Revisiting the excitation/inhibition imbalance hypothesis of ASD through a clinical lens

Abstract

Autism spectrum disorder (ASD) currently affects 1 in 59 children, although the aetiology of this disorder remains unknown. Faced with multiple seemingly disparate and noncontiguous neurobiological alterations, Rubenstein and Merzenich hypothesized that imbalances between excitatory and inhibitory neurosignaling (E/I imbalance) underlie ASD. Since this initial statement, there has been a major focus examining this exact topic spanning both clinical and preclinical realms.

The purpose of this article is to review the clinical neuroimaging literature surrounding E/I imbalance as an aetiology of ASD. Evidence for E/I imbalance is presented from several complementary clinical techniques including magnetic resonance spectroscopy, magnetoencephalography and transcranial magnetic stimulation. Additionally, two GABAergic potential interventions for ASD, which explicitly attempt to remediate E/I imbalance, are reviewed.

The current literature suggests E/I imbalance as a useful framework for discussing the neurobiological etiology of ASD in at least a subset of affected individuals. While not constituting a completely unifying aetiology, E/I imbalance may be relevant as one of several underlying neuropathophysiologies that differentially affect individuals with ASD. Such statements do not diminish the value of the E/I imbalance concept—instead they suggest a possible role for the characterization of E/I imbalance, as well as other underlying neuropathophysiologies, in the biologically-based subtyping of individuals with ASD for potential applications including clinical trial enrichment as well as treatment triage.

Introduction

Autism spectrum disorder (ASD) is characterized by social and communication impairments, as well as restricted and stereotyped behaviors.¹ Current prevalence estimates suggest that 1 in 59 children aged 8 years old are diagnosed with ASD,² though the estimated prevalence has increased steadily over the last decade.^3,4 The pathophysiology of ASD has been intensely scrutinized, leading to the identification of many seemingly disparate neurobiological alterations. In an effort to consolidate the multitude of neurobiological perturbations observed in ASD into a potential unifying etiology that ultimately leads to the characteristic constellation of behavioral phenotypes; seminal work from Rubenstein and Merzenich suggested imbalances between excitatory and inhibitory neurosignaling (E/I imbalance) as a potential neuropathophysiology.⁵ The proposal of this hypothesis, along with the rapid technical development of clinical imaging-related modalities, has led to intense investigation of ASD in humans, drawing physiological/biochemical inference from the clinical modalities in a fashion informed by the Rubenstein and Merzenich model. Given the grounding of the Rubenstein and Merzenich thesis, the current review will not discuss preclinical literature unless it is of direct relevance to the clinical-based discussion. Instead, the purpose of the current review is to summarize the recent clinical neuroimaging evidence that examines E/I imbalance in individuals with ASD. This review will present data that are interpretable in terms of E/I imbalance within ASD from three modalities: magnetic resonance spectroscopy (MRS), electrophysiology (primarily magnetoencephalography, MEG), and transcranial magnetic stimulation (TMS). For the majority of studies reviewed, subjects were high functioning individuals with ASD with no concurrent intellectual disability, epilepsy or other neuropsychiatric disorder. As such, the resultant conclusions may not necessarily generalize to individuals with ASD that exhibit more severe neurological, neuropsychiatric or cognitive perturbations.

This review concludes that while the existing literature contains considerable inconsistencies (both at the intra- and intersubject levels) with E/I imbalance as the sole unifying neurobiological aetiology of ASD—it may be that a subset of individuals with ASD exhibit such E/I imbalance as their primary pathology, and therefore E/I imbalance may potentially be used as a stratification ‘biomarker’ (a biologically-based marker) for clinical trial enrichment or early treatment triage.

MRS

Advances in MRS have allowed for the direct non-invasive in-vivo estimation of both glutamate and also gamma-aminobutyric acid (GABA),⁶ the principal excitatory and inhibitory neurotransmitters in the central nervous system respectively (Figure 1). Such non-invasive brain tissue assays are crucial, as blood plasma levels of glutamate and GABA may not accurately reflect the concurrent neural levels,^7,8 though this lack of fidelity has recently been disputed.⁹ Glutamate, or more commonly “Glx” (the summation of the overlapping glutamate and glutamine resonances), may be estimated using standard STEAM or PRESS sequences at 3 T. Moreover, at 7 T the glutamate and glutamine peaks may become independently resolvable, although the vast majority of research has not yet exploited 7 T hardware but rather operates at the more clinically established 3 T. GABA may be estimated at 3 T via “edited” spectroscopy techniques such as MEGA-PRESS,⁶ capitalizing on the J-coupling inherent to the GABA molecule. MEGA-PRESS also allows estimation of Glx.

Figure 1.

Example spectra of neurometabolites collected using STEAM (top) and MEGA-PRESS edited (bottom) spectroscopy at 3 T. Top trace shows the neurometabolites present in a short TE STEAM spectrum, with black denoting the actual collected MRS spectrum, green denoting the baseline estimation for quantification, and red denoting the fitted neurometabolite contribution estimates. Note that several separate resonances correspond to Glx and that within a domain (especially ~2.2–2.5 ppm) multiple resonances overlap. Bottom trace shows the finalized edited spectrum from a MEGA-PRESS sequence. Note that only one component of Glx is apparent but that many overlapping peaks have been eliminated. Additionally, the simplicity of MEGA-PRESS derived edited spectra permits relatively simpler quantification of the neurometabolites resonances as compared to STEAM spectra. Cr/PCr = Creatine/Phosphocreatine; Glx = Glutamine/Glutamate; MI = Myo-inositol; Cho = Choline; NAA = N-acetylaspartate; GABA+=GABA with co-edited macromolecular contamination. MRS, magnetic resonance spectroscopy.

Of note, although absolute quantification of Glx or GABA (through comparison to a water reference spectrum) is, at least in principle, possible, the vast majority of studies rather attempt relative quantification. While some studies report Glx and GABA levels in arbitrary or institutional units (a.u. or i.u. respectively), most prefer to recognize the spatial sensitivity variation of modern MRI radiofrequency signal detection coils by “normalizing” the Glx or GABA resonance integrals to a reference metabolite derived from the same spatial location.

Readily available (high signal-to-noise ratio) choices include creatine (Cr, sometimes notated Cre) or N-acetyl aspartate (NAA). Gaetz et al¹⁰ reported slightly superior coefficients of variation using Cr compared to NAA as the normalizing denominator in healthy adults.

NAA is less-preferred for the normalization of Glx/GABA in studies of ASD that include children, as NAA levels may differentially mature between cohorts.¹¹ Potential differential Cr level maturation (although Cr is widely assumed to be stable), as well as potential differences in Cr levels between diagnostic groups, may similarly confound the interpretation of Cr-normalized studies, but nonetheless Cr is the most widely adopted measure in reported studies.^10,12–18 Unfortunately, it remains unclear how differences in the reported metric (including normalization) for either Glx or GABA account for, or least contribute to, the observed variance within the literature examining MRS in ASD.

Reports of alterations to Glx/glutamate levels in ASD appear largely inconsistent. Such inconsistencies likely stem from¹ the different MRS methodologies employed,² differences in the exact metric reported (see above), as well as³ the heterogeneous nature of the subjects’ demographics and characterization between studies. Increased Glx and glutamate levels have been reported within cortical structures such as frontal lobe,⁹ anterior cingulate cortex (ACC),^9,19,20 temporal lobe,²¹ and occipital lobe¹⁷ in ASD. Contrary to these observations of increased levels in ASD, reduced Glx and/or glutamate levels in ASD have also been reported for many of the same regions (frontal lobe,^18,22 ACC,^23,24 occipital lobe²²). Further contributing to the confusion, several studies have reported no significant alterations of Glx and/or glutamate levels in ASD for these same cortical structures (frontal lobe,^16,25–33 ACC,^15,34 temporal lobe,²² parietal lobe^{23,26,28,29,35}).

A similar level of inconsistency has been reported in studies of the cerebellum, with increases,⁹ decreases²² or alternatively no change^24,30,34 in Glx and/or glutamate levels observed in ASD. This theme of inconsistent reports surrounding Glx and/or glutamate level alterations in ASD extends to subcortical regions. Increased Glx/glutamate levels in ASD have been reported for basal ganglia^9,36 as well as hippocampus.³⁵ Contrary to this, studies targeting similar (but not identical) voxel placement have observed no alteration of Glx/glutamate levels within the basal ganglia²⁵ and hippocampus¹⁹ of individuals with ASD. Separately, decreased glutamate levels in basal ganglia have been observed for ASD.^26,27 The most consistent finding revealed by Glx/glutamate-related MRS of ASD is the lack of an alteration to thalamic Glx and/or glutamate levels.^23,36,37 However, even this somewhat stable observation was recently challenged by Hegarty and colleagues,³⁸ who observed decreased Glx levels in the thalamus of affected as compared to unaffected co-twins.

As such, the changes to Glx/glutamate levels in ASD seem far from concrete. Insight into these inconsistencies arises from the corresponding preclinical literature that also employed MRS Goncalves and colleagues³⁹ observed decreased glutamate levels in mice heterozygous for NF1 (NF1^+/-) mice as compared to wild-type (WT) littermates in both hippocampus and striatum, but not prefrontal cortex. As such, alterations to glutamate levels seem regionally heterogenous even within a fixed genetic insult. Moreover, Horder and colleagues²⁷ used MRS to probe various preclinical murine models with relevance to ASD for glutamate level alterations. Increases, decreases, as well as no change in regional glutamate levels were observed between mouse models relevant to ASD and their WT littermates, and such changes were often regionally heterogeneous. Such non-invasive studies are complemented by more traditional preclinical studies, where glutamate levels are directly quantified thorough techniques such as high performance liquid chromatography. Even within these studies that utilize direct quantification of ex vivo tissue, increases,⁴⁰ decreases⁴¹ and no change^42–44 to glutamate levels are observed in relevant murine models. Moreover, regionally heterogenous glutamate levels alterations have been reported within a murine model of relevance to ASD.⁴⁵ Thus, assessment of glutamate in ASD must be regarded as variable.

There are several caveats to the above MRS observations. First, many of the above studies examined the Glx component of the MRS spectrum as a proxy for glutamate. As previously mentioned, the Glx signal is a combination of glutamate as well as glutamine resonances, and so it is currently not possible to resolve each neurometabolite’s contribution to the Glx signal within the 3 T MRS spectrum. Although several analysis pipelines nominally allow for the model-based estimation of glutamate from these spectra,⁴⁶ such quantification relies on reference basis sets, for which accuracy, relevance and completeness cannot be guaranteed. While 7 T MRS sequences are more able to separately resolve the glutamate and glutamine peaks, the availability of such ultrahigh field magnets to the clinical research community is currently sparse.

An additional caveat to estimating glutamate levels via the Glx signal arises from glutamine’s role as glutamate’s precursor in the Glutamate–Glutamine cycle (see Waagepetersen et al.⁴⁷ for comprehensive review). The Glutamate–Glutamine cycle recycles released synaptic glutamate via uptake into astrocytes and subsequent conversion to glutamine. This glutamine is then shuttled back to the presynaptic terminal where it is converted back to glutamate.⁴⁷ Therefore, the Glutamate–Glutamine cycle further confounds the interpretation of Glx’s relation to glutamate levels. For instance, glutamate level alterations caused by a perturbed efficacy of phosphate-activated glutaminase, the enzyme that converts glutamine into glutamate, would be invisible to Glx-based metrics. Furthermore, separate from neurosignaling, glutamate plays multiple functions in cell metabolism, such as linking the tricarboxylic acid (TCA) cycle and amino-acid synthesis,⁴⁸ as well as taking part in the malate–aspartate shuttle.⁴⁹ As such, even when estimated without glutamine contamination, the relative role of glutamate in neurosignaling vs metabolism is not resolvable.

Results of GABA-related MRS studies in ASD prove more consistent, at least within region. First, studies examining the occipital lobe have consistently reported no change in GABA+ levels between ASD and age-matched TD controls (^10,13,17 though this study did observe decreased GABA+/Glu ratio,⁵⁰ ). These studies reported either relative cortical GABA+ (Cr normalization^10,17) or institutional units^13,50. The GABA+ notation (as opposed to GABA) recognizes the possible contamination of the GABA signal from co-edited macromolecules, which occurs during “conventional” MEGAPRESS approaches. Some attempts to suppress macromolecular contamination within the MEGAPRESS framework have been proposed⁵¹ (and in those cases the notation GABA is used instead).

Similarly, no alterations to GABA or GABA+ levels have been observed in ASD within the striatum^25,27 and cerebellum.^30,34 On the other hand, decreased GABA+ levels in ASD have been consistently reported in pre/post-central gyrus locations.^10,13 Studies examining frontal regions (including prefrontal cortex and ACC) conflict on the presence of GABA+ level alterations in ASD. While studies of young children^25,34 or adults^27,30 report no alterations to GABA/GABA+ levels in ASD; studies sampling children approximately 10 years old report significantly/near significantly decreased GABA+/Cr levels in ASD.^16,18 Brix and colleagues¹⁵ is the exception to this generalization, which reported no significant differences in GABA+ (with water used as an internal concentration reference), as well as GABA+/Cr levels, between approximately 10 year old individuals with ASD and their TD peers.¹⁵ However, Brix and colleagues did observe a negative correlation between GABA+/Cr levels and autism symptom severity metric, a correlation later recapitulated in another study.¹⁶

Insight into this possibly developmentally regulated GABA level decrease in ASD arises from MRS studies of the temporal lobe. GABA+/Cr has been consistently reported as reduced within temporal lobe of approximately 12-year-old children with ASD compared to TD peers^10,12,14 (Figure 2a). Additionally, the neurotypical maturational increase (at least as inferred from cross-sectional studies) of GABA+/Cr levels is lacking in similarly aged children/adolescents with ASD.¹² This previously reported differential GABA+/Cr level maturational trajectory has replicated with larger sample size (Figure 2b & c). Of note, the perturbed maturation of relative cortical GABA+ levels in ASD, highlighted within Figure 2, may explain why studies of younger children with ASD fail to exhibit atypical GABA levels—the decreased GABA levels in ASD become more apparent with increased developmental age as a consequence of different developmental trajectory rates.

Figure 2.

Relative Superior Temporal Gyrus GABA+/Cr levels diverge in children/adolescents with ASD from neurotypical levels, due to a lack of typical maturational increases. (a) GABA+/Cr levels are decreased in individuals with ASD (red) as compared to TD (blue), especially when focusing on children/adolescents aged over 10 years old. Data presented are mean ± standard error of the mean. (b) TD children/adolescents exhibit a maturational increase in GABA+/Cr levels. (c) Children/adolescents with ASD fail to exhibit the typical maturational increases of GABA+/Cr levels. Lines represents within diagnostic group linear regressions of GABA+/Cr with age for both b and c. Note – presented data is cross-sectional, and thus some of the scatter can be ascribed to interindividual differences. # p < 0.10, * p < 0.05, ** p < 0.01. ASD, autism spectrum disorder; GABA, gamma-aminobutyricacid.

One of the aforementioned MRS studies that targeted temporal lobe GABA in ASD¹² also examined the same temporal lobe voxel in adults, and observed only marginal alterations to GABA levels. This observation of similar temporal lobe GABA levels between adults with ASD and their TD peers has since been recapitulated by an independent laboratory.⁵² It is currently unclear how/when GABA levels in individuals with ASD recover to/arrive at neurotypical levels, but it appears likely that there is a critical window during development during which deficient GABA levels have functional sequelae (e.g. in compromising the establishment of intact and optimal neuronal circuitry), which might not be reversible by later restoration of GABA levels alone. True (multidecade) longitudinal studies of GABA and circuit function developmental trajectories in ASD are warranted.

As with MRS-based estimates of glutamate, GABA levels derived from MRS also contain several caveats. As mentioned above, frequently GABA is estimated in the form of a “GABA+” signal, which integrates the combination of the GABA signal as well as potential signals from contaminating co-edited macromolecules. Recent advances in the field have produced MRS sequences that allow for macromolecule-suppressed GABA estimates,⁵¹ though such sequences involve a considerable loss of signal to noise, with approximately a two-thirds decrease in the resultant GABA signal. It is not immediately clear how much of this signal decimation results from losing sensitivity to actual underlying GABA levels compared with macromolecule suppression alone. Moreover, as stated above, the field remains considerably inconsistent regarding the exact metric reported (i.e. GABA+ in institutional units, or relative GABA+ normalized by concomitant Cr, water or NAA), and as such, how the variance in these reported metrics contributes to the inconsistent observations within the literature remains unclear. In fact, there are also differences within these reviewed studies regarding the exact MR sequence utilized to collect the spectra, including scan time and other technical parameters (while a recent multisite study⁵³ has showed the robustness of the most widely used MEGAPRESS sequence across sites and platforms, it is unclear what influence the technical choices made in the reported ASD literature impose).

Additionally, and perhaps most importantly, MRS-based GABA estimates (similar to Glx) are not able to resolve neurosignaling pools of GABA from pools that are involved in metabolism. Within both GABAergic presynaptic neurons, as well as astrocytes, GABA may be removed from the synaptic pool by the GABA shunt and directed into the tricarboxylic acid cycle.⁵⁴ This GABA shunt allows for both the degradation and creation of GABA, via the use of GABA-aminotransferase (GABA-T) and Glutamate decarboxylase (GAD)⁴⁷ respectively. Of note, there are two isoforms of GAD, GAD65 and GAD67. While it has previously been hypothesized that the isoform of GAD that generated the GABA molecule specifics GABA’s role within the cell⁵⁵ (GAD65 is associated with neurotransmission; GAD67 is associated with the GABA shunt), this hypothesis has since been disputed.⁵⁶

In summary, there is promising emerging evidence of decreased GABA at least within peri-Rolandic and temporal regions in ASD, along with a difference in the developmental trajectories of such levels between ASD and typical development (allowing optimal delineation of groups in later childhood and adolescence). Still, it appears there exists considerable regional specificity in this observation. As such, the role of such highly region-specific decreased relative cortical GABA+ levels requires additional examination, especially with regard to the functional outcome of such an observed phenotype. Of note, studies have also started addressing either electrophysiological or even functional correlates of these GABA disturbances.

As such, several caveats limit direct interpretation of MRS-derived GABA in ASD—(1) the lack of absolute quantification and thus the reliance on an assumption of a developmentally and cross-cohort stable denominator (usually creatine), (2) the co-editing of macromolecules in the MEGAPRESS sequence, noted in the terminology “GABA+,” suggesting true GABA levels may be overestimated, (3) the difficulty resolving glutamate from glutamine components in either PRESS or MEGAPRESS approaches, and (4) the fundamental difficulty interpreting signals of Glx or GABA as indices of neural signaling vs “mere” metabolites (assumed in most of the above studies is that neurotransmitter “relevant” GABA levels scale with “total” measurable GABA). With such limitations in MRS methodology, alternative insights into E/I imbalance are sought from other (perhaps more indirect) technologies.

Electrophysiology

Although direct measurement of neurosignaling-related E/I balance through techniques such as MEG, or its electrical counterpart electroencephalography (EEG) is not achievable, recent advancements surrounding the biological generators of electrophysiological signals allow the field to test E/I balance indirectly. Gamma-band electrophysiological activity (30–100 Hz) is thought to represent a functional readout of current E/I balance within local neural circuits, due to its strict dependence on GABA neurosignaling⁵⁷—including the GABA_A receptor time constant.⁵⁸ Moreover, the balance of excitation and inhibition within local circuits (contained within the tissue sampled by either voxel-based MRS (above) or point-spread function resolution-limited “sources” sampled by EEG/MEG or stimulated by TMS) and, by proxy, γ-band activity has been posited as a potential zone of convergence for the many seemingly disparate and noncontiguous neurobiological alterations observed in ASD.⁵⁹

Alterations to γ-band activity have been extensively studied in ASD ⁶⁰. Reports generally follow a pattern of decreased early gamma-band activity in response to simple sensory stimuli, but a complex pattern for resting-state (RS) alterations ^14,61. There are two forms of sensory evoked γ-band activity: transient γ-band activity, and steady-state γ-band activity, which may differ in what they represent regarding local circuit activity. Transient γ-band activity exemplifies how the local circuit responds to a normal stimulus and embodies typical circuit functioning. Typical paradigms elicit transient responses from a single “punctate” sensory stimulus (e.g. brief (~50–300 ms) auditory stimulus, visual checkerboard reversal (2 Hz) or isolated somatosensory “tap”).

On the other hand, stimulus evoked steady-state γ-band activity exemplifies local circuit activity when driven to a higher-level than with typical stimuli. Steady state stimuli might include amplitude-modulated tones (with a modulation frequency of ~40 Hz and a duration of ~1 s⁶²), or visual gratings (of appropriate spatial frequency and extended static display duration⁶³) or piezo-electric vibro-tactile somatosensory stimuli.⁶⁴

Therefore, stimulus evoked transient γ-band activity represents typical local circuit activity, whereas stimulus evoked steady-state γ-band activity represents how “strongly” the circuit can be driven to respond when stimulated at that frequency.

Altered auditory-related γ-band activity within ASD has been of particular focus within the clinical literature. Of note, while it is possible to measure γ-band activity within the auditory system using EEG, unless high-density electrode array as well as multicompartment volume conduction models are explicitly used, it remains difficult to separate hemispheric signals. This is important, as multiple MEG studies have reported a hemisphere-specificity to altered auditory cortex electrophysiological responses in ASD.^65–67 As such, this review will focus on MEG-based studies of γ-band activity.

Reduced transient auditory-evoked γ-band activity has been consistently reported in ASD^12,67–70 (Figure 3a). Of note, these studies presented tones 45 dB above their individually determined auditory detection threshold, so hearing-sensitivity differences could not account for altered γ-band responses. Furthermore, transient auditory-evoked γ-band activity deficits are greater between adults with ASD and their age-matched TD peers, as opposed to the corresponding comparison involving children, due to differential maturational trajectories of transient auditory-evoked γ-band activity.¹² This altered developmental trajectory of transient auditory-evoked γ-band coherence within ASD has been replicated after approximately doubling the sample size (Figure 3b & c), akin to the confirmation of decreased relative cortical GABA+ level maturation in ASD in Figure 2.

Figure 3.

Transient auditory γ-band responses diverge in children/adolescents with ASD from neurotypical levels, due to a lack of typical maturational increases. (a) While not significantly different within children/adolescents, transient auditory γ-band responses are significantly different between ASD and TD in adults. Data presented are mean ± standard error of the mean (b) TD children/adolescents exhibit maturational increases in the transient auditory γ-band responses. (c) Children/adolescents with ASD fail to exhibit the typical maturational increases of transient auditory γ-band responses. Line represents within diagnostic group linear regressions of transient auditory γ-band activity with age for both b and c. Note—presented data are cross-sectional, as opposed to longitudinal, although a recent longitudinal study report analogous findings.⁶⁹ * p < 0.05

Similarly, steady-state auditory-evoked γ-band activity appears decreased in ASD.⁶² Of note, decreased auditory-evoked γ-band responses are also exhibited by parents of individuals with ASD, and as such, may represent a heritable endophenotype of ASD.^70,71

A caveat to the above studies arises when speech stimuli are used as opposed to the simple tones or tone bursts of the previous studies. McFadden and colleagues⁷² reported that parents of children with ASD exhibited increased γ-band activity in response to spoken words. It remains unclear what this increase in γ-band activity in response to words truly represents. It may be that increased stimulus complexity leads to increased stimulus evoked γ-band activity in ASD, as parents with ASD often exhibit a broad autism phenotype.⁷³ Alternatively, such changes may be homeostatic or compensatory, and may underlie the reason the parents are not as impacted as their offspring.

Of note, auditory-evoked γ-band activity correlates to ASD symptom severity^67,71 and language impairment.⁷² Additionally, a recent study suggests that transient γ-band activity may even predict future optimal outcome, indexing individuals that subsequently become subthreshold for a clinical ASD diagnosis, as well as accounting for a small but significant 5% of the variance in verbal intelligence 2–3 years later.⁶⁹ While accounting for only a small part of the total variance in future verbal ability in a general ASD population, it may be that this ability of auditory evoked γ-band activity to predict later functional outcome is actually bimodal. As such, within the subset of the ASD population with E/I imbalance as their primary etiology, evoked γ-band activity may account for a higher fraction of the variance, though this effect is diluted by the inclusion of individuals with ASD who have a different primary etiology. This same may be true for the above correlations between ASD symptom severity and auditory stimulus evoked γ-band activity; while the correlation may be observable in a general population of individuals with ASD, the true relationship may be bimodal. Some individuals with ASD may exhibit no E/I imbalance, and so exhibit no coupling between ASD symptom severity and auditory stimulus evoked γ-band responses, whilst other with E/I imbalance may exhibit considerable coupling between ASD symptom severity and auditory stimulus evoked γ-band responses. As such, E/I imbalance as a basis for stratification may be a useful tool for clinical trial enrichment and or early treatment triage.

Despite the aforementioned advances in GABA-estimating MRS, recent studies have begun to non-invasively directly test the relationship of GABA levels and γ-band activity within the same individuals. Several studies have reported that γ-band activity correlates to underlying GABA+ levels in TD healthy individuals,^12,74–77 though this finding is not without controversy.^78,79 Such correlations (converging evidence) between independent assays, each somewhat relating to E/I balance, lends credence to each methodology, despite their above-mentioned limitations, as well as supporting the notion of E/I balance as a unifying framework for interpreting the findings from disparate clinical tools. The observed correlations between γ-band activity and underlying GABA+ levels are supported by findings from another neuroimaging modality, positron emission tomography (PET). PET-derived GABA_A receptor density has been observed to correlate to γ-band activity from the corresponding brain-region.⁸⁰

While the coupling of neurochemistry and electrophysiology seems plausible within TD individuals, less attention has been paid to how neuropsychiatric disorders may affect such a relationship. Within ASD, a recent study reported that individuals with ASD lack the developmentally regulated coupling of transient auditory-evoked γ-band responses and underlying relative cortical GABA+/Cr levels.¹² Indeed, this observation remains true after subsequently revisiting this analysis after considerably increasing the subject pool (Figure 4). Moreover, covarying for the effect of age does not negate the significant correlation (Figure 4a). A recent preclinical study recapitulated this finding of differential coupling, reporting that while WT littermates exhibited coupling of neurochemistry and electrophysiology, mice with a genetic insult relevant to ASD did not.⁴³

Figure 4.

Transient auditory γ-band responses are coupled to underlying relative cortical GABA in TD children/adolescents, but not in children/adolescents with ASD. (A) Transient auditory γ-band responses are coupled to underlying Superior Temporal Gyrus GABA+/Cr levels in TD children/adolescents. (B) Children/adolescents with ASD lack the analogous coupling of. Line represents within diagnostic group linear regressions of transient auditory γ-band activity with GABA+/Cr for both a and c. ** p < 0.01, †, p < 0.05 after covarying age in correlation.

TMS

TMS is a method for non-invasive brain stimulation, whereby intracranial electrical currents, large enough to depolarize a small population of neurons, are generated by rapidly changing extracranial magnetic fields produced by an electromagnetic coil placed on the scalp.⁸¹ Different TMS paradigms have been developed to probe cortical excitability, inhibitory control, and plasticity respectively, and have been used to explore the neurophysiology of ASD, generally among individuals without intellectual disability. TMS can be applied in single pulses, pairs of pulses, or repeated trains of pulses (rTMS). When single pulse TMS is applied in primary motor cortex (M1) at suprathreshold intensities, it activates corticospinal outputs, producing a twitch in a peripheral muscle (a motor evoked potential, MEP), which has been used in TD and ASD populations as an index of corticospinal excitability.⁸² In addition to excitability TMS has been used to probe intracortical inhibitory and facilitatory processes using paired pulse stimulation protocols.^83–86 Finally, trains of repeated TMS pulses (rTMS) at various stimulation frequencies and patterns can induce a lasting modification of activity in the targeted brain region, which can outlast the effects of the stimulation itself. The after effects of rTMS are thought to relate to activity-dependent changes in the effectiveness of synaptic connections between cortical neurons, reflecting cortical plasticity mechanisms.^87–90 For the purposes of this review, we will focus on single and paired pulse studies relevant to E/I imbalance (See⁹¹ for a review of the use of TMS in ASD).

In ASD, six independent studies have used single pulse TMS to probe baseline levels of corticospinal excitability and have shown no difference in either motor threshold (the lowest intensity of stimulation required to induce a MEP) or size of MEP in response to a suprathreshold pulse of TMS between individuals with ASD and neurotypical individuals.^92–97 These published data suggest that baseline M1 excitability is not affected in ASD. The majority of these studies were in adults with ASD, however, Enticott et al., 2010 included adolescents as well.

Conventional paired pulse TMS protocols involve applying two consecutive magnetic pulses through the same TMS coil in rapid succession over primary motor cortex at various interpulse intervals. The outcome measure is the degree of effect of the first pulse “conditioning stimulus” (CS) on the second pulse “test stimulus” (TS).^83–86 When the interpulse interval between a subthreshold CS and suprathreshold TS is 1–6 ms, the resulting MEP suppression is thought to reflect GABA_A receptor mediated short-interval intracortical inhibition.^83,98 When the interpulse interval is increased to 10–25 ms, the net result is facilitatory, making this paired pulse paradigm a putative index of intracortical facilitation, which is thought to be mediated by a combination of receptor types including n-methyl-d-aspartate glutamate receptors,⁹⁹ GABA_A receptors,^100–102 and noradrenaline receptors.^103–109 Two suprathreshold pulses delivered at an interpulse interval of 50–200 ms is used to evaluate GABA-mediated long interval intracortical inhibition.^85,110–113 A number of studies have been conducted using these paradigms to probe intracortical inhibition and facilitation in ASD. Two studies report no significant difference in response to the short-interval intracortical inhibition paradigm between ASD and neurotypical adolescents and adults.^97,114 Three studies employed the intracortical facilitation paradigm and found no significant difference between ASD and neurotypical adolescents and adults.^92,93,97 Three studies have reported mixed results with some ASD adolescents and adults showing impaired intracortical inhibition and others showing typical responses.^92,93,115 Thus, abnormal intracortical inhibition may be present in a subgroup, but this alteration of cortical physiology does not appear to be consistently demonstrable in all individuals with ASD.

In summary, the findings from the above-mentioned literature using TMS as an investigational device partially support the theories suggesting excitation/inhibition imbalance in ASD. However, what the studies above reveal most clearly is the variability of the findings. Other than no abnormality in baseline corticospinal excitability, all of the paired pulse paradigms resulted in mixed responses both within and across studies. One should note that the sample sizes in the studies are relatively small (ranging from 5 to 36) and represent a small subgroup of the overall ASD population. Specifically,¹ the aforementioned studies either did not document or did not exclude individuals on psychoactive medications²; all studies excluded individuals with intellectual disability³; all studies excluded individuals with a history of seizures or abnormal EEG findings⁴; All studies included only adults or adolescents. No study has been conducted in younger children with ASD. Sadly, no studies acquired concurrent multimodal data, such as MEG or MRS to help account for heterogeneous responses.

E/I balance as a therapeutic target

With the aforementioned evidence for E/I imbalance in ASD, it is perhaps not surprising several clinical trials have targeted either glutamatergic or GABAergic neurosignaling systems for their intervention. Unfortunately, many of these compounds failed to show efficacy during clinical trials.¹¹⁶ While the clinical trial failures may be due to glutamatergic/GABAergic compounds ultimately being ineffective at treating ASD, alternatively it may be that the lack of efficacy of these compounds for treating ASD derives from a situation where E/I imbalance is only one of several primary aetiologies causing ASD. As such, glutamatergic or GABAergic compounds would only be effective in that subset of individuals with E/I imbalance. Hence, clinical trial enrichment (delineated specifically by the presence of markers of E/I imbalance) could be hypothesized to optimize detection of efficacy of glutamatergic or GABAergic compounds for the treatment of ASD.

Two GABAergic therapeutics for ASD are currently undergoing clinical trials and show promising efficacy in both preclinical and clinical settings. Bumetanide has shown promise in early preclinical and clinical settings for treating both behavioral^117–122 and neural^122–125 aspects of ASD and related disorders. STX209 (also known as either R-Baclofen, or arbaclofen) has also shown promise in treating ASD^126,127 or the related genetic disorder Fragile X Syndrome,¹²⁸ though more recent Phase 3 studies suggested that their efficacy is limited to children.¹²⁹ Such findings corroborate aforementioned insights from MR spectroscopy, where the disruption of GABA levels specifically within a developmental window (as opposed to alterations to GABA levels across the lifespan of individuals with ASD) may lead to the reduced auditory γ-band responses exhibited by adults with ASD. If this were the case, GABAergic treatments would only be effective at restoring auditory γ-band responses during this critical developmental window. A severe confound in the Phase 3 clinical trials of STX209 for Fragile X Syndrome were the sizable placebo effects observed, negating depiction of any limited improvements caused by STX209.¹²⁹

Corresponding preclinical murine studies have provided greater consistency in the efficacy of STX209 or its racemic counterpart baclofen. STX209 has been reported to recover behavioral,^130–133 protein level,^131,133 electrophysiology,^43,134–137 though such observations of efficacy have been contradicted.¹³⁸ It remains unclear that if the reported failure of STX209/baclofen to normalize behavior or neural activity is due to only a single dose being tried,¹³⁸ with at least one study suggesting genotype-dependent effects of STX209 at certain doses.¹³² Additionally, continuous dosing with STX209 potentially induces a tolerance for seizure reduction,¹³⁶ due to possible mGluR5 expression alterations.¹³¹

Clinical trial designs may be compromised by choice of outcome measure or, particularly in the case of heterogeneous ASD, insufficiently restrictive inclusion criteria, wherein a broad population is included rather than a more homogeneous subpopulation, perhaps enriched in the appropriateness of their underlying biological aetiology/pathophysiology for the mode of action of the tested pharmaceutical. It is hoped that some combination of the assays discussed in this review might provide the basis for such inclusion enrichment.

Interventions that target the glutamate neurosignaling system are also being investigated with regards to ASD. Unfortunately, a repeated theme of these drug trials are positive initial open-label results followed by mixed or null findings during more rigorous clinical trials. For example, memantine demonstrated promising open-label results in initial studies.^139–141 Subsequently, a single-blinded study by Karahmad and colleagues¹⁴² demonstrated promising decreases in several core symptoms. However, a concurrent double blinded placebo-controlled study failed to demonstrate significant symptom reduction greater than that observed in the control condition,¹⁴³ though this study was confounded by sizeable placebo effects. Of note, a double-blinded placebo-controlled study did observed reduced irritability, stereotypic behaviors as well as hyperactivity/non-compliance behavior in children with ASD when using memantine as an adjunct to risperidone-based interventions.¹⁴⁴

A similar pattern was observed for the drug riluzole. Initial open-label reports observed promising signs of riluzole-based interventions for ASD-related disorders.^145,146 A more rigorous follow-up study by Wink and colleagues¹⁴⁷ demonstrated no significant effect of riluzole on the Clinical Global Impression Improvement Scale or the Aberrant Behavior Checklist Irritability subscale when using a double-blind placebo-controlled design. Of note though, riluzole has been shown to increase GABA levels in prefrontal cortex of individuals with ASD, whilst also normalizing their functional connectivity.³³ As with memantine, riluzole has been observed to be a promising adjunct to risperidone-based interventions for ASD during a double-blinded placebo-controlled study.¹⁴⁸

Limitations of studies to date

A clear limitation of the studies reviewed above are the relevant current technological limits, especially with regards to MRS-based neurometabolite estimation. Due to the overlap of different neurometabolite resonances within the spectra, whether that be glutamine and glutamate or GABA and macromolecules, the field is unable to resolve precisely which components we are examining. Moreover, current MRS techniques are unable to compartmentalize their results, and instead combine metabolic and neurosignaling pools. Therefore, whether any MRS-detected GABA or glutamate level perturbation reflects a neurotransmission, as opposed to a metabolic, phenotype is unknown. Several of these limitations can be addressed with the implementation of higher field strength (to differentiate between glutamine and glutamate) or tailored MRS sequences (to mitigate macromolecular contamination in GABA estimates),⁵¹ though such systematic changes require time to permeate throughout the literature. Additionally, there probably exist significant caveats to these newer methodologies that the field may not yet be fully aware of.

Studies of γ-band activity in ASD also have their own limitations. γ-band activity is, at best, a proxy-measure of E/I balance, and alterations to this electrophysiological activity in ASD may be attribute to other biological or technological factors, instead of E/I imbalance per se. Additionally, what biological components transient vs steady-state stimuli differentially target is, as of yet, unknown. Moreover, how the biological processing of these stimuli relates to more complex stimuli (such as words or even linguistical constructs) is also unclear. While improved paradigm design, along with sensor technology advancements, may partially resolve some of the existing issues, γ-band activity still fundamentally remains a proxy-measure.

While the field is beginning to understand the immediate biological effects of different TMS paradigms, the primary underlying biological substrates for longer-term alterations is not immediately evident. Similarly, it may be (and currently seems likely) that different biological processes are affected by single as opposed to repeated sessions of TMS. Lastly, studies involving the above modalities, as well as the reviewed drug-development studies, may be flawed due to the fact that ASD potentially arises from several different primary aetiologies, of which E/I imbalance is just one. Such a situation would lead observations of the presence/effect of E/I imbalance to be chronically diluted throughout the literature. Such an issue could easily be addressed through clinical trial enrichment specifically for those demonstrating E/I imbalance, though this does raise the issue of circular logic in studies examining the aetiology of ASD. Such circular logic would not be the case for clinical drug studies, as there the outcome measure would be one of symptom-severity reduction. Until clinical trial enrichment though, studies of E/I imbalance in ASD using the above modalities still continue to suffer from the biological heterogeneity of ASD.

Conclusion

In conclusion, there is ample evidence of E/I imbalance in ASD. That said, the heterogeneity of the behavioral phenotype and the multitude of genetic mutations associated with ASD would challenge the existence of a single “unifying” pathophysiological aetiology. Nonetheless, perhaps E/I imbalance provides a framework for defining at least a subpopulation of the ASD spectrum, permitting the more rapid advance of an appropriately tailored intervention. While alterations to MRS-derived glutamate levels in ASD seem inconsistent between studies, alterations to GABA levels are more consistent across studies (and consistent in their variation across brain regions, with peri-Rolandic and temporal regions most implicated). Furthermore, indirect evidence for E/I imbalance indirectly arises from examination of γ-band activity in ASD, with decreased auditory γ-band response replicated across multiple studies from independent laboratories. Such auditory γ-band responses appear to correlate to underlying GABA level estimates in TD children, but not children with ASD. Moreover, TMS has shown heterogeneous findings both within and across individuals, but do suggest that there exists E/I imbalance in a subset of individuals with ASD. Lastly, while many glutamatergic or GABAergic compounds have failed clinical trials for treatment of ASD, two potential pharmacological interventions for ASD premised on interacting with E/I imbalance show clinical and preclinical promise. Ultimately, clinical trial enrichment should be implemented during further studies of these interventions to target those individuals with ASD that exhibit E/I imbalance. In summary, while the E/I imbalance hypothesis may have its limitations in applicability, it nonetheless provides a useful guiding framework for the informed integration of insights from clinical modalities, themselves not usually as specifically interpretable as their preclinical and cellular counterparts.