Dynamic excitatory-inhibitory differences in autistic and non-autistic adults: Evidence from a pharmacological challenge with arbaclofen

Abstract

Background:

Brain function is the dynamic output of coordinated excitatory and inhibitory (E-I) activity. E-I alterations, arising from differences in excitatory glutamate and inhibitory GABA pathways, are implicated in the development and heterogeneity of autism, and are consequently targets for pharmacological support options. Existing tools, such as Magnetic Resonance Spectroscopy, are limited in capturing the dynamic nature of E-I regulation. The aperiodic 1/f exponent of the EEG power spectrum has shown sensitivity to E-I perturbations in animals and neurotypical humans, but its applicability to neurodiverse populations remains underexplored.

Methods:

Therefore, as proof-of-concept, this study tested the hypotheses that (i) the aperiodic 1/f exponent of resting-state EEG changes following a pharmacological E-I challenge with arbaclofen (STX209), a GABAB receptor agonist; and (ii) dynamic responsivity to GABAergic challenge is different in autism. Participants were 40 adults, 15 autistic. EEG was recorded at rest after randomised, double-blind administration of a placebo, 15 mg of arbaclofen, and 30 mg of arbaclofen. Aperiodic 1/f exponents were extracted.

Results:

As predicted, in both groups the aperiodic 1/f exponent significantly increased following a high (30 mg) dose of arbaclofen, replicating the effect observed in animals. Furthermore, a lower (15 mg) dose showed a different response pattern across groups, with aperiodic exponents tending to increase in autistic individuals but decrease in non-autistic individuals, suggesting differences in GABAergic responsivity.

Conclusions:

These findings support the aperiodic 1/f exponent as a metric for dynamic E-I regulation and provide preliminary evidence of distinct homeostatic E-I dynamics in autism.

Keywords

excitation inhibition EEG autism arbaclofen neurodiversity homeostasis

Introduction

Excitatory and inhibitory (E-I) signalling is fundamental to brain function. It involves numerous chemical transmitter systems, with glutamate as the most abundant excitatory neurotransmitter in the postnatal brain and Gamma-aminobutyric acid (GABA) as the primary inhibitory neurotransmitter. A tightly organised E-I relationship emerges during early development (Dorrn et al., 2010) and adapts to brain states such as rest and sleep (Okun and Lampl, 2008; Niethard et al., 2016). Coordinated E-I is crucial for precise sensory response (Heiss et al., 2008; Mariño et al., 2005; Poo and Isaacson, 2009; Tao and Poo, 2005; Wehr and Zador, 2003), as well as higher cognitive functions such as memory (Lim and Goldman, 2013; Rubin et al., 2017; Vogels et al., 2011), information processing, and social behaviour (Yizhar et al., 2011).

In neurotypical brains, E-I activity is dynamically regulated to maintain balance and prevent excess excitation or inhibition, which could saturate or silence neuronal networks (Turrigiano, 2012). Differences in E-I regulation have been implicated in neurodevelopmental conditions, such as autism (Sohal and Rubenstein, 2019). Genes linked to autism, such as neurexins, neuroligins, and SHANK3, affect synapse generation and function, disrupting components of the E-I pathway (Cao and Tabuchi, 2017; Südhof, 2008). These synaptic perturbations lead to altered E-I activity (Toro et al., 2010) and also interfere with synaptic scaling mechanisms that regulate homeostasis (Turrigiano, 2012), as observed in autism-relevant animal models (Tatavarty et al., 2020).

Despite this evidence, tools for assessing E-I signalling in the living human brain remain limited. E-I signalling is a complex, dynamic process (Bridi et al., 2020; Brunwasser and Hengen, 2020), yet traditional methods for studying it, such as Magnetic Resonance Spectroscopy (MRS) and Positron Emission Tomography (PET), are expensive, invasive, and lack the temporal resolution needed to capture E-I dynamics (Lystad and Pollard, 2009; Pfister et al., 2014). This limitation is especially relevant because E-I temporal dynamics can change without affecting static (average) E-I properties, highlighting the need for tools that can measure rapid, dynamic processes (Bruining et al., 2020; Litwin-Kumar and Doiron, 2012; Sohal and Rubenstein, 2019; Szücs and Huerta, 2015; Turrigiano, 1999). EEG, with its superior temporal resolution, safety, and affordability, offers a promising alternative for capturing E-I processes on the necessary timescales in humans.

The aperiodic component of the Electroencephalography (EEG) power spectral density (PSD) has been proposed as a proxy for E-I signalling (Gao et al., 2017; He et al., 2019; Molina et al., 2020; Peterson et al., 2023; Tran et al., 2020; Veerakumar et al., 2019). The EEG PSD follows a 1/f-like power law, where the aperiodic component represents non-oscillatory brain activity observed in the absence of prominent oscillations (Donoghue et al., 2020a, 2020b). This component can be quantified using the aperiodic 1/f exponent, equivalent to the slope of the PSD (Donoghue et al., 2020b). At the same time, changes in E-I balance may be relatively local and can be partially compensated over development, so any influence of E-I perturbations on adult scalp EEG is likely to be indirect rather than one-to-one (Nelson and Valakh, 2015). We, therefore, interpret the aperiodic exponent as a measure that is sensitive to circuit states shaped by E-I interactions, rather than a precise read-out of E-I balance. In earlier work (Ahmad et al., 2022), we outlined the theoretical rationale for using the aperiodic 1/f exponent as a marker for E-I dynamics, emphasising its translational potential. Building on that foundation, this study provides the first direct empirical test of this metric in human neurodiverse populations.

To test whether the aperiodic exponent reflects dynamic E-I changes, it is necessary to experimentally perturb E-I activity and observe corresponding changes in the metric (Whelan, Daly et al., 2024). While previous research has demonstrated sensitivity of the aperiodic exponent to E-I modulation in animals (Gao et al., 2017) and in neurotypical adults (Waschke et al., 2021), its application to neurodiverse populations has not been established.

This study serves as a proof-of-concept (PoC), aimed at bridging preclinical evidence with translational applications in humans. We used the GABA_B receptor agonist arbaclofen to directly increase global inhibition (Nance et al., 2011; Padgett and Slesinger, 2010) in adults with and without autism. We hypothesised that arbaclofen would increase the aperiodic 1/f exponent relative to placebo, consistent with findings in animals (Gao et al., 2017) and neurotypical adults (Waschke et al., 2021). We further predicted that the dose-dependent change in the 1/f slope would differ between autistic and non-autistic individuals, reflecting theories of altered E-I dynamics in autism (Sohal and Rubenstein, 2019).

Materials and method

Design, participants, procedure

This research was conducted in accordance with the Declaration of Helsinki, at the Institute of Psychiatry, Psychology, and Neuroscience (IoPPN) at De Crespigny Park, SE5 8AF, London, UK (Feb 2018–Mar 2020). The Medicines and Health care products Regulatory Authority confirmed the study design was not a Clinical Trial as defined under EU/UK law, but for transparency the study was registered on https://clinicaltrials.gov/ct2/show/NCT03594552.

Forty participants (25 non-autistic and 15 autistic) took part in this double-blind, placebo- controlled, repeated measures study. All participants were invited to take part in three study visits, whereby they were given a single dose of arbaclofen (STX209; low dose = 15 mg, high dose = 30 mg) or a placebo, the order of which was randomised. The randomisation was generated by Prof McAlonan using https://www.random.org/. Study visits were separated by a minimum of 1 week to allow adequate washout of arbaclofen and to minimise potential carry-over effects between sessions. A total of 104 study visits were completed. EEG data were collected 3 hours post-drug and within the half-life of arbaclofen (Berry-Kravis et al., 2017). A medic was present for all study visits and monitored participants regularly for adverse side effects. At their discretion, in response to potential side effects, medics were then (and only then) able to access unblinding information. We noted that participants were more likely to experience known side effects of arbaclofen (nausea and dizziness) at the higher dose; therefore, our ethics committee approved an amendment to ensure that the order of administration was adjusted so that the high dose of arbaclofen was always after the low dose. This way, participants who experienced particularly uncomfortable side effects could avoid exposure to the higher dose of arbaclofen at a later visit.

Details regarding recruitment and diagnostic screening procedures have been provided previously (Huang et al., 2022). Briefly, autistic participants were either recruited from National Autism and Attention Deficit Hyperactivity Disorder Service for Adults (NAASA) at the South London and Maudsley National Health Service (NHS) Foundation Trust, where diagnosis is a clinical decision supported by information from the Autism Diagnostic Interview-Revised, where an informant is available and/or Autism Diagnostic Observation Schedule of current features. Where diagnosis information was provided from another clinic, their diagnostic process was reviewed by an experienced clinician at the screening interview. Participants gave informed consent, according to the Declaration of Helsinki, for a protocol as approved by the King’s College London Ethics Committee (Institutional Review Board). Exclusion criteria included IQ < 70, known autism-related genetic syndromes (e.g. fragile X syndrome or 22q11 deletion syndrome), medications directly affecting GABA or glutamate, significant comorbid psychiatric illness, epilepsy, known allergies to medication components and MRI-related contraindications. A priori power calculations, informed by previous single-dose drug studies in autism, suggested that a sample size of n = 16 per group would provide adequate power (80%) to detect significant drug × group interactions and within-group drug effects at α = 0.05. However, the study concluded earlier than anticipated due to COVID-19 restrictions, resulting in slightly smaller sample sizes in some instances. Therefore, findings are interpreted with awareness of this limitation.

Demographic data, including age, biological sex, full-scale intelligence quotient and autism quotient, are provided in Table 1. Biological sex and full-scale IQ did not significantly differ between groups. As expected, there was a significant difference in autistic traits (AQ scores) between the autistic and non-autistic group. All participants were adults between the ages of 19 and 53. Mean age of the autism group was ~9 years older than the neurotypical group and although this was a statistically significant difference, all participants were over 18 years and age was not related to aperiodic exponent at any dose in this study (all p’s > 0.05) and so was unlikely to contribute to the pattern of results reported.

Table 1.

Demographic data.

Measure	N (M/F)	Age	FSIQ	VIQ	PIQ	AQ
Non-autistic	9/16	28.88 (1.59)	122.32 (2.36)	121.41 (2.88)	118.18 (1.95)	15.78 (1.60)
Autistic	9/6	37.80 (2.53)	117.33 (2.77)	117.15 (2.52)	114.00 (3.60)	35.87 (1.81)
p-Value	0.140	0.006*	0.183	0.274	0.321	<0.001**

Demographic variables for each group are given, including number (N) of males and females (M/F). Means (standard errors) are given for age, FSIQ, VIQ, PIQ and AQ. A chi-square test was used to compare the ratio of males to females between groups; between-group t-tests were conducted to compare age, IQ, and AQ (bottom row).

AQ: autism quotient; FSIQ: full-scale intelligence quotient; PIQ: performance intelligence quotient; VIQ: verbal intelligence quotient.

Asterisks indicate statistical significance: * = p < .05; ** = p < .001.

EEG data acquisition

Scalp EEG signals were collected using a 64-channel standard actiCAP (EASYCAP GmbH) with a sampling rate of 5 kHz and amplified by a BrainAmp amplifier (Brain Products GmbH). Electrode placements followed the international 10–20 system. Impedances between the scalp and electrodes were kept below 15 kilohms. Data were recorded relative to an FCz reference, and a ground electrode was located at FPz. Participants took part in a resting-state protocol at the start of the EEG session, seated in a darkened room in front of a stimulus computer. The resting-state paradigm consisted of 6 × 1 minute trials that were either ‘eyes-open’ or ‘eyes-closed’, presented alternately, the order of which was counterbalanced across participants. During ‘eyes-closed’ trials, participants were asked to close their eyes; during ‘eyes-open’ trials, participants were asked to look at a 1-minute sand-timer.

EEG data pre-processing

Raw EEG data were downsampled to 250 Hz. MATLAB R2023b (The MathWorks, Inc., Natick, MA, USA) and EEGLAB v2024.2.1 (Swartz Center for Computational Neuroscience, University of California San Diego, La Jolla, CA, USA) were used for preprocessing. Data were band-pass filtered using zero-phase, Hamming-windowed FIR filters (high-pass ~1 Hz; low-pass ~35 Hz). Channels with very low variance were identified and removed, and additional bad channels were detected using the FASTER algorithm. A further moving-window variance procedure (5-s windows, 2-s step) was applied to detect intermittently noisy channels, which were removed if their maximum window variance exceeded 200 times their median variance. This threshold ensured only extreme residual noise was removed. Independent Component Analysis (ICA) was conducted using EEGLAB’s extended Infomax algorithm (runica). Independent components were classified with ICLabel (Pion-Tonachini et al., 2019), and components were removed if they were predominantly muscle, eye, heart, line noise, or channel noise (any of these classes ⩾0.70), or if they were classified as ‘Other’ with high probability (⩾0.97) in combination with low ‘Brain’ probability (⩽0.20). Finally, flat and bad channels were interpolated, and data were re-referenced to the average of all channels. Data were then segmented into two separate files, one containing three eyes-open trials and one containing three eyes-closed trials. All files were then converted from .set to .mat.

PSDs were computed in Python 3.8 using the Neuro Digital Signal Processing toolbox (NeuroDSP 2.1.0; University of California San Diego, San Diego, CA, USA). Channels for analysis were selected based on topographic maps of the aperiodic 1/f exponent at placebo across groups (see Figure 1(a) and (b)), which showed maximal exponents over central-midline scalp sites, consistent with previous work (Jacob et al., 2021; Wang et al., 2022) and likely reflecting contributions from midline sensorimotor and posterior regions that exhibit robust 1/f structure in resting EEG. To capture this maximum while minimising potential EMG contamination from frontalis, temporalis and neck muscles, analyses were focused on a centro-central region of interest and averaged across FC3, FC1, FC2, FC4, C3, C1, Cz, C2, C4, CP3, CP1, CPz, CP2 and CP4 for all analyses. For each trial, these channels were averaged to form a single time series.

Figure 1.

Topographic maps of aperiodic exponents across the scalp under the placebo condition. Scalp distribution of aperiodic fit parameters indicates a midline, central to posterior parietal maximum for the aperiodic exponent in eyes-closed trials (a) and a midline maximum for the aperiodic exponent in eyes-open trials (b).

The PSD for each trial was then estimated using Welch’s method (2-s windows, 50% overlap), restricted to 3–32 Hz with a consistent frequency grid enforced across trials. This 3–32 Hz range was chosen to exclude very low frequencies dominated by slow drifts and non-neural artefacts, and very high frequencies that are more susceptible to EMG contamination and line noise, while retaining a band in which the log-log spectrum is approximately linear and suitable for estimating a single aperiodic exponent. Aperiodic exponents were extracted using the Fitting Oscillations and One-Over-F (FOOOF 1.0.0) model (Donoghue et al., 2020b). Model parameters, selected based on extensive data quality checks and agreement between two experienced EEG researchers, were: peak_width_limits = [1, 8], min_peak_height = 0.15, max_n_peaks = 8, peak_threshold = 1.5, and aperiodic_mode = ‘fixed’. In the present data, these settings restricted oscillatory peaks to plausible widths, ensured that only peaks clearly rising above the aperiodic background were modelled, limited the number of peaks to avoid overfitting noise, and used a fixed aperiodic component (a single 1/f slope without a knee) in the 3–32 Hz range.

Quality control

Because arbaclofen can reduce skeletal muscle tone and cranial EMG contributes high-frequency activity to EEG, additional quality control analyses were conducted to assess whether EMG could plausibly confound the planned aperiodic analyses. First, during preprocessing, independent components were classified with ICLabel and components labelled as muscle were removed; for each dataset, the number of muscle ICs rejected was quantified and inspected across dose and group. Second, to probe potential EMG-related changes in the scalp signal, log(10)-transformed power in a high-frequency band (20–32 Hz) was computed over frontal, temporal, occipital and central regions of interest, separately for eyes-closed and eyes-open trials, and analysed with linear mixed-effects models. Third, FOOOF fit quality was evaluated via the coefficient of determination ( $R^{2}$ ); only spectra with $R^{2} > 0.90$ were retained for the primary aperiodic exponent models, which led to the exclusion of three eyes-open observations (one autistic placebo, one autistic 15 mg, one non-autistic 15 mg); all remaining sessions met the R² cutoff. Because the linear mixed-effects models used here can accommodate missing cells, all participants were retained in the final models. Across these checks, there was no evidence that dose systematically reduced the number of muscle ICs and no consistent pattern of dose-related decreases in high-frequency power, suggesting that EMG is unlikely to have systematically biased the reported aperiodic results. Full details of these analyses are provided in Supplemental Results (S1-2).

Statistical analysis

Data were first quality-filtered at the level of the FOOOF fits: only epochs with aperiodic 1/f exponents and goodness-of-fit $R^{2} > 0.90$ were retained for analysis. To reduce the influence of extreme values while preserving the original scale, aperiodic exponents were winsorised within trial type (eyes open vs eyes closed) at the 5th and 95th percentiles, and these winsorised values were used as the outcome. Linear mixed-effects models were fitted in R using lmer, with a random intercept for subject. An initial model included fixed effects of group (autistic, non-autistic), dose (0, 15, 30 mg arbaclofen), trial (eyes closed, eyes open), and all interactions. The primary model included fixed effects of group, dose, and trial, plus the group × dose interaction:

aperiodic_winsorised ~ group * dose + trial + (1 | subject)

Fixed effects were evaluated using Type III tests with Satterthwaite-approximated degrees of freedom. Estimated marginal means were obtained using emmeans. Prespecified follow-up contrasts tested (i) within-group differences from placebo (15 vs 0 mg; 30 vs 0 mg), (ii) group differences in dose-related change at each active dose, and (iii) overall dose effects collapsed across group. For these contrasts, we report estimated marginal mean differences, 95% confidence intervals, unadjusted p-values, and standardised d-equivalent effect sizes derived from the contrast $t$ -statistics and denominator degrees of freedom.

Results

A linear mixed-effects model was first fitted with fixed effects of Group (autistic, non-autistic), Dose (0, 15, 30 mg) and Trial (eyes-closed, eyes-open), including all interactions, and a random intercept for subject. In this full model, there were significant main effects of Dose, F(2, 147.85) = 11.26, p < 0.001, and Trial, F(1, 143.99) = 12.19, p = 0.001, as well as a Group × Dose interaction, F(2, 147.85) = 4.03, p = 0.020. Group, Group × Trial, Dose × Trial, and the three-way interaction were not significant (all p ⩾ 0.487), providing no evidence that the effect of dose differed by trial type. Estimated marginal means from the full model showed that, across groups and doses, aperiodic exponents were higher in eyes-closed trials (EMM = 1.36, SE = 0.05, 95% CI [1.27, 1.46]) than in eyes-open trials (EMM = 1.27, SE = 0.05, 95% CI [1.17, 1.36]).

Because trial type did not interact with dose or group, the main hypothesis concerning how arbaclofen dose affects aperiodic exponents in autistic versus non-autistic participants was addressed with a simplified model including fixed effects of Group, Dose and Trial (main effects only) plus the Group × Dose interaction, and a random intercept for subject. In this reduced model, the main effects of Dose, F(2, 152.88) = 11.46, p < 0.001, and Trial, F(1, 148.98) = 11.54, p = 0.001, remained significant, Group again was not significant (F(1, 38.30) = 0.01, p = 0.921), and the Group × Dose interaction was significant, F(2, 152.89) = 4.19, p = 0.017, indicating that dose-related changes differed between groups (see Table 2).

Table 2.

Aperiodic 1/f exponents of autistic and non-autistic participants, at placebo, 15 mg of arbaclofen and 30 mg of arbaclofen.

Dose	Non-autistic (n = 25)	Autistic (n = 15)	Total
Placebo	1.287 (0.061) [1.165, 1.409]	1.248 (0.077) [1.093, 1.403]	1.268 (0.049) [1.169, 1.366]
15 mg of arbaclofen	1.214 (0.061) [1.091, 1.336]	1.314 (0.076) [1.160, 1.467]	1.264 (0.049) [1.166, 1.362]
30 mg of arbaclofen	1.457 (0.065) [1.328, 1.586]	1.369 (0.079) [1.211, 1.526]	1.413 (0.051) [1.311, 1.515]
Total	1.319 (0.056) [1.205, 1.433]	1.310 (0.071) [1.165, 1.455]

Estimated marginal means (and standard errors), with 95% confidence intervals in brackets, of aperiodic 1/f exponents for each group at placebo, 15 mg of arbaclofen and 30 mg of arbaclofen. Totals are given for each group across drug doses (bottom row) and for each drug dose, across groups (rightmost column)

Planned contrasts were used to characterise the Dose and Group × Dose effects (see Table 3 for full statistics and confidence intervals). In the non-autistic group, aperiodic exponents were higher at 30 mg than placebo (Δ = 0.17, p < 0.001, d = 0.59), with a small non-significant decrease at 15 mg (Δ = −0.07, p = 0.091). In the autistic group, exponents were also higher at 30 mg than placebo (Δ = 0.12, p = 0.024, d = 0.37), with a small, non-significant increase at 15 mg (Δ = 0.07, p = 0.186). Collapsing across groups, 30 mg arbaclofen significantly increased aperiodic exponents relative to placebo (Δ = 0.15, p < 0.001, d = 0.67), whereas 15 mg showed no overall effect (Δ ≈ 0, p = 0.902). The Group × Dose interaction at 15 mg reflected a tendency for exponents to decrease in the non-autistic group but increase in the autistic group (contrast of dose-related changes: Δ = 0.14, p = 0.035, d = 0.34), whereas the corresponding interaction at 30 mg was not significant (p = 0.482; Figure 2).

Table 3.

Results from planned contrasts on aperiodic 1/f exponents.

Planned contrast	ΔEMM (SE) [95% CI]	t	p	d-Equivalent
Autistic 15 mg vs placebo	0.065 (0.049) [−0.032, 0.162]	1.33	0.186	0.218
Autistic 30 mg vs placebo	0.121 (0.053) [0.016, 0.225]	2.28	0.024*	0.371
Non-autistic 15 mg vs placebo	−0.073 (0.043) [−0.159, 0.012]	−1.70	0.091	0.271
Non-autistic 30 mg vs placebo	0.170 (0.046) [0.079, 0.262]	3.68	<0.001**	0.593
Main effect of 15 mg	−0.004 (0.033) [−0.069, 0.061]	−0.12	0.902	0.020
Main effect of 30 mg	0.145 (0.035) [0.076, 0.215]	4.13	<0.001**	0.670
Group × Dose interaction at 15 mg	0.139 (0.065) [0.010, 0.268]	2.12	0.035*	0.344
Group × Dose interaction at 30 mg	−0.050 (0.070) [−0.189, 0.089]	−0.71	0.482	0.114

ΔEMM = difference in estimated marginal means from the mixed-effects model. d indexes the standardised effect size derived from the contrast test statistic, with values of ~0.20, ~0.50 and ~0.80 interpreted as small, medium and large effects, respectively.

Asterisks indicate statistical significance: * = p < .05; ** = p < .001.

Figure 2.

Aperiodic responses to drug dose in autistic and non-autistic people. (a) Mean aperiodic 1/f exponents (±Standard error of the mean) for non-autistic (blue) and autistic (orange) participants across placebo, 15 mg and 30 mg arbaclofen, averaged over eyes-open and eyes-closed trials. Points show individual participants. (b) Raincloud plot showing aperiodic 1/f exponents at placebo and 15 mg arbaclofen. (c) Raincloud plot showing aperiodic 1/f exponents at placebo and 30 mg arbaclofen. In panels (b) and (c), half-violins depict the distribution, boxplots show median and interquartile range, and light lines link repeated measures within individuals. Higher values indicate steeper aperiodic exponents.

Discussion

This study provides PoC evidence that the aperiodic 1/f exponent of the EEG power spectrum is responsive to E-I challenge with the GABA agonist arbaclofen, and that, on average, responsivity differs between autistic and non-autistic people. Thus, our results support the utility of the aperiodic 1/f exponent as a metric sensitive to E-I dynamics and suggest differences in E-I responsivity to pharmacological modulation between autistic and non-autistic individuals. These findings may reflect differences in the homeostatic regulation of E-I dynamics in autism.

In brief, across both groups, the highest dose of arbaclofen elicited a shift (increase) in aperiodic 1/f exponents, steepening the slope. As arbaclofen increases global inhibition by blocking glutamate release pre-synaptically, and acts as a GABA_B receptor agonist post-synaptically (Nance et al., 2011; Padgett and Slesinger, 2010), these results are in line with animal findings of steeper aperiodic exponents with increased central inhibition (Gao et al., 2017). These results add rigour to existing findings using broader-action pharmacological manipulations of E-I (Waschke et al., 2021). Furthermore, as EEG was recorded at rest, our results were not confounded by task effects (Molina et al., 2020; Waschke et al., 2021).

Overall, there were no case-control group differences in aperiodic 1/f exponents between autistic and non-autistic people under placebo (baseline) or any drug condition. This is broadly consistent with accounts emphasising heterogeneity in E-I mechanisms in autism rather than a uniform shift in baseline excitation (Sohal and Rubenstein, 2019). Differences from studies reporting altered aperiodic exponents in autistic infants and children (Shuffrey et al., 2022; Manyukhina et al., 2022) may reflect variation in developmental stage, cognitive ability, and analytic approaches. In particular, prior findings have been reported in younger or lower-IQ samples, whereas the present study examined adults with above-average IQ. Developmental changes in E-I pathways (Dorrn et al., 2010) and age-related reductions in aperiodic 1/f exponents (Voytek et al., 2015), together with the heterogeneity of autism (Dickinson et al., 2016; Loth, 2016; Loth et al., 2021; Mottron and Bzdok, 2020), may contribute to variability in findings across studies.

In addition, the aperiodic 1/f exponent is a dynamic measure that captures E-I activity over time. In the placebo condition at rest, we likely capture the E-I system in a baseline state of flux. Introducing a pharmacological challenge with arbaclofen may impose different demands on the E-I systems of autistic and non-autistic individuals. In line with this, we observed a different drug response pattern between groups at 15 mg of arbaclofen; on average, aperiodic 1/f exponents steepened in the autistic group but flattened in the non-autistic group. One explanation for this could be that the compensatory mechanisms in autistic adults, which maintain E-I homeostasis at baseline, are more easily disrupted at a lower dose compared to their non-autistic peers, because their dynamic regulation is different (Sohal and Rubenstein, 2019).

That is, E-I circuits are not static; they constantly adjust in response to the environment. For example, to prevent sensory inputs from destabilising E-I circuits, tightly controlled E-I dynamics are established early in development to offset destabilisation (Dorrn et al., 2010), avoid runaway or silent activity (Litwin-Kumar and Doiron, 2012), and fine-tune sensory response (Heiss et al., 2008; Mariño et al., 2005; Poo and Isaacson, 2009; Tao and Poo, 2005; Wehr and Zador, 2003). Maintenance of E-I homeostasis is achieved by a set of synaptic plasticity mechanisms, such as synaptic scaling; whereby E or I synaptic strength is adjusted up or down to stabilise firing rate (Turrigiano, 1999, 2012, 2011; Turrigiano and Nelson, 2004). Our results suggest this homeostatic response may be different in autism.

An alternative, but not mutually exclusive, possibility is that the 15 mg effect in the autistic group reflects a broader difference in pharmacological sensitivity rather than a process specific to E-I challenge. Autistic individuals are often reported to show atypical dose-response profiles and tolerability to psychotropic medications (Aishworiya et al., 2022; Carthy et al., 2023; Davico et al., 2023). Recent pharmaco-neuroimaging work also suggests systematically different patterns of network-level responsivity to acute GABAergic, serotonergic and μ-opioid challenges in autistic versus non-autistic adults (Whelan, Franca et al., 2024). However, broader differences in pharmacological responsivity need not be independent of E-I mechanisms. E-I balance is a fundamental property of circuit regulation that shapes neural gain and responsiveness (Sohal and Rubenstein, 2019; Turrigiano, 2012; Nelson and Valakh, 2015), and altered E-I dynamics in autism could therefore influence responses to pharmacological perturbation more generally. In this sense, apparently ‘general’ differences in drug responsivity may partly reflect underlying differences in circuit regulation. Future studies should test this directly by examining whether individual differences in aperiodic dynamics predict responsivity not only to GABAergic challenge, but also to pharmacological probes acting on other receptor systems.

Application of the aperiodic 1/f exponent

The utility of the aperiodic 1/f exponent, as a metric of E-I, for psychiatric research has been questioned (Bruining et al., 2020), as there is no absolute value to indicate whether someone has an E or I dominant regime. In addition, theoretical and biophysical work suggests that several mechanisms may contribute to 1/f-like spectral scaling, including the morphology and filtering properties of dendrites and the capacitive nature of the extracellular medium (Buzsáki et al., 2012). More recent modelling studies have begun to explore how the aperiodic exponent relates to synaptic E-I balance and indicate that this relationship may be non-linear and context dependent rather than strictly one-to-one (Brake et al., 2024). These considerations imply that inter-individual variability in the aperiodic exponent will inevitably reflect a mixture of E-I-related and other factors, and that pharmacological manipulations such as arbaclofen are likely to influence only part of this variance.

However, at rest, ‘somehow the unstable stuff of which we are composed has learned the trick of maintaining stability’ (Cannon, 1932). We would argue that the utility of this metric lies in its ability to capture a shift from stability, that is, brain dynamics. We have shown that, in combination with single-dose drug challenge designs which in our lab we have termed ‘shiftability studies’, the aperiodic 1/f exponent can expose dose-dependent differences in E-I flux (Whelan, Daly et al., 2024). Importantly, this study did not include clinical outcome measures, so our data do not indicate whether larger or smaller aperiodic shifts are beneficial. At present, we therefore interpret these changes as indices of target engagement and individual differences in neural responsivity to E-I relevant pharmacology, rather than as markers of treatment response. Prior clinical trials of arbaclofen in autism have produced largely negative or mixed results (Berry-Kravis et al., 2017), suggesting that more refined approaches to matching individuals to treatment are needed. If future work can link aperiodic ‘shiftability’ to clinical benefit or tolerability, it could provide one way to improve stratification and trial design for arbaclofen and related compounds.

Potential contribution of muscle activity

A further consideration is whether the observed changes in aperiodic exponent could be explained by arbaclofen-related changes in scalp muscle activity, given that baclofen-class drugs can reduce muscle tone (Ertzgaard et al., 2017) and high-frequency EEG is vulnerable to EMG contamination (Pope et al., 2022; Whitham et al., 2007). Several aspects of the design and control analyses argue against a purely myogenic account. First, the main analyses were deliberately restricted to central electrodes, which are relatively distant from frontalis, temporalis and neck muscles and therefore less susceptible to EMG than lateral frontal or temporal leads. Second, when high-frequency (20–32 Hz) power was examined over frontal, temporal and occipital scalp regions, there was no consistent dose-related reduction that would be expected if arbaclofen simply dampened muscle activity; instead, small increases in this band were seen at some sites. These increases could in principle relate to neural beta-band changes (see Ahmad et al., 2022), but beta activity was not a prespecified outcome, and the present quality-control analyses are not designed to support strong conclusions about its physiological meaning. Third, the number of independent components labelled as ‘muscle’ by ICLabel and removed during preprocessing was stable across doses. Taken together, these observations suggest that while residual EMG contamination can never be completely excluded, it is unlikely that the dose-dependent modulation of the aperiodic exponent in the central ROI is driven primarily by changes in muscle activity.

Limitations

This study provides PoC evidence for the feasibility of using the aperiodic exponent to study E-I dynamics in neurodiverse populations. However, arbaclofen is not without side effects, which can impact data collection. In our study, these were restricted to its known side effects (particularly dizziness and nausea). We adapted the study design to minimise the chances of discomfort; if a participant had uncomfortable side effects at a low dose, they did not attend the high dose visit (see methods). To accommodate this sample size variation, results were interpreted with their corresponding effect sizes, which were consistently medium to large. Furthermore, the current sample size far exceeds that used in the preclinical study we sought to translate (Gao et al., 2017). Indeed, as this analysis was primarily undertaken to assess translation of animal evidence for 1/f as a measure of E-I dynamics in humans (autistic and non-autistic): to that end, the results from this sample support our aim. Finally, although the study was pre-registered for the main drug effects, the specific group-by-dose interaction patterns were not pre-specified. Given the modest sample size and multiple contrasts, the group-specific findings should be regarded as preliminary.

In conclusion, here we provide PoC evidence that the aperiodic 1/f exponent is responsive to E-I pharmacological challenge in autistic and non-autistic people. Critically, these results were achieved with a non-invasive, cheap method in combination with a completely passive task that is not cognitively demanding. This opens up multiple opportunities for future research. Within autism, for example, there may be scope to extend this tool to ensure those with intellectual disability are included in research into E-I. We show that, in humans, the aperiodic 1/f exponent indexes individual E-I dynamics and opens the potential for this metric to be adopted in research which target E-I homeostasis for potential pharmacological support options for neurodevelopmental and/or other psychiatric conditions with E-I alterations. Because the aperiodic 1/f signal is observed at multiple different scales (i.e. Local Field Potential, Electrocorticography and EEG) and across species, it could help bridge the translational gap that so often separates preclinical and clinical neuroscience research.

Supplemental Material

sj-docx-1-jop-10.1177_02698811261449378 – Supplemental material for Dynamic excitatory-inhibitory differences in autistic and non-autistic adults: Evidence from a pharmacological challenge with arbaclofen

Supplemental material, sj-docx-1-jop-10.1177_02698811261449378 for Dynamic excitatory-inhibitory differences in autistic and non-autistic adults: Evidence from a pharmacological challenge with arbaclofen by Claire Louise Ellis, Jumana Ahmad, Alexia Zoumpoulaki, Mihail Dimitrov, Hester E. Velthuis, Andreia C. Pereira, Nichol M. L. Wong, Maria F. Ponteduro, Lukasz Kowalewski, Alison Leonard, Pilar Garces, Qiyun Huang, Eileen Daly, Declan G. M. Murphy and Gráinne M. McAlonan in Journal of Psychopharmacology

Footnotes

ORCID iDs

Claire Louise Ellis

Mihail Dimitrov

Andreia C. Pereira

Nichol M. L. Wong

Qiyun Huang

Eileen Daly

Author contributions

Concept and design of study: GMA, DM, JA, CLE; Data acquisition: MD, HV, ACP, NMLW, MFP, LK, AL and CLE. Analysis or interpretation of data: CLE, JA, DM, GMA. Drafting of the manuscript: CLE. Critical revision of the manuscript for important intellectual content: All authors. Technical support: AZ, MD, PG, QH. Supervision: JA, GMA, DM.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was funded by an Independent Investigator Award (G.M.M.) from the Brain and Behaviour Research Foundation and by Clinical Research Associates, L.L.C. (CRA), an affiliate of the Simons Foundation. Support is also acknowledged from Autistica and the Institute for Translational Neurodevelopment at King’s College London and EU- AIMS (European Autism Interventions)/ AIMS-2-TRIALS, an Innovative Medicines Initiative Joint Undertaking under Grant Agreement No. 777394. In addition, this paper represents independent research part funded by the NIHR-Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.

Declaration of conflicting interests

The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Prof. Murphy has received consultancy fees from F. Hoffmann–La Roche and Servier. Prof McAlonan has received consultancy fees from Greenwich Biosciences and funding for an Investigator-Initiated study from Compass Pathways Ltd. Prof McAlonan and Prof Murphy are supported by the Institute for Translational Neurodevelopment, the MRC Centre for Neurodevelopmental Disorders and the NIHR-Maudsley Biomedical Research Center. P. Garces is an employee of F. Hoffmann–La Roche Ltd. There are no other declarations.

Supplemental material

Supplemental material for this article is available online.

References

Ahmad

Ellis

Leech

, et al. (2022) From mechanisms to markers: Novel noninvasive EEG proxy markers of the neural excitation and inhibition system in humans. Translational Psychiatry 12(1): 467.

Aishworiya

Valica

Hagerman

, et al. (2022) An update on psychopharmacological treatment of autism spectrum disorder. Neurotherapeutics 19(1): 248–262.

Berry-Kravis

Hagerman

Visootsak

, et al. (2017) Arbaclofen in fragile X syndrome: Results of phase 3 trials. Journal of Neurodevelopmental Disorders 9(3): 3.

Brake

Duc

Rokos

, et al. (2024) A neurophysiological basis for aperiodic EEG and the background spectral trend. Nature Communications 15: 1514.

Bridi

MCD

Zong

Min

, et al. (2020) Daily oscillation of the excitation-inhibition balance in visual cortical circuits. Neuron 105(4): 621–629.e4.

Bruining

Hardstone

Juarez-Martinez

, et al. (2020) Measurement of excitation-inhibition ratio in autism spectrum disorder using critical brain dynamics. Scientific Reports 10(1): 1–15.

Brunwasser

Hengen

(2020) Currently unstable: Daily ups and downs in E-I balance. Neuron 105(4): 589–591.

Buzsáki

Anastassiou

Koch

(2012) The origin of extracellular fields and currents-EEG, ECoG, LFP and spikes. Nature Reviews Neuroscience 13(6): 407–420.

Cannon

(1932) The Wisdom of the Body. W. W. Norton & Company.

10.

Cao

Tabuchi

(2017) Functions of synapse adhesion molecules neurexin / neuroligins and neurodevelopmental disorders. Neuroscience Research 116: 3–9.

11.

Carthy

Ross

Murphy

(2023) Psychotropic medication prescribing in people with autism spectrum disorders with and without psychiatric comorbidity. BJPsych Advances 29(2): 131–140.

12.

Davico

Amianto

Gaiotti

, et al. (2023) Pharmacological treatments in autism spectrum disorder: A narrative review. Journal of Psychopathology 29(2): 98–111.

13.

Dickinson

Jones

Milne

(2016) Measuring neural excitation and inhibition in autism: Different approaches, different findings and different interpretations. Brain Research 1648: 277–289.

14.

Donoghue

Dominguez

Voytek

(2020a) Electrophysiological frequency band ratio measures conflate periodic and aperiodic neural activity. ENeuro 7(6): 1–14.

15.

Donoghue

Haller

Peterson

, et al. (2020b) Parameterizing neural power spectra into periodic and aperiodic components. Nature Neuroscience 23(12): 1655–1665.

16.

Dorrn

Yuan

Barker

, et al. (2010) Developmental sensory experience balances cortical excitation and inhibition. Nature 465(7300): 932–936.

17.

Ertzgaard

Campo

Calabrese

(2017) Efficacy and safety of oral baclofen in the management of spasticity: A rationale for intrathecal baclofen. Journal of Rehabilitation Medicine 49(3): 193–203.

18.

Gao

Penzes

(2016) Common mechanisms of excitatory and inhibitory imbalance in schizophrenia and autism spectrum disorders. Current Molecular Medicine 15(2): 146–167.

19.

Gao

Peterson

Voytek

(2017). Inferring synaptic excitation/inhibition balance from field potentials. NeuroImage 158: 70–78.

20.

Donoghue

Sowman

, et al. (2019) Co-Increasing neuronal noise and beta power in the developing brain. bioRxiv. DOI: 10.1101/839258.

21.

Heiss

Katz

Ganmor

, et al. (2008) Shift in the balance between excitation and inhibition during sensory adaptation of S1 neurons. Journal of Neuroscience 28(49): 13320–13330.

22.

Huang

Pereira

Velthuis

, et al. (2022) GABA B receptor modulation of visual sensory processing in adults with and without autism spectrum disorder. Science Translational Medicine 14(626): eabg7859.

23.

Jacob

Roach

Sargent

, et al. (2021) Aperiodic measures of neural excitability are associated with anticorrelated hemodynamic networks at rest: A combined EEG-fMRI study. NeuroImage 245(January): 118705.

24.

Lim

Goldman

(2013) Balanced cortical microcircuitry for maintaining information in working memory. Nature Neuroscience 16(9): 1306–1314.

25.

Litwin-Kumar

Doiron

(2012) Slow dynamics and high variability in balanced cortical networks with clustered connections. Nature Neuroscience 15(11): 1498–1505.

26.

Loth

(2016) Identification and validation of biomarkers for autism spectrum disorders. Nature Reviews Drug Discovery 15(1):70–73.

27.

Loth

Ahmad

Chatham

, et al. (2021) The meaning of significant mean group differences for biomarker discovery. PLoS Computational Biology 17(11): 1–16.

28.

Lystad

Pollard

(2009). Functional neuroimaging: A brief overview and feasibility for use in chiropractic research. The Journal of the Canadian Chiropractic Association 53(1): 59–72.

29.

Manyukhina

Prokofyev

Galuta

, et al. (2022) Globally elevated excitation–inhibition ratio in children with autism spectrum disorder and below-average intelligence. Molecular Autism 13(1): 1–14.

30.

Mariño

Schummers

Lyon

, et al. (2005) Invariant computations in local cortical networks with balanced excitation and inhibition. Nature Neuroscience 8(2): 194–201.

31.

Molina

Voytek

Thomas

, et al. (2020) Memantine effects on electroencephalographic measures of putative excitatory/inhibitory balance in schizophrenia. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 5(6): 562–568.

32.

Mottron

Bzdok

(2020) Autism spectrum heterogeneity: Fact or artifact? Molecular Psychiatry 25(12): 3178–3185.

33.

Nance

Huff

Martinez-Arizala

, et al. (2011) Efficacy and safety study of arbaclofen placarbil in patients with spasticity due to spinal cord injury. Spinal Cord 49(9): 974–980.

34.

Nelson

Valakh

(2015) Excitatory/inhibitory balance and circuit homeostasis in autism spectrum disorders. Neuron 87(4): 684–698.

35.

Niethard

Hasegawa

Itokazu

, et al. (2016) Sleep-stage-specific regulation of cortical excitation and inhibition. Current Biology 26(20): 2739–2749.

36.

Okun

Lampl

(2008) Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities. Nature Neuroscience 11(5): 535–537.

37.

Padgett

Slesinger

(2010) GABAB receptor coupling to G-proteins and ion channels. Advances in Pharmacology 58: 123–147.

38.

Peterson

Rosen

Belger

, et al. (2023) Aperiodic neural activity is a better predictor of Schizophrenia than neural oscillations. Clinical EEG and Neuroscience 54(4): 434–445.

39.

Pfister

Kaynig

Botha

, et al. (2014) Visualization in connectomics. In: Hansen

Chen

Johnson

, et al. (eds) Scientific Visualization. London: Springer, pp. 221–245.

40.

Pion-Tonachini

Kreutz-Delgado

Makeig

(2019) ICLabel: An automated electroencephalographic independent component classifier, dataset, and website. NeuroImage 198: 181–197.

41.

Poo

Isaacson

(2009) Odor representations in olfactory cortex: “Sparse” coding, global inhibition, and oscillations. Neuron 62(6): 850–861.

42.

Pope

Lewis

Fitzgibbon

, et al. (2022) Managing electromyogram contamination in scalp recordings: An approach identifying reliable beta and gamma EEG features of psychoses or other disorders. Brain Behav 12(9): e2721.

43.

Rubin

Abbott

Sompolinsky

(2017) Balanced excitation and inhibition are required for high-capacity, noise-robust neuronal selectivity. Proceedings of the National Academy of Sciences of the United States of America 114(44): E9366–E9375.

44.

Shuffrey

Pini

Potter

, et al. (2022) Aperiodic electrophysiological activity in preterm infants is linked to subsequent autism risk. Developmental Psychobiology 64(4): e22271.

45.

Sohal

Rubenstein

JLR

(2019) Excitation-inhibition balance as a framework for investigating mechanisms in neuropsychiatric disorders. Molecular Psychiatry 24(9): 1248–1257.

46.

Südhof

(2008) Neuroligins and neurexins link synaptic function to cognitive disease. Nature 455(7215): 903–911.

47.

Szücs

Huerta

(2015) Differential effects of static and dynamic inputs on neuronal excitability. Journal of Neurophysiology 113(1): 232–243.

48.

Tao

Poo

(2005) Activity-dependent matching of excitatory and inhibitory inputs during refinement of visual receptive fields. Neuron 45: 829–836.

49.

Tatavarty

Torrado Pacheco

Groves Kuhnle

, et al. (2020) Autism-associated Shank3 is essential for homeostatic compensation in Rodent V1. Neuron 106(5): 769–777.

50.

Toro

Konyukh

Delorme

, et al. (2010) Key role for gene dosage and synaptic homeostasis in autism spectrum disorders. Trends in Genetics 26(8): 363–372.

51.

Tran

Rolle

Gazzaley

, et al. (2020) Linked sources of neural noise contribute to age-related cognitive decline. Journal of Cognitive Neuroscience 32(9): 1813–1822.

52.

Turrigiano

(2011) Too many cooks? Intrinsic and synaptic homeostatic mechanisms in cortical circuit refinement. Annual Review of Neuroscience 34: 89–103.

53.

Turrigiano

(2012) Homeostatic synaptic plasticity: Local and global mechanisms for stabilizing neuronal function. Cold Spring Harbor Perspectives in Biology 4(1): 1–18.

54.

Turrigiano

(1999) Homeostatic plasticity in neuronal networks: The more things change, the more they stay the same. Trends in Neurosciences 22(5): 221–227.

55.

Turrigiano

Nelson

(2004) Homeostatic plasticity in the developing nervous system. Nature Reviews Neuroscience 5(2): 97–107.

56.

Veerakumar

Tiruvadi

Howell

, et al. (2019) Field potential 1/f activity in the subcallosal cingulate region as a candidate signal for monitoring deep brain stimulation for treatment-resistant depression. Journal of Neurophysiology 122(3): 1023–1035.

57.

Vogels

Sprekeler

Zenke

, et al. (2011) Inhibitory plasticity balances excitation and inhibition in sensory pathways and memory networks. Science 334(6062): 1569–1573.

58.

Voytek

Kramer

Case

, et al. (2015) Age-related changes in 1/f neural electrophysiological noise. Journal of Neuroscience 35(38): 13257–13265.

59.

Wang

Sun

, et al. (2022) Separating the aperiodic and periodic components of neural activity in Parkinson’s disease. European Journal of Neuroscience 56(6):4889–4900.

60.

Waschke

Donoghue

Fiedler

, et al. (2021) Modality-specific tracking of attention and sensory statistics in the human electrophysiological spectral exponent. ELIFE 10: e70068.

61.

Wehr

Zador

(2003) Balanced inhibition underlies tuning and sharpens spike timing in auditory cortex. Nature 426(6965): 442–446.

62.

Whelan

Daly

Puts

, et al. (2024) Editorial perspective: Bridging the translational neuroscience gap in autism- development of the “shiftability” paradigm. Journal of Child Psychology Psychiatry 65(6): 862–865.

63.

Whelan

Franca

LGS

Dimitrov

, et al. (2024) Autistic neural shiftability: The distinct pharmacological landscape of the autistic brain. medRxiv. DOI: 10.1101/2024.11.08.24316969.

64.

Whitham

Pope

Fitzgibbon

, et al. (2007) Scalp electrical recording during paralysis: Quantitative evidence that EEG frequencies above 20 Hz are contaminated by EMG. Clinical Neurophysiology 118(8): 1877–1888.

65.

Yizhar

Fenno

Prigge

, et al. (2011) Neocortical excitation/inhibition balance in information processing and social dysfunction. Nature 477(7363): 171–178.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.18 MB