How Brain Networks Shift Over Your Lifetime

This new emerging research with novel insights, published in Nature Communications by Mousley and colleagues, maps how the brain’s structural network organization changes non-linearly from birth to 90 years of age. Using diffusion MRI and graph theory, the authors identify four topological turning points—around ages 9, 32, 66, and 83—that mark shifts in how different brain regions are wired together and how efficiently they can communicate.

Why does this matter for biofeedback and neurofeedback? Because every time we ask the brain to learn new regulatory patterns—whether by training heart-rate variability, breathing patterns, or EEG rhythms—we are interacting with a network that is itself evolving. Biofeedback and neurofeedback are methods that give people real-time information about their own physiology or brain activity (for example, heart-rate, skin conductance, or EEG), and reward movement in a desired direction so that the nervous system gradually self-trains better regulation. We often talk about “neuroplasticity” as if it were a single thing, but this paper shows that the shape of plasticity changes depending on where you are in the lifespan.

From infancy through childhood, the brain is busily pruning and strengthening connections; in adolescence and young adulthood, it becomes more globally efficient and “small-world”, supporting complex cognition. In midlife, efficiency gradually declines and networks become more segregated, while aging brings increased modularity and changes in the importance of specific nodes. This multiscale reorganization is the backdrop against which we design and interpret neurofeedback and biofeedback interventions.

In the sections that follow, we will briefly unpack what the authors did and what they found, then explore how these age-specific network landscapes might inform clinical practice—especially EEG-based neurofeedback.

Methods

Rather than follow a single cohort over time, the authors assembled a very large cross-sectional dataset: 4,216 diffusion MRI scans spanning infancy (birth) to 90 years. These came from nine major projects, including neonatal, child, adult, and aging cohorts. All scans were processed with a consistent pipeline to reconstruct white-matter pathways and generate individual structural connectomes—basically, 90-node brain networks in which each node is a cortical or subcortical region and each edge reflects the number and strength of connecting fibers.

Several key methodological steps are worth highlighting:

1. Tractography and parcellation

   • Fiber tracking was performed using generalized q-sampling imaging and deterministic tractography, with five million streamlines per brain and consistent tracking parameters.

   • All networks used an AAL90 atlas, but scaled variants were applied for neonates and toddlers so that the same set of regions could be compared across dramatic brain growth in early life.
     

2. Harmonisation across datasets

   • Because scans came from different sites and protocols, the team used a “double ComBat” harmonisation: first across atlases, then across datasets. This aims to remove site-specific biases while preserving variance linked to age, sex, and neurodivergence status.
     

3. Controlling for density vs. analysing connectivity

   • For connectivity itself (how many edges and how strong), they used variable-density networks that reflected age-related changes in raw density.

   • For comparing topology (the pattern of connections) across ages, they created density-controlled networks: each brain was thresholded to exactly 10% density and edge weights were rescaled to 0–1. This avoids the trivial effect that “more edges usually means more efficiency” when comparing different ages.
     

4. Graph-theoretical measures

   • They computed 12 metrics spanning:

     • Integration (global efficiency, characteristic path length, small-worldness, average strength)

     • Segregation (modularity, core/periphery structure, k-core, s-core, local efficiency, clustering coefficient)

     • Centrality (betweenness centrality, subgraph centrality)

   • Generalized additive models (GAMs) were used to capture non-linear age trajectories for each metric, controlling for sex, dataset, and atlas.
     

5. Manifold learning and turning points

   • Here’s the clever part. The authors wanted to find lifespan phases in the joint behaviour of all metrics rather than in each metric separately.

   • They used Uniform Manifold Approximation and Projection (UMAP) to project the 11 age-predicted metrics (k-core was excluded) into a 3-dimensional “topological manifold.”

   • To make sure findings weren’t an artifact of specific parameters, they generated 968 UMAPs with different settings (e.g., number of neighbours, minimum distance). For each manifold, they:

     • Averaged metrics within each single year of age

     • Fitted polynomials through the 3D trajectory

     • Extracted ages at which the curve changed direction (“turning points”)
       

6. Epoch-wise analyses

   • Ages where turning points clustered (9, 32, 66, 83) defined five epochs. Within each epoch, the authors:

     • Correlated each metric with age to understand direction of change

     • Used LASSO regression to identify which metrics best predicted age

   • Finally, they ran a Principal Components Analysis across all ages to summarise integration, segregation, and centrality into three interpretable components, comparing these between epochs.
     

This combination of graph theory, manifold learning, and regularised regression offers a genuinely multivariate, lifespan-wide picture of structural brain organisation.

Results

Lifespan connectivity

Before topology, they looked at basic connectivity using variable-density networks. Brain networks were very dense but weakly weighted at birth, became sparser in late childhood, then showed another high-density phase around 30 years before declining again into late old age. Across the entire lifespan, node strength (how strongly each region is connected) increased in a fairly linear way, highlighting a shift from many weak connections to fewer but stronger ones over time.

Global integration and segregation

In density-controlled networks, global efficiency peaked around 29 years, with the mirror pattern for characteristic path length: shortest on average in the late 20s and then gradually increasing. Small-worldness showed multiple peaks and valleys but generally indicated that human brain networks maintain a small-world configuration (high clustering, short paths) across life.

Segregation showed a different story. Modularity was lowest around 31 and highest in late life; core/periphery organisation peaked around 20 and dipped in midlife. Local metrics (clustering coefficient and local efficiency) increased almost linearly, suggesting that neighbouring regions become increasingly tightly knit, even as global integration waxes and wanes.

Centrality

Betweenness centrality showed a U-shaped pattern with a minimum around 31 years, rising again in older age. Subgraph centrality steadily increased across life, indicating that certain nodes become increasingly important in local network loops and motifs.

Five topological epochs

UMAP-derived turning points at 9, 32, 66, and 83 years defined five epochs, each with its own fingerprint:

1. Epoch 1 (0–9 years): “Infancy into childhood”

   • Decreasing global integration, increasing local segregation, relatively stable centrality.

   • Small-worldness showed the strongest correlation with age; clustering coefficient was the best age-predictor.

   • Many regions (over half of the network) showed age-related increases in clustering, suggesting that neighbourhood-level specialisation is a key developmental feature in early life.
     

2. Epoch 2 (9–32 years): “Adolescence” (broadly defined)

   • All metrics correlated with age. Integration increased; global modularity decreased; local segregation increased; centrality measures showed complex but systematic shifts.

   • Small-worldness both strongly predicted age and showed the steepest directional change, capturing the co-evolution of efficiency and specialisation.

   • Around 32, directionality flips: global efficiency stops increasing and begins to decline; modularity and betweenness centrality start to rise.
     

3. Epoch 3 (32–66 years): “Adulthood”

   • Decreasing global efficiency and small-worldness, increasing segregation (especially clustering and local efficiency), minimal change in centrality.

   • Local efficiency was the strongest predictor of age, highly correlated with clustering, with effects spread across most of the brain.

   • Networks become less globally integrated but more locally cohesive.
     

4. Epoch 4 (66–83 years): “Early aging”

   • Fewer metrics correlated significantly with age, but modularity increased, integration measures tended to decline, and centrality increased.

   • Modularity was the key predictor of age, consistent with networks fragmenting into more clearly separated modules.
     

5. Epoch 5 (83–90 years): “Late aging”

   • Only subgraph centrality remained significantly associated with age, and only in a handful of occipital and parietal regions.

   • The age–topology relationship itself seems to weaken, possibly reflecting both biological variability and statistical power limits.
     

PCA confirmed that the biggest overall topological shift occurs at around 32 years, with substantial though somewhat different shifts at 9, 66, and 83, aligning with known milestones in white-matter maturation, cognitive development, and aging.

Discussion

At first glance, this paper might seem far removed from a neurofeedback clinic. There are no electrodes on the scalp, no training sessions, just large-scale tractography and graph theory. But in practice, what Mousley and colleagues offer is a kind of lifespan wiring diagram that can inform how we think about brain self-regulation at different ages.

A key message is that there is no single “normal” trajectory of brain network organisation. Instead, the brain moves through a sequence of epochs, each with distinct priorities:

• early life: building clustered neighbourhoods while pruning long-range efficiency,

• adolescence and young adulthood: maximising small-world efficiency and flexible integration,

• midlife: gradual shift towards stronger local organisation and reduced global integration,

• aging: greater modularity and a growing importance of particular nodes.
  

For people seeking help—whether for attention, mood, trauma, or performance—it may be more helpful to ask “Which topological epoch is this brain in?” than simply, “How old is this person?” A 15-year-old, a 25-year-old, and a 30-year-old might all sit in the waiting room, but their networks are sitting at very different points along the small-world trajectory described here.

For clinicians and referring professionals, this has several implications:

• In the broad “adolescent” epoch (9–32 years), the brain is becoming more globally integrated and small-world. This is a period of enormous plasticity—but also vulnerability. From a biofeedback or neurofeedback standpoint, this might be an especially receptive window for interventions that rely on synchronising large-scale networks, such as training frontal midline theta for executive control or modulating alpha networks for anxiety and sensory gating.

• In midlife (32–66), where local efficiency and clustering dominate, training may need to respect a nervous system that is less focused on rapid global integration and more on stabilising local loops. Here, work targeting specific functional circuits (for example, fronto-insular networks for interoception and emotional awareness, or fronto-parietal connections for working memory) may align well with the brain’s “preferred” organisational direction.

• In aging, increasing modularity and node centrality highlight the importance of maintaining communication hubs and cross-module connectors. Neurofeedback aimed at preserving flexibility in frontal and parietal hubs, or enhancing sensory-motor integration, could be conceptualised as supporting the bridges between modules that gradually erode with time.
  

From the perspective of neurofeedback practitioners, these findings encourage us to think beyond individual frequency bands and electrode sites and to ask how a given protocol might tap into—or strain against—the network tendencies of a particular epoch. For instance, SMR (sensori-motor rhythm) training at C3/C4 or Cz is often associated with stabilising sensorimotor and thalamocortical circuits that support behavioural inhibition and sleep. For a child in Epoch 1, this may dovetail with the brain’s ongoing consolidation of local circuits; for a young adult in Epoch 2, SMR might help anchor a network whose integration is peaking and, at times, overshooting into hyper-arousal or impulsivity.

Similarly, alpha up-training over posterior sites (Pz, P3/P4, or Oz) is frequently used to promote relaxation and internal focus. In midlife (Epoch 3), when global integration is gently declining and local connectivity is strengthening, such protocols could be interpreted as supporting efficient local processing within visual-parietal networks while easing strain on long-distance fronto-parietal circuits that are becoming more expensive to maintain.

An interesting thread in the paper is the weakening linkage between age and topology after about 66, and particularly after 83. This resonates with the clinical reality that older adults tend to diverge in trajectories: some remain cognitively resilient well into their 80s and 90s, while others show early decline. For biofeedback and neurofeedback, this variability suggests that assessments (qEEG, cognitive testing, autonomic measures) become increasingly important with age; “age-norms” may tell us less, and individual baselines more. Protocols might focus less on “normalising” to external norms and more on supporting each person’s own optimal pattern—maintaining flexibility, preserving hubs, and stabilising sleep, balance, and emotional regulation.

At a more conceptual level, this study reinforces an idea that’s very natural to anyone who works with self-regulation: the brain does not simply decline with age; it reorganises. Early life prioritises exploration and redundancy; adolescence optimises connectivity; midlife quietly reinforces local specialisation; and aging emphasises modularity and reliance on key nodes. Neurofeedback can be seen as a way of gently nudging these ongoing reorganisations in a more adaptive direction—whether that means helping a teenager rein in runaway integration that presents as anxiety and rumination, or helping a 70-year-old maintain cross-module communication that underpins flexible problem-solving.

Brendan’s perspective

When I read a paper like this, I imagine a time-lapse of the brain’s connectome: edges lighting up, fading, re-routing, whole modules pulsing as life unfolds. For neurofeedback clinicians, the temptation is always to jump straight to “So what should I train?” Let’s walk through what these five epochs might mean for day-to-day EEG work, with an emphasis on individualisation.

Epoch 1 (0–9): scaffolding local networks

We don’t typically train very young children with full-blown EEG protocols, but by late childhood it becomes feasible. Here, the topology story is: decreasing global integration, increasing clustering, very active local re-wiring.

That suggests a few principles:

• Keep protocols simple and stabilising. For example, SMR training at Cz (12–15 Hz) or C3/C4 can support sensorimotor stability, sleep, and attention without demanding heavy long-range integration.

• Be cautious with protocols that strongly drive coherence across distant regions or heavily push faster beta (16+Hz) over frontal sites—these might add load to systems that are still pruning and consolidating.

• Biofeedback targeting breathing and heart-rate variability can be especially helpful in this epoch, supporting autonomic regulation that in turn shapes network development.
  

In practice, I’d be thinking: “How do I help this child’s brain settle into effective local loops—sensorimotor, frontal-striatal, and posterior alpha networks—so that later integration has solid foundations?”

Epoch 2 (9–32): surfing the small-world wave

This is the big one. The paper frames 9–32 as a single epoch of increasing integration and small-worldness. Clinically, this is exactly the window where we see:

• onset or consolidation of anxiety disorders, mood disorders, and attentional problems,

• explosive learning capacity,

• big swings in identity, social context, and stress.
  

Neurofeedback here can be bold—but structured. Some ideas:

• For attentional and executive challenges:

  • Frontal midline theta (Fmθ) inhibition (or enhancement, depending on the case) at FCz (4–7 Hz with good 12–15 Hz SMR support) to strengthen executive networks and working memory.

  • Reduce excessive high beta (22–30 Hz) if accompanied by worry, rigidity, or insomnia.
    

• For anxiety and over-arousal:

  • Posterior alpha (9–12 Hz) up-training at POz, sometimes combined with gentle down-training of fast beta. This can help re-balance internal vs external focus and calm hyper-integrated vigilance networks.
    

• For emotional regulation and trauma-related patterns:

  • Training at fronto-central sites (F3, Fz, F4, FCz) with protocols that encourage stable mid-range frequencies (13–18 Hz) and reduce excessive slow activity (2–8 Hz) linked to dysregulation or dissociation.
    

Because this epoch’s networks are naturally becoming more efficient and interconnected, neurofeedback can act like steering the trajectory of a rocket that’s already firing hard: small adjustments in protocol and context (sleep, relationships, load) can have outsized long-term effects.

Epoch 3 (32–66): tuning circuits, protecting flexibility

By midlife, large-scale integration is gently declining while local efficiency and clustering keep rising. Subjectively, many people describe this as the phase where they “know who they are”—but also feel stretched by chronic stress, caregiving, and work load.

For EEG work, I tend to think less in terms of “building new capacity” and more in terms of:

• relieving chronic over-activation in specific circuits,

• supporting recovery and restorative processes,

• preserving flexibility between modules.
  

Examples:

• For chronic stress and burnout:

  • Alpha-theta protocols (e.g., at Pz, eyes closed) to promote deep relaxation and memory processing, provided the client is stable enough for the intensity.

  • Heart-rate variability biofeedback combined with simple SMR or low-beta training to consolidate a calmer baseline.
    

• For cognitive complaints (word-finding, “brain fog”):

  • Fronto-parietal support with protocols at F3/F4 and P3/P4, focusing on low-beta (13–18 Hz) stability, while inhibiting over-driving high beta.

  • If qEEG shows under-engaged alpha networks, moderate alpha reinforcement in posterior regions may help optimise idle/engaged transitions.
    

Because networks are already more segregated, I’m careful not to “over-specialise” further. Excessive coherence training within a module that’s already quite tight could, in theory, worsen rigidity. Instead, I like protocols that support switching—helping the brain move between task-positive, default mode, and salience networks more smoothly (for example, by stabilising mid-range frequencies at midline sites).

Epochs 4 & 5 (66+): preserving bridges and self-efficacy

In early aging (66–83), modularity increases and the network leans more heavily on specific nodes—like a train system that depends more on a few key junctions. After about 83, the link between age and overall topology weakens, and individual variability dominates.

Three practical principles stand out:

1. Assess, don’t assume

   • qEEG baselines, cognitive screening, gait, sleep, and autonomic measures become crucial here. Two 75-year-olds may be in very different places in terms of network health.

   • Protocols should be lighter, with more frequent small adjustments based on subjective response.
     

2. Protect hubs and cross-module communication

   • Frontal midline and parietal sites are common focuses: gentle training to stabilise low-beta or SMR while avoiding pushing too hard into high-beta territory, which can exacerbate anxiety or insomnia.

   • Exercises that coordinate sensory and motor systems—like combining neurofeedback with balance training or gentle movement—may indirectly support structural connections that are starting to weaken.
     

3. Prioritise function and wellbeing over “normalisation”

   • At this stage, success might look like better sleep, fewer falls, improved confidence, or reduced anxiety—not a perfect z-score on a normative database.

   • Biofeedback (breathing, HRV) pairs nicely with EEG to support cardiovascular health, which in turn is tightly linked to white-matter integrity.
     

Across all epochs, individualisation is the golden rule. The turning points identified by Mousley and colleagues give us a map of the terrain, but each person still walks a unique path. Neurofeedback and biofeedback are at their best when they honour both: the general principles of network development and the unique story encoded in each brain.

Conclusion

This study shows that human brain networks don’t simply mature and then decline. Instead, they move through distinct topological epochs, marked by turning points around 9, 32, 66, and 83 years of age. Each epoch has its own balance of global integration, local segregation, and nodal centrality, aligning with known milestones in cognition, behaviour, and health.

For biofeedback and neurofeedback practitioners, these findings invite us to think developmentally about our protocols: to design training that respects where the brain is in its natural organisational journey, and to adjust our expectations of plasticity, integration, and stability across the lifespan. The take-home message is hopeful: throughout life, the brain keeps rewiring itself, and carefully tuned feedback-based interventions can help that reorganization bend toward resilience, flexibility, and wellbeing.

Reference

Mousley, A., Bethlehem, R. A. I., Yeh, F.-C., & Astle, D. E. (2025). Topological turning points across the human lifespan. Nature Communications, 16, 10055. https://doi.org/10.1038/s41467-025-65974-8