When Jie Zheng and colleagues threaded hair-thin electrodes into the brains of epilepsy patients at Children’s Hospital Boston, they expected to find separate groups of neurons handling separate jobs: one set for facts, another for personal memories. Instead, the recordings, published in Nature in 2025, revealed that individual medial temporal lobe neurons carried both content signals (the “what” of an experience) and context signals (the “when” and “where”) simultaneously. The cells were not choosing sides. They were doing both things at once.
That finding lands squarely in the middle of a question neuroscience has been circling for years: Are the two kinds of long-term memory that textbooks have separated since the 1970s really housed in different brain circuits? A convergence of fMRI studies, network mapping, and now single-neuron data collected through May 2026 suggests the answer is no, or at least not in the clean way students have been taught.
The textbook version and why it stuck
The distinction traces back to the work of Endel Tulving, who in 1972 proposed that episodic memory (reliving your tenth birthday party) and semantic memory (knowing that Paris is the capital of France) are fundamentally different systems. The idea gained traction partly because of dramatic clinical cases. The Canadian patient known as K.C., who suffered hippocampal damage in a motorcycle accident, lost virtually all episodic memory but could still recall general facts. If the same circuits handled both, how could one type survive while the other was destroyed?
That logic shaped curricula for decades. But newer imaging technology has made it possible to look at what healthy brains actually do during retrieval, and the picture that emerges is far messier than two tidy boxes on a diagram.
Overlapping circuits, not separate modules
A selective meta-analysis published in 2017, comparing dozens of neuroimaging experiments, found that episodic and semantic retrieval rely on overlapping default-network regions, including medial prefrontal cortex, posterior cingulate, and lateral parietal areas. The differences that did surface were concentrated mainly in the hippocampus and parahippocampal cortex, which showed somewhat greater engagement during episodic recall. But the dominant pattern across the brain was shared activation, not separation.
A separate fMRI and behavioral study, published in eLife, tested a category that sits between the two traditional types: personal semantic knowledge, meaning facts about your own life (“I grew up in Ohio”) that carry autobiographical weight without replaying a specific scene. The researchers found substantial neural overlap between personal semantics and full autobiographical retrieval, suggesting the boundary between knowing and reliving is blurry even within a single person’s memory.
Network-level mapping has reinforced the point. Researchers charting cortico-hippocampal connections in the human brain found that these circuits overlap extensively with the default mode network, the large-scale system linked to internal thought, self-referential processing, and memory retrieval. Because both fact recall and experiential reliving draw on the same distributed architecture, the traditional diagram of two cleanly separated memory modules does not match what the brain actually does.
The single-neuron recordings from the Nature study sharpen the case further. Content signals and context signals mixed within the same cells rather than staying segregated in separate populations. That result is difficult to reconcile with any model that places facts and episodes in strictly different circuits.
Where the overlap breaks down
The convergence is large, but it is not total, and the remaining differences matter. Research published in Nature Communications found that some network elements preferentially track semantic content while others encode time-linked context. The open question is whether those partial dissociations reflect genuinely separate memory systems or simply different processing demands within a single, flexible system.
Intracranial EEG recordings from neurosurgical patients have shown measurable interactions between episodic and semantic neural signals during retrieval, as reported in a study using multivariate decoding of depth electrode data. These results capture timing dynamics that fMRI cannot, but the patient populations are small and drawn from clinical cohorts undergoing epilepsy monitoring, which limits how broadly the findings can be generalized.
There is also the question of what happens to memories over time. An fMRI study led by Lifanov and colleagues, published in Nature Communications in 2023, found that hippocampal reinstatement becomes less perceptually specific and more gist-like as memories age. That suggests the brain actively converts episodic detail into semantic-like summaries. But no longitudinal follow-up has tracked how this process unfolds across months or years in the same individuals, so it remains unclear whether the shift is gradual, stepwise, or dependent on how often a memory is revisited.
And then there is the clinical evidence that originally motivated the two-system model. Patients like K.C. really did lose episodic memory while retaining semantic knowledge. One possibility is that the hippocampus acts as a hub whose damage disrupts the contextual binding that makes a memory feel like reliving, while leaving the broader default-network scaffolding intact enough to support factual retrieval. That would be consistent with overlapping circuits that have partially specialized subcomponents, rather than with two independent systems.
What the methods can and cannot show
The strongest claims in this body of work rest on primary imaging and electrophysiology data collected directly from human participants. The 2017 meta-analysis synthesized results across many independent experiments, giving it broad statistical weight. The single-neuron recordings carry a different kind of authority: they reveal what individual brain cells do, not just which regions show increased blood flow on a scanner. Together, these two evidence types are harder to dismiss than either would be alone.
At the same time, the methods come with familiar caveats. fMRI measures blood-oxygen-level changes, not neural firing directly, and its spatial smoothing can blur subtle distinctions between nearby subregions. Intracranial recordings offer exquisite temporal precision but are limited to clinical samples and restricted electrode placements. Meta-analyses depend on how studies are selected and how tasks are sorted into “episodic” versus “semantic” bins, categories that the new data themselves call into question. Raw single-neuron spike-timing datasets from the Nature study have not been publicly linked for independent re-analysis as of June 2026, though Nature’s data-availability policies may lead to their release.
Contextual reviews, such as a recent synthesis of the default mode network’s role in memory and internal mentation published in Current Opinion in Behavioral Sciences, help frame the findings but do not generate new data. Readers should treat them as guides to interpretation rather than as primary evidence, especially where they speculate beyond what the experiments directly show.
Why it matters beyond the lab
If episodic and semantic memory share most of their neural machinery, the implications ripple outward. In Alzheimer’s research, the disease’s early assault on the hippocampus has long been framed as an attack on episodic memory specifically. A shared-circuit model raises the possibility that semantic deficits in early Alzheimer’s are underdiagnosed because clinicians are not looking for them with the right tests. In education, the assumption that rote fact-learning and experiential learning engage separate brain systems has influenced curriculum design; overlapping circuits suggest the two approaches may reinforce each other more than previously thought.
One hypothesis that emerges from the evidence is testable in principle: if the strength of cortico-hippocampal connectivity predicts how much semantic and episodic processing overlap, then individuals with tighter coupling in these networks should show more blending between factual recall and experiential reliving. People with weaker coupling might display clearer behavioral and neural separation. Longitudinal work could also examine whether changes in this connectivity over time track the transformation of rich episodes into sparser, gist-like knowledge.
For now, the safest conclusion is a middle-ground one. The classic textbook picture of two neatly segregated memory systems is not supported by current human neuroscience data. Episodic and semantic retrieval rely on a shared, distributed network that includes the hippocampus, medial prefrontal cortex, posterior cingulate, and lateral parietal regions, with modest biases rather than hard boundaries. Within that network, some subcircuits and temporal dynamics do appear to specialize more in contextual detail or in abstracted content.
As new tools make it possible to link single-neuron firing, mesoscale network activity, and behavior on the same trials, the field is moving beyond the episodic-semantic dichotomy toward models that treat memory as a continuum. Those models will need to explain not just where in the brain different kinds of information are stored, but how the same circuits flexibly shift between reliving the past and extracting the knowledge that persists after the details fade.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
