Spatiotemporal dynamics of hippocampal-cortical networks underlying the unique phenomenological properties of trauma-related intrusive memories

Trauma-related intrusive memories (TR-IMs) are involuntary, sensory-rich recollections that often lack contextual details and can evoke a strong sense of reliving (“here-and-now”), contributing to adverse outcomes across disorders. Conceptual models emphasize their phenomenological properties: the warning signal hypothesis highlights heightened emotional intensity as a learned cue for threat, while dual-representation theory posits imbalance between contextual (C-rep) and sensory (S-rep) systems. Neurobiologically, TR-IMs have been linked to sensory cortices, salience network (SN), and hippocampal dysfunction, alongside large-scale network alterations in PTSD (elevated SN/sensory activity; reduced DMN/HPC connectivity; disrupted DMN–SN anticorrelation). Episodic memory research shows dynamic, functionally distinct interactions of anterior hippocampus (aHPC) with affective/cognitive networks (DMN, SN) and posterior hippocampus (pHPC) with sensory and posterior-medial systems supporting visuospatial detail. The study’s objective was to identify dynamic spatiotemporal patterns of intrinsic a/pHPC–cortical co-activation associated with specific TR-IM phenomenological properties, hypothesizing aHPC–cognitive network dynamics for cognitive-affective properties and pHPC–sensory/posterior-medial dynamics for sensory-perceptual properties.

Prior work on TR-IMs has largely examined frequency and intensity rather than distinct phenomenological features. The warning signal hypothesis frames TR-IMs as affectively intense cues to threat. Dual-representation theory proposes interacting contextual (C-rep) and sensory (S-rep) memory systems, with TR-IMs marked by strong S-reps and deficient C-reps. Neurocircuitry models of PTSD implicate amygdala–hippocampal–mPFC circuits and large-scale networks: increased SN and sensory activity, decreased DMN and hippocampal connectivity, and disrupted DMN–SN anticorrelation. Episodic memory literature supports anterior–posterior hippocampal specialization: aHPC with limbic/prefrontal (emotion, gist, self-reference) and pHPC with posterior-medial/sensory systems (visuospatial detail, mental imagery). Prior neuroimaging on intrusions implicates SN and prefrontal regions during encoding/retrieval of negative autobiographical memories. However, intrinsic neural correlates of specific TR-IM properties remain understudied, and the hippocampus is often treated as a unitary, static structure rather than dynamically interacting subregions.

Design: Cross-sectional neuroimaging study with ecological momentary assessment (EMA) preceding resting-state fMRI, examining dynamic co-activation patterns (CAPs) of a/p hippocampus with cortical networks and their associations with TR-IM properties. Participants: 99 trauma-exposed adults enrolled; inclusion: DSM-5 Criterion A trauma exposure and ≥2 TR-IMs/week over past month. Procedures: Two weeks of daily EMAs (3 semi-random surveys/day via MetricWire) assessing presence of TR-IMs and 18 items (0–4 Likert) adapted from Autobiographical Memory Questionnaire (AMQ), grouped into vividness, visual detail, reliving (here-and-now), emotional intensity, fragmentation, intrusiveness. Visit 2: clinical interview, self-report questionnaires, and 13-min eyes-open rs-fMRI. Usable fMRI data: 84 participants (exclusions: excessive motion n=10, structural abnormalities n=2, poor alignment n=1, fell asleep n=2). Demographics (N=84): mean age 31.1±9.7 years; gender: women 69%, men 18%, non-binary 13%; race/ethnicity: NH White 67%, bi-/multiracial 19%, NH Black 6%, Asian 4%, Hispanic/Latino 1%; PTSD diagnosis 75% (n=63), CAPS-5 total 33.7±11.4; LEC-5 total 12.1±7.0; total number of TR-IMs over EMA 23.1±25.6. Clinical measures: CAPS-5 for PTSD; AMQ completed at Visit 2 for retrospective memory qualities (0–4 Likert). MRI acquisition: 3T Siemens Prisma with 64-channel coil; HCP Young Lifespan protocols; 13-min eyes-open rs-fMRI. Preprocessing: fMRIPrep v20.2.7; further processing in CONN (regressing WM/CSF signals, scrubbing motion outliers FD>0.5 mm, high-pass filter 0.01 Hz). Co-activation pattern (CAP) analysis: a/pHPC seeds defined from prior work to maximize A–P segregation; TbCAP toolbox. Union seed approach: select volumes exceeding Z>1 for aHPC and/or pHPC to ensure HPC activation. K-means clustering of selected volumes; consensus clustering determined optimal k=4 CAPs. Metrics per participant: (1) count (number of supra-threshold volumes per CAP), (2) persistence (probability of remaining in a CAP across consecutive volumes). Statistical analysis: Partial correlations between CAP metrics and EMA-averaged TR-IM properties controlling for age and sex; multiple-comparison correction via FDR at two levels: across all tests (4 CAPs × 6 properties = 24; FDR_total) and within-property across CAPs (FDR_property). Properties with significant effects entered into multiple linear regression (all CAPs as predictors) for specificity. Robustness tested with linear mixed models (LMMs) on repeated EMA measures: univariate (each CAP metric separately) and multivariate (simultaneous CAP metrics), with subject-specific random intercepts; FDR applied to univariate LMMs. Additional correlations: CAP metrics with conventional clinical measures (total TR-IM frequency, retrospective AMQ properties, CAPS-5 severity) and moderation by PTSD diagnosis. Variables were mean-centered and scaled. Imaging-static connectivity checks with HPC and CAP networks reported in SI.

CAP characterization: Four CAPs identified. CAP1: aHPC with DMN activation and deactivation of SN/VAN and DAN; most frequent and persistent (t>3.82, p<0.001). CAP2: activation of visual cortex and SPL (DAN); dominated by pHPC activation [pHPC mean count 32.2 (50%) ±7.8 vs aHPC 22.7 (36%) ±7.5; t=11.3, p<0.001]. CAP3: activation in SPL/SMG/MFG/dlPFC (DAN/FPN) with visual cortex deactivation. CAP4: co-activation of visual and sensorimotor cortices with dACC/AI (SN/VAN) and DMN deactivation (PCC/Precuneus, ANG, SFG). CAP1 associated more with aHPC than pHPC [aHPC 35.5 (45%) ±11.1 vs pHPC 28.1 (37%) ±8.1; t=6.27, p<0.001]. Associations with TR-IM properties (EMA): - Visual features: positively associated with CAP4 count (partial r=0.33, p=0.002, FDR_total<0.05); specificity confirmed in multiple regression (b=0.39, t=2.69, p=0.009). LMMs: univariate b=0.28, t=3.11, p=0.003 (FDR_total<0.05); multivariate b=0.37, t=2.77, p=0.007. - Emotional intensity: inversely associated with CAP1 count (partial r=-0.32, p=0.003, FDR_total<0.05) and CAP1 persistence (partial r=-0.30, p=0.007, FDR_emotion<0.05). LMMs: count b=-0.23, t=-3.17, p=0.002; persistence b=-0.21, t=-2.86, p=0.005. Specificity in multivariate models not confirmed (count p=0.530; persistence p=0.108). - Reliving (here-and-now): positively associated with CAP2 persistence (partial r=0.28, p=0.009, FDR_reliving<0.05); specificity confirmed (b=0.25, t=2.22, p=0.029). LMMs: univariate b=0.21, t=2.66, p=0.009; multivariate b=0.20, t=2.24, p=0.028. - No significant associations reported for vividness, fragmentation, or intrusiveness with CAP metrics after correction. Associations with conventional measures: No associations between CAP metrics and total number of TR-IMs (|r|<0.12, p>0.288) or PTSD symptom clusters/total severity (|r|<0.15, p>0.166); PTSD diagnosis did not moderate CAP–TR-IM associations (p>0.140). Weak, non-FDR-surviving links with retrospective reports: reliving with fewer CAP1 occurrences (r=-0.24, p=0.031) and emotional intensity with less CAP1 persistence (r=-0.24, p=0.034).

Findings support that distinct hippocampal–cortical dynamics underlie different phenomenological properties of TR-IMs. Reduced frequency and stability of aHPC–DMN co-activation (CAP1) related to greater emotional intensity, suggesting disruption of intrinsic aHPC–DMN interactions that typically support affective, schematic aspects of autobiographical memory and DMN–SN anticorrelation. This aligns with the warning signal hypothesis and literature on altered DMN/SN dynamics in PTSD. Sensory (visual) features were uniquely linked to increased frequency of co-activation of hippocampus with sensory cortices and VAN/SN (CAP4), supporting roles for sensory cortex and salience/attention networks in bottom-up, multimodal sensory-driven capture posited by dual-representation theory’s S-rep system. Reliving (here-and-now) was associated with more persistent pHPC–visual cortex co-activation (CAP2), consistent with pHPC’s role in visuospatial detail and mental imagery; persistence (rather than frequency) may reflect being “stuck” in sensory-driven reconstruction (sensory replay) that biases spontaneous re-experiencing. Null associations with static connectivity and with conventional symptom counts underscore the value of dynamic measures and ecological sampling for capturing neural substrates of specific TR-IM properties. Overall, results suggest interacting sensory-perceptual and cognitive-affective systems, with potential hyper-co-activation patterns, contribute to the heterogeneity of TR-IM phenomenology and highlight dynamic HPC–network interactions as mechanistic targets.

The study demonstrates that dynamic a/p hippocampal–cortical co-activation patterns differentially map onto key phenomenological properties of trauma-related intrusive memories: emotional intensity relates to diminished aHPC–DMN dynamics, visual features to increased HPC–sensory/VAN–SN co-activation, and reliving to persistent pHPC–visual coupling. These insights clarify neural substrates beyond conventional symptom measures and advocate for ecological assessments to identify mechanism-linked biomarkers. Dynamic HPC–network interactions emerge as viable intervention targets; neuromodulatory approaches (e.g., neurofeedback, non-invasive brain stimulation) that engage identified networks, paired with fine-grained phenomenological assessments, may enable individualized, transdiagnostic treatments. Future work should probe task-based dynamics, differentiate flashbacks, examine multiple sensory modalities, and assess potential sex differences.

- Imaging was limited to resting state and occurred days to weeks after EMA completion, introducing temporal separation. - EMA did not strictly anchor to a single index trauma; participants with multiple traumas may have referenced different events, increasing within-subject variability. - The study did not distinguish TR-IMs from flashbacks; mechanisms specific to dissociation or nowness need clarification. - Intrinsic (resting) dynamics were studied; task-based paradigms (symptom provocation, prompted retrieval, or probing spontaneous memories during scan) are needed for finer mechanistic inference. - Sensory-perceptual assessments focused on visual features; other modalities (somatosensory, auditory, olfactory, interoceptive) were not evaluated. - The sample was predominantly female, limiting analyses of sex differences despite known effects on PTSD symptoms. - No associations were found with retrospective measures, highlighting potential recall bias but also leaving questions about generalizability across assessment methods.