Functional Neurological Disorder Through the Lens of Population Neuroscience
Functional Neurological Disorder (FND) is an umbrella term for a group of neurological conditions characterised by symptoms such as tremor, weakness, dystonia, gait disturbance and functional seizures that cannot be adequately explained by the structural neurological disease processes that would conventionally produce them.
Historically, those diagnosed with FND have often been dismissed or led to believe that their symptoms are primarily psychological in nature, reflected in terms such as hysteria, conversion disorder and psychogenic movement disorder. Thankfully, our understanding has progressed considerably. FND is now recognised as a genuine disorder of nervous system functioning, and research increasingly implicates altered communication within and between neural networks involved in sensorimotor processing, attention, salience, emotion, interoception and the sense of agency. Predictive-processing models have added further depth, proposing that altered weighting of sensory information, expectations and predictions may contribute to the generation of functional symptoms.
This represents enormous progress. But I believe developments in population neuroscience allow us to take the question one level deeper. If communication and processing within these networks are altered in FND, what might that dysfunction actually look like at the level of the neural populations performing those computations?
This is where I believe neural population dynamics, representational geometry and neural trajectories could become important to our understanding of FND.
Looking beyond structure
There is an interesting parallel with the evolution of pain science. We now understand that pain does not map neatly onto tissue damage. People can experience significant pain without sufficient structural pathology to explain its severity, while substantial structural abnormalities can exist with surprisingly little pain. Pain emerges from a complex interaction between sensory information and factors including context, expectation, attention, previous experience and emotional state.
An absence of visible structural pathology is therefore not the same thing as an absence of biological mechanism.
FND arguably presents us with a similar challenge. A tremor, functional seizure or episode of limb weakness is an observable output of the nervous system. If conventional structural examination cannot explain that output, it does not follow that there is no neurological process producing it. It may simply mean that the relevant dysfunction exists at a level of nervous system organisation that our conventional clinical tools are not yet capable of observing.
I'm reminded of an analogy David Attenborough once used when discussing agnosticism: If you remove the top of a termite mound, the termites have no awareness that a human is standing above them observing them because they do not possess the sensory apparatus required to perceive us.
The analogy is obviously not scientific evidence, but the principle is useful: our inability to perceive something does not demonstrate its absence.
MRI, EMG, EEG and neurological examination are incredibly powerful tools, but each interrogates particular aspects of nervous system function. A structurally normal MRI tells us something important, but it cannot tell us that the computations occurring across billions of interacting neurons are normal.
Population neuroscience may give us a way of thinking about what could be happening beyond that level.
From neurons to neural populations
The neuron remains the fundamental cellular unit of the nervous system. However, increasingly, systems neuroscience recognises that understanding individual neurons alone is insufficient to explain complex behaviour. Saxena and Cunningham described this emerging perspective as the neural population doctrine: the idea that understanding neural computation requires us to study coordinated activity across populations of neurons rather than assigning individual neurons simple, fixed functions.
These populations contain neurons with different and often mixed selectivity. Together, their activity creates an enormous high-dimensional neural state space. At any moment, the combined activity of the population represents a particular point within that space. As population activity changes over time, it moves through that space, forming a neural trajectory.
Crucially, population activity does not explore every theoretically possible state randomly. It tends to occupy structured, lower-dimensional regions of the wider state space, commonly described as neural manifolds. These geometries constrain and organise the trajectories available to the neural population and, ultimately, the computations and behaviours that emerge.
This adds another dimension to our understanding of the nervous system. Sensory inputs are not simply received, processed and converted into outputs through a linear feedback loop. They interact with the current state of enormous neural populations, whose collective activity evolves through structured neural state spaces over time.
It is at this level that I hypothesise some of the currently unexplained features of FND may exist.
A population-level hypothesis of FND
To be clear, there is currently no evidence demonstrating that abnormal neural manifolds or neural trajectories are the mechanisms behind FND. What follows are my ideas based upon applying established principles from population neuroscience to what we already know about FND.
There is growing evidence suggesting that altered communication between neural networks is key in FND. Research has implicated systems involved in sensorimotor integration, attention, salience, emotion, interoception and agency, while predictive-processing accounts propose altered relationships between sensory evidence, expectations and predictions.
My question is: what does that altered communication and computation actually look like?
My hypothesis is that these established network-level abnormalities are underpinned by changes in the geometry and dynamics of the neural populations within and between those networks. Changes in sensory processing, attention, prediction and internal state may alter population activity and neuronal tuning. In turn, the representational geometry generated by those populations may change, altering the neural trajectories that can be readily accessed and stabilised.
The structural components may remain largely intact, and the large-scale networks may still be identifiable, but the population-level computation occurring within and between them may be different.
If neural trajectories ultimately contribute to the generation of behaviour, altered trajectories could produce altered behavioural outputs: tremor, weakness, dystonia, gait disturbance, functional seizures or changes in sensation.
This would not replace existing models of FND. It would provide a potential population-level description of what those models are observing.
Predictive processing might tell us that sensory information is being weighted differently. Network neuroscience might tell us where communication is altered. Population neuroscience may eventually help us understand what that altered computation physically looks like as patterns of neural activity evolving through time.
That, for me, is the interesting next question.
Could the same principle apply beyond FND?
Since beginning to think about neurological rehabilitation through this lens, I have found myself applying the same principle to musculoskeletal injury within my practice also.
ACL injury provides a good example. An ACL rupture is clearly a structural injury, yet its consequences extend far beyond the knee. Loss and alteration of afferent information changes the sensory information reaching the central nervous system, and neuroimaging studies have demonstrated altered cortical activation during knee movement following ACL reconstruction, including differences in regions involved in sensory processing, motor planning and visual-motor control.
Again, we already know that the brain changes following ACL injury. My hypothesis concerns what those changes might look like at the population level.
If the information entering the system changes, neuronal tuning and population activity may change with it. If population activity changes sufficiently, the geometry of the neural state space occupied during movement may change, and therefore the trajectories generated within that space may also change.
The resulting movement and experience may look different because the neural computation producing it is different.
I am not suggesting that ACL injury and FND are the same pathology. Clearly they are not. I am suggesting that they may share a fundamental systems principle: alter the information and constraints acting upon a neural system, and you may alter the population dynamics through which that system generates behaviour.
What does this mean for rehabilitation?
This way of thinking was one of the ideas that led me to develop Geometry-Informed Rehab (GIR), of which I published my first article in July 2025.
If behaviour emerges from neural population dynamics, then rehabilitation should not focus solely on strengthening muscles or reconnecting neural pathways. We should also consider how the experiences we create during rehabilitation influence the population dynamics from which movement emerges.
Rather than asking only, “How do I make this patient move better”, we can ask, “How do I create the conditions in which this person’s nervous system can learn, discover and access better movement solutions over time?”
That means manipulating sensory information, attention, movement variability, rhythm, task constraints, expectation, and environmental demands, whilst still building fundamental physical capacities such as strength and stamina.
Interestingly, many established approaches to FND rehabilitation already do this by redirecting excessive attention away from movement, retraining automatic movement and changing habitual movement patterns.
GIR does not necessarily argue against these approaches. Instead, it offers a hypothesis for what may be occurring underneath them at the level of neural population dynamics.
Where this could take us
This way of thinking has fundamentally changed how I approach complex neurological and musculoskeletal cases. When somebody presents with FND, I can acknowledge the evidence we already have for altered neural network function while also considering what that might mean at a deeper computational level. Their nervous system may be structurally capable of producing normal movement while currently occupying population states and trajectories that repeatedly generate an undesirable solution.
Ultimately, the research needs to catch up with these ideas. We do not yet know whether people with FND demonstrate altered representational geometries or neural trajectories, whether particular symptoms correspond to particular population dynamics, or whether successful rehabilitation changes those dynamics.
But these are now testable questions.
I believe that future research will increasingly identify functional changes in neural population geometry and dynamics across conditions such as FND, chronic pain and musculoskeletal injury, and that understanding these changes will eventually allow us to develop more targeted rehabilitation approaches.
Population neuroscience has not solved FND. But it may provide a new level of explanation for something we already know: a nervous system does not need to be structurally damaged to function differently. Perhaps the next step is understanding what that difference actually looks like, and my hypothesis is that part of the answer lies in the geometry and trajectories of neural population activity.