Mechanisms of Recovering Forgotten Memories

Post by Anastasia Sares

The takeaway

This study in fruit flies showed it is both possible to recover forgotten memories and to create false memories. However, true and false memories resulted from different processes in the brain. 

What's the science?

There has been considerable debate about the recovery of latent (or silent) memories in humans: events from the past that someone had previously forgotten. Along with that debate is the one surrounding false memories, which can be created through the power of suggestion. This week in Nature Neuroscience, Yang and colleagues investigated the neural mechanisms underlying the recovery of true and false memories in the humble fruit fly.

How did they do it?

The authors performed a conditioning task with flies, where they exposed them once to an odor paired with a mild shock, and once to another odor that was not paired with a shock. This is a common experiment, and we already know that after about 24 hours, the flies lose their memory of the pairing and return to normal behavior if they are not reminded about the experience. We can tell the flies have forgotten the association because if they are put in a container with two arms, one with the odor that was paired with the shock and the other which was not, they show no preference, spending equal amounts of time in each arm (see image). This forgotten association is a latent memory that could potentially be re-activated if it is still stored somewhere in the brain of the fly. The authors tested different conditions for trying to re-activate the memory: reminding the flies of the odor throughout the day (without shock), and varying things like texture of the enclosure or lighting along with those reminders. They also tested to see if there were ways for a fly to form a false memory, learning to avoid an odor that had never been paired with a shock.

The authors then used genetic techniques to selectively block the activity of certain brain cells in the fly (a fruit fly brain has around 150,000 neurons, which have been mapped and documented). With this, they hoped to understand which cells were involved in retrieving forgotten memories. Finally, they observed the brain of the fly (pinned to a microscope slide), as they presented it the same protocol (one smell paired with a shock and another smell not paired with a shock) and tested its responses later in the day. They used calcium imaging to observe where brain activity was happening in the fly.

What did they find?

If flies were reminded of the aversive odor in between the initial experience and testing, they were able to recover and retain the shock-odor association—but only if important parts of the original context were also replicated (same floor texture, lighting).  On the other hand, if the flies were reminded of the unpaired odor in between the initial learning and testing, they created a false memory, associating the unpaired odor with the shock instead, and preferring the odor that had originally been paired with the shock!

Contextual cues were necessary to help the flies retrieve aversive memories. When reminded of the texture or lighting where they had first experienced the shock, the flies were more likely to avoid the odor. When the flies were placed in a completely different environment (different texture/lighting), they acted neutrally. This indicates that, in order to recover lost memories, it might be necessary to re-create the original context of the memory in a multi-sensory way.

As for the cell-level interactions that occur to recover these memories, the authors found specific neurons that were necessary for the reactivation of memories, even though they weren’t involved in the initial learning. These neurons developed their connections even as the original memory was being forgotten. Stimulating these neurons allowed the flies to recover the aversive memory even in the absence of reminders, as well as during changes in the context (lighting, texture). However, stimulating these neurons did not lead to false memories, which meant that false memories are developed through an entirely different pathway.

It was a different set of neurons that turned out to be responsible for creating false memories (a similar class of neurons, but still distinct). Still, false memories were fragile and could be interfered with by changing the context. This means that even though it’s possible to form false memories, these can only form in contexts very similar to the true memory.

What's the impact?

This work shows that there are different brain mechanisms for retrieving silent memories versus creating false memories. It’s an important distinction, but of course this study was performed in fruit flies, so more work would be needed to understand these kinds of memories in other animals or humans.

Access the original scientific publication here. 

Neuronal Hyperactivity Recruits the Immune System to Eliminate Synapses

Post by Shalana Atwell

The takeaway

Neuronal hyperactivity in the adult brain can trigger local synapse loss by recruiting the immune protein complement component 1q (C1q), which initiates a cascade that flags and removes synapses at specific terminals. In mouse hippocampal circuits, C1q-dependent pruning requires nearby antibody-producing B cells that secrete immunoglobulins, especially IgM, to mark vulnerable synapses for removal.

What's the science?

Synapse loss is tightly linked to cognitive decline in diseases such as Alzheimer’s disease (AD). Prior work showed that microglia and C1q mediate synaptic pruning; however, the mechanisms by which specific synapses are tagged and whether adaptive immune cells play a role in this process have not been characterized. This week in Science, Crowley and colleagues asked whether changing neuronal activity is sufficient to drive C1q-dependent synapse loss in adult hippocampal circuits and how immune proteins contribute to activity-dependent pruning. 

How did they do it?

The authors used chemogenetic tools called DREADDs (Designer Receptor Exclusively Activated by Designer Drugs) to bidirectionally control activity in defined hippocampal pathways. They injected an excitatory DREADD virus into the medial entorhinal cortex to selectively activate perforant pathway neurons that project to the dentate gyrus, and an inhibitory DREADD virus into the same pathway in an Alzheimer’s mouse model (J20 amyloid model) to dampen its hyperactivity. For five days, animals received daily injections of the designer drug clozapine-N-oxide (CNO) to specifically activate these engineered receptors, and the authors confirmed efficacy through in vivo neural recording (Neuropixels) and immediate early genes (e.g. c-FOS) related to neuronal activity. To further validate their findings, they used other models of hyperexcitability (kainic acid and subthreshold pentylenetetrazole) that induce seizure-like network activity and then measured C1q deposition and synapse density.

To uncover molecular changes, they applied multiple forms of spatial transcriptomics (Visium and Xenium) to hippocampal sections, focusing on differentially expressed genes and cell types around the dentate gyrus in activated versus control hemispheres. These analyses unexpectedly highlighted immunoglobulin genes, so the authors used RNAscope, immunohistochemistry, and high-resolution microscopy to visualize B-lineage cells and different antibody isotypes in cells adjacent to hyperactive circuits. Finally, they used several transgenic and pharmacological tools targeting B cells and antibodies (e.g. IgM knockout mice) to test whether secreted, antigen-specific immunoglobulins are required for activity-dependent C1q deposition and synapse loss.

What did they find?

Inducing hyperactivity in the perforant pathway in healthy mice produced a focal increase in C1q deposition in the activated projection zone and targeted loss of specific presynaptic terminals (VGLUT2-positive). This pattern suggests that local circuit activity can spatially direct where C1q accumulates and which synapses are selectively removed, rather than causing global synaptic damage. Inhibiting perforant pathway hyperactivity in J20 amyloid mice reduced C1q accumulation, decreased amyloid-β immunoreactivity, and partially restored postsynaptic markers, supporting the relationship between hyperactivity and C1q deposition in disease. In mice lacking a functional C1q protein (C1qa knockout), chemogenetic activation increased c-Fos expression but no longer produced VGLUT2 terminal loss, demonstrating that C1q itself is required for this form of activity-dependent presynaptic pruning.

Spatial transcriptomics in activated wild-type brains revealed strongly upregulated immunoglobulin genes, indicating local antibody-secreting cells near the hyperactive pathway. RNAscope, spatial transcriptomics, and immunostaining identified B-lineage cells adjacent to activated hippocampal regions. Imaging revealed that IgM localized near C1q-positive VGLUT2 terminals, suggesting that locally produced IgM antibodies help recruit C1q to specific synapses during periods of increased activity. Functionally, mice lacking B cells or IgM, or with restricted B-cell receptor specificity (B cells all recognize the same kind of antigen), failed to show the typical activity-dependent C1q deposition and VGLUT2 loss after hyperactivity. This indicates that antigen-specific, secreted IgM from B-lineage cells is required to enable C1q-mediated synapse elimination in the perforant pathway.

What's the impact?

This study is the first to demonstrate that neuronal hyperactivity is sufficient to drive C1q-dependent synapse loss in the adult brain and that local B-lineage cells and their IgM antibodies are key enablers of this process. By revealing that adaptive immune cells can respond to circuit activity and tag synapses for C1q-mediated pruning, it broadens our understanding of how innate and adaptive immunity collaborate to sculpt adult neural circuits. Further, the work suggests that in disorders with early hippocampal hyperexcitability such as Alzheimer's disease, modulating neuronal activity or interrupting IgM-C1q interactions at synapses could be promising strategies to limit synapse loss and preserve cognitive function.

Access the original scientific publication here.

How Bilingual Brains Handle Grammar in Two Languages at Once

Post by Amanda Engstrom

The takeaway

The bilingual brain must compute different grammar rules for different languages at millisecond resolution. This computation relies on a shared brain network spanning the left frontal and temporal lobes, using flexible, reusable rules rather than a separate network for each language.

What's the science?

Human language requires the speaker to alter the pronunciation of words to account for changing information such as the plural form of an object, but the phonological rules for this are different across languages. This raises the question as to whether the brain utilizes language-specific neural systems or if it relies on a shared abstract mechanism that generalizes across linguistic contexts. Prior brain imaging work on this question has produced mixed results and has mostly examined language comprehension, leaving open whether the same holds true during the split-second demands of planning and producing speech. This week in Journal of Neuroscience, Chen and colleagues aimed to resolve this by using magnetoencephalography (MEG) to track millisecond-scale brain activity in Spanish/English bilinguals, using a task designed to separate the mental act of changing a word's grammatical form from other factors like the word's meaning or how it sounds, allowing them to pinpoint when and where this computation occurs in the brain as bilinguals produced singular and plural nouns in both languages.

How did they do it?

The authors recruited 23 highly proficient Spanish-English bilinguals to complete a phrase-completion task: participants saw a written noun (e.g., "boat"), then heard a spoken cue ("one," "two," or "say") prompting them to produce the singular form, plural form, or simply repeat the word aloud. The task varied language, whether grammatical change was needed, and whether that change altered the word's sound, all independently of one another.

While participants performed the task, the authors recorded brain activity using MEG, a non-invasive technique that detects magnetic fields from active neurons at a millisecond timescale, critical for capturing the split-second planning between hearing a cue and speaking. They also collected structural MRI scans and combined them with MEG data through source localization, which estimates where in the brain each signal originated using each participant's own anatomy. First, they compared brain activity when transforming versus simply repeating a word. Second, they trained a computer algorithm to recognize that activity pattern in one language (e.g., English) and then tested whether it could detect the same pattern in the other (Spanish). Successful detection across languages ("cross-decoding") would show that English and Spanish rely on a common neural code, rather than separate systems.

What did they find?

The authors found that transforming a word's grammatical form (e.g., "boat" to "boats") activated the left frontal and temporal regions much more strongly than simply repeating a word, even for words that don't change sound when pluralized (like "tuna/tuna"). This suggests that these brain regions carry out grammatical transformation as an abstract computation, rather than one tied to a specific sound pattern. Comparing English and Spanish separately, both languages activated nearly the same brain regions, with Spanish showing activity about 110 milliseconds earlier than English. A decoding algorithm trained to detect grammatical transformation in one language successfully detected it in the other, suggesting that English and Spanish rely on a shared brain mechanism for this computation rather than two separate systems, with the timing difference simply reflecting how quickly each language can be processed. Finally, this same pattern held even for words that don't exist in either language, and cognates (like "chocolate/chocolate") showed no added advantage over other real words. This suggests that the brain's grammar system is highly generalizable: it can apply grammatical rules to entirely novel words and doesn't depend on having a matching or overlapping vocabulary entry to work with.

What's the impact?

This study found that Spanish-English bilinguals engage in a shared left frontal-temporal network to process the grammatical transformation of pluralizing a word in both languages, suggesting the bilingual brain uses flexible, reusable rules rather than separate networks per language. These findings support a growing body of work showing that the brain favors efficient, generalizable abstract computations that can be reused across different contexts. Further, this study demonstrates how bilingualism can be used to interrogate general principles of neural organization. 

Access the original scientific publication here