By understanding the signals hidden in the electrical chatter of the brain, scientists gain information about sleep, aging and other processes.
In January 2020, at a symposium on sleep, Janna Landner presented discoveries that can help us find the boundaries between wakefulness and unconsciousness in the activity of the human brain. For patients in a coma or under anesthesia, it is very important that doctors are able to make this distinction correctly. And this is much more complicated than it might seem - after all, the human brain in the phase of REM sleep produces the same familiar, smoothly pulsing waves as during wakefulness.
However, Landner argued that the answers to these questions lie not in ordinary brain waves - but in an aspect of brain activity that scientists usually ignore. This is a random background noise.
Some researchers were skeptical about this claim. "They said: you mean there is useful information hidden in the noise?" Says Landner, an anesthesiologist at the University Medical Center Tübingen in Germany, who recently received a postdoctoral fellowship at the University of California, Berkeley. "I said: yes, to whom the noise, and to whom - the signal."
Landner belongs to a growing group of neuroscientists who are inspired by the idea that noise in the electrical activity of the brain may hold clues to the mysteries of how it works. What was once thought to be the neurological equivalent of annoying television static can suddenly have a profound effect on how we study the brain.
Bradley Wojtek has heard from various skeptics that there is nothing about brain noise worth exploring. However, the results of his independent study of changes in this noise during the aging process, as well as information from the literature regarding statistical trends in irregular brain activity, convinced him that neuroscientists were missing something. And he spent several years helping scientists rethink the data they collected.
“It's not enough just to speak to a group of scientists with a statement: I think we are doing something wrong,” said Wojtek, associate professor of cognitive science and data science at the University of California, San Diego. “They need to be given a new working tool,” an improved one, or just another.
Bradley Wojtek
Together with neuroscientists at the Universities of California at San Diego and Berkeley, Wojtek developed a program that isolates periodic oscillations - such as alpha waves, which he has actively studied in waking and sleeping people - hiding in aperiodic brain activity. As a result, neuroscientists have a new tool that allows them to study both periodic waves and aperiodic brain activity in order to separate their roles in behavior, recognition and disease.
They call the phenomenon studied by Voytek and other scientists differently. Some call it "1 / f bias" or "scale-free activity." Wojtek promotes the name "aperiodic signal" or "aperiodic activity".
And this is not just some whim of the brain. The patterns that scientists are looking for are associated with a phenomenon that scientists began to find in various complex systems, generated by both nature and technology, in 1925. This statistical structure mysteriously manifests itself in so many different contexts that some scientists consider it to be one of the undiscovered laws of nature.
Although in the past 20 years, works have been published where scientists have looked for and described arrhythmic brain activity, none of them could understand what it exactly is. However, biologists today have the tools to better isolate aperiodic signals in new experiments, as well as to study old data more deeply. Thanks to Wojtek's algorithm and other methods, a whole galaxy of works has appeared in recent years, professing the idea of treasures of knowledge hidden in aperiodic activity, capable of revolutionizing the study of aging, sleep, child development, etc.
What is aperiodic activity?
Our bodies enjoy familiar rhythms of heartbeat and breathing - steady cycles necessary for survival. However, a rhythm is heard in the brain that seemingly has no regularities, but is just as important for life - and it can hide the keys to clues to behavior and consciousness.
When a neuron sends a signal to another neuron using a compound such as glutamate , the receiving end is more likely to be activated. This situation is called agitation. Conversely, when a neuron secretes a neurotransmitter such as gamma-aminobutyric acid, or GABA, the likelihood of activation of the host side is reduced - this is called inhibition, or suppression. Everything is good in moderation: too much excitement leads to seizures, too much suppression is characteristic of sleep, and in more serious cases, coma.
To study the delicate balance of arousal and suppression, scientists measure the electrical activity of the brain using electroencephalography, EEG. Cycles of excitement and suppression form waves, different forms of which are associated with different states of consciousness. For example, brain waves with a frequency of 8 to 12 Hz produce alpha waves associated with sleep.
But the brain's output is not a perfect smooth curve. Rising to highs and falling to lows, activity charts jump to and fro. Sometimes there is no regularity in the work of the brain at all, and it becomes more like an electrical noise. It does have a truly random component, white noise, but some components exhibit a more interesting statistical structure.
It is these imperfections that spoil the smoothness of the curve, as well as noise, that Wojtek and other scientists are interested in. “It is, of course, random, but there are different accidents,” he said.
Not all noises were created equal. In the above spectrograms, low frequencies are at the bottom, high frequencies are at the top. The brighter the color, the greater the intensity. On the left is white noise, whose signal intensity does not change with frequency. In the center is pink noise, 1 / f, whose intensity at high frequencies drops at a certain rate. The brown noise on the right has a much deeper intensity drop.
To quantify the aperiodic activity, the scientists split the raw EEG data much like a prism splits a sunbeam into a rainbow. They first applied Fourier analysis. Any graph of changes in data over time can be expressed in terms of the sum of trigonometric functions, which, in turn, can be expressed in terms of frequency and amplitude. Amplitude versus frequency can be plotted on a power spectrum plot.
The amplitudes of the power spectrum are usually placed in logarithmic coordinates, since they have a large spread. For completely random white noise, the power spectrum curve will be relatively flat and horizontal, with zero slope - after all, it is about the same at all frequencies. Brain activity data show curves with a negative slope, when at low frequencies the amplitudes are higher, and at high frequencies the intensity decreases exponentially. This form is called 1 / f, hinting at the inverse ratio of frequency and amplitude. Neuroscientists are interested in what the horizontalness or slope of this graph can tell about the processes taking place in the brain.
Analyzing the EEG in this way is like looking at a sound recording made on a railway bridge thrown over a highway, as he says Lawrence Ward , a cognitive neuroscientist at the University of British Columbia. The rumble of tires from randomly passing cars produces aperiodic background sounds, and the whistles of scheduled trains every 10 minutes will give a periodic signal with peaks exceeding the background noise in volume. Sudden single noises such as horn horns or car collisions will produce a noticeable burst of sound, contributing to the 1 / f slope.
Scientists have been familiar with this phenomenon since 1925, from workJohnson of Bell Telephone Laboratories, who studied vacuum tube noise. German scientist Hans Berger published the first human EEG just four years later. In the decades that followed, neuroscience became fascinated by the noticeable periodic waves present in brain activity. At the same time, fluctuations of the 1 / f type are found in all kinds of electrical noises, stock market activity, biological rhythms, and even in music - and no one knows why.
The aperiodic 1 / f brain activity (top) is converted into a set of waves of different frequencies (in the middle) using the Fourier transform, and then the power spectrum is plotted on the graph (bottom).
Perhaps because of the universality of this phenomenon, many biologists have dismissed the idea that useful signals can be extracted from the characteristics of 1 / f activity. They believed that scientific instruments could be the cause of this noise, as Biyu Khe , associate professor of neuroscience, neurobiology and physiology at New York University's Grossman School of Medicine, wrote in her 2014 review .
However, He and others have debunked these suspicions by experimenting with controlled noise from measuring instruments. This noise turned out to be much less than the activity of the brain. In a 2010 paper in Neuron magazine, He and colleagues also foundthat, although EEG plots, seismic waves of the earth's crust and fluctuations in stock prices show 1 / f trends, their higher-order statistical structures are different. This work challenged the idea that aperiodic signals are created by some unified law of nature.
The issue, however, has not yet been finally resolved. Ward has found mathematical similarities in various contexts, and believes that they must be based on something fundamentally common. Either way, Ward and He agree that it is worth getting deeper into brain sensing.
“For decades, brain activity in a 1 / f slope was considered unimportant and was often simply removed from analysis to emphasize periodic fluctuations,” He wrote in a 2014 paper. "However, there has been a growing body of evidence that non-periodic brain activity is a major contributor to brain function."
New signals from noise
Wojtek stumbled upon the topic of aperiodic signals almost by accident: first he wanted to build a model that removes white noise from the EEG. But, delving deeper into the jungle of code that works with data, he became interested in what they contained.
A 2015 study by Wojtek with his research supervisor Robert Knight , a professor of neuroscience at Berkeley, described how more aperiodic activity occurs in the brains of older people than in younger adults. Wojtek and Knight saw that white noise begins to dominate in the brain as we age. They also found a correlation between this noise and age-related memory impairment.
Wojtek wanted to give neuroscientists software that could automatically separate periodic and aperiodic features in data, incl. collected long ago, and help researchers find meaningful 1 / f trends. And he and the team wrote such a program.
The request for such a tool was immediately apparent. After publicationprograms on the site biorxiv.org on April 11, 2018, it was downloaded almost 2000 times in a month - quite a lot for a niche computational tool from the field of neuroscience. In November of that year, Wojtek made a presentation to the neuroscience community describing how to use this program. Due to its great popularity, he organized a seminar where, together with the team, he helped dozens of interested scientists to deal with the program. As a result of the workshop and the ensuing messaging, new collaborations began to form.
One of these has been linked to a study on the symptom of arousal during sleep, published byLandner in July 2020 in eLife magazine. Using this program, Landner and his colleagues found that in the aperiodic noise recorded in the EEG of subjects, high-frequency activity in the REM sleep phase fell faster than during wakefulness. In other words, the slope of the power spectrum was greater.
Spectrogram of brain activity during sleep. White graph tracks changes in the slope of the spectrum
In their work, Landner et al. Argue that aperiodic signals can serve as a unique characteristic suitable for describing the state of human consciousness. Such a new marker could help in the administration of anesthesia and the treatment of people in coma.
Other publications using the Wojtek code include studies of the efficacy of medication for attention deficit disorder and differences in brain activity in autistic individuals by gender. The code was first published in the peer-reviewed journal Nature Neuroscience in November 2020. The work of the code was demonstrated on simulated data.
Natalie Shavoronkov , a postdoc at Wojtek's laboratory, usually studies periodic oscillations such as alpha waves. They are, in her words, "more beautiful than aperiodic signals." However, recently, turning to the study of the brain of babies and the electrical signals that characterize their cognitive development, she faced a problem: babies do not produce elegant alpha waves. How and when these waves begin to appear is an open question.
Using Wojtek's algorithm, she analyzed open EEG data from the brains of babies. In a new paper published in Developmental Cognitive Neuroscience, he and Wojtek described the significant changes they discovered in the first seven months of a baby's life. However, more research needs to be done to understand whether this activity is related to the involvement of children in tasks related to brain development, or simply arises from the increase in gray matter density.
Wojtek's code has spawned a lot of new research, but this is not the only example of aperiodic noise analysis. In 2015, when Nvidia's Haiguang Wen and the University of Michigan's Zhongming Liu were at Purdue University, they published another example of an approach to isolating periodic and aperiodic components of brain activity - irregular-resampling auto-spectral analysis (IRASA). And Biyu He worked on this issue even before both of these tools appeared - as did the recently departed neuroscientist Walter Freeman, whose work was inspired by Wojtek. This task, by the way, can be performed manually, although it will take much longer.
The importance of having a tool to make it easier for neuroscientists to analyze data is that the data itself is simply a collection of numbers collected over a period of time. By itself, the graph does not say anything about whether the brain is working properly or not.
“In neuroscience, interpretation is key. Based on this, we make decisions on treatment or drug development, ”said Wojtek. According to him, the huge amount of data accumulated in the literature can potentially generate new ideas after processing them in a new way. "We didn't process this data deeply enough."
What does it mean?
A major obstacle to the study of these aperiodic signals is that no one knows exactly what gives rise to them. More research is needed to clarify the contributions of various neurotransmitters, neural circuits, and the interactions of neuronal networks, says Sylvain Baile , professor of neuroscience and neurosurgery, biomedical technology and computer science at McGill University.
“The reasons and sources have not yet been determined,” Baile said. "However, research must be carried out in order to accumulate knowledge and observation."
One theory is that aperiodic signals reflect the delicate balance between arousal and suppression that the brain requires for healthy and vigorous activity. Too much excitement can overload the brain; too much suppression can put it to sleep.
Knight believes that such an explanation is not far from the truth. "I would not say that I am sure that this is due to changes in the ratio of arousal and suppression, but I think that this is the most likely explanation," - he said.
An alternative explanation is that aperiodic signals are a consequence of the physical organization of the brain.
Based on how 1 / f behavior manifests itself in other physical systems, Ward concludes that there is a certain structural-hierarchical system in the brain that generates aperiodic activity. This, for example, could be a consequence of how a huge number of neurons are grouped, then forming larger regions that work in unison.
This brain activity may be ideal for processing sensory data, as such data often exhibit 1 / f fluctuations. In a 2018 study published byin The Journal of Neuroscience, examines how the brain predicts sounds whose structure contains 1 / f, and how aperiodic activity participates in the processing and prediction of natural stimuli. It is not surprising that any music, from Bach to jazz, can also contain 1 / f features - after all, the human brain creates music.
Wojtek said that in order to test the hypotheses of the origin of aperiodic signals, it is necessary to carefully study the various types of neuronal activity. Neuroscientists can then try to link brain regions to general physiology to better understand which neural mechanisms generate specific patterns of activity, and to predict how aperiodic and periodic signals should look like in various brain disorders.
Wojtek also hopes to do more large-scale research by applying his code to existing datasets, which will bring previously unseen signals to light.
Landner and Knight are currently analyzing data from comatose patients at the University of Alabama to see if this brain activity correlates with coma development. They predict that when a person comes out of a coma, an increase in high-frequency brain activity will manifest itself in the form of a change in the slope of the 1 / f graph. The preliminary results look promising, she said.
For Baile, aperiodic signals from the brain are somewhat reminiscent of dark matter - an invisible frame of the Universe that interacts with normal matter only through gravity. We do not know what it consists of, and what its properties are, but it is present on the heavenly background, imperceptibly keeping the Milky Way from decaying.
Scientists have not yet figured out what causes these aperiodic signals, but they too may be a reflection of the vital auxiliary structure of the universe contained in our heads. Something mysterious can help distract our minds from half-asleep.