde and retrograde tracing study in the rat. J. Comp. Neurol. 496(3):349–368.
Amaral DG & Cowan WM (1980) Subcortical afferents to the hippocampal formation in the monkey. J. Comp. Neurol. 189(4):573–591.
Zheng J, et al. (2017) Amygdala-hippocampal dynamics during salient information processing. Nat Commun 8:14413.
Kim WB & Cho JH (2017) Synaptic Targeting of Double-Projecting Ventral CA1 Hippocampal Neurons to the Medial Prefrontal Cortex and Basal Amygdala. J. Neurosci. 37(19):4868–4882.
Xu C, et al. (2016) Distinct Hippocampal Pathways Mediate Dissociable Roles of Context in Memory Retrieval. Cell 167(4):961–972 e916.
Jimenez JC, et al. (2020) Contextual fear memory retrieval by correlated ensembles of ventral CA1 neurons. Nat Commun 11(1):3492.
Coltheart M & Caramazza A (2006) Cognitive neuropsychology twenty years on (Psychology Press).
McIntosh ARJm (1999) Mapping cognition to the brain through neural interactions. 7(5–6):523–548.
Oostenveld R, Fries P, Maris E, & Schoffelen JM (2011) FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Intell. Neurosci. 2011:156869.
Oehrn CR, et al. (2014) Neural communication patterns underlying conflict detection, resolution, and adaptation. J. Neurosci. 34(31):10438–10452.
Delorme A & Makeig S (2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J. Neurosci. Methods 134(1):9–21.
Johnson EL, et al. (2018) Dynamic frontotemporal systems process space and time in working memory. PLoS Biol. 16(3):e2004274.
Maris E & Oostenveld R (2007) Nonparametric statistical testing of EEG- and MEG-data. J. Neurosci. Methods 164(1):177–190.
Foster BL, Dastjerdi M, & Parvizi J (2012) Neural populations in human posteromedial cortex display opposing responses during memory and numerical processing. Proc. Natl. Acad. Sci. U. S. A. 109(38):15514–15519.
Brincat SL & Miller EK (2015) Frequency-specific hippocampal-prefrontal interactions during associative learning. Nat. Neurosci. 18(4):576–581.
Barnett L & Seth AK (2014) The MVGC multivariate Granger causality toolbox: a new approach to Granger-causal inference. J. Neurosci. Methods 223:50–68.

There is NO Competing Interest.

FigureS1legend.docx
Figure S1 legend
Fig.S1.pdf
Figure S1

Download PDF

Version 1

posted

You are reading this older preprint version

Read the latest preprint version →

Functional Specialization and Coordination in the Amygdala-Hippocampus Circuit Support Working Memory

Archived Versions:

Version 1

Abstract

Figures

Significance

Introduction

Results

Task, behavior and recording channels

Increased theta-alpha power in the amygdala and hippocampus

Functional specialization: Representation in the amygdala is specific during encoding

Functional specialization: Hippocampal representation is stable from encoding to maintenance

Functional coordination: Coordination of the amygdala and hippocampus in representational reinstatement

Functional coordination: Directional information transfer from the hippocampus to the amygdala tracks WM processing

Functional specialization and coordination in the amygdala-hippocampal circuit support successful WM outcomes

Recordings in the epileptic seizure onset zone

Discussion

Methods

Subject characteristics

Task

Channel localization

Channel selection

Data acquisition and preprocessing

Time-frequency analysis

Representational dissimilarity analysis

Representational similarity between encoding and maintenance

Coordinated reinstatement analyses between the amygdala and the hippocampus

Inter-regional phase synchronization

Granger causality analysis

Data And Code Availability

Declarations

References

Additional Declarations

Supplementary Files

Archived Versions:

Version 1