From Correlation to Causation: Understanding Episodic Memory Networks

The paper examines how distributed brain regions coordinate to support episodic memory—encoding, consolidation, and retrieval—and how non-invasive brain stimulation (NIBS), particularly TMS, enables causal tests of these processes. Motivated by clinical relevance to disorders with memory impairment (e.g., Alzheimer’s disease, schizophrenia, PTSD), the review synthesizes historical lesion evidence (e.g., H.M.) and neuroimaging findings with recent stimulation studies to identify functional roles and network interactions among hippocampus/MTL and fronto-parietal cortices. The goal is to progress from correlational to causal understanding, outlining promising cortical stimulation targets to modulate deeper episodic memory networks and suggesting future research directions.

The review traces the development of episodic memory theory and neural models: (1) Foundational psychological frameworks distinguished episodic from semantic memory (Tulving) and proposed modular processing (Fodor; Moscovitch). Early PET/fMRI work showed left prefrontal engagement during encoding and right prefrontal during retrieval, with material/process-specific activations forming large-scale networks. (2) Mechanistic roles across stages: Encoding involves sensory cortices, hippocampal binding, and PFC/parietal attentional-control; emotional memory benefits from amygdala–hippocampus interactions. Short-term maintenance relies on hippocampal–neocortical reactivation and parietal attention to internal representations, overlapping with working memory (DLPFC). Consolidation encompasses synaptic and systems processes, prominently during NREM sleep via coupling of hippocampal ripples, thalamo-cortical spindles, and cortical slow oscillations; alternative contextual binding accounts emphasize persistent hippocampal involvement and time-extended context effects. Evidence suggests rapid cortical (parietal, PFC) engram formation alongside hippocampus, with shifting contributions over time. Retrieval is described by transfer-appropriate processing and hippocampal indexing, but also by constructive/reconstructive reinstatement shaped by attention/control (PFC) and parietal mechanisms (AtoM model). (3) Network frameworks: The MTL memory system (hippocampus, PRC, PHC, ERC) as core storage/transformation hub; the PMAT framework posits two interacting networks—Posterior Medial (PHC, RSC, PCC, angular gyrus, precuneus, anterior thalamus, mPFC) processing context, and Anterior Temporal (PRC, aVTC, amygdala, lateral OFC) processing item/concept—with hippocampus and vmPFC as integrative nodes. (4) Resting-state network perspectives: The Default Mode Network (DMN) overlaps with PMAT components, subdividing into subnetworks (e.g., DMN-A hippocampal, DMN-B frontal; PM, AT, and MP subnetworks) with distinct hippocampal long-axis coupling and roles (emotion/value, social cognition, control/self-referential processing). Data-driven mapping further delineates a Medial Temporal Network (MTN) interfacing visual, DMN, and hippocampus. Collectively, episodic memory emerges from coordinated, dynamic large-scale network interactions rather than isolated regions.

This is a systematic-narrative review of TMS studies targeting episodic memory networks. The authors conducted systematic searches in PubMed and Web of Science to identify studies that used TMS to modulate regions implicated in episodic memory (primarily cortical sites with network connections to MTL). Inclusion, screening, and additional methodological details are described in the supplementary materials. A summary of included studies is provided (Fig. 4), and key parameters (target region, stimulation protocol, participant demographics, stimulation timing relative to encoding/retrieval, neuroimaging and behavioral outcomes) are tabulated (Table 1). The review also outlines TMS principles (electromagnetic induction, depth limitations), stimulation protocols (single/paired pulse; rTMS with frequency-dependent effects; theta-burst stimulation including cTBS/iTBS), neuronavigation for precise targeting (including MRI-guided frameless stereotaxy), and integration with neuroimaging/EEG for target localization and state-informed application.

• Parietal cortex as a causal access point to hippocampal networks:
  • Left lateral parietal cortex (LPC)/inferior parietal regions with strong hippocampal connectivity: Beta-frequency rTMS (e.g., 20 Hz) enhanced cortical–hippocampal connectivity and improved associative memory with effects lasting up to 24 hours after multi-day stimulation; effects replicated across several studies. cTBS at similar targets can also acutely enhance episodic memory.
  • Angular gyrus (AG): cTBS reduced free recall of autobiographical memories and decreased episodic detail generation for past/future simulation, consistent with AG’s role in contextual detail integration within PM networks.
  • Precuneus: cTBS decreased source memory errors (improved context retrieval); beta rTMS enhanced memory and neural activity including in prodromal Alzheimer’s disease.
  • Right PPC: 1 Hz rTMS before retrieval improved non-verbal recognition memory; stimulation before encoding often showed no effect, suggesting timing- and lateralization-dependent roles.
• Frontal cortex findings are mixed with partial lateralization:
  • Encoding: Left DLPFC stimulation before/during encoding variably disrupted or improved memory depending on protocol and individual strategy/baseline; paired-pulse during encoding impaired performance, and some rTMS protocols showed null effects.
  • Retrieval: Right DLPFC stimulation effects varied by frequency (e.g., 1 Hz improved recognition; 50 Hz disrupted; cTBS sometimes null), indicating frequency-, timing-, and task-dependence.
• Timing matters:
  • Before/during encoding: Cerebellar theta (but not beta) improved encoding; hippocampal-network-targeted cTBS before encoding increased reinstatement of naturalistic patterns and improved accuracy for typical patterns; right DLPFC during encoding enhanced under full attention but impaired under divided attention; immediate post-encoding right DLPFC disrupted late-stage encoding.
  • Before/during retrieval: Left AG cTBS before retrieval reduced free recall (no effect on cued recall/word-pairs); cTBS can improve item retrieval relative to sham or beta; right PPC before retrieval improved performance; 1 Hz right DLPFC during pre-retrieval enhanced retrieval.
  • Pre-/post-task blocks: Multi-day posterior-medial network stimulation increased MTL connectivity during autobiographical retrieval and improved memory; some effects persisted 24 hours but waned by 1 week in certain studies.
• Network-consistent interpretation:
  • Effects align with PMAT/DMN frameworks: Parietal targets (AG, precuneus, LPC) modulate PM-network processes and hippocampal coupling; vmPFC/hippocampus posited as integrators though mPFC is underexplored by TMS.
• Clinical indications and populations:
  • Older adults and MCI: rTMS to precuneus or parietal targets improved memory and modulated hippocampal connectivity; some studies reported enhanced AVLT scores in MCI after weeks of rTMS.
• Method/technology insights:
  • Neuronavigation improves targeting precision; TMS–fMRI can capture timing-specific connectivity changes; frequency labels (excitatory/inhibitory) from motor cortex do not generalize straightforwardly to episodic memory systems.
• Convergent pattern noted by authors: Stimulation of the left PPC prior to memory tasks consistently tended to improve performance across studies, despite heterogeneity elsewhere.

The synthesis indicates that episodic memory is supported by distributed, interacting networks wherein cortical parietal regions provide causal network access to hippocampal/MTL mechanisms. TMS results move beyond correlational imaging by demonstrating that modulating left LPC/AG/precuneus can alter hippocampal connectivity and memory behavior, consistent with PMAT and DMN subnetworks. Variability across prefrontal findings reflects complex roles of executive control, attentional states, and protocol/timing dependencies. Timing relative to encoding/retrieval is critical, mirroring theoretical stage-specific processes (e.g., control/attention early, reinstatement later). Network-informed targeting and state-dependent stimulation (integrating EEG/fMRI) offer pathways to refine causal tests and personalize interventions. The review argues for a dynamic, state- and stage-sensitive perspective on episodic memory networks, emphasizing that stimulation effects are contextually constrained by individual connectivity, oscillatory states, and task demands.

This review integrates psychological theory, neurobiological models, and causal TMS studies to outline how episodic memory emerges from coordinated hippocampal–cortical networks. It identifies parietal cortex nodes (left LPC, AG, precuneus) as promising stimulation targets to modulate hippocampal connectivity and memory, with timing and protocol choices shaping outcomes. The work highlights translational potential for enhancing memory in aging and MCI, while underscoring the need for precision approaches. Future research should: (1) personalize targeting via precision fMRI and state-informed EEG/MEG; (2) employ closed-loop stimulation; (3) extend causal tests to underexplored stages (short-term maintenance, consolidation, including sleep); (4) investigate mPFC/vmPFC roles with TMS; and (5) explore emerging deep-reaching noninvasive methods (e.g., temporal interference electrical/magnetic stimulation) to directly modulate MTL structures.

Key limitations include variability of TMS outcomes due to individual differences in structural/functional connectivity, fluctuating brain states/oscillations, and reliance on motor-threshold-based dosing that may not generalize to non-motor targets. Many findings depend on specific timing and protocols, limiting generalizability. Current methods indirectly affect deep MTL structures; TES approaches often lack focal precision. Research has focused more on encoding/retrieval than on short-term maintenance and consolidation. Detailed search/inclusion criteria are in supplementary materials, limiting reproducibility assessment within the main text. Some behavioral improvements show short-lived durability (e.g., effects waning after a week), and result heterogeneity complicates firm conclusions on lateralization and frequency effects.