More than a century ago, the anatomist Santiago Ramón y Cajal unveiled his incredibly detailed drawings of neurons. From these, he posited that neurons were the main building blocks of the brain and that memories might be encoded in the connections between neurons.1 Since then, neuroscientists have been trying to answer how neurons and the networks they weave capture something so complex, so illusory, as a memory.

Our three-pound brains are home to nearly 100 billion neurons (almost equivalent to the number of stars in the milky way) that communicate across short and long distances, shaping our behavior and identities. Intercellular communication is not unique to neurons, but the sheer scale of their interactions distinguish neurons from other cells. “Neurons are in the business of interacting and connecting,” said Erin Schuman, a neuroscientist at the Max Planck Institute for Brain Research. 

Neurons are in the business of interacting and connecting.

 —Erin Schuman, Max Planck Institute for Brain Research

Unlike most cells, neurons expand beyond their spherical cell bodies into a complex labyrinth of protrusions that evoke visions of naked trees in winter. Axons and dendrites sprout from the cell body, branching out in different directions in search of connections. Dendritic branches are densely populated with even smaller budding protrusions called spines, which make synapses with other neurons to receive important messages. Researchers have demonstrated how, at its most basic level, memory is modulated by strengthening weakening synaptic connections.2

A single neuron can receive messages through several thousand independent synapses spread across the dendritic arbor. “We estimate that around 80% of the volume of a neuron is not in the cell body but rather in the dendrites and the axons,” said Schuman.

When Schuman started studying synapses in the 1980s and 1990s, she was curious how neurons, with their tiny cell bodies and expansive volumes, maintained thousands of independent connections, many of which were far from the cell body.

          Photos of Erin Shuman (top) and Kelsey Martin (bottom)
In the 1990s, pivotal studies from Erin Schuman (top) and Kelsey Martin (bottom) demonstrated that local translation in dendrites was important for synaptic plasticity.
(top) Gilles Laurant; (bottom) University of California Los Angeles

A heretical hypothesis 

In the 1960s, scientists studying memory formation in goldfish and rodents demonstrated that long-term but not short-term storage of memories requires new protein production.3,4 Shortly after, scientists identified and characterized a cellular analogue of long-term memory: long-term potentiation (LTP), the persistent strengthening of a synaptic connection, and therefore communication, between two neurons. LTP results from repeated stimulation and induces new protein production and new synaptic connections. However, the mechanisms by which the cell body communicates with its distal compartments to regulate gene expression remained an area of contention until the late 1990s.

Eric Kandel, a neuroscientist at Columbia University, found that squirting a drug that blocks protein synthesis onto neurons from the sea slug Aplysia did not prevent a short-term spike in activity in response to a chemical stimulus measured in electrophysiological recordings.4 This activity lasted for an hour before gradually decaying back to baseline. These findings suggested that activated synapses could use proteins already lying around while waiting an hour or so for new proteins to arrive from the cell body.

However, things did not quite add up. Synaptic construction sites can be hundreds of microns away from protein factories in the cell body, and the time it would take for a message to reach these factories, produce new proteins, and ship them back out did not align with the rapid demands that synapses place on the neuron. 

In 1982, Oswald Steward and William Levy, then neuroscientists at the University of Virginia School of Medicine, published a study proposing that it was inefficient for the neuron to produce proteins in bulk when there is no clear demand. Instead, they argued that it made more sense for protein synthesis to be regulated locally based on need.5 In support of their hypothesis, Steward and Levy provided electron microscopy data showing clusters of ribosomes called polyribosomes residing far away from the cell body, floating around the base of spines. In 1988, researchers identified the first mRNA in dendrites, transcripts for the microtubule-associated protein (MAP2), soon followed by the discovery of mRNA encoding the protein kinase CaMKII.6

Although these molecular breadcrumbs suggested local protein synthesis, with little other evidence to support Steward and Levy’s heretical hypothesis, the field largely adhered to dogma. “People thought that it was crazy,” said Schuman. “They thought that protein synthesis happened in the cell body and that all the proteins that were needed in the axons and dendrites were trafficked out from the cell body.” Scientists explained away observational studies like Steward and Levy’s as the result of a leaky nucleus—ribosomes and mRNA that, by chance, found themselves lost and floating around in cytoplasmic space. 

When Schuman started her lab at the California Institute of Technology in the 1990s, she was studying synaptic plasticity—the strengthening or weakening of synapses – in rat brain slices. She was curious if brain-derived neurotrophic factor (BDNF), a chemical messenger she knew to be important for the growth and survival of developing neurons, also had a functional role in mature neurons. To her surprise, BDNF dramatically increased synaptic activity for hours, similar to what was seen with LTP, but activity was completely blocked when she bathed the slices in a protein synthesis inhibitor.7 Unlike LTP, BDNF-induced activity required immediate access to new proteins.

For these experiments, Schuman used slices of the hippocampus, a brain region deeply entrenched in learning and memory. Electrophysiologists turn to the hippocampus for complex questions because of its unique architecture. The hippocampus is organized with cell bodies in one layer that send their axons and dendrites to other layers. Instead of a spaghetti nest of neurons, they see a beautifully segmented view of the cell. Schuman cut right through these projections, severing distant compartments from their cell bodies. When she bathed the disconnected dendrites in BDNF, they still exhibited synaptic plasticity, and therefore immediate protein synthesis, despite being physically separated from the protein factories.8 Rather than making use of premade proteins until reinforcements arrive, as was suggested with LTP, these findings suggested that synapses could produce their own proteins on demand.

Schuman’s experiments provided a functional role for the floating space debris in neurons and jump started a new frontier in neuroscience: local translation. Following Schuman’s findings, others showed that various forms of synaptic plasticity also require local translation.9 Given that synaptic plasticity is considered the cellular analogue of memory, this implied that local translation may also be fundamental for memory storage in the brain. The work of Schuman and others highlighted how neurons could synthesize proteins locally, but many questions regarding the specificity and underlying mechanisms remained. 

Sluggish memory and synaptic specificity

Kelsey Martin, a neuroscientist at the University of California, Los Angeles, joined Eric Kandel’s lab at Columbia University in the early 1990s as a postdoctoral researcher. Martin was intrigued by the neuron’s incredible polarity. “I became really interested in how a single neuron that has a single nucleus with thousands of different compartments regulates gene expression in a way that you get synapse specificity,” said Martin.

To answer this complex question, Martin turned to the relatively simple Aplysia. Kandel and his colleagues had previously characterized the Aplysia’s gill-withdrawal reflex wherein touching the siphon induces a reflexive retraction of the siphon and the gill.4 The group demonstrated how repetition of a stimulus transformed the short-term manifestation of this response into a longer-lasting memory and impacted behavioral response to subsequent stimulation. Similar to what behaviorists had demonstrated in more complex models of vertebrate learning, long-term but not short-term memory in the slug required the synthesis of new proteins. “That behavior was seen as a really tractable preparation to be able to ask, ‘how do you understand the difference between short and long-term memories?’” said Martin.

To study the molecular mechanisms driving this behavior, Martin took an even more reductionist approach. Kandel and his colleagues worked out the relatively simple neural circuitry driving the gill response, namely sensory neurons from the siphon that connect to motor neurons in the gill. In a dish, Martin cultured a single sensory neuron with a bifurcated axon that formed synapses with two separate motor neurons. This preparation allowed Martin to test whether individual synapses could be modified independently or whether modifications occurred cell wide. 

When Martin applied five spaced squirts of the neurotransmitter serotonin onto one of the sensory neuron synapses—targeted using a tiny electrode—it induced long-lasting synaptic activity and also increased protein synthesis specifically at the stimulated synapse.10 Martin observed the same effects when she deprived the synapse of its cell body. These findings corroborated Schuman’s results and suggested that local translation occurs in a spatially restricted manner, specific to the site of activity.

A tsunami of evidence in the late 1990s and early 2000s made it clear that neurons could engage in local translation, but it was still unclear how neurons achieved such speed and precision.


          Infographic showing the localization of mRNA from the neuronal cell body to distant spines.

3’ UTRs encode information about mRNA localization, translation efficiency, and stability (1). mRNA is packaged with translational machinery into transport granules, which hitch a ride on motor proteins that drive along microtubule highways (2).

          Infographic showing the localization of mRNA from the neuronal cell body to distant spines.
© ikumi kayama, studio kayama

Following stimulation, cruising transport granules arrive at synapses and unload their contents. The mRNA anchors to the neuronal spine and undergoes translation (3). With repeated stimulation, synapses undergo synaptic plasticity whereby protein accumulation leads to spine head enlargement and strengthened connectivity (4). 

See full infographic: WEB | PDF

Conserved but unique

Although synaptic plasticity requires local protein translation, the extent of RNA localization was unclear. Throughout the 1990s and early 2000s, scientists catalogued several individual mRNAs located in distal compartments using in situ hybridization.6 “At the time, we didn’t have the capacity to fully explore the transcript space,” said Schuman. The commercialization of next generation sequencing technologies in the early 2000s changed that.

Going back to the hippocampus, Schuman and her team analyzed mRNA content using deep sequencing.11 Prior estimates that approximately 100 mRNA strands resided in axons and dendrites were not just a little off; they were way off. Schuman and her team identified around 2,550 mRNAs, many of which encoded for structural elements, receptors, and signaling proteins—the building blocks of a synapse. 

RNA localization beyond the nucleus and cell body is not unique to neurons. “It is conserved from bacteria to us,” said Young Yoon, a neurobiologist at Albert Einstein College of Medicine. Although they are tiny single-cell organisms like bacteria and yeast exhibit RNA localization around the cell membrane or at the budding end in yeast.12  The first reports of asymmetric mRNA distribution came in 1983 when researchers studying early embryonic development in the sea squirt discovered mRNA encoding actin in the cytoplasm far from the nucleus amongst muscle-forming cells.12 Subsequent studies in frogs, drosophila, chickens, bacteria, and plant cells further supported a role for mRNA localization in embryogenesis as well as in fundamental cell processes like polarization, differentiation, and cell migration. 

However, compared to other cells, neurons pose a unique challenge to cell biology. With thousands of independently modified synapses, RNA localization and translation must be precise to maintain synaptic specificity while avoiding mislocalization or improper protein production. “In the neuron where you have such huge distances, I think the neuron has evolved a more elaborate model,” said Robert Singer, a cell biologist at Albert Einstein College of Medicine.

Intracellular shipping routes

By the end of the 1990s, it became increasingly apparent that Steward and Levy’s cytoplasmic space debris was not an anomaly but one of thousands of molecules participating in a wide-spread shipping network. Local translation makes sense from a logistics point of view. A single mRNA transcript can produce numerous protein copies, labor that the cell body can outsource to far away regions in the neuron. Furthermore, with protein building materials nearby, far away synapses can rapidly respond to incoming information and infrastructure demands without waiting for the cell body to complete their supplies order. 

The rules by which mRNAs leave the cell body and traverse the neuron is still an active area of research. The 3’ and 5’ untranslated regions (UTR) of mRNA encode information that affect the molecule’s stability, localization, and translation efficiency.6 The 3’ UTR serves as a kind of shipping label that contains information about its end destination, which is used by different molecules like RNA binding proteins (RBP) to orchestrate localization and translational control. Using RNA-sequencing technologies, scientists are beginning to decode how differences in mRNA 3’ UTR sequences relate to an mRNAs’ destination. 6,13 

          Neurons labeled in pink on a black background. White dots depict the expression patterns of two mRNAs throughout the neurons.
Erin Schuman and her colleagues identified around 2550 mRNA molecules that localize to the dendrites and axons of neurons. These fluorescence in situ hybridization images show the expression patterns of Ca2+-calmodulin-dependent protein kinase (CamkIIa), which is highly enriched in dendrites, and histone H3-3B (His3), which is abundant in the cell body.
Fusco CM, et al. Nat Commun. 2021;12:6127T.23 CC-BY 4.0

Moving down the assembly line, components needed for translation, including mRNA, ribosomes, and regulatory proteins, get packed and loaded into RNA transport granules.12 These transport granules orchestrate the spatiotemporal transport and translation of mRNA and protect the transcripts from degradation. 

Delivery of the RNA granules is powered by motor proteins that move these packages along the neuron’s vast microtubule and actin cytoskeleton road network to reach far away compartments.12 Relatively little is known about what is under the hood of these delivery trucks and how they are manufactured. In a recent study, scientists found that the adaptor protein APC links mRNAs with G-rich 3’ UTRs to kinesin motor proteins.14 Alternatively, some mRNA take the economical route and hitchhike on organelles like endosomes and mitochondria that are also zooming along the microtubule roads.12 

Packed and ready to go, RNA granules hit the road in search of incoming orders from synapses.

Decentralizing genomic control

How experiences like a cherished holiday with a loved one, your child’s first steps, or a car accident get encoded and maintained as memories is fundamental to our identities. Over the course of hours, days, and years, as we form, recall, and reconsolidate these memories, messages, or the lack thereof, between neurons prompt signaling cascades that orchestrate local and global responses in the neuron. 

There is now plenty of evidence of translational machinery traveling along axons and dendrites, but how different messages and signaling cascades trigger transport granules to arrive at synapses and unload their cargo remains an open question.15 Scientists hope that by understanding the ongoings of specific mRNAs in individual synapses, we will get closer to understanding how memories are formed and maintained and how neurological diseases take shape.

Much of our understanding to date comes from stitching together findings from cell culture studies that track the lives of a handful of the known mRNA species. Methods for real-time high-resolution tracking in living cells shed further light on local translation dynamics. 

Returning to the bifurcated sea slug system, Martin and her team tested whether mRNA are selectively delivered to activated spines. When they activated a spine on one side, they observed a broad dispatching of mRNA encoding the neuropeptide sensorin, but the mRNA was only translated into protein at the site of activation.16 “One of the things that transcriptional regulation does is set the neuron in a state of readiness,” said Martin. This allows neurons to quickly respond to local cues to change their protein compositions. 

Singer’s team published a series of papers between 2014 and 2016 that provide a more detailed picture of the spatial and temporal kinetics of local translation.12 Using single-molecule fluorescence in situ hybridization (FISH), Singer and his team tracked mRNA encoding the cytoskeletal protein β-actin (ActB mRNA) in live neurons. Soon after activating a single spine, RNA granules containing ActB mRNA arrived at the scene and unloaded their contents.17 “They’re sort of cruising around in this quiescent state looking for action,” said Singer. 

In a subsequent study, Yoon developed a procedure to artificially induce action by shining a laser on a single synapse, which released pools of the excitatory neurotransmitter glutamate that binds to and activates a spine.18 Within a few minutes, the first ActB mRNA arrived and adhered to the base of the activated spine before undergoing translation.

Instead of tracking individual mRNA activity, Schuman and her team cast a wider net. In combination with RNA sequencing to characterize all mRNA, they used a technique called Ribo-seq to capture and sequence mRNA that are actively undergoing translation.19 While most proteins are synthesized in both the cell body and distal compartments, they found around 800 mRNA transcripts that exhibited more translation in axons and dendrites, suggesting that local translation contributes significantly to the local proteome. “This was another surprise for us because the thinking was that most of the mRNA are still primarily translated in the cell body, but actually, for a huge number of mRNA, their primary source of synthesis is local,” said Schuman. Understanding how translational control is regulated at synapses will be critical to understanding how memories are formed and maintained. 


Like vertebrates, sea slugs engage in simple forms of learning and memory, and they have provided valuable insights into synaptic plasticity. Researchers cultured a bifurcated sensory neuron with two motor neurons and found that repeated stimulation of a synapse on one of the motor neurons but not the other specifically strengthened that synapse.

          Infographic showing the selective strengthening of synapses that received stimulation.
© ikumi kayama, studio kayama
See full infographic: WEB | PDF

Reinforcing synapses through auto-renewal

Memories can last a lifetime, but mRNAs and proteins that support synapses—the physical manifestation of memories—have half-lives on the order of hours to days. Interested in how transient gene expression supports long-term changes in plasticity, Singer and his colleagues track single mRNAs from synthesis to decay in living neurons. 

Singer and Yoon are still tracking the life of the well-characterized ActB mRNA, but this is just one of thousands of mRNA that localize to synapses to support their formation and maintenance. Whether the spatial and temporal kinetics of ActB mRNA is broadly applicable to the local translation of all mRNA is an active area of investigation.12 However, it is becoming increasingly clear that local translation is not a one-size-fits-all mRNA mechanism. 

In a paper published earlier this year, Singer and his colleagues shed light on how specific mRNA are differentially regulated and why this is important for the maintenance of memories.20 In contrast to ActB mRNA packages that cruise around seeking out activity, mRNA encoding immediate early genes (IEGs) are activity-regulated, meaning that they are not transcribed until they are needed. Scientists have extensively characterized Arc as an important multifunctional IEG that regulates different forms of synaptic plasticity and is necessary for the storage of long-term memories.21 In fact, genetically modified mice that lack Arc are unable to form long-term memories. Furthermore, mRNA coding for IEG like Arc are very short-lived with a half-life of one hour. “We are trying to answer this question of how do you develop a long-term memory from a short-lived RNA,” said Sulagna Das, a neuroscientist at Albert Einstein College of Medicine and coauthor of the paper.

It’s like when you memorize a poem. You must go over and over it to strengthen it, strengthen your synapses, so that you can make a stable connection of that poem in your brain. 

 —Robert Singer, Albert Einstein College of Medicine

Using high-resolution imaging in cultured mouse neurons, Das and her colleagues tracked the life cycle of fluorescently-tagged Arc following stimulation.20 Similar to what others in the field have shown, a brief stimulus triggered transcription of Arc mRNA in the cell body and its export to the stimulated spine. However, the maintenance of that “memory” long after the stimulation ends requires additional cycles of transcription to replenish the short-lived Arc transcripts. Das and her colleagues discovered that following an initial transcriptional cycle, a portion of the locally translated Arc protein returns to the nucleus to reactivate Arc mRNA transcription, thus initiating a positive feedback loop. These cycles go on and on to strengthen the synaptic contact. 

“It’s like when you memorize a poem,” said Singer. “You must go over and over it to strengthen it, strengthen your synapses, so that you can make a stable connection of that poem in your brain.”

Neuroscientists have come a long way in understanding the mesmerizing molecular orchestration of local translation in neurons, but as one question gets answered, many more crop up. However, it is no longer a question of whether translation occurs in axons and dendrites but how local translation unfolds to shape health and disease.

Lost in translation 

Translational control drives many processes, including neuronal development, learning and memory, and axonal repair. To manage local translation across huge distances, the neuron has evolved an elaborate model with specialized packages and sophisticated road networks. With so many moving parts to this intricate system, it is easy to imagine how disruption at any point in the pipeline could have consequences for neuronal activities like memory formation and retrieval. 

Because local translation is required for synaptic plasticity, and synaptic plasticity is the cellular analogue for memory, local translation may be required for memory. At present, there is little evidence to support or refute this. This is largely because scientists do not yet have the tools that allow them to block protein synthesis locally without affecting protein synthesis in the cell body. 

In an earlier study from 2002, scientists removed the 3’ UTR region of CaMKIIα mRNA in mice, leading to a near complete reduction in dendritic expression of the transcript.22 Although these mice exhibited diminished performance in learning and memory tasks, results that are consistent with local translation of CaMKIIα being necessary for memory, the experimental design makes it impossible to determine whether the deficits result from an acute absence of CaMKIIα during the behavioral tests or from long-term knockdown since birth.

So far, breakthroughs in scientists’ understanding of local translation in neurons have come from experimental designs leveraging in vitro or ex vivo preparations. To really answer the question of how neurons encode memories and the contributions of local translation to this process, researchers need to get in vivo. In addition to causal experiments, valuable information will be derived from in vivo live tracking of multiple mRNA molecules as they are transcribed, transported, and translated during a learning and memory task. To watch all of these molecules in motion, “that is the ultimate frontier,” said Yoon.

Future studies focused on delineating general versus local contributions to gene regulation will shed light on the extent of the contribution of the local transcriptome, translatome, and proteome in health and disease. Insights gleaned from these experiments will be critical to begin to understand how memory and identity can get lost in translation.

There has been a big shift in scientists’ understanding of the neuron over the last 30 years that will impact how research unfolds over the coming decades. “When I look at this younger generation of scientists, it’s an accepted truth that there’s local translation,” said Martin. “That was not the case in the 1990s, and even early 2000s.” 


  1. Jones EG. Santiago Ramón y Cajal and the Croonian Lecture, March 1894. Trends Neurosci. 1994;17:190-192.
  2. Mayford M, et al. Synapses and memory storage. Cold Spring Harb Perspect Biol. 2012;4:a005751.
  3. Flexner JB, et al. Memory in mice is affected by intracerebral puromycin. Science. 1963;141:57-59.
  4. Kandel ER. The molecular biology of memory storage: A dialogue between genes and synapses. Science. 2001;294:1020-1038.5.    
  5. Steward O & Levy WB. Preferential localization of polyribosomes under the base of dendritic spines in granule cells of the dentate gyrus. J Neuro. 1982;2:284-291.
  6. Holt CE, et al. Local translation in neurons: Visualization and function. Nat Struct Mol Biol. 2019;26:557-566.
  7. Kang H & Schuman EM. Long-lasting neurotrophin-induced enhancement of synaptic transmission in the adult hippocampus. Science. 1995;267:1658-1662.
  8. Kang H & Schuman EM. A requirement for local protein synthesis in neurotrophic-induced hippocampal synaptic plasticity. Science. 1996;273:1402-1406.
  9. Sutton MA & Schuman EM. Dendritic protein synthesis, synaptic plasticity, and memory. Cell. 2006;127:49-58.
  10. Martin KC, et al. Synapse-specific, long-term facilitation of aplysia sensory to motor synapses: a function for local protein synthesis in memory storage. Cell. 1997; 91:927-938.
  11. Cajigas IJ, et al. The local transcriptome in the synaptic neuropil revealed by deep sequencing and high-resolution imaging. Neuron. 2012;74:453-466.
  12. Das S, et al. Intracellular mRNA transport and localized translation. Nat Rev Mol Cell Biol. 2021;22:483-504.
  13. Mendonsa S, et al. Massively parallel identification of mRNA localization elements in primary cortical neurons. Nat Neuro. 2023;26:394-405.
  14. Baumann SJ, et al. APC couples neuronal mRNAs to multiple kinesins, EB1, and shrinking microtubule ends for bidirectional mRNA motility. PNAS. 2022;119:e2211536119.
  15. Bourke AM, et al. De-centralizing the central dogma: mRNA translation in space and time. Mol Cell. 2023;83:452-468.
  16. Wang DO, et al. Synapse- and stimulus-specific local translation during long-term neuronal plasticity. Science. 2009;324:1536-1540.
  17. Buxbaum AR, et al. Single β-Actin mRNA detection in neurons reveals a mechanism for regulating its translatability. Science. 2014;343:419-422.
  18. Yoon YJ, et al. Glutamate-induced RNA localization and translation in neurons. PNAS. 2016;113:E6877-E6886.
  19. Glock C, et al. The translatome of neuronal cell bodies, dendrites, and axons. PNAS. 2021;118:e2113929118.
  20. Das S, et al. Maintenance of a short-lived proteins required for long-term memory involves cycles of transcription and local translation. Neuron. 2023;S0896-6273(23)00267-2.
  21. Plath N, et al. Arc/Arg3.1 is essential for the consolidation of synaptic plasticity and memories. Neuron. 2006;52:437-444.
  22. Miller S, et al. Disruption of dendritic translation of CaMKIIα impairs stabilization of synaptic plasticity and memory consolidation. Neuron. 2002;36:507-519.
  23. Fusco CM, et al. Neuronal ribosomes exhibit dynamic and context-dependent exchange of ribosomal proteins. Nat Commun. 2021;12:6127.