ABOVE: TLR9 activation through genomic DNA fragments formed during learning promote the formation of memory. ©istock, vitacopS

Cooperating neurons form the physical structure of a memory, but how these assemblies form and persist as long-term memories still stumps scientists. After an event, brain activity increases in the hippocampus, which consolidates information and holds short-term memories. However, long-term memories are stored in the cortex, which means that the two regions must be able to communicate.

That suggested to Jelena Radulovic, a neuroscientist at Albert Einstein College of Medicine, that the hippocampus must maintain activity to conduct this transfer. By exploring this idea, her team identified an immune process in hippocampal neurons that could help explain how memories are maintained. The team published the findings in Nature.1  

The researchers investigated fear-based memories in mice by exposing animals to an electric shock during a contextual fear conditioning task and studying the change in gene expression after 96 hours. “What surprised us there was that we saw still a lot of activity, but completely different than what we saw in the early time points,” Radulovic said.

About one week after fear exposure, the team observed increased inflammatory gene expression. Although they thought it was an artifact at first, closer inspection showed that these neurons expressed more genes for nucleic acid-sensing immune responses, specifically toll-like receptor 9 (TLR9), which recognizes extranuclear DNA. They also saw greater TLR9 protein abundance in neurons by confocal microscopy of hippocampal sections.

In neurons, both stress and activation lead to double stranded DNA (dsDNA) breaks predominantly in the mitochondria and genome, respectively.2,3 To determine the source of dsDNA, the team isolated and sequenced extranuclear DNA from neurons after fear conditioning and showed that these fragments belonged to genomic DNA. Additionally, immunofluorescent labeling of a histone variant that denotes dsDNA breaks confirmed that these breaks occurred specifically in neurons.

Fluorescent microscopic image showing a white and purple (DNA with histones) dot separate from blue (nuclei) areas.
DNA fragments associated with histone variant (white and purple) escape the nucleus in neurons after learning to promote memory
The Radulovic Lab

Prior studies indicated that activity-induced dsDNA breaks promoted early response genes but did not investigate later time points.4 To better understand the source and dynamics of these dsDNA fragments, the team analyzed these breaks over a time course.

One hour after conditioning, the researchers showed that channels formed in the nuclear membrane that permitted the dsDNA fragments to exit the nucleus near the endoplasmic membrane where TLR9 also resides; a small number of these channels remained visible through the 96-hour observation period. The team confirmed by using microscopy that TLR9 and dsDNA fragments associated together. Starting at six hours after fear conditioning, the dsDNA localized near the centrosome, where the team also identified a DNA damage repair enzyme.

To explore the function of dsDNA breaks and TLR9 activation further, the team generated mice with the TLR9 gene knocked out (TLR9 KO) specifically in neurons using adeno-associated viral delivery of Cre-recombinase. TLR9 KO animals exhibited impaired memory and learning after fear training compared to wild type mice.

While fear conditioning induced expression in genes related to ER proteins, vesicle transport, and interleukin-6 and TLR9, TLR9 KO animals failed to express these genes. Absence of TLR9 in neurons also reduced the localization of DNA repair proteins and the production of extracellular structures previously shown to be important for memory.5

Jacob Raber, a neuroscientist at the Oregon Health and Science University who was not involved with the study, said that the finding linking the immune system to the formation of memory was striking. “It drives home the idea that you need immune activation,” he said. “If you don't have any, it's not good. If you have too much, it's not good. If it's chronic, it's not good.”

Raber said that it would be important to see this mechanism explored in other learning modalities that don’t involve a negative stimulus. He also noted that it would also be interesting to explore the influence and effect on this pathway in the context of neurodegeneration, something that Radulovic’s team intends to study.

“We hope that this study will provide the mechanism for us to understand how we maintain memories over a long time,” Radulovic said, noting that there are many downstream pathways from the inflammatory mediators they observed that could aid in memory development, including the development of structures called perineuronal nets. “Now we just have to dig deeper in each of them and see what exactly they do,” she said.


  1. Jovasevic V, et al. Formation of memory assemblies through the DNA-sensing TLR9 pathway. Nature. 2024;628: 145-153
  2. Maatouk L, et al. TLR9 activation via microglial glucocorticoid receptors contributes to degeneration of midbrain dopamine neurons. Nature Commun. 2018;9:2450
  3. Crowe SL, et al. Rapid phosphorylation of histone H2A.X following ionotropic glutamate receptor activation. Eur J Neurosci. 2006;23(9):2351-2361
  4. Madabhushi R, et al. Activity-induced DNA breaks govern the expression of neuronal early-response genes. Cell. 2015;161(7):1592-1605
  5. Gogolla N, et al. Perineuronal nets protect fear memories from erasure. Science. 2009;325(5945):1258-1261.