Memories are created in a matter of seconds. From the moment the brain receives a sensory input (i.e. sight, sound, smell, etc.), neurons across the brain activate. Connections formed between these neurons give rise to dynamic neural networks called engrams. For example, when exploring a new city, an engram forms and continuously updates as you walk down various streets and turn corners. The moment you finally encounter the landmark you have been searching for, there is a burst of neural activity. Neurons that were activated seconds prior also increase their firing. Your brain consolidates this information into a mental map of how to get to the landmark. Engram formation, therefore, depends not only on neurons firing simultaneously but also on those that activate immediately before and after. This is known as behavioral timescale learning. Researchers at the Max Planck Florida Institute for Neuroscience now may have begun to uncover how behavioral timescale learning integrates memories across several seconds.
When a neuron activates, an action potential is generated. First, an electrical or chemical input stimulates a dendritic branch on the neuron. If the stimulus is strong enough, a branch becomes activated. The signal travels through the cell body and into the neuron’s axon. The activated axon releases chemical messengers called neurotransmitters to activate other cells in the network. Neural activation lasts just two milliseconds before the cell resets to allow another action potential to be generated.
Generating action potentials is the basis of all brain activity. During learning, action potentials transmit signals that encode new experiences. A key region involved in this process is the hippocampus. Here, the brain consolidates short-term memories into long-term memory. Jain et. al, therefore, began their study by isolating hippocampal tissue samples from a group of mice. Using electrical stimulation, the team induced action potentials and measured changes that occur naturally during learning. To mimic behavioral timescale learning, Jain et. al separated each electrical input by approximately one second, representing the extended period for integration into an engram.
When a new memory is being formed, action potentials repeatedly fire from one neuron to another. Increased firing strengthens their connections, making them more efficient and more likely to fire together in the future. This process is driven by synaptic plasticity. Synapses are gaps between neurons that facilitate the transfer of information. When a neuron receives repeated inputs, its synapse undergoes structural changes that make it easier for the neuron to depolarize and produce an action potential. This type of synaptic plasticity is called long-term potentiation.
Long-term potentiation is crucial for forming long-term memories. Repeated stimulation from a presynaptic neuron (the neuron sending the signal) to a postsynaptic neuron (the neuron receiving the signal) triggers molecular changes. First, neurotransmitters released by the presynaptic neuron bind to receptors on the postsynaptic neuron. When neurotransmitters bind to these receptors, channels open that allow calcium ions to enter the neuron. The influx of calcium induces an increase of calmodulin-dependent protein kinase II (CaMKII) enzymes. These enzymes play a pivotal role in recruiting additional receptors to the synapse. More receptors lead to increased sensitivity and stronger synapses. As such, neurons that fire together become increasingly connected over time.
Canadian psychologist Donald Hebb first proposed the concept that “neurons that fire together wire together” in 1949. Activating just one neuron in an engram activates the entire network and triggers a memory. It was first thought that neurons must fire within milliseconds to wire together. Now, behavioral timescale learning suggests that a broader time window may exist, extending seconds or even minutes. This emerging theory revolutionizes our understanding of how neural connections influence our memory.
In their study, Jain et. al manipulated the time between sensory inputs to replicate behavioral timescale learning. Specialized biosensors enabled the team to observe microscopic changes at individual synapses within the hippocampus. Their first observation found that disrupting the function of CaMKII enzymes in the long-term potentiation pathway halted learning. This led researchers to speculate that behavioral timescale learning may be mediated by the activity of CaMKII enzymes.
However, even with highly sensitive biosensors, Jain et. al were unable to detect CaMKII activation during the simulation experiment. To their surprise, the activation of CaMKII appeared delayed, occurring tens of seconds after learning. In addition to influencing synaptic plasticity, the delayed activation also seemed to induce signaling changes in other parts of the neuron. Increased activation of these enzymes was especially apparent in dendrite branches, suggesting that this enzyme may play a role in processing information within dendrites receiving these inputs. These surprising findings reveal that CAMKII activation may be part of a larger mechanism that primes neurons to integrate repeated inputs.
These processes are so fast that they seem to be instantaneous. Behavioral timescale learning argues that these small delays, however, have tremendous implications for how our brains consolidate multiple information over time into a memory. Future studies are needed to fully uncover these underlying mechanisms. This will not only help us to better understand learning, but perhaps may reveal new approaches for preventing memory disorders, such as Alzheimer’s disease. The next time you forget something, remember—it could be a sign that our brains are working in ways we don’t fully understand yet..
