A New Model of Synaptic Plasticity: Neurons Depend on Their Neighbours for Learning

Post by Laura Maile

The takeaway

Plasticity, or strengthening of synapses over time, is dependent on a complex interaction involving both excitatory and inhibitory inputs from a network of nearby neurons. When we learn, we rely not on single inputs and outputs, but on dynamic communication between networks of neurons.  

What's the science?

Synaptic plasticity is the collection of changes to both excitatory and inhibitory connections between neurons that occur when we learn. Historically, the understanding has been that this plasticity functions at the single synapse level, relying on the activity of a single presynaptic neuron and the response of its partner across the synapse. Hebb’s theory of learning states that when a presynaptic neuron repeatedly fires and activates a neighboring neuron, their connection is strengthened. More recent evidence has shown that learning and plasticity are more complex, integrating both excitatory and inhibitory information from other neighboring synapses, dependent on the circuitry of nearby networks. Scientists have not yet agreed upon a framework to explain this concept of interdependent synaptic plasticity in biological models. This week in Nature Neuroscience, Agnes and colleagues describe a new model of synaptic plasticity that relies on the activity of multiple neighboring synapses.  

How did they do it?

The authors created a theoretical model consisting of a set of rules integrating the timing, strength, distance, and identity of excitatory and inhibitory inputs to describe the interaction between sets of neurons during learning. They first utilized two excitatory neurons isolated from other inputs and presented different stimulation patterns to the pre and post-synaptic neurons. To increase the system's complexity, they introduced neighboring synapses to the same excitatory postsynaptic neuron. Next, to determine the influence of multiple neighboring inputs, they modeled a single postsynaptic neuron with several presynaptic inputs spaced apart uniformly. The authors then sought to understand how synaptic plasticity influences the receptive fields of neurons, or their ability to respond to different stimuli. To do this, they simulated a neuron receiving eight different inputs, to mimic the input of eight different frequencies of sound. They mimicked the learning period by using inhibitory inputs to gate or limit the time the postsynaptic neuron could be influenced by excitatory inputs. They next wanted to test their rules on a more spatially and structurally complex dendritic tree, mimicking the tree-like organization of synapses onto a single neuron. To do this, they connected dendritic compartments that could be independently activated, to a single neuron.  

What did they find?

Their model demonstrated that long-term potentiation (LTP), the process of synaptic strengthening upon repeated activation of a presynaptic neuron, can be initiated by the presynaptic neuron, and increases when the pre and postsynaptic neurons fire in synchrony. When neighboring synapses also increase their firing, the postsynaptic neuron showed increased LTP, in a fashion dependent on time and distance from the postsynaptic neuron. When multiple neighboring neurons targeted the same postsynaptic neuron at equal distances, their distance and ability to influence one another drove competition.

In experiments modeling receptive fields, the authors demonstrated that learning occurred when the gating inhibitory neurons were shut down, allowing the postsynaptic neuron to experience strong stimulation by excitatory inputs. With dendritic tree modeling, they showed that the plasticity of the excitatory synapses was dependent on inhibitory gating, distance from the cell body, and co-activity of surrounding inputs. Inhibition was found to directly influence excitatory synaptic plasticity. Inhibitory plasticity is slower than excitatory, and has strong control over excitatory plasticity, preventing too much change in excitatory weights and stabilizing learning. Finally, they created a model of a neuronal network, in which setpoints are used to balance LTP with synaptic weakening, creating a stable network that allows for learning without too much runaway excitation.  

What's the impact?

This study found that synaptic plasticity depends on a network of nearby synapses. The model developed can help explain how clusters of synapses develop and strengthen into stable systems. This work helps neuroscientists better represent and understand the complex dynamics of neural connections that change over time as we learn.  

Access the original scientific publication here.

Neuralink’s Brain Chip: Understanding Implantable Brain-Computer Interfaces

Post by Shahin Khodaei 

What is Neuralink’s brain chip?

In January 2024, Neuralink, the neurotechnology company owned by Elon Musk, implanted a “brain chip” in a human for the first time. This human is a man named Noland Arbaugh, who has tetraplegia (paralysis in both arms and both legs) due to a spinal cord injury. Neuralink’s coin-sized device was inserted into his skull with the help of a surgical robot, with microscopic wires implanted into the brain tissue to record neural activity. Information from the device is then wirelessly transmitted to a receiving unit for processing. Two months in, the device has given Mr. Arbaugh the ability to use his brain activity to move a computer cursor with enough dexterity to play online chess and the video game Civilization VI.

The Neuralink device is an example of brain-computer interface (BCI) technology. In short, BCIs allow direct communication between the central nervous system (CNS) and a computer. As a result, BCI technologies expand the natural outputs of the CNS, which would normally involve the use of muscles – for example, moving your tongue and mouth to speak, or using your hands to manipulate objects. With BCIs, an entirely new set of artificial outputs from the CNS becomes possible. By directly monitoring brain activity, these devices have been used to type, move a computer cursor, or operate a robotic arm.  

What are the components of a BCI system?

BCI systems consist of four key components: 1) signal acquisition, 2) feature extraction, 3) feature translation, and 4) device output. 

1) Signal acquisition: Any BCI system begins by measuring signals from the CNS using sensors. For implanted devices, the sensors are electrodes that are surgically placed under the skull, either on the surface of the brain or penetrating the brain tissue. Brain activity can also be measured in non-invasive ways – for example, by placing sensors on the scalp to measure electrical signals known as the electroencephalogram. In either case, such signals are small in size and therefore need to be amplified, and then converted into a digital format that can be processed by a computer.

2) Feature extraction: Signals recorded from the CNS are rich with information. However, not all the information within the signal is relevant for a particular BCI. In this component, the relevant features of the signal are extracted for further processing. The extracted feature should have a strong correlation with the user’s intentions. For an implanted device, the extracted feature is often the activity pattern of groups of neurons around the sensors.  

3) Feature translation: The extracted features are then given to a translation algorithm. This algorithm is designed to translate the relevant features of the signal into commands that reflect the user’s intent. For example, a specific activity pattern may be translated to “move the computer cursor upward”, and another pattern to “move the computer cursor downward”. In this way, the user’s goal is deduced from their brain activity.

4) Device output: The commands from the translation algorithm go on to operate an external device. Depending on the nature of the BCI system, the final output may be the movement of a computer cursor, the operation of a robotic arm, steering an electric wheelchair, etc.

In some instances, there may be a fifth component, where the BCI system delivers input back to the brain to modulate the CNS. This input may be delivered by directly applying electrical currents into the brain tissue through the implanted electrodes, or by non-invasive methods such as transcranial magnetic stimulation.

Progress in BCI technology

Major advances have been made since the first report of BCI technology in the 1960s when electrical signals recorded from the scalp were used to control a slide projector. The main goal of BCI research and development so far has been to assist people affected by stroke, spinal cord injury, or CNS disorders such as amyotrophic lateral sclerosis. The most common use of BCIs has been to replace natural CNS output that is lost to injury, which was reported as early as 2006. In that year, Hochberg et al. published a study in Nature reporting that a patient with tetraplegia (similar to Mr. Arbaugh) used an implanted BCI device to control both a computer cursor and a robotic arm.

Hochberg’s 2006 device implanted 96 electrodes into the participant’s brain. Nearly 20 years later, the BCI implant from Neuralink can measure brain activity using more than 3000 electrodes placed in brain tissue. This represents a significant improvement in the signal acquisition component of BCIs. With ongoing innovations in machine learning, the signals can be also processed in new ways for better feature extraction and translation. Together, these technologies will likely advance the capabilities of BCIs.

There have also been exciting uses for BCIs to not just replace, but to restore natural CNS output. In a 2023 study published in Nature, researchers developed a BCI system where the device output was electrical stimulation of the spinal cord. Specifically, the stimulation activated areas of the spinal cord that controlled muscles involved in walking. Using this strategy in a participant with tetraplegia, their device translated brain activity into leg movements to restore the participant’s ability to walk. 

BCIs also have the potential to either enhance or supplement the natural outputs of the CNS. In this way, researchers have explored how BCIs can help the general population. For example, a BCI can improve performance in tasks that require intense concentration by detecting brain activity that indicates loss of attention and playing a sound to restore concentration. 

Potential risks associated with BCI devices

Often, implanted BCI devices require an invasive and high-risk open brain surgery to place sensors into the brain. There is an unavoidable risk of damage to the brain area where the implant is placed, as well as possible complications such as infection, bleeding, and brain swelling. While these risks may be acceptable for users with severe disabilities who can greatly benefit from implanted BCIs, they discourage most individuals from getting an implant. It is worth mentioning that newer minimally invasive BCI devices are currently being developed. An example of this is the Stentrode sensor developed by Synchron. Instead of open brain surgery, the Stentrode is inserted into the brain through the jugular vein, using a minimally invasive endovascular surgery. 

If and when BCIs become more broadly used, there will be a growing risk to user’s privacy and safety. These devices are likely to measure brain activity in increasing detail as the hardware and software evolve. If the recordings of brain activity are not immediately discarded, it is critical that they are stored safely and privately. This is particularly important as the BCI field attracts more private companies, whose business interests may not align with the user’s expectation of data privacy. Additionally, the digital components of the BCI system, in particular the device output, may be vulnerable to threats such as hacking.  

Takeaway

Implanted BCI technologies show great potential in assisting individuals with neurological injury or disease. As the tools evolve to improve all components of BCI systems, BCIs could become an important technology to not only replace and restore function for people with disabilities but also to enhance and supplement performance in the general population. ​

References +

He et al. Brain–computer interfaces. 2020. Neural Engineering. Access the publication here.

Hochberg et al. Neuronal ensemble control of prosthetic devices by a human with tetraplegia. Nature. 2006. Access the publication here.

Lorach et al. Walking naturally after spinal cord injury using a brain-spine interface. Nature. 2023. Access the publication here.

Maiseli et al. Brain-computer interfaces: trend, challenges, and threats. Brain Informatics. 2023. Access the publication here.

Mitchell et al. Assessment of safety of a fully implanted endovascular brain-computer interface for severe paralysis in 4 patients: the Stentrode with thought-controlled digital switch (SWITCH) study. 2023. JAMA Neurology. Access the publication here.

Musk and Neuralink. An integrated brain-machine interface platform with thousands of channels. Journal of Medical Internet Research. 2019. Access the publication here.

Oi. Neuralink: Musk’s firm says first brain-chip patient plays online chess. 2024. BBC. Access the publication here.

Shih et al. Brain-computer interfaces in medicine. Mayo Clinic Proceedings. 2012. Access the publication here.

The Presence of an Onscreen Instructor Improves Learning

Post by Natalia Ladyka-Wojcik

The takeaway

Students show better learning outcomes and greater synchrony of brain activity, despite lower visual attention to material, when an onscreen instructor is present during video lectures of multimedia information.  

What's the science?

The COVID-19 pandemic prompted a significant shift towards online learning using multimedia instructional videos — a shift that persists in educational settings today. The multimedia learning framework suggests that combining pictures and words in instructional videos can create a more interactive learning context which in turn boosts motivation and attention among students. Within this framework, previous research supports the principle of embodiment: that human facial expressions and voice are critical for students’ socio-emotional responses and sustained attention during learning. However, it remains unclear whether the onscreen image of the instructor itself called the “image principle”, actually engages students’ cognitive processing of the information to be learned. This week in PNAS, Ping Li and colleagues investigated whether seeing the instructor on-screen during online learning promotes learning and attention, or instead serves as a distractor.

How did they do it?

The authors recruited over 70 college-aged participants to watch video lectures of novel information in biology, psychology, and meteorology while undergoing functional magnetic resonance imaging (fMRI). Their eye movements were tracked as they looked around the screen. Critically, participants were excluded if they had prior expertise in one of the topics presented, thus ensuring that the material to be learned was introduced during the study. Participants were shown video lectures with either an on-screen human instructor, an animated instructor, or no on-screen instructor. Finally, the participants completed a post-learning assessment to determine how much information they had learned from the lecture videos, in addition to a series of tests of their cognitive and socio-emotional abilities.

Using the fMRI and eye movement data collected, the authors studied neural synchrony; a correlational measure of brain activity over time amongst a group of people sharing a particular context (in this case, the presence of an instructor during video lecture watching). The authors also used eye movement fixations to the on-screen video lectures as a measure of attention during learning.

What did they find?

In support of the image principle of multimedia learning, the authors found that students learned better with an on-screen instructor than without. However, a caveat is that learning was not affected by whether the on-screen instructor was human or animated, which suggests that the embodiment type of the instructor does not affect learning performance. Curiously, the authors found that students were more attentive to the instructional slides – as measured by eye movements – without a human instructor onscreen but this did not improve learning performance, reflecting a trade-off between visual distractors and socio-emotional processing.  

The authors also found higher neural synchrony between participants during learning when an instructor was present, compared to when no instructor was on-screen. This neural synchrony was most evident in regions of the brain involved in cognitive processing, as well as in socio-emotional processing. Importantly, greater eye movement and neural synchrony predicted better learning of the information, especially among participants with higher cognitive and socio-emotional abilities. Together, these findings strongly support the inclusion of an on-screen instructor during multimedia learning.

What's the impact?

This study found that the presence of on-screen instructors during online video learning benefits students’ comprehension of material, providing higher socio-emotional cues and outweighing the impacts of visual distraction. Although these benefits were found to vary across individuals, the study broadly highlights the importance of creating engaging and interactive learning opportunities, even in online educational settings.  

Access the original scientific publication here.