Genetic basis of dynamic brain states reveals cellular and disease associations

Scientists used large-scale brain imaging and genetic data to show that the way our brain switches between different activity patterns is inherited. These patterns are linked to specific cell types like neurons and support cells. They are also connected to traits like education, sleep, and risks for diseases like Alzheimer's.

For decades, neuroscientists have relied on "static" measures of brain activity. These are essentially single snapshots of which brain regions are talking to each other. However, the brain is not a static circuit. It is a highly dynamic system that constantly shifts its configuration. Understanding these shifts is critical to understanding everything from healthy cognition to the breakdown of neural networks in disease.

While we know the brain moves through different functional states, the biological "why" behind individual differences has remained elusive. We knew that people spend different amounts of time in different brain states. We did not know if those differences were hardwired into our DNA. This study from the University of Cambridge closes that gap. It establishes dynamic brain states as measurable, heritable biological markers.

Moving beyond the brain snapshot

The standard approach to analyzing resting-state fMRI (functional magnetic resonance imaging) involves looking at functional connectivity (the temporal correlation between the activity of different brain regions). Traditionally, researchers used "sliding-window" methods. These take averages over fixed periods of time to see how connections change. The problem is that these windows are often arbitrary. If the window is too short, the signal is noisy. If it is too long, you wash out the very transitions you are trying to study.

Furthermore, static measures fail to capture the "fractional occupancy" (FO) of brain states. FO is the actual proportion of time an individual spends in a specific neural configuration. Without a way to quantify these temporal dynamics, it is difficult to determine if a person's unique brain activity pattern is a result of environment or fundamental biological programming. As seen in, the researchers needed a way to move from simple snapshots to a structured model of how the brain cycles through recurring, whole-brain configurations.

Decoding dynamics with Hidden Markov Models

To solve this, the authors implemented a Hidden Markov Model (HMM) framework. Think of an HMM as a way to uncover a hidden sequence of events (the brain states) by observing a visible stream of data (the fMRI signal). The model assumes the brain is in one of $k$ discrete states at any moment. Each state is defined by a specific pattern of activation and a unique covariance matrix (a mathematical description of how different regions coordinate their activity).

The mechanism operates in two primary layers: 1. The Observation Model: This defines the "signature" of each state. For each state, the model learns a mean activation vector and a connectivity pattern. This creates a template for what that state looks like. 2. The Transition Model: This governs the "switching" logic. It calculates the probability of moving from one state to another. This allows the model to capture the temporal flow of brain activity.

By applying this to data from 52,335 participants, the authors segmented the brain's activity into 12 distinct, recurrent states . This allowed them to transform complex, continuous fMRI time series into a concise set of quantitative phenotypes: the fractional occupancy of each state.

Linking genes to neural rhythms

The core of the study lies in a massive genome-wide association study (GWAS) to see if these occupancy levels were tied to specific genetic variants. The authors report that brain state dynamics are indeed significantly heritable. Using GCTA-GREML, they found heritability estimates ($h^2$) ranging from approximately 0.0355 to 0.1332. For context, State 9 showed the highest heritability at 0.1332 [Figure 2A]. This indicates that over 13% of the variation in this brain state can be attributed to common genetic factors.

Beyond mere inheritance, the paper identifies the specific biological machinery driving these dynamics. By integrating genetic data with single-nucleus RNA sequencing, the authors performed gene-set enrichment analysis. This helps pinpoint which cells are responsible for these patterns [Figure 2C]. The results show a heavy enrichment for specific neuronal populations. These include GABAergic and glutamatergic signaling (the primary inhibitory and excitatory neurotransmitter systems in the brain). They also found a contribution from oligodendrocytes (cells that create the insulating myelin sheath). This suggests that white matter plasticity helps coordinate large-scale network synchronization.

The study also demonstrates that these states are not isolated. They share a "pleiotropic" architecture, meaning a single genetic locus can influence multiple brain states simultaneously. For instance, the authors report strong colocalization at the APOE gene cluster between States 1 and 4 .

They also found strong signals at the PAX8 locus across many states [Extended Data Fig. 4].

Assessing the causal chain

One of the most ambitious parts of the work is determining whether these brain states are merely correlated with life traits or if they sit in a causal chain. The authors used Mendelian randomization (MR)—a method that uses genetic variants as "natural experiments" to infer directionality—to test this.

The paper finds that educational attainment and sleep duration exert causal effects on specific brain states [Figure 4C]. This suggests that the way our brains cycle through activity patterns is shaped by these factors. However, the authors note a nuance in the sleep data. Their MR-Egger analysis indicated significant pleiotropy for sleep duration. This implies that the genetic effects on sleep might involve multiple, diverse biological pathways rather than a single, clean mechanism. Furthermore, the authors report significant genetic correlations between brain state occupancy and several clinical phenotypes, including neurodegenerative diseases and personality traits like extraversion [Figure 4A].

Limits of the dynamical map

Despite the breadth of the study, there are important boundaries to consider. First, while the authors successfully replicated some findings in East and South Asian subsamples, they note that these groups had small sample sizes. This limits the statistical power for robust cross-ancestry conclusions [Extended Data Fig. 8].

Second, the connection to disease is still emerging. The authors report that while genetic correlations with neurological conditions were large, they did not survive rigorous False Discovery Rate (FDR) correction. This means the statistical signal is present but not yet strong enough to confirm a direct link to clinical disease. Finally, the causal inferences for sleep duration were complicated by the aforementioned pleiotropy.

Verdict: A new intermediate phenotype

Is this ready for the clinic? Not yet. But as a tool for research, it is a major advancement. The authors have effectively moved brain dynamics from a descriptive curiosity to a biologically grounded "intermediate phenotype." This acts as a measurable bridge between our DNA and our clinical health.

If you are building models to predict disease risk or cognitive decline, the fractional occupancy of HMM-derived states is a high-value feature. It captures temporal complexity that static connectivity misses. It is also rooted in the very cells that drive brain function. The analysis pipeline is available at https://github.com/amir-ebneabbasi/Brain-dynamics-GWAS.

Figures from the paper