How the brain’s wiring changes

The brain’s wiring changes throughout life, with regions communicating differently in older age. Wiring can also look different in people with certain neurological conditions. To understand these shifting patterns — their vulnerabilities and variations — researchers introduce a mathematical approach¹ that maps how the organ’s functional connectivity changes.

The approach, a measure called node persistence, identifies specific regions most impacted by these alterations, many of which are targets for experimental non-invasive medical therapies.

Researchers at the the Institute of Mathematical Sciences in Chennai, Max Planck Institute for Mathematics in the Sciences, Germany, and Nanyang Technological University, Singapore used persistent homology (PH), a method from topological data analysis (TDA), which examines the overall “shape” of complex data rather than individual connections.

In this case, PH picks out robust patterns in how different regions talk to each other during brain activity. Rather than focusing on single links, PH examines large-scale structures — such as connected groups, loops, and voids — and tracks how long these features, often referred to as ‘holes’, persist as the threshold for connectivity changes.

By looking for these mathematical 'holes,' we can see where the network's coordination is breaking down, says Areejit Samal at IMSc, Chennai.

The study analysed resting-state fMRI scans from 72 elderly people and 153 young people, plus 820 people with and without autism. The team tracked connectivity across three nested levels: a global level (the brain as a whole), a mesoscopic level (seven major resting-state networks), and individual regions.

They found that young adults show more complex and longer-lasting topological features than older adults, while people with autism show less persistent one-dimensional structures than typically developing individuals. “These results indicate that the brain-wide functional connectivity shifts both with age and in autism,” says IMSc’s Madhumita Mondal.

At the mesoscopic level, ageing particularly influences the somatomotor, dorsal attention, salience/ventral attention, and default mode networks, whereas autism-related differences are concentrated in the somatomotor, salience/ventral attention, and default mode networks.

The somatomotor network, which is closely linked to motor performance, stood out in both cases.

“Our method identified multiple regions within this network in both ageing and ASD. While the involvement of this network is supported by existing literature, it was striking that these regions emerged naturally from a data-driven, topological analysis rather than being predefined or explicitly targeted,” Mondal adds.

Narrowing down to individual regions, the authors applied node persistence to identify the regions contributing most strongly to connectivity differences. The team pinpointed 108 specific regions showing connectivity changes in ageing and 27 in autism. Many of these regions are linked to key functions such as movement, language, memory and social cognition.

When the researchers compared their findings with clinical data on stimulation techniques, they found that regions showing significant topological shifts overlapped with those where clinical stimulation improves motor function in the elderly or alleviates symptoms in ASD.

This concordance between predicted target regions and the stimulation sites commonly used in three prominent non-invasive techniques — transcranial direct current stimulation, transcranial alternating current stimulation and transcranial magnetic stimulation — could help develop predictive tools, according to neuroscientist Dipanjan Roy at the Indian Institute of Technology Jodhpur, who was not involved in the study.

An acknowledged limitation is that the analysis currently focuses on one-dimensional topological features, potentially missing more complex network structures. Future work could examine higher-dimensional patterns. The approach could also be extended to other neurological conditions and help predict which patients are most likely to respond to stimulation of a given region.

The next step is to test whether combining node persistence, node frequency and time-varying measures can better predict age in new data or help distinguish clinical subgroups, Roy adds. “It will also be essential to compare the predictive power of these combined measures with existing PH-based metrics and with standard graph-theoretical approaches.”