We first examined neuronal activity in the AI and the PCC/precuneus and tested whether activity in the PCC/precuneus is suppressed compared to activity in the AI. Previous studies have suggested that power in the high-gamma band (80–160 Hz) is correlated with fMRI BOLD signals (Hermes et al., 2017; Hutchison et al., 2015; Lakatos et al., 2019; Leopold et al., 2003; Mantini et al., 2007; Schölvinck et al., 2010), and is thought to reflect local neuronal activity (Canolty and Knight, 2010). Therefore, we compared high-gamma band power (see ‘Methods’ for details) in the AI and PCC/precuneus electrodes during both encoding and recall and across the four episodic memory tasks. Briefly, in the VFR task, participants were presented with a sequence of words and asked to remember them for subsequent recall (‘Methods’, Appendix 1—tables 1, 2 and 6, Figures 1a and 2).

We next examined high-gamma power in the CATVFR task. In this task, participants were presented with a list of words with consecutive pairs of words from a specific category (e.g., JEANS-COAT, GRAPE-PEACH, etc.) and subsequently asked to recall as many as possible from the original list (‘Methods’, Appendix 1—tables 1, 3 and 7, Figure 1b; Qasim et al., 2023).

The PALVCR task also consisted of three periods: encoding, delay, and recall (‘Methods’, Appendix 1—tables 1, 4 and 8, Figure 1c). During encoding, a list of word-pairs was visually presented, and then participants were asked to verbally recall the cued word from memory during the recall periods.

We next examined high-gamma power in the WMSM task. Participants performed multiple trials of a spatial memory task in a virtual navigation paradigm (Goyal et al., 2018; Jacobs et al., 2016; Lee et al., 2018) similar to the Morris water maze (Morris, 1984; ‘Methods’, Appendix 1—tables 1, 5 and 9, Figure 1d). Participants were shown objects in various locations during the encoding periods and asked to retrieve the location of the objects during the recall period.

We next used replication BF analysis to estimate the degree of replicability of high-gamma power suppression of the PCC/precuneus compared to the AI during the memory encoding periods of the four tasks. We used the posterior distribution obtained from the VFR (primary) dataset as a prior distribution for the test of data from the CATVFR, PALVCR, and WMSM (replication) datasets (Ly et al., 2019; see ‘Methods’ for details). We used the encoding time windows for which we most consistently observed decrease of PCC/precuneus high-gamma power compared to the AI. These correspond to 110–1600 ms during the VFR task, 570–790 ms in the CATVFR task, 2010–2790 ms in the PALVCR task, and 3150–4690 ms in the WMSM task. We first averaged the high-gamma power across these strongest time windows for each task and then used replication BF analysis to estimate the degree of replicability of high-gamma power suppression of the PCC/precuneus compared to the AI.

Findings corresponding to the high-gamma power suppression of the PCC/precuneus compared to AI were replicated in the PALVCR (BF 5.16e+1) and WMSM (BF 2.69e+8) tasks. These results demonstrate very high replicability of high-gamma power suppression of the PCC/precuneus compared to AI during memory encoding. The consistent suppression effect was localized only to the PCC/precuneus, but not to the mPFC node of the DMN or the dPPC and MFG nodes of the FPN (Appendix 1—figures 1–3).

In contrast to memory encoding, a similar analysis of high-gamma power did not reveal a consistent pattern of increased high-gamma power in AI and suppression of the PCC/precuneus across the four tasks during memory recall (Figure 3).

We examined whether AI and PCC/precuneus high-gamma power response during the encoding and recall periods are enhanced or suppressed compared to the baseline periods. High-gamma power in the AI was increased compared to the resting baseline during both the encoding and recall periods, and across all four tasks (ps<0.05, Figure 3). This suggests an enhanced role for the AI during both memory encoding and recall compared to resting baseline.

In contrast, high-gamma power in the PCC/precuneus was reduced compared to the resting baseline in three tasks—VFR, PALVCR, and WMSM—providing direct evidence for PCC/precuneus suppression during memory encoding (Figure 3). We did not find any increased high-gamma power activity in the PCC/precuneus, compared to the baseline, during memory retrieval (Figure 3). These results provide evidence for PCC/precuneus suppression compared to both the AI and resting baseline during externally triggered stimuli during encoding.

High-gamma power for other brain areas compared to resting baseline were not consistent across tasks (Appendix 1—figures 1–3).

We next examined directed information flow from the AI to the PCC/precuneus and mPFC nodes of the DMN during the memory encoding periods of the VFR task. We used PTE (Lobier et al., 2014) to evaluate directed influences from the AI to the PCC/precuneus and mPFC and vice versa. Informed by recent electrophysiology studies in nonhuman primates, which suggest that broadband field potentials activity, rather than narrowband, governs information flow in the brain (Davis et al., 2020; Davis et al., 2022), we examined PTE in a 0.5–80 Hz frequency spectrum to assess dynamic-directed influences of the AI on the DMN.

Directed information flow from the AI to the PCC/precuneus (F(1, 264) = 59.36, p<0.001, Cohen’s d = 0.95) and mPFC (F(1, 208) = 13.96, p<0.001, Cohen’s d = 0.52) was higher than the reverse (Figure 4a).

Next, we examined directed influences of the AI on PCC/precuneus and mPFC during the recall phase of the verbal episodic memory task. During memory recall, directed information flow from the AI to the PCC/precuneus (F(1, 264) = 43.09, p<0.001, Cohen’s d = 0.81) and mPFC (F(1, 211) = 21.94, p<0.001, Cohen’s d = 0.65) was higher than the reverse (Figure 4a).

We next probed directed information flow between the AI and FPN nodes during the encoding periods of the VFR task. Directed information flow from the AI to the dPPC (F(1, 1143) = 11.69, p<0.001, Cohen’s d = 0.20) and MFG (F(1, 1245) = 21.69, p<0.001, Cohen’s d = 0.26) was higher than the reverse during memory encoding of the VFR task (Figure 5a).

Directed influences from the AI to the dPPC (F(1, 1143) = 17.47, p<0.001, Cohen’s d = 0.25) and MFG (F(1, 1246) = 42.75, p<0.001, Cohen’s d = 0.37) were higher than the reverse during memory recall of the VFR task (Figure 5a).

To examine the specificity of the AI-directed information flow to the DMN and FPN, we conducted a control analysis using electrodes implanted in the IFG (BA 44). The IFG serves as an ideal control region due to its anatomical adjacency to the AI, its involvement in a wide range of cognitive control functions including response inhibition (Cai et al., 2014), and its frequent co-activation with the AI in fMRI studies. Furthermore, the IFG has been associated with controlled retrieval of memory (Badre et al., 2005; Badre and Wagner, 2007; Wagner et al., 2001), making it a compelling region for comparison.

Our analysis revealed a striking contrast between the AI and IFG in their patterns of directed information flow. While the AI exhibited strong directed influences on both the DMN and FPN, the IFG showed the opposite pattern. Specifically, both the DMN and FPN demonstrated higher influence on the IFG than the reverse during both encoding and recall periods, and across all four memory experiments (Appendix 1—figures 4 and 5).

To quantify this difference more precisely, we calculated the net outflow for both regions, defined as the difference (PTE(out) – PTE(in), see ‘Methods’ for details). This analysis revealed that the AI’s net outflow was significantly higher than that of the IFG during both encoding and recall phases, a finding replicated across all four experiments (all ps<0.001) (Appendix 1—figure 6).

These results not only highlight the unique role of the AI in orchestrating large-scale network dynamics during memory processes but also demonstrate the specificity of this function compared to an anatomically adjacent and functionally relevant region. The consistent pattern across diverse memory tasks and experimental phases underscores the robustness of the AI’s role as an outflow hub during memory formation and retrieval.

We next examined whether directed information flow from the AI to the DMN and FPN nodes during the memory tasks differed from the resting-state baseline. Resting-state baselines were extracted immediately before the start of the task sessions and the duration of task and rest epochs were matched to ensure that differences in network dynamics could not be explained by differences in duration of the epochs. Directed information flow from the AI to both the DMN and FPN was higher during both the memory encoding and recall phases and across the four experiments compared to baseline in all but two cases (Appendix 1—figures 7 and 8).

To further elucidate the task-specific role of the AI, we compared its net outward directed influence during memory tasks to that observed during resting state. We quantified this influence as the difference between outgoing and incoming information flow (PTE(out) – PTE(in)). This analysis revealed that the AI’s net outflow was significantly enhanced during both encoding and recall phases of memory tasks compared to resting state in all but one case (ps<0.05) (Appendix 1—figure 9). This pattern was consistently observed across all four experiments. These findings provide strong evidence for enhanced role of AI-directed information flow to the DMN and FPN during memory processing compared to the resting state.

We examined memory effects by comparing PTE between successfully recalled and forgotten memory trials. However, this analysis did not reveal differences in directed influence from the AI on the DMN and FPN or the reverse between successfully recalled and forgotten memory trials during the encoding as well as recall periods in any of the memory experiments (all ps>0.05) (Appendix 1—figures 10 and 11).

fMRI studies have suggested that the AI acts as an outflow hub with respect to interactions with the DMN and FPN (Sridharan et al., 2008). To test the potential neural basis of this finding, we calculated net outflow (PTE(out) – PTE(in)) as the difference between the total outgoing information and total incoming information.

We next directly compared the phase locking values (PLVs) (see ‘Methods’ for details) between the AI and the PCC/precuneus and mPFC nodes of the DMN and also the dPPC and MFG nodes of the FPN for the encoding and the recall periods compared to resting baseline. However, narrowband PLV values did not significantly differ between the encoding/recall vs. rest periods in any of the delta-theta (0.5–8 Hz), alpha (8–12 Hz), beta (12–30 Hz), gamma (30–80 Hz), and high-gamma (80–160 Hz) frequency bands. These results indicate that PTE, rather than phase synchronization, more robustly captures the AI dynamic interactions with the DMN and the FPN.

Dynamic interactions between the AI and the DMN and FPN are hypothesized to shape human cognition (Cai et al., 2016; Cai et al., 2014; Dosenbach et al., 2008; Dosenbach et al., 2006; Menon, 2015b; Menon and Uddin, 2010). Although fMRI research has suggested that the AI plays a pivotal role in the task-dependent engagement and disengagement of the DMN and FPN across diverse cognitive tasks (Menon and Uddin, 2010; Sridharan et al., 2008), the neuronal basis of these results or the possibility of their being artifacts arising from slow dynamics and regional variation in the hemodynamic response inherent to fMRI signals remained unclear. To address these ambiguities, our analysis focused on casual interactions involving the AI and leveraged the high temporal resolution of iEEG signals. By investigating the directionality of information flow, we aimed to overcome the temporal resolution limitations of fMRI signals, providing a more mechanistic understanding of the AI’s role in modulating the DMN and FPN during memory formation. To assess reproducibility, we scrutinized network interactions across four different episodic memory tasks involving VFR, CATVFR, PALVCR, and WMSM tasks (Solomon et al., 2019).

We employed PTE, a robust metric of nonlinear and nonstationary dynamics, to investigate dynamic interactions between the AI and four key cortical nodes of the DMN and FPN. PTE assesses the ability of one time series to predict future values of another, estimating time-delayed directed influences, and is superior to methods like phase locking or coherence as it captures nonlinear and nonstationary interactions (Bassett and Sporns, 2017; Hillebrand et al., 2016; Lobier et al., 2014). PTE offers a robust and powerful tool for characterizing information flow between brain regions based on phase coupling (Hillebrand et al., 2016; Lobier et al., 2014; Wang et al., 2017) and has been successfully utilized in our previous studies (Das et al., 2022c; Das and Menon, 2020; Das and Menon, 2021; Das and Menon, 2022b; Das and Menon, 2023).

Informed by recent electrophysiology studies in nonhuman primates, which suggest that broadband field potentials activity, rather than narrowband, governs information flow in the brain (Davis et al., 2020; Davis et al., 2022), we first examined PTE in a 0.5–80 Hz frequency spectrum to assess dynamic-directed influences of the AI on the DMN and FPN. Our analysis revealed that AI exerts stronger influences on the PCC/precuneus and mPFC nodes of the DMN than the reverse. A similar pattern also emerged for FPN nodes, with the AI displaying stronger directed influences on the dPPC and MFG than the reverse. Crucially, this asymmetric pattern of directed information flow was replicated across all four memory tasks. Moreover, this pattern also held during the encoding and recall of memory phases of all four tasks.

Replication, a critical issue in all of systems neuroscience, is particularly challenging in the field of intracranial EEG studies, where data acquisition from patients is inherently difficult. Compounding this issue is the virtual absence of data sharing and the substantial complexities involved in collecting electrophysiological data across distributed brain regions (Das and Menon, 2022b). Consequently, one of our study’s major objectives was to reproduce our findings across multiple experiments, bridging verbal and spatial memory domains and task phases. To quantify the degree of replicability of our findings across these domains, we employed replication BF analysis (Ly et al., 2019; Verhagen and Wagenmakers, 2014). Our analysis revealed very high replication BFs related to replication of information flow from the AI to the DMN and FPN (Table 1). Specifically, the BFs associated with the replication of direction of information flow between the AI and the DMN and FPN were decisive (BFs > 100), demonstrating consistent results across various memory tasks and contexts.

Our analysis revealed a striking contrast between the AI and IFG in their patterns of directed information flow. While the AI exhibited strong directed influences on both the DMN and FPN, the IFG demonstrated an inverse relationship. Specifically, both the DMN and FPN exerted higher influence on the IFG than vice versa, a pattern that held consistent across both encoding and recall periods, and throughout all four memory experiments (Appendix 1—figures 4 and 5). Our analysis also revealed that the AI’s net outflow was significantly higher than that of the IFG during both encoding and recall phases, a finding replicated across all four experiments.

Furthermore, we compared the AI’s net outward directed influence during memory tasks to that observed during resting state. This analysis showed that the AI’s net outflow was significantly enhanced during both encoding and recall phases of memory tasks compared to resting state, consistently across all four experiments. This task-specific enhancement suggests that the AI’s role in coordinating large-scale network dynamics is specifically amplified during memory processes.

These results not only highlight the unique role of the AI in orchestrating large-scale network dynamics during memory processes but also demonstrate the specificity of this function compared to an anatomically adjacent and functionally relevant region implicated in cognitive control (Badre et al., 2005; Badre and Wagner, 2007; Cai et al., 2014; Wagner et al., 2001). The consistent pattern across diverse memory tasks and experimental phases underscores the robustness of the AI’s role in memory-related network interactions.

Our analysis of local neuronal activity revealed a consistent and specific pattern of high-gamma power suppression in the PCC/precuneus compared to the AI during memory encoding across all four episodic memory tasks. This finding aligns with the typical deactivation of DMN nodes during attention-demanding tasks (Wen et al., 2013), while also extending our understanding of the DMN’s role in episodic memory formation (Buckner et al., 2008; Menon, 2023).

Importantly, this suppression effect was confined to the PCC/precuneus within the DMN, with no parallel reductions observed in the mPFC. Moreover, suppression of the PCC/precuneus was stronger compared to the dPPC and MFG nodes of the FPN (Appendix 1—figures 12 and 13). Bayesian replication analysis substantiated the high degree of replicability of this PCC/precuneus suppression effect across tasks (BFs > 5.16e+1). These findings extend previous fMRI studies reporting DMN suppression during attention to external stimuli (Bressler and Menon, 2010; Raichle et al., 2001; Seeley et al., 2007) and complement optogenetic research in rodents’ brains demonstrating AI-induced suppression of DMN regions (Menon, 2023).

High-gamma activity (80–160 Hz) is a reliable indicator of localized, task-related neural processing, often associated with synchronized activity of local neural populations and elevated neuronal spiking (Canolty and Knight, 2010). High-gamma activity (typically ranging from 80 to 160 Hz) has been reliably implicated in various cognitive tasks across sensory modalities, including visual (Lachaux et al., 2005; Tallon-Baudry et al., 2005), auditory (Crone et al., 2001; Edwards et al., 2005), and across cognitive domains, including working memory (Canolty et al., 2006; Mainy et al., 2007) and episodic memory (Daitch and Parvizi, 2018; Sederberg et al., 2007). The suppression we observed during encoding likely reflects functional down-regulation of the PCC/precuneus, potentially to minimize interference from internally oriented processes during the encoding of external information.

In contrast, during memory recall, we observed different patterns of activity. In the three verbal tasks (VFR, CATVFR, and PALVCR), PCC/precuneus activity showed enhanced responses compared to the AI in the 1–1.6 s window prior to word production. However, it is crucial to note that our analysis was time-locked to word production rather than the onset of internal retrieval processes. In the spatial memory task WMSM, the PCC/precuneus exhibited an earlier onset and enhanced activity compared to the AI. This task may provide a clearer window into recall processes: findings align with the view that DMN nodes may play a crucial role in triggering internal recall processes. However, the precise timing of internal retrieval initiation remains a challenge in verbal tasks, potentially limiting our ability to capture the full dynamics of regional activity, and its replicability, during early stages of recall.

The observed high-gamma suppression in the PCC/precuneus during encoding, but not recall, likely reflects the distinct cognitive demands of these memory phases. Encoding primarily involves externally driven processes, requiring attention to and processing of incoming stimuli. In contrast, recall is predominantly internally driven, relying on the retrieval and reconstruction of stored information. This dissociation in PCC/precuneus activity aligns with its known role in the DMN, which typically shows deactivation during externally oriented tasks and activation during internally directed cognition. This pattern of activity underscores the flexible and context-dependent functioning of brain regions within large-scale networks, adapting their engagement to support different aspects of memory processing.

Notably, our findings reveal a robust and consistent directed influence exerted by the AI on all nodes of both the DMN and the FPN, extending across all four memory tasks and both memory encoding and recall phases. These directed influences were prominently manifested in broadband signals. Interestingly, such directed influences were not observed in the high-gamma frequency range (80–160 Hz). This absence aligns with current models positing that high-gamma activity is more likely to reflect localized processing, while lower-frequency bands are implicated in longer-range network communication and coordination (Bastos et al., 2015; Das et al., 2022c; Das and Menon, 2020; Das and Menon, 2021; Das and Menon, 2023; Miller et al., 2007). More generally, our findings emphasize that it is crucial to differentiate between high-gamma activity (f > 80 Hz) and sub-high-gamma (f < 80 Hz) fluctuations as these signal types are indicative of different underlying physiological processes, each with distinct implications for understanding neural network dynamics.

Our analysis revealed no significant differences in directed connectivity between successfully recalled and forgotten memory trials, suggesting that the reported effects may not be specific to successful memory formation and may be related to attentional or other general cognitive processing rather than memory processing per se. While our study provides valuable insights into the interactions between the AI and the DMN and FPN during cognitive tasks involving verbal and spatial information processing during memory tasks, it is crucial to acknowledge that these interactions may not be unique to memory processes. The AI’s directed influence on the DMN and FPN could reflect a more general role in coordinating attentional resources, which are essential for various cognitive functions, including memory formation (Menon and Uddin, 2010; Uddin, 2015). To disentangle the specific contributions of memory recall and attention, future studies should incorporate carefully designed control tasks that do not involve memory components. It is also important to note that successful memory recall likely involves the coordinated activity of multiple brain systems beyond the triple network model investigated here. For instance, the medial temporal lobe, including the hippocampus and adjacent cortical regions, plays a crucial role in episodic memory formation and retrieval (Burgess et al., 2002; Moscovitch et al., 2016). Future studies will need to investigate a broader set of brain areas during successful and unsuccessful memory trials to gain a more comprehensive understanding of the neural circuits supporting distinctions between successfully recalled and forgotten memory trials.

Our results reveal a consistent pattern of directed information flow from the AI to both the DMN and FPN, persisting across externally triggered encoding and internally driven free recall. This pattern underscores the AI’s robust and versatile role in modulating large-scale brain networks across diverse task contexts, aligning with the triple network model’s conceptualization of the AI as a critical hub for attentional and cognitive control (Menon, 2011; Menon, 2023). However, the persistence of AI-driven information flow during internally triggered free recall was unexpected, given the view of the DMN’s dominance in internal cognition. This reproducible pattern, observed across both externally and internally driven tasks in all four experiments, reinforces the AI’s crucial role in orchestrating network dynamics over extended time periods.

We did not detect an opposing pattern of greater directed influences from the DMN during recall, as might be expected given the internally driven nature of free recall. Several factors may contribute to this unexpected result. First, in the three verbal recall tasks, our PTE analysis was time-locked to word production onset, which may not capture the dynamics of network interactions during recall, particularly in the early retrieval initiation stage whose precise onset is unknown. This limitation is especially relevant for understanding the DMN’s role, which might be more prominent in the initiation of recall rather than the selection of verbal output. Secondly, the PTE method requires relatively long time series for robust estimation of information flow. The brief windows associated with the initiation of individual recall events may not provide sufficient data for detecting subtle shifts in network dynamics, potentially masking transient increases in DMN influence.

Moreover, the consistent AI-driven information flow during recall might reflect the SN’s ongoing role in monitoring and evaluating retrieved information, even during internally driven processes. This interpretation aligns with Sestieri and colleagues’ observation of sustained SN activity across all phases of memory search tasks (Sestieri et al., 2014) and suggests a more complex view of the AI’s function in both externally driven and internal cognitive processes.

Intriguingly, as noted above, while we observed PCC/precuneus suppression during encoding and enhancement during recall, the AI maintained its directed influence on this DMN node during encoding and recall. This apparent discrepancy between local activity (suppression) and network-level communication highlights the complex nature of brain network dynamics. It is likely that PTE-based network interactions examined in this study at the time scale of about 2 s miss subtle changes in directed interactions that occur during internally driven initiation of memory recall. Furthermore, our directed connectivity analysis used broadband signals (0.5–80 Hz), while power analysis of local neuronal activity focused on the high-gamma band (80–160 Hz). These different frequency ranges may capture distinct aspects of neural processing, with broadband connectivity reflecting more general, sustained network interactions.

To further elucidate these dynamics, future studies should consider employing techniques that can capture rapid changes in directed network interactions, investigating the temporal evolution of network interactions leading up to and following recall events, exploring the relationship between different frequency bands in connectivity and local activity measures, and developing methods to better estimate the onset of internal retrieval processes in verbal tasks. These approaches could provide valuable insights into the transition between externally driven and internally driven processes and offer a more precise understanding of the AI and PCC/precuneus’s differential roles in coordinating network dynamics across different memory phases.

Beyond information flow along individual pathways linking the AI with the DMN and FPN, our PTE analysis further revealed that the AI is an outflow hub in its interactions with the DMN and the FPN regardless of stimulus materials. As a central node of the SN (Menon and Uddin, 2010; Seeley et al., 2007; Sridharan et al., 2008), the AI is known to play a crucial role in influencing other networks (Menon and Uddin, 2010; Uddin, 2015). Our results align with findings based on control theory analysis of brain networks during a working memory task. Specifically, Cai et al., 2021 found higher causal outflow and controllability associated with the AI compared to DMN and FPN nodes during an n-back working memory task. Controllability refers to the ability to perturb a system from a given initial state to other configuration states in finite time by means of external control inputs. Intuitively, nodes with higher controllability require lower energy for perturbing a system from a given state, making controllability measures useful for identifying driver nodes with the potential to influence overall state dynamics. By virtue of its higher controllability relative to other brain areas, the AI is well-positioned to dynamically engage and disengage with other brain areas. These findings expand our understanding of the AI’s role, extending beyond attention and working memory tasks to incorporate two distinct stages of episodic memory formation. Our study, leveraging the temporal precision of iEEG data, substantially enhances previous fMRI findings by unveiling the neurophysiological mechanisms underlying the AI’s dynamic regulation of network activity during memory formation and cognition more generally.

Our findings bring a novel perspective to the seminal model of human memory proposed by Atkinson and Shiffrin, 1968. This model conceptualizes memory as a multistage process, with control mechanisms regulating the transition of information across these stages. The observed suppression of high-gamma power in the PCC/precuneus and enhancement in the AI during the encoding phase may be seen as one neurophysiological manifestation of these control processes. The AI’s role as a dynamic switch, modulating activity between the DMN and FPN, aligns with active processing and control needed to encode sensory information into short-term memory. On the other hand, the transformations observed during the recall phase, particularly the discernible lack of DMN suppression patterns, may correspond to the retrieval processes where internally generated cues steer the reactivation of memory representations during recall. These results provide a novel neurophysiological model for understanding the complex control processes underpinning human memory functioning.

Our study, while revealing important insights into network dynamics during memory processes, has several limitations that provide avenues for further investigation. Although our computational methods suggest directed influences, direct causal manipulations, such as targeted brain stimulation during memory tasks, are needed to establish definitive causal relationships between network nodes. The PTE method, while powerful, cannot reliably capture rapid shifts in network dynamics. Subsequent research should employ techniques with higher temporal precision to map these changes.

To determine whether our observed network dynamics are memory-specific or reflect more general cognitive processes, additional work should compare directed connectivity patterns across memory and non-memory tasks. Our analysis approach, necessitated by limited multi-task participation, precluded robust within-subject analyses. Future studies should aim for more consistent multi-task participation to enable individual-level analyses of network dynamics across tasks.

In the free recall verbal tasks, precisely timing the onset of internal retrieval processes remains challenging. Experimental designs with cued recall similar to the WMSM task could provide crucial insights into early stages of memory retrieval. This approach could help clarify the roles of different networks, especially the DMN, during the initiation of recall versus the execution of verbal output. The dissociation we observed between local activity and network-level communication warrants further investigation. Further studies are needed to determine the relationship between different frequency bands in connectivity and local activity measures to better understand how these distinct aspects of neural processing contribute to memory formation and retrieval.

Despite these limitations, our findings provide a robust foundation for investigations into the electrophysiological basis of large-scale brain network interactions during memory formation and recall. By addressing these limitations, subsequent studies can further refine our understanding of how these networks dynamically coordinate to support episodic memory and other cognitive functions. Such investigations may reveal a more dynamic interplay between the SN, DMN, and FPN, where their relative influences shift rapidly depending on the specific cognitive demands of the task.

Our study provides novel insights into the neural dynamics underpinning episodic memory processes across four diverse memory experiments. We discovered that the AI, a key node of the SN, exerts a strong and consistent directed influence on both the DMN and FPN during memory encoding and recall. This finding extends the applicability of the triple network model to episodic memory processes in both verbal and spatial domains, highlighting the AI’s crucial role as an outflow hub that modulates information flow within and between these cognitive networks.

Importantly, we observed a dissociation between local activity and network-level communication in the PCC/precuneus node of the DMN. The suppression of high-gamma power in this region during encoding, but not during recall, suggests a context-specific functional regulation that varies across memory phases. This finding reveals the intricate and dynamic interplay between local neural activity and large-scale network communication, and highlights the multifaceted nature of brain mechanisms underlying human memory processing.

The robust replicability of our findings across multiple memory tasks and modalities enhances the reliability and generalizability of our results, addressing a critical need in human intracranial EEG research. Our results reinforce the concept that memory operations rely on the concerted action of widely distributed brain networks (Mesulam, 1990), extending beyond traditional memory-specific regions.

By elucidating the electrophysiological basis of directed information flow within the triple network model, our study advances the understanding of neural circuit dynamics in human memory and cognition. Our findings provide a template for understanding the neural basis of memory impairments in neurological and psychiatric disorders. For instance, the disruption of these network interactions could contribute to memory deficits in conditions such as Alzheimer’s disease, where dysfunctions in the SN, DMN, and FPN are now being increasingly documented (Bonthius et al., 2005; Guzmán-Vélez et al., 2022).

iEEG recordings from 249 patients shared by Kahana and colleagues at the University of Pennsylvania (UPENN) (obtained from the UPENN-RAM public data release) were used for analysis (Jacobs et al., 2016). Patients with pharmaco-resistant epilepsy underwent surgery for removal of their seizure onset zones. iEEG recordings of these patients were downloaded from a UPENN-RAM consortium-hosted data-sharing archive (http://memory.psych.upenn.edu/RAM). These data were recorded at eight hospitals: Thomas Jefferson University Hospital; University of Texas Southwestern Medical Center; Emory University Hospital; Dartmouth College Hospital; University of Pennsylvania Hospital; Mayo Clinic; National Institutes of Health; and Columbia University Hospital. Prior to data collection, research protocols and ethical guidelines were approved by the Institutional Review Board at the participating hospitals and informed consent was obtained from the participants and guardians (Jacobs et al., 2016).

Details of all the recording sessions and data pre-processing procedures are described by Kahana and colleagues (Jacobs et al., 2016). Briefly, iEEG recordings were obtained using subdural grids and strips (contacts placed 10 mm apart) or depth electrodes (contacts spaced 5–10 mm apart) using recording systems at each clinical site. iEEG systems included DeltaMed XlTek (Natus), Grass Telefactor, and Nihon-Kohden EEG systems. Electrodes located in brain lesions or those which corresponded to seizure onset zones or had significant interictal spiking or had broken leads were excluded from analysis.

Anatomical localization of electrode placement was accomplished by co-registering the postoperative computed CTs with the postoperative MRIs using FSL (FMRIB [Functional MRI of the Brain] Software Library), BET (Brain Extraction Tool), and FLIRT (FMRIB Linear Image Registration Tool) software packages. Preoperative MRIs were used when postoperative MRIs were not available. The resulting contact locations were mapped to MNI space using an indirect stereotactic technique and OsiriX Imaging Software DICOM viewer package.

We used the insula atlas by Faillenot and colleagues to demarcate the AI (Faillenot et al., 2017), downloaded from http://brain-development.org/brain-atlases/adult-brain-atlases/. This atlas is based on probabilistic analysis of the anatomy of the insula with demarcations of the AI based on three short dorsal gyri and the posterior insula (PI), which encompasses two long gyri. To visualize iEEG electrodes on the insula atlas, we used surface-rendering code (GitHub: https://github.com/ludovicbellier/InsulaWM; Bellier, 2022) provided by Llorens et al., 2023. We used the Brainnetome atlas (Fan et al., 2016) to demarcate the PCC/precuneus, the mPFC, the dPPC, and the MFG. The dorsal anterior cingulate cortex node of the SN was excluded from analysis due to lack of sufficient electrode placement. Out of 249 individuals, data from 177 individuals (aged from 16 to 64, mean age 36.3 ± 11.5, 91 females) were used for subsequent analysis based on electrode placement in the AI and the PCC/precuneus, mPFC, dPPC, and MFG.

Original sampling rates of iEEG signals were 500 Hz, 1000 Hz, 1024 Hz, and 1600 Hz. Hence, iEEG signals were downsampled to 500 Hz, if the original sampling rate was higher, for all subsequent analysis. The two major concerns when analyzing interactions between closely spaced intracranial electrodes are volume conduction and confounding interactions with the reference electrode (Burke et al., 2013; Frauscher et al., 2018). Hence, bipolar referencing was used to eliminate confounding artifacts and improve the signal-to-noise ratio of the neural signals, consistent with previous studies using UPENN-RAM iEEG data (Burke et al., 2013; Ezzyat et al., 2018). Signals recorded at individual electrodes were converted to a bipolar montage by computing the difference in signal between adjacent electrode pairs on each strip, grid, and depth electrode and the resulting bipolar signals were treated as new ‘virtual’ electrodes originating from the midpoint between each contact pair, identical to procedures in previous studies using UPENN-RAM data (Solomon et al., 2019). Line noise (60 Hz) and its harmonics were removed from the bipolar signals using band-stop filters at 57–63 Hz, 117–123 Hz, and 177–183 Hz. Finally, each bipolar signal was Z-normalized by removing mean and scaling by the standard deviation. For filtering, we used a fourth-order two-way zero phase lag Butterworth filter throughout the analysis. iEEG signals were filtered in the broad frequency spectrum (0.5–80 Hz) as well as narrowband frequency spectra delta-theta (0.5–8 Hz), alpha (8–12 Hz), beta (12–30 Hz), gamma (30–80 Hz), and high-gamma (80–160 Hz).

We first filtered the signals in the high-gamma (80–160 Hz) frequency band (Canolty et al., 2006; Helfrich and Knight, 2016; Miller et al., 2009) using sequential band-pass filters in increments of 10 Hz (i.e., 80–90 Hz, 90–100 Hz, etc.), using a fourth-order two-way zero phase lag Butterworth filter. We used these narrowband filtering processing steps to correct for the 1/f decay of power. We then calculated the amplitude (envelope) of each narrow band signal by taking the absolute value of the analytic signal obtained from the Hilbert transform (Foster et al., 2015). Each narrow band amplitude time series was then normalized to its own mean amplitude, expressed as a percentage of the mean. Finally, we calculated the mean of the normalized narrow band amplitude time series, producing a single-amplitude time series. Signals were then smoothed using 0.2 s windows with 90% overlap (Kwon et al., 2021) and normalized with respect to 0.2 s pre-stimulus periods by subtracting the pre-stimulus baseline from the post-stimulus signal.

PTE is a nonlinear measure of the directionality of information flow between time series and can be applied to nonstationary time series (Das and Menon, 2021; Lobier et al., 2014). Note that the information flow described here relates to signaling between brain areas and does not necessarily reflect the representation or coding of behaviorally relevant variables per se. The PTE measure is in contrast to the Granger causality measure, which can be applied only to stationary time series (Barnett and Seth, 2014). We first carried out a stationarity test of the iEEG recordings (unit root test for stationarity [Barnett and Seth, 2014]) and found that the spectral radius of the autoregressive model is very close to one, indicating that the iEEG time series is nonstationary. This precluded the applicability of the Granger causality analysis in our study.

Given two time series {xi} and {yi}, where i=1,2,...,M, instantaneous phases were first extracted using the Hilbert transform. Let {xip} and {yip} , where i=1,2,...,M, denote the corresponding phase time series. If the uncertainty of the target signal {yip} at delay τ is quantified using Shannon entropy, then the PTE from driver signal {xip} to target signal {yip} can be given by, (1)PTEx→y=∑ip(yi+τp,yip,xip)log(p(yi+τp|yip,xip)p(yi+τp|yip))

where the probabilities can be calculated by building histograms of occurrences of singles, pairs, or triplets of instantaneous phase estimates from the phase time series (Hillebrand et al., 2016). For our analysis, the number of bins in the histograms was set as 3.49×STD×M−1/3 and delay τ was set as 2M/M±, where STD is average standard deviation of the phase time series {xip} and {yip} and M± is the number of times the phase changes sign across time and channels (Hillebrand et al., 2016). PTE has been shown to be robust against the choice of the delay τ and the number of bins for forming the histograms (Hillebrand et al., 2016). In our analysis, PTE was calculated for the entire encoding and recall periods for each trial and then averaged across trials.

Net outflow was calculated as the difference between the total outgoing information and total incoming information, that is, net outflow  = PTE(out) − PTE(in). For example, for calculation of PTE(out) and PTE(in) for the AI electrodes, electrodes in the PCC/precuneus, mPFC, dPPC, and MFG were considered, that is, PTE(out) was calculated as the net PTE from AI electrodes to the PCC/precuneus, mPFC, dPPC, and MFG electrodes, and PTE(in) was calculated as the net PTE from the PCC/precuneus, mPFC, dPPC, and MFG electrodes to AI electrodes. Net outflow for the PCC/precuneus, mPFC, dPPC, and MFG electrodes was calculated similarly.

We used PLV to compute phase synchronization between two time series (Lachaux et al., 1999). We first calculated the instantaneous phases of the two signals by using the analytical signal approach based on the Hilbert transform (Bruns, 2004). Given time series x(t),t=1,2,...,M, its complex-valued analytical signal z(t) can be computed as(2)z(t)=x(t)+ix~(t)=Ax(t)eΦx(t),

where i denotes the square root of minus one, x~(t) is the Hilbert transform of x(t), and Ax(t) and are the instantaneous amplitude and instantaneous phase respectively and can be given by(3)Ax(t)=[x(t)]2+[x~(t)]2andϕx(t)=arctanx~(t)x(t).

The Hilbert transform of x(t) was computed as(4)x~(t)=1πPV∫−x∞x(τ)t−τdτ,

where PV denotes the Cauchy principal value. MATLAB function ‘hilbert’ was used to calculate the Hilbert transform in our analysis. Given two time series x(t) and y(t), where t=1,2,...,M, the PLV (zero-lag) can be computed as(5)PLV≜|E[ei(ϕx(t)−ϕy(t))]|,

where ϕy(t) is the instantaneous phase for time series y(t), | ⋅ | denotes the absolute value operator, E[⋅] denotes the expectation operator with respect to time t, and i denotes the square root of minus one. PLVs were then averaged across trials to estimate the final PLV for each pair of electrodes.

Statistical analysis was conducted using mixed-effects analysis with the lmerTest package (Kuznetsova et al., 2017) implemented in R software (version 4.0.2, R Foundation for Statistical Computing). Because PTE data were not normally distributed, we used BestNormalize (Peterson and Cavanaugh, 2020), which contains a suite of transformation-estimating functions that can be used to optimally normalize data. The resulting normally distributed data were subjected to mixed-effects analysis with the following model: PTE ~ Condition + (1|Subject), where Condition models the fixed effects (condition differences) and (1|Subject) models the random repeated measurements within the same participant, similar to prior iEEG studies (Das and Menon, 2021; Hoy et al., 2021; Salamone et al., 2021). Before running the mixed-effects model, PTE was first averaged across trials for each channel pair. ANOVA was used to test the significance of findings with FDR-corrections for multiple comparisons (p<0.05). Linear mixed-effects models were run for encoding and recall periods separately. Similar mixed-effects statistical analysis procedures were used for comparison of high-gamma power across task conditions, where the mixed-effects analysis was run on each of the 0.2 s windows.

For effect size estimation, we used Cohen’s d statistics for pairwise comparisons. We used the lme.dscore() function in the EMAtools package in R for estimating Cohen’s d.

We used replication BF (Ly et al., 2019; Verhagen and Wagenmakers, 2014) analysis to estimate the degree of replicability for the direction of information flow for each frequency and task condition and across task domains. Analysis was implemented in R software using the BayesFactor package (Rouder et al., 2009). Because PTE data were not normally distributed, as previously, we used BestNormalize (Peterson and Cavanaugh, 2020) to optimally normalize data. We calculated the replication BF for pairwise experiments. We compared the BF of the joint model PTE(task1 + task2) ~ Condition + (1|Subject) with the BF of individual model as PTE(task1) ~ Condition + (1|Subject), where task1 denotes the VFR (original) task and task2 denotes the CATVFR, PALVCR, or WMSM (replication) conditions. We calculated the ratio BF(task1 + task2)/BF(task1), which was used to quantify the degree of replicability. We determined whether the degree of replicability was higher than 3 as BF of at least three indicates evidence for replicability (Jeffreys, 1998). A BF of at least 100 is considered ‘decisive’ for the degree of replication (Jeffreys, 1998). Same analysis procedures were used to estimate the degree of replicability for high-gamma power comparison of DMN and FPN electrodes with the AI electrodes across experiments.