Associative memory, the ability to learn and remember relationships between different stimuli, is fundamental to daily life. Impairments in this cognitive function, exacerbated by aging and neurological disorders, necessitate a thorough understanding of its neural underpinnings. While the hippocampus and MTL are implicated in associative memory encoding and retrieval, the precise cellular mechanisms remain unclear. This study aimed to elucidate these mechanisms at the single-cell level in the human MTL. Based on theories proposing temporally precise neural binding, the hypothesis was that individual stimuli within associative memories are encoded by distinct, functionally specialized neurons, which interact transiently during encoding and retrieval. Object-location associations, critical for knowing item locations, were chosen as a model for this investigation. The study investigated whether encoding and retrieval of object-location memories correlate with simultaneous neural activity during hippocampal ripples, high-frequency oscillations believed to synchronize neural activity across brain regions. Ripple-locked coactivity of object and place cells could potentially underlie encoding and retrieval by inducing and reactivating synaptic connections between neurons representing different memory elements. This coactivity could also produce conjunctive memory representations in downstream neurons responding only to specific stimulus combinations. Prior research established the involvement of hippocampal ripples in various cognitive functions, including memory encoding, retrieval, and consolidation in both animal and human studies, particularly the link between ripple-locked place cell activity and navigational paths. However, whether hippocampal ripples interconnect functionally different neuron types to form associative memories remained unexplored. To address this, the researchers used single-neuron and intracranial EEG recordings from MTL in epilepsy patients during an associative object-location memory task in a virtual environment.

A substantial body of research points to the hippocampus and surrounding MTL regions as crucial for encoding and retrieving associative memories. However, the underlying neural mechanisms remain poorly understood. Several theories attempt to explain how neural circuits encode associative memories. The "conjunctive hypothesis" proposes that conjunctive representations exist, where neurons encode unique combinations of multiple stimuli, only responding when those stimuli are presented together. In contrast, the "coactivity hypothesis" suggests that stimulus-specific neurons, representing individual memory elements, temporarily coactivate during encoding and retrieval. Encoding coactivity establishes synaptic connections, while retrieval coactivity reflects mutual activation via pre-existing connections. Studies in rodents and primates have provided evidence supporting both hypotheses, showing hippocampal neurons encoding specific object-location or scene-eye movement associations. In humans, hippocampal neurons have also been observed to respond to unique stimulus combinations. However, direct evidence for the coactivity hypothesis in human associative memory at the single-neuron level has been lacking, particularly the role of hippocampal ripples in mediating this coactivity.

This study involved intracranial EEG and single-neuron recordings from the MTL of 30 epilepsy patients (16 female; age 19–61 years) undergoing treatment for pharmacologically intractable epilepsy. Ethical approval was obtained, and all participants provided informed consent. Data were collected using Ad-Tech macroelectrodes and Behnke-Fried microelectrodes (Ad-Tech) which contained bundles of nine platinum-iridium microelectrodes that protruded from the tip of the macroelectrode. Recordings were performed at sampling rates of 2 kHz (macroelectrodes) and 30 kHz (microelectrodes). Patients performed an associative object-location memory task in a virtual environment created using Unreal Engine 2. The task involved an initial encoding phase where participants learned the locations of eight objects. Subsequently, they performed multiple test trials consisting of an inter-trial interval (ITI), a cue period (2-second object presentation), a retrieval phase (self-paced navigation to the remembered location), feedback (1.5-second emoticon), and a re-encoding phase (self-paced collection of the object from its correct location). Hippocampal ripples were identified using established algorithms, rigorously excluding interictal epileptic discharges (IEDs). Ripple characteristics (rate, duration, frequency) were analyzed across different trial phases. Single neurons were classified as object cells (increased firing rates for a specific object) or place cells (increased firing rates at a specific location). Coactivity between object and place cells during hippocampal ripples was analyzed using a coactivity z-score, considering various time bins relative to ripple peaks. Three statistical tests were employed to evaluate coactivity significance: comparison against chance, against baseline coactivity, and against non-associative cell pairs. Detailed statistical analyses included repeated measures ANOVAs, linear mixed models, partial correlations, one-sample t-tests, Rayleigh tests, and cluster-based permutation tests, all performed in MATLAB.

The study identified 120 object cells (11% of all neurons) and 109 place cells (10%) across various MTL regions, with object cells most prevalent in the entorhinal cortex, parahippocampal cortex, and temporal pole, and place cells concentrated in the entorhinal cortex, hippocampus, and parahippocampal cortex. Object cells exhibited strong responses to preferred objects during the first second after cue onset, and place cells showed robust firing within their place fields. Crucially, during both retrieval and re-encoding, object and place cells representing associative information (preferred object located within the place field) showed significantly increased coactivity during hippocampal ripples. This ripple-locked coactivity was particularly prominent during later periods of the task, suggesting a role in memory stabilization or updating rather than initial memory formation. The ripple-locked timing of coactivity shifted between retrieval (slightly after ripple peaks) and re-encoding (slightly before ripple peaks), potentially reflecting task-dependent information flow changes. Furthermore, ripple rates correlated with behavioral state and memory performance, increasing before successful retrieval and after unsuccessful retrieval, highlighting their functional importance in encoding and retrieval. The increased ripple-locked coactivity was predominantly observed during periods of immobility.

The findings provide strong support for the coactivity hypothesis of associative memory, demonstrating that stimulus-specific neurons representing individual memory elements coactivate during hippocampal ripples during encoding and retrieval of object-location associations. This ripple-locked coactivity, particularly during later task phases, suggests a role in memory consolidation or updating. The observed shift in coactivity timing between retrieval and re-encoding points to a flexible interplay between MTL neurons and hippocampal ripples, potentially reflecting changes in information flow direction. The correlation between ripple rates and memory performance extends previous findings, strengthening the link between hippocampal ripples and human memory processes. While the study primarily focuses on the coactivity hypothesis, the results do not exclude the conjunctive hypothesis. It is plausible that coactivity between object and place cells might induce conjunctive coding in downstream neurons. The study's findings extend the understanding of hippocampal ripple function in awake humans, suggesting a role beyond memory consolidation and retrieval to include associative memory formation and updating.

This study provides compelling evidence for the involvement of hippocampal ripples in human associative memory. The ripple-locked coactivity of stimulus-specific neurons representing distinct memory elements strengthens with learning and is crucial for both encoding and retrieval. The observed temporal shift in coactivity suggests task-dependent flexibility in information processing. Future research should investigate the causal role of ripple-locked coactivity in synaptic plasticity and explore the potential of ripple modulation for therapeutic interventions in memory disorders.

The study's correlational design does not establish causality between ripple-locked coactivity and synaptic connection formation. Future studies using causal manipulations are needed. The use of relatively broad (100 ms) time windows for coactivity analysis might not fully capture the precision of spike-timing-dependent plasticity. The precise nature of the high-frequency events identified as "ripples" in the study and their relationship to other oscillations requires further investigation. Furthermore, the study sample consisted of epilepsy patients, potentially limiting the generalizability of the findings to the wider population.