A Brain Orchestra … Playing at a Range of Unsynchronized Tempos

Patterns Recur, But Don’t Line Up; Transitions are Preserved 

“The core question that we wanted to answer is whether there is anything common across the various timescales in the brain,” Jun said. When they analyzed the data, they saw two primary phenomena. First, “there are commonalities between the EEG and fMRI,” Jun said. The same spatial patterns showed up no matter which timescale they looked at. The article refers to this as a “timescale-overarching spatial principle.”  

However, “when we overlaid the fMRI changes and the EEG changes, if we are indeed measuring the same thing, one of the spatial patterns that is shown at the fMRI, at some point, should also be shown in EEG … but that was not the case,” Jun said. In other words, there’s not a lot of overlap, in a given snapshot of time, between specific spatial connectivity states when looking across timescales. “At a single timepoint, state A, I will say, is active in the fMRI. But states B, C, D, and E are active in different timescales of EEG,” Jun elaborated. This draws a distinction from the alternative, where one might predict connectivity in the slow stream based on what is seen in the faster streams. The data actually appear more as though the timescales are totally disconnected. “They are working together, but in a different way,” Jun said. 

The second finding involved paying attention to the changes between these patterns and the probabilities that one state progressed to another or that a specific sequence of states showed up. “That changing pattern itself, the switching pattern, is also preserved across EEG and fMRI,” Jun said. In other words, if pattern A often precedes pattern B in one channel of the EEG, the same holds true in fMRI. These findings applied not only between EEG and fMRI but also across the five frequencies into which the EEG signal was divided. The paper refers to this as a “timescale-overarching temporal principle.” As such, the patterns are not in sync moment to moment, but the same repertoire of possible states and the same grammar of transitions are shared. In fact, all 126 possible brain connectivity states were seen, demonstrating the significant activity occurring in the resting-state brain. 

The Orchestra 

Sadaghiani provided helpful analogies to explain the underlying behavior. In one case, she likened the brain to a symphony orchestra and pointed out that within a violin section, which might be 30 violins, they don’t all play the exact same music. Instead, the first violins might be “communicating” or synchronizing with the French horns, while the second violins “talk” to the double bass, and each could be playing a distinct rhythm. “They might be contributing to multiple voices at the same time,” Sadaghiani said. “Some might be much more rapid than others.” That analogy also works, because we don’t need to seat faster violins in a special uptempo music section; likewise, it may be possible for subpopulations of any one brain domain, composed of millions of neurons, to distinguish themselves by playing a different rhythm. In fact, the data suggest that subsets are regularly playing at each of the six “rhythms” studied. No analogy is perfect, however. Most orchestras play one coherent tune, no matter how many parts. In contrast, the brain data collected by Jun and Sadaghiani appears more like similar pieces of music, or at least the same musical language, played at different tempos with no synchronization among the tempos. 

It is unclear from the analysis whether there are specialized brain regions that only work within individual timescales, or if switching timescales can occur. The physical brain architecture may provide clues. Sadaghiani pointed to the fibers that connect neurons. “The brain has both thick and thin ones, and myelinated and non-myelinated,” she said. “Think about it as electrical cables, some of which have really good insulation, and some that don’t … I think that speaks to some neurons preferably doing things really fast and others doing things preferably really slowly,” she said. “I think there is a reason for that … we just need different speeds.” The paper also notes that faster rhythms are more prominent in outer layers of the cortex, whereas slower rhythms tend to be deeper in the brain. 

The group confirmed their findings first with an independent and updated concurrent fMRI-EEG dataset—24 more individuals studied at the University of Illinois at Urbana-Champaign—and then with a much larger (443 individuals), independent, EEG-only dataset. 

Future Ideas 

Validating their results in the larger EEG-only dataset provides a nice advantage for future studies. “fMRI is not available to everyone,” Jun said, such as individuals with metal implants. Additionally, EEG is much more affordable. She estimated $50 for the same quantity of data one could acquire with $600-800 of fMRI. “We’re saying that this alternative is just as good or even better than the fMRI, so we can use this EEG with high reliability and confidence,” Jun said. 

The paper notes that “inter-individual variations in multi-timescale profiles could yield more sensitive markers of cognitive function and transdiagnostic risk than traditional single-timescale approaches,” meaning the tools could eventually find use in diagnostics across a range of conditions, though many studies would be needed to demonstrate diagnostic utility. 

There is an idea that timescale-resolved connectivity interrogation studies could provide a framework for biofeedback-based behavior modification, particularly in conditions such as anxiety, depression, PTSD, etc. “We would have this canonical sequence that we would say is desirable from the healthy controls, and we would train the clinical populations to move their brain pattern that way,” Jun explained. “I think it is giving us that clinical intervention possibility.” They are currently writing a grant aimed at this potential intervention. Other studies, she says, have compared “how the transition pattern or the sequence itself differs from healthy controls and neurodegenerative diseases like Parkinson’s,” primarily using fMRI, though EEG could be explored.  

While the current work used resting data, task-based connectivity is up next. “Our currently working manuscript is addressing whether these preserved temporal regularities that we defined in this PNAS paper are something that we can see changes between the rest and task, and we’re trying to answer that with different methodological approaches like a multifractality approach,” Jun said.  

It’s not yet clear what’s governing coalescence of these asynchronous patterns into unified cognition. However, Jun and Sadaghiani wrote a paper last year (The Journal of Neuroscience, DOI: 10.1523/JNEUROSCI.1939-24.2025) identifying polymorphisms in specific genes related to neurotransmitter proteins, which have effects on pattern residence time and transition probability. These neurotransmitters may at least regulate the brain landscape and its ability to transition between states. 

Sadaghiani had some closing thoughts. “One of the major conclusions we were trying to push with this paper is that when we think about understanding the basis of cognition and … issues that come up with cognition and different disorders, that we need multimodal measurements,” she said. “We can't just do the slow thing and then the fast thing, and then the other thing that listens to the individual neurons … if we do all of those things all the time separately, one at a time, we may just never be able to put the pieces back together.”