This article features moderately technical neuroscientific language. We’ve front-loaded the technical detail in the first section, with the latter sections focused on implications and what this means for intervention approaches.
The normalisation paradox — when helping one network harms another
In a recent study, Huang et al. (2026) administered arbaclofen — a GABAB receptor agonist (not GABA, the inhibitory neurotransmitter, but GABAB, a protein that GABA binds to) — to 24 non-autistic and 15 autistic adults across 93 study visits. Participants received placebo, 15mg, or 30mg doses in randomised, double-blind conditions. The research team measured brain network responses via EEG three hours post-administration, during the drug’s active physiological window.
The premise followed established excitatory-inhibitory (E-I) imbalance theory. Autistic brains show disrupted GABA signalling across rodent models, post-mortem analyses, magnetic resonance spectroscopy, and GABA-dependent sensory functions. Previous research relied on static baseline comparisons. This study tested dynamic response — how autistic brain networks actually shift when GABA function is pharmacologically challenged.
The researchers divided cortical activity into seven functional networks: frontoparietal (planning, cognitive control), default mode (self-referential thought), limbic (emotion, memory), ventral attention (salience detection), somatomotor (movement, body sensation), dorsal attention (spatial focus), and visual (sensory processing). They measured phase-amplitude coupling within and between these networks at baseline and under drug conditions.
What they discovered demolished simple normalisation models.
Autistic participants showed higher baseline coupling across most networks, particularly pronounced in limbic regions. When researchers administered arbaclofen, different networks responded in fundamentally contradictory ways. Sensory networks — visual and somatomotor — required high dose (30mg) to shift toward non-autistic measurements. Cognitive networks — frontoparietal, default mode, attention systems — barely responded to either dose. The limbic network demonstrated a paradoxical U-shaped response: measurements normalised at low dose (15mg) but returned to atypical patterns at high dose (30mg).
This wasn’t measurement noise. The pattern held across participants and replicated across multiple coupling frequencies.
For any individual autistic person, no single dose normalised all networks simultaneously.
The dose that helped sensory processing made limbic processing worse. The dose that normalised emotional circuitry left sensory networks atypical.
The researchers had proven normalisation logically impossible through this mechanism.
Not difficult. Not requiring personalised dosing. Impossible — because improving one network actively destabilises another within the same brain.
Measuring brain network responses through phase-amplitude coupling
Brain activity operates at different frequencies simultaneously. Slow waves — theta (4-8 Hz) and alpha (8-12 Hz) — coordinate large-scale network activity across distributed regions. Fast waves — beta (12-30 Hz) and gamma (above 30 Hz) — handle local, fine-grained computational processes. Phase-amplitude coupling measures whether the timing of slow waves controls the strength of fast waves.
High coupling means tight coordination. Slow rhythms strongly dictate when fast rhythms can fire. Low coupling means loose coordination. Fast rhythms operate more independently from network-level organisation. The researchers measured four coupling combinations: theta-beta, theta-gamma, alpha-beta, and alpha-gamma.
PAC (Phase-Amplitude Coupling) reflects information transfer between scales. Slow waves organise global network states. Fast waves execute local computation. Coupling coordinates between them. Crucially, PAC depends on GABA circuits — inhibitory interneurons generate the coupling mechanism. This makes PAC a direct measure of GABAergic function across different organisational levels.
At placebo baseline, autistic participants showed significantly higher theta-beta coupling across nearly the entire brain during eyes-closed resting state. The frontoparietal network, default mode network, limbic network, ventral attention network, somatomotor network, and visual network all demonstrated elevated coupling. Only the dorsal attention network showed comparable measurements between groups. The limbic network stood out — autistic participants showed higher coupling across all four frequency combinations measured.
“Normalised” in this context means measurements shifted closer to non-autistic baseline values. The research framework assumed non-autistic (“neurotypical“) patterns represent optimal function and deviation indicates dysfunction requiring correction.
Under low-dose arbaclofen (15mg), autistic participants showed minimal change in most networks. Sensory networks remained atypically high. Cognitive networks remained unchanged. But the limbic network responded dramatically — all four coupling measures shifted toward non-autistic baselines. Between-network connectivity involving limbic regions also normalised.
Under high-dose arbaclofen (30mg), the pattern reversed. Sensory networks — somatomotor and visual — shifted toward non-autistic measurements. The dose that failed to help at 15mg succeeded at 30mg for these regions. But limbic network measurements, which had normalised at low dose, returned to atypical patterns. The very networks that responded positively to 15mg showed renewed dysfunction at 30mg.
Cognitive networks — frontoparietal, default mode, attention systems — showed minimal response to either dose. The intervention that affected sensory and limbic processing left higher-order networks essentially unchanged.
The researchers documented this through rigorous statistical methods, controlled for multiple comparisons, and verified the pattern held when controlling for age differences between groups. The contradictory responses weren’t artifacts. They represented genuine differential GABAergic responsivity across functional networks within individual autistic brains.
Inherent individual autistic coherence vs forced normalisation
The findings reveal autistic brain organisation operating under different principles than researchers assumed. If networks within a single brain require opposite interventions simultaneously, the system isn’t incoherent or broken — it’s balanced around different trade-offs.
Consider what internal coherence requires. A functional brain coordinates information transfer between local processing and global network states. Different networks handle different computational demands. Sensory systems process external input. Limbic systems regulate internal states. Cognitive systems integrate across domains. These networks must coordinate whilst maintaining specialised functions.
Non-autistic brains achieve this coordination through specific patterns of coupling between slow network rhythms and fast local processing. The researchers assumed these patterns represent universal optima — the correct way to organise information flow.
But autistic brains demonstrate tighter coupling in baseline conditions. Higher phase-amplitude coupling means stronger coordination between network-level organisation and local computation. This could serve different computational purposes. Perhaps autistic cognition benefits from tighter integration between scales. Perhaps the processing demands autistic systems handle require different coordination patterns.
The drug intervention tested whether loosening coupling — making it more similar to non-autistic patterns — improves function. What the intervention actually demonstrated: forcing individual networks toward non-autistic states creates contradictions that completely break internal coherence.
Sensory networks loosened coupling at high dose, shifting toward non-autistic measurements. But this required a dose that simultaneously disrupted limbic networks that had achieved stability at low dose. There’s no configuration where all networks match non-autistic patterns simultaneously because the autistic system isn’t organised around those patterns as its stable state.
The researchers noted this creates problems for clinical translation. They frame it as complexity requiring precision medicine approaches — finding the right dose for each individual. But the data shows something more fundamental:
Individual autistic people don’t need personalised “therapeutic” doses towards typicality. They need interventions that don’t assume non-autistic patterns represent optimal function for autistic systems.
The backwards assumption operates throughout the framework. The paper describes autistic measurements as “aberrant,” “atypical,” “altered,” requiring “rescue” and “normalisation.” The limbic network’s return to baseline autistic patterns at high dose is described as “re-emergence” of dysfunction. The language encodes the assumption that deviation from non-autistic measurements indicates pathology.
But if autistic networks require contradictory interventions when forced toward non-autistic states, perhaps those states aren’t optimal for autistic organisation. Perhaps autistic coherence operates through different coupling dynamics that serve the system’s computational requirements. Forcing components toward non-autistic parameters doesn’t fix dysfunction — it creates dysfunction by fighting the system’s organisational logic.
The researchers acknowledge this partially. They note GABAB receptors can cause both inhibition and disinhibition through complex pre and post-synaptic effects. They recognise different networks might have different GABAB receptor densities or different functional configurations. They suggest “functioning” of GABAB receptor circuits differs between networks in autistic brains.
But they maintain the framework that non-autistic = target. They don’t consider that autistic “functioning” might ALREADY represent coherent organisation rather than distributed dysfunction.
They document that normalisation proves impossible whilst maintaining normalisation as the conceptual goal.
Square pegs and pharmacological interventions
The logical impossibility the research demonstrates extends beyond GABAB agonism. Any intervention premised on making autistic brain function match non-autistic brain function faces the same fundamental problem — it fights the internal coherence of autistic systems.
Stimulant medications prescribed for autistic individuals typically aim to make attention patterns match non-autistic attention norms. Sustained cybernetic attention on externally-determined tasks. Reduced distractibility by tangential information. Suppressed interest-driven exploration in favour of assigned priorities.
These interventions assume autistic attention represents dysfunction requiring correction toward non-autistic parameters.
But if autistic brain networks balance different trade-offs than non-autistic networks, forcing attention systems toward non-autistic patterns might create the same contradictions this study documented. Improving sustained focus on assigned tasks whilst destabilising the tight coupling between interest and processing depth that serves autistic cognition. The dose that makes an autistic person look attentive in classroom settings might disrupt the neural organisation that enables their deep systematic exploration.
Applied Behaviour Analysis operates on identical assumptions. The intervention systematically rewards non-autistic behaviour patterns and suppresses autistic behaviour patterns. Eye contact instead of sensory regulation. Scripted social responses instead of genuine but atypical interaction. Compliance with external demands instead of autonomous coherence-seeking. ABA assumes autistic behaviour indicates dysfunction requiring correction toward non-autistic norms.
Our latest article on the recent research on ABA outcomes demonstrated the framework creates trauma without producing genuine skill development. This study’s findings suggest why — forcing behavioural conformity to non-autistic patterns fights the neural organisation supporting autistic coherence. You can train external compliance through sufficient reinforcement and punishment. You cannot make that compliance align with the system’s internal logic.
Social skills training follows the same model. Learn non-autistic interaction norms. Suppress autistic communication patterns. Practice non-autistic emotional expression. Mask non-autistic-incompatible responses. The intervention assumes autistic social behaviour represents deficit requiring correction toward non-autistic function.
But if autistic neural organisation operates through different coupling dynamics, perhaps autistic social behaviour emerges from that organisation coherently. Perhaps the systematic, interest-focused, direct communication autistic people demonstrate serves their neural architecture effectively.
Forcing those patterns toward non-autistic norms doesn’t fix dysfunction — it creates internal contradictions between neural organisation and behavioural demands.
The square peg metaphor captures this precisely. The problem isn’t that autistic brains are broken non-autistic brains requiring repair. The problem is systems designed for non-autistic neural organisation demanding autistic systems conform. Pharmacological interventions that assume non-autistic measurements represent universal optima. Behavioural interventions that assume non-autistic patterns represent correct development. Educational systems that assume non-autistic learning styles represent proper cognition. Employment and workplace infrastructure that assumes non-autistic processing represents productive work.
All of these share the backwards assumption this research exposes. They treat autistic difference as deviation from correct function requiring normalisation. They don’t recognise autistic systems might have their own coherence that the presence of a round hole and its normalisation efforts actively disrupt.
The researchers frame their findings as explaining clinical challenges — why arbaclofen trials showed mixed results, why autistic individuals report paradoxical medication effects, why dose sensitivity complicates treatment. They suggest precision medicine approaches targeting specific networks might succeed where broad interventions fail.
But the data, as I’ve said, suggests something more fundamental. Autistic brain networks don’t need interventions forcing them toward non-autistic states. They need environments compatible with their organisational logic. Systems that work with autistic coupling dynamics rather than fighting them. Demands that align with autistic processing rather than requiring constant internal contradiction.
The dose that helps one autistic network whilst harming another isn’t a dosing problem requiring better calibration or, worse, individual optimisation. It’s evidence that the entire normalisation framework operates backwards. Autistic brains aren’t failed non-autistic brains. They’re systems organised around different principles. Interventions that ignore those principles don’t just fail — they prove logically impossible.
And now it is proven.
Citations
Huang et al. (2026) — Differential GABA dynamics across brain functional networks in autism
