Progress in Neuropsychopharmacology & Biological Psychiatry The contribution of [1H] magnetic resonance spectroscopy to the study of excitation-inhibition in autism

ndings from resonance spectroscopy (MRS) the di ﬀ erences in the excitatory glutamate and inhibitory γ -aminobutyric acid (GABA) balance may be both a key biomarker and a tractable treatment target in ASD.


Introduction
Autism spectrum disorder (ASD) is a highly heterogeneous lifelong neurodevelopmental syndrome characterised by deficits in social reciprocity and communication, and by restricted interests and stereotyped behaviours (DSM-5;American Psychiatric Association, 2013). Its prevalence is growing (Lai et al., 2014) and, in the UK alone, adults with ASD cost the economy over £32 billion/year, i.e. more than heart disease, cancer and stroke combined (Buescher et al., 2014). Therefore, there is a pressing need to identify the underlying biological mechanisms that would provide a basis for the development of ASD stratification markers and treatment options. Progress is happening however, based on accumulating evidence that multiple genetic and environmental risk factors for ASD converge to disrupt the balance between glutamate-mediated excitatory and γ-aminobutyric acid (GABA)-mediated inhibitory neurotransmission; and this may inform treatment targets for the disorder (Nelson and Valakh, 2015;Rubenstein and Merzenich, 2003).

Evidence for excitation-inhibition imbalance in autism spectrum disorder
A balanced interaction between glutamate and GABA is essential for synaptic maturation, refinement of neuronal circuitry and subsequently the regulation of cognition, emotion and behaviour (Lujan et al., 2005). An imbalance in excitation-inhibition (E-I) caused by disruption to any component of the glutamate and GABA signalling systems could therefore have potentially far-reaching consequences. Indeed, based on observations such as an increased incidence of epilepsya disorder of E-I balancein individuals with ASD (Bolton et al., 2011;Muhle et al., 2004), and post-mortem evidence of altered expression of proteins involved in glutamate and GABA biosynthesis and neurotransmission in the brains of individuals with ASD (Blatt and Fatemi, 2011;Fatemi et al., 2002Fatemi et al., , 2009Fatemi et al., , 2014Oblak et al., 2009Oblak et al., , 2010Oblak et al., , 2011Purcell et al., 2001;Shimmura et al., 2013;Yip et al., 2007), E-I imbalance has been suggested to be a 'common final pathway' in ASD (Nelson and Valakh, Genetic studies are in line with this proposal. For example, studies of ASD susceptibility genes have repeatedly identified chromosome 15q11-13 as a particular region of interest as 1-3% of all ASD cases were reported to be associated with duplications in this region (Moreno-De-Luca et al., 2013;Urraca et al., 2013). This chromosomal region contains genes for three GABA A receptor subtype genes (i.e. β3, γ3 and α5). Mutations in this region, especially in the β3 subunit, convey a high risk for ASD and have been linked to repetitive symptoms and insistence on sameness (Chaste et al., 2014;Isles et al., 2016;Shao et al., 2003). Direct evidence for the link between GABA A -β3 and ASD has also been generated from a β3 knock-out mouse. These mice have behaviours analogous to the condition; they also have seizures, which are a frequent comorbid complication of ASD caused by E-I imbalance (DeLorey et al., 1998). In addition, we and others have reported that environmental risk factors for ASD, such as prenatal exposure to maternal inflammation or the anti-epilepsy medication valproate, disrupt gene and protein expression across GABA pathways and trigger behavioural features which mimic the human condition (Fatemi et al., 2008;Fukuchi et al., 2009;Giovanoli et al., 2014;Meyer et al., 2011;Wei et al., 2016).
There is also good evidence for genetic abnormalities in glutamatergic signalling pathways in ASD and related conditions. This has, in part, been built on the advances made in fragile X syndrome, the most common genetic form of ASD, which is caused by a single gene mutation affecting fragile X mental retardation protein (FMRP) (Verkerk et al., 1991). FMRP negatively regulates postsynaptic glutamate receptor synthesis (including metabotropic glutamate receptor 5; mGluR5) by binding to ribosomes and stalling translation of target mRNAs (Darnell et al., 2011). mRNA translation at the synapse is then increased (Bear et al., 2004), leading to excess excitatory activity, enhanced long term depression, and ASD-like pathology in animal models (Lozano et al., 2014;Yu and Berry-Kravis, 2014). Lower levels of FMRP are found in individuals with idiopathic ASD (i.e. ASD without a known cause) who have a concomitant increase in mGluR5 . In addition to alterations to metabotropic receptors in ASD, genetic aberrations have also been linked to ionotropic glutamate receptors, namely N-methyl-D-aspartate (NMDA) receptors. However, the picture is complex, and both increases and decreases in NMDA receptor function have been associated with the ASD phenotype (Gandal et al., 2012).
GABA and glutamate signalling in ASD may also be indirectly derailed by mutations in genes coding for proteins that stabilise GABAergic and glutamatergic synapses. For example, alterations of neurexin and neuroligin cell adhesion molecules (Ching et al., 2010;Jamain et al., 2003), postsynaptic density protein 95 (PSD-95) and/or Shank scaffold proteins (Uchino and Waga, 2015;Xing et al., 2016) confer a high risk for ASD. These molecules anchor pre-and post-synaptic neuron terminals and disruptions of the molecular assembly of the synapse can potentially cause E-I imbalance (Futai et al., 2007;Südhof, 2008). Consistent with this, neurexin 1-α (Nrxn1-α) knock-out mice show a reduction in spontaneous and evoked neurotransmitter release (Kattenstroth et al., 2004;Missler et al., 2003;Zhang, 2005) accompanied by behavioural abnormalities that resemble some of the core ASD symptoms of social impairment and inflexibility/stereotypy (Grayton et al., 2013). Similarly, mice with mutations in the neuroligin 3 gene (Nlgn3) exhibit brain-region specific synaptic dysfunction (e.g. a shift of synaptic transmission towards inhibition in the somatosensory cortex and towards excitation in the hippocampus) and impairments in social behaviour that mimic ASD (Burrows et al., 2015;Etherton et al., 2011aEtherton et al., , 2011bJaramillo et al., 2014;Radyushkin et al., 2009;Tabuchi et al., 2007). Such "multiple" E-I anomalies, that can be not only brain region-but even neural circuit-specific, as shown in multiple animal models of ASD (Horder et al., 2018;Lee et al., 2017;Nelson and Valakh, 2015) further exemplify the complexities of defining and measuring the E-I (im)balance in ASD.
Finally, the metabolism of glutamate and GABA is tightly inter-linked so an alteration in either metabolite may potentially affect the other. The enzyme glutaminase catalyses the glutamate to glutamine cycle, allowing glutamine to be stored in astrocytes or converted to glutamate in both glutamatergic and GABAergic neurons (Rowley et al., 2012). In excitatory neurons, glutamate is transported into vesicles via vesicular glutamate transporters, whereas in inhibitory neurons, glutamate is first converted to GABA by the glutamic acid decarboxylase and then transported to vesicles via vesicular GABA transporters. Upon release, the neurotransmitters are taken up by high affinity membrane transporters and returned into neurons and surrounding glia, where they are recycled. Thus, GABA, glutamate and glutamine are in constant flux. In ASD however, the levels of the multiple enzymes controlling glutamine-glutamate-GABA cycles are altered (Fatemi et al., 2002;Shimmura et al., 2013;Yip et al., 2007), and so glutamine-glutamate-GABA metabolism is likely to be atypical in the ASD brain. How these diverse E-I anomalies might act mechanistically has been the subject of a number of recent comprehensive reviews (Foss-Feig et al., 2017;Gao and Penzes, 2015;Gatto and Broadie, 2010;Lee et al., 2017;Mullins et al., 2016;Nelson and Valakh, 2015;Uzunova et al., 2016). For example, Yizhar et al. (2011) have shown that increase of a cellular E-I balance in the mouse medial prefrontal cortex (mPFC) elicits elevated rhythmic high-frequency (gamma) activity (a disease correlate in clinical conditions; Orekhova et al., 2007) and social dysfunction. Furthermore, it was recently shown that reducing the E-I ratio by either reducing the excitability of pyramidal neurons or by increasing the excitability of inhibitory parvalbumin-positive neurons in the mPFC rescued social deficits in mice lacking the CNTNAP2 gene, which has been implicated in ASD (Selimbeyoglu et al., 2017). Maintenance of E-I balance is thus essential to network function as not only functional, but also structural properties of neural circuits are defined by inter-neuronal communication.

[1H] Magnetic resonance spectroscopy evidence for E-I imbalance in ASD
The hypothesis that the glutamine-glutamate-GABA system is altered in ASD has also been addressed by using [1H] magnetic resonance spectroscopy (MRS) as a non-invasive method to measure tissue levels of E-I metabolites in the living human brain (Ford and Crewther, 2016;Schur et al., 2016). In brief, [1H]MRS uses the nuclear magnetic resonance properties of hydrogen atoms to generate a frequency spectrum (as opposed to the image in MR imaging) (Govindaraju et al., 2000;Juchem and Rothman, 2014;Stagg et al., 2011) in which different metabolites are identified by the position of their signal peak, i.e. chemical shift, along the frequency axis (in units of 'parts per million'; ppm). [1H]MRS metabolite levels can then be quantified by calculating the area under the corresponding peaks or by fitting the in vivo spectra to a basis set, i.e. linear combinations of measured or simulated model spectra of metabolites Naressi et al., 2001;Provencher, 1993). [1H]MRS is not without limitations however. For an adequate signal-to-noise ratio, standard single voxel [1H]MRS data is acquired from a predefined and relatively large brain volume-unit (usually around 4-8 cm 3 ) (Stagg et al., 2011;Zhu and Barker, 2011) compared with conventional MR image resolution (typically around 1-10 mm 3 ) (Blüml, 2013). Therefore, both grey and white matter is usually sampled within the same voxel. To avoid errors when quantifying metabolites, these so-called 'partial volume' effects need to be corrected by segmenting spectroscopic voxel according to different tissue contents (Dager et al., 2008;Quadrelli et al., 2016). As shown in Table 1, most of the studies reported using tissue segmentation, however, exact approaches varied considerably between studies which might contribute to inconsistencies in the literature. Such partial volume effects are minimised in [1H]MRS imaging (MRSI), a less frequently used extension of [1H]MRS that allows data from multiple adjacent small (1 cm 3 and below) voxels to be acquired simultaneously (Blüml, 2013;Oz et al., 2014;Schneider, 2016;Zhu and Barker, 2011). Table 1 Summary of [1H]MRS studies reporting Glx, Glu and/or GABA+ levels in ASD.  (8) Estimated concentration refers to water-or external reference-scaled metabolites.
However, MRSI is often associated with reduced spectral resolution, non-uniform water suppression and voxel bleeding artefact (i.e. signal contamination from neighbouring voxels into the voxel-of-interest), therefore single voxel [1H]MRS remains the technique of choice (Astrakas and Argyropoulou, 2016;Blüml, 2013;Duarte et al., 2012;Jansen et al., 2006). This large volume also means that [1H]MRS samples tissue metabolites within both neurons and glia and in both intra-and extra-cellular space (Dager et al., 2008). Therefore, to account for possible variation in voxel tissue composition, some studies report metabolite concentrations as ratios to total creatine (creatine + phosphocreatine; tCr), which is assumed to be present at constant concentration in brain tissue and cerebrospinal fluid (Gussew et al., 2012). However, as an alternative to tCr ratios, water has been used as a reference because tCr itself can be altered in neurodevelopmental conditions (Alger, 2010). Finally, depending on technical factors, such as the magnetic field strength, it is not always possible to differentiate between glutamate and glutamine due to the overlap of their resonant peaks, and the combined signal termed Glx is commonly reported (Table 1), and thus is only a proxy measure of glutamate. Similarly, the GABA signal usually also has a contribution from 'macromolecules', and is denoted as GABA+ (Edden et al., 2012;Henry et al., 2001;Mikkelsen et al., 2017;Near et al., 2011).
By definition, ASD is a neurodevelopmental condition and how the brain matures is acknowledged to be quite different to even age matched controls (Amaral et al., 2008;Aoki et al., 2012;Redcay and Courchesne, 2005). Thus, results of [1H]MRS studies in ASDas other MR modalitiesmay vary depending on the age group sampled. For example, Corrigan et al. (2013) have reported lower Glx in the cerebral white matter of 3-4-year-old children; however, there was no change in Glx when the same children were scanned again at age 6-7 and 9-10 years; and no difference in grey matter Glx at any age in childhood (Table 1). However, even in adulthood, when the brain may be expected to have fully matured, [1H]MRS studies of glutamate have generated conflicting findings. Lower Glx/Glu has been reported in the anterior cingulate cortex in ASD adults (Bernardi et al., 2011;Tebartz van Elst et al., 2014) (Table 1), but others have observed no significant difference in Glx between ASD adults and controls in this cortical region (Endres et al., 2017;Libero et al., 2015Libero et al., , 2016 (Table 1). Horder et al. (2013Horder et al. ( , 2018 reported lower Glx/Glu in the basal ganglia of adults with ASD, but unchanged cortical levels (Table 1).
In addition to differences in study samples and brain regions examined, different [1H]MRS sequences for quantification of Glx and/or Glu have been employed in the studies reviewed here (Table 1). While conventional [1H]MRS sequences (e.g. STEAM, PRESS) were used by most, unedited (equivalent to a PRESS) (Horder et al., 2018;Robertson et al., 2015) as well as the difference MEGA-PRESS spectra (edit ON-edit OFF subtraction; please see below) (Ajram et al., 2017;Cavalho Pereira et al., 2018) were also used to quantify Glx and/or Glu (Table 1).
Reduction in GABA in the somatosensory cortex has been associated with altered tactile function in children with ASD (Puts et al., 2017). Interestingly, developmental tactile perception abnormalities due to a loss of presynaptic inhibition of somatosensory neuron transmission have been shown to result in ASD-related social interaction deficits in mice (Orefice et al., 2016). However, even in the absence of group differences in GABA+, a role for GABA in ASD has still been argued based on findings that [1H]MRS GABA+ levels in ASD are inversely correlated with scores on the Autism Spectrum Screening Questionnaire (Brix et al., 2015) and with the communication and delayed development scores of the Autism Diagnostic Interview -Revised (Cavalho Pereira et al., 2018). More difficult to disentangle are the results from Robertson et al. (2015). In their study, occipital GABA+ measures were equivalent in ASD and control groups, but the ASD group had a deficit in the performance of a binocular rivalry test of visual function thought to rely upon visual cortical E-I balance. The authors went on to show that a positive correlation between GABA+ and performance on a binocular rivalry task in control group was not evident in the ASD group, and suggested that the results supported an impairment in 'inhibitory signalling' in ASD (Robertson et al., 2015). This may well be the case, but because [1H]MRS captures the metabolites within a composite of cell types and does not capture active neurotransmission, it is impossible to say with certainty.

E-I flux in ASD
One perspective on this constellation of [1H]MRS findings in ASD, is that their range may mirror the etiological and phenotypical heterogeneity of the spectrum as well as the variety in the typically developing control groups used for comparison. For example, [1H]MRS measures of glutamate and GABA+ differ even within the typically developing population (Bogner et al., 2010;Brix et al., 2017;Kaiser et al., 2005;O'Gorman et al., 2011;Sailasuta et al., 2008).
Another consideration for the field is that, in the main, indices of glutamate and GABA have been assessed separately in ASD; yet the results have been interpreted in terms of E-I balance (Dickinson et al., 2016). It could be argued that a true imbalance can only be uncovered if the level of each metabolite is reported. Thus, if a GABA decrease is accompanied by a glutamate decrease, then the system may remain balanced. Moreover, until recently, the logic of trying to capture metabolites within a glutamine-glutamate-GABA cycle that is constantly changing or in 'flux' in cohorts with and without ASD, has not been questioned. We now suggest that replication difficulties in cross-sectional studies of glutamate and GABA pathways in ASD arise, at least in part, because they sample only one part of an E-I system, which is a 'moving target'. As described earlier, the regulatory enzymes controlling this dynamic system are altered in ASD, and so E-I flux in ASD is therefore very likely to be atypical.
Supporting this hypothesis, there is now evidence that [1H]MRS is sensitive to changes in E-I flux elicited by pharmacological probes that challenge the system. This comes initially from studies in rats, which have established that [1H]MRS can reliably capture changes in glutamate and GABA levels following pharmacological interventions (Waschkies et al., 2014). Specifically, [1H]MRS was used to track the E-I response to pharmacologically relevant doses of compounds with a known mechanism of action on GABAergic and glutamatergic neurotransmission, namely vigabatrin, 3-mercaptopropionate, tiagabine, methionine sulfoximine, and riluzole. Clear drug dose-[1H]MRS effect relationships were demonstrated, with changes as low as 6% in glutamate and 12% in GABA reliably measured using [1H]MRS. The authors concluded that quantitative [1H]MRS of glutamate and GABA may usefully reveal E-I 'target engagement'in other words, the interaction of a drug with its biological goal (Waschkies et al., 2014). We have recently extended this approach to humans. Using [1H]MRS, we found that the anti-glutamate and pro-GABA drug riluzole increased the proportion of GABA+ relative to the total level of GABA+ and Glx metabolites in the prefrontal cortex in ASD, but decreased it in controls, despite a comparable baseline between groups (Ajram et al., 2017).
This difference in E-I 'responsivity' in ASD has important implications for the development of pharmacological treatments for ASD, as it suggests that drugs may not act in the ASD brain the same way as in the typically developing brain. That is, selecting a compound for a trial/ treatment in ASD based on its mechanism of action in the typical brain may not always be appropriate. Moreover, the response to pharmacological treatments in ASD may itself be heterogeneous. This may, for instance, explain why somebut not allindividuals with ASD can have paradoxical responses to GABA A /benzodiazepine receptor agonists; usually these medications have a sedative effect, but 'excitation' has at times been reported in individuals with ASD (Lemonnier et al., 2012;Marrosu et al., 1987). It may also explain some of the trial disappointments in ASD. For example, lamotrigine, an anticonvulsant, was proposed to be a potential treatment for ASD because it inhibits synaptic glutamate release in excitatory neurons. Unfortunately, it did not improve clinical outcomes in children with ASD (Belsito et al., 2001). A similar failure followed the use of D-cycloserine, a partial agonist at the NMDA glutamate receptor (Belsito et al., 2001;Minshawi et al., 2016;Posey et al., 2004).

Conclusions
We conclude that disruption within E-I systems could serve as a biomarker for ASD. However, we extend this observation to suggest that E-I systems in the ASD brain are pharmacologically atypical. However, the picture is far from straight-forward. This means that we need better strategies, potentially including [1H]MRS, to examine the biological response or target engagement to proposed treatments in sub-groups of individuals with ASD. The identification of more pharmacologically homogeneous sub-groups within the spectrum may inform a new generation of biologically-driven clinical trials and avoid the time and expense (and failure) of un-stratified large-scale studies.

Declarations of interest
None.