GABA and glutamate response to social processing: a functional MRS feasibility study

Several studies have suggested that atypical social processing in neurodevelopmental conditions (e.g. autism) is associated with differences in excitation and inhibition, through changes in the levels of glutamate and gamma‐aminobutyric acid (GABA). While associations between baseline metabolite levels and behaviours can be insightful, assessing the neurometabolic response of GABA and glutamate during social processing may explain altered neurochemical function in more depth. Thus far, there have been no attempts to determine whether changes in metabolite levels are detectable using functional MRS (fMRS) during social processing in a control population. We performed Mescher–Garwood point resolved spectroscopy edited fMRS to measure the dynamic response of GABA and glutamate in the superior temporal sulcus (STS) and visual cortex (V1) while viewing social stimuli, using a design that allows for analysis in both block and event‐related approaches. Sliding window analyses were used to investigate GABA and glutamate dynamics at higher temporal resolution. The changes of GABA and glutamate levels with social stimulus were largely non‐significant. A small decrease in GABA levels was observed during social stimulus presentation in V1, but no change was observed in STS. Conversely, non‐social stimulus elicited changes in both GABA and glutamate levels in both regions. Our findings suggest that the current experimental design primarily captures effects of visual stimulation, not social processing. Here, we discuss the feasibility of using fMRS analysis approaches to assess changes in metabolite response.

the mental state of others from these non-verbal social clues. 2 Several studies have suggested that social processing differences observed in various neurodevelopmental and psychiatric conditions (e.g.10] It is generally acknowledged that proton MRS ( 1 H-MRS) is the only neuroimaging technique that allows for non-invasive in vivo quantification of neurometabolites within pre-defined regions of the brain. 11 1H-MRS is a well established technique for robust measurement of multiple neurometabolites in a single acquisition and has been extensively applied in both non-clinical and clinical groups. 12,13These MRS-measured metabolites can be directly linked to brain function and responses including excitatory and inhibitory tone. 14,15Two metabolites that are of particular interest are the primary excitatory neurotransmitter glutamate and the primary inhibitory neurotransmitter gamma-aminobutyric acid (GABA).
While the vast majority of MRS studies of glutamate and GABA have been performed at rest, being able to determine dynamic changes in glutamate/GABA balance is of increased interest to research in health and disease.MRS of glutamate and GABA (and particularly that of the latter) requires tailored techniques due to its low concentrations in the brain (and thus, low signal-to-noise ratio, SNR) and overlap of more highly concentrated metabolites.J-difference editing is mostly typically used to measure GABA, and provides simultaneous measures of glutamate and glutamine (Glx).Despite their low concentration and overlap with more highly concentrated metabolites, recent developments in 1 H-MRS techniques and MR instruments have led to increased temporal resolution and robust quantification of glutamate and GABA. 16 an attempt to capture dynamic metabolic changes due to stimulation, 2 MRS data acquired across completion of a task paradigm can be probed for metabolite fluctuations within windows of varying task demands.This emerging technique is called functional MRS (fMRS). 17Recent Jdifference editing fMRS works have demonstrated the feasibility of measuring activation-induced neurometabolite responses during various types of physiological stimulation, including in visual, motor, cognitive, and pain domains, 18,19 with an effective temporal resolution as low as 12 s. 19,20RS is of particular interest for neurodevelopmental conditions where E/I balance is thought to be shifted, such as in autism and attention-deficit/hyperactivity disorder (ADHD), since it enables the study of GABA and glutamate change directly in response to a stimulus.However, higher temporal resolution comes at a cost of lower SNR from the low number of transients used compared with typical MRS measurement.Two main types of fMRS paradigm are used to probe metabolite fluctuations during stimulation: block and event-related designs.Block design paradigms hold the advantage of robust metabolite quantification by averaging over a large number of stimuli/trials and thus high SNR, but require long fMRS measurements in a block.By contrast, in event-related designs, stimuli are time-locked with the fMRS acquisition and therefore more directly reflect the brain response to the stimulus; however, this comes at the cost of low SNR.Thus, both approaches have their pros and cons (described in more detail in References 19 and 21).
With the ultimate goal of investigating altered E/I balance and its relation to social functioning in autism, we are first interested in investigating whether fMRS can be used to assess metabolic changes during social processing in a non-clinical adult population.Thus far, there have been no attempts to determine whether changes in metabolite levels are detectable using fMRS during social processing.Here, we performed fMRS in non-clinical young adults to assess the temporal resolution of GABA+ (GABA + macromolecules) and Glx (glutamate + glutamine) response during social processing in vivo, with a paradigm allowing for assessment in both block and event-related responses.We measured metabolite dynamics in two brain regions: right superior temporal sulcus (STS), a region considered to play a major role in social perception and cognition 22 ; and the occipital cortex containing V1, an early visual processing brain region that is not associated with social processing, serving as a control.
We hypothesized that social stimuli (faces) would modulate GABA and glutamate levels in the STS region, while non-social stimuli (objects) would elicit no change.We subsequently explore both block/event-related analyses as well as dynamic sliding window and changepoint analyses to determine best analytical approaches to observe relevant changes, accounting for potential metabolite lag after stimulus presentation.

| Participants
All studies were performed in accordance with the procedures approved by the Swinburne University Human Research Ethics Committee in line with the National Statement on Ethical Conduct in Human Research.Data were acquired from non-clinical adults aged 18-40 years.Participants were free from psychiatric, genetic, and neurological conditions, medications, nicotine and caffeine (abstained 24 h), and recreational drugs (abstained 1 week).All data were acquired after written informed consent was obtained.After structural image acquisition, fMRS was performed using the Mescher-Garwood point resolved spectroscopy (MEGA-PRESS) sequence 23 with acquisition parameters recommended by expert consensus (T E /T R 68/2000 ms, 2048 data points). 16,17,24Editing pulses were placed at 1.9 (edit-ON) and 4.8 (edit-OFF) ppm.Voxels were placed in the right STS (4 Â 3 Â 2 cm 3 ) and midline V1 (3 Â 3 Â 3 cm 3 ).The acquisition sequence was consistent across participants, beginning with MRS acquisition in the STS region, followed by the V1 region.During MRS data acquisition, participants were presented with four stimulus blocks: (1) resting MRS (fixation cross, 256 transients); (2) fMRS (dynamic faces or objects, 200 transients); (3) fMRS resting MRS (fixation cross, 200 transients); (4) fMRS (dynamic faces or objects, 200 transients, order counterbalanced).The four stimulus blocks were repeated for both STS and V1 voxels (Figure 1).Prescans were run prior to each functional and rest block to ensure field homogeneity and accurate centre frequency.

| MR data acquisition
Within each functional block (second and fourth blocks), social (faces) and non-social (objects) stimuli were presented as sub-blocks (24 s stimulus-ON, 16 s stimulus-OFF), making these data a combined block and event-related design (example of stimuli used in Figure S1). 22,25A recent fMRI study has shown that the dynamic face stimuli used in this study reliably activate the STS, a central social brain region, 22 while the objects do not.Stimuli were 3 s videoclips of dynamic faces or dynamic objects.There were eight 3 s videos presented in each sub-block, and there were a total of 10 sub-blocks for each functional stimulus block (see Figure 1).The total time for each functional block was 6 min 40 s.

| MRS data analysis
Using an in-house modified version of Gannet, 26 data were pre-processed 27 and fit using standard frequency and phase correction and fitting approaches within block and expressed relative to internal reference total creatine (creatine + phosphocreatine, tCr). 28Using total Cr (tCr) as internal reference allows for minimization of the influence of changes during the data acquisition (e.g.scanner drift, change in linewidth and chemical shift displacement), 29 which is beneficial to a within-subject design.Glutamine and glutamate at 2.2-2.4 ppm were quantified as Glx.We should note that with MEGA-PRESS the GABA signal at 3 ppm is co-edited with macromolecule signal, and hence will be referred to as GABA+.
MRS quantification therefore yielded metabolite levels corresponding to each voxel location relative to tCr, for example GABA+/tCr and Glx/tCr.Schematic representations of analyses performed are in Figure 2. Voxel heatmaps were generated in Osprey to illustrate voxel placement on the Montreal Neurological Institute's 152 brain template (MNI152) using default parameters. 30,31The analysis and quantification scripts can be found in the Open Science Framework at https://osf.io/y7jqp/.
F I G U R E 1 fMRS experimental design.In each MRS voxel of STS and V1, a total of four MRS acquisition blocks were performed.During functional blocks (FUNC1 and FUNC2), participants were asked to passively watch 3 s movie clips of dynamic faces or dynamic objects (24 s stimulus-ON, 16 s stimulus-OFF) (order counterbalanced).

Static window analysis
Data were analysed in rest (REST1 and REST2) and functional (FUNC1 and FUNC2) blocks.Due to the nature of block analysis, we did not separate for T R times corresponding to stimulus type.

Event-related analysis
Since each functional block contains both stimulus-ON and stimulus-OFF blocks, we further analysed differences within block by comparing T R periods corresponding to stimulus-ON (24 s stimulus-ON blocks of 3 s movie clips of dynamic faces or dynamic objects) and stimulus-OFF (16 s rest) blocks, allowing for event-related analysis. 32We then halved the stimulus-ON and stimulus-OFF for each FUNC block and REST block to gain better temporal resolution while retaining sufficient SNR for MRS spectral fitting (ON1, ON2, OFF1, OFF2 for FUNC1; ON3, ON4, OFF3, OFF4 for FUNC2).

Sliding window analysis
Metabolite levels were calculated by averaging the spectra over a window width of 50 transients, with a step size of five transients, regardless of the stimulus type (faces or objects).This allows for creating effective time courses of metabolite levels as they change through time, with an effective temporal resolution of 1 min 40 s for 50 transients and effective time course of 10 s for five transients.

| Statistics
Data were analysed in R (Version 4.1.1).Data were first screened for data skewness/kurtosis using Shapiro-Wilk tests.Outliers were removed based on the absolute deviation from the median of each metabolite with a threshold of 2.5. 33,34I G U R E 2 A, Static (block) analysis based on stimulus paradigm of rest blocks and functional blocks.B, Event-related analysis based on stimulus presentation T R times within each functional block.The stimulus-ON and OFF periods were divided into ON1, ON2, OFF1, and OFF2 for FUNC1, and ON3, ON4, OFF3, and OFF4 for FUNC2.C, Sliding window analysis with window width of 50 transients and step size of five transients.

Blocks (static) analysis
Data between averaged spectra blocks (static) were compared for changes in metabolite levels across stimulus blocks (REST or FUNC) using a one-way ANOVA test.A Kruskal-Wallis non-parametric statistic test was used in subsequent analysis to compare metabolite levels across stimulus blocks with different stimulus types (faces or objects) due to data skewness.

Event-related analysis
Linear mixed-effects models were performed using the R package lmerTest 35 to assess the relationship between metabolite levels and stimulus-ON/stimulus-OFF within functional blocks, and to investigate whether there was a possible lag in metabolite responses within the functional blocks (i.e.response delay and bleed into subsequent stimulus-OFFs).The sub-blocks of stimulus-ONs or stimulus-OFFs of each functional block were added as the fixed effect.The participants were included as intercept for random effect in the model.This analysis was done irrespective of stimulus types.

Sliding window analysis
For sliding window analysis, a linear mixed-effects model 35 was used to investigate the effect of stimulus type and stimulus block on metabolite levels.The new variables to represent the combination of session and stimulus type were used as fixed effects, that is, REST1, FUNC1 + faces, FUNC1 + objects, REST2, FUNC2 + faces, and FUNC2 + objects.The times of measurement from the start of the experiment were included as fixed effects along with their interaction, and participants were included as a random effect.
A changepoint analysis (R mcp 36 package) was performed on sliding window data to detect when metabolite levels first changed in pattern.
This was an exploratory analysis to identify potential time points where metabolite levels change in pattern (i.e. from plateau segment to a jointed slope) across all participants.This method also allows for autocorrelations between data points to be accounted for, and our hypothesis that changes in metabolite response would be more gradual would be reflected by the probability curve in changepoint analysis, rather than exhibiting abrupt shifts.The mean metabolite levels were calculated for each timepoint.A piecewise regression model was used with an autoregressive (AR) component to account for autocorrelation between datapoints.The model had two segments, each representing a different relationship between the metabolite and time, with a changepoint separating the two segments.In the first segment, the metabolite levels were modelled as a constant and an AR(1), to model the dependence of the current value on the previous value.In the second segment, the metabolite levels were modelled as a function of time and an AR(1) process, but without an intercept.7][38] The spread of the posterior distribution reflects uncertainly or the variability of the changepoint in the data, where the total probability is equal to 1. 36 Additional Bayesian analyses were also performed for hypothesis testing against null hypothesis using brms 39 and bayesanova 40 R, with full details in the Supporting Information.

| RESULTS
Data were acquired from 11 non-clinical adults aged between 18 and 40 years (mean age 26.5 years old; six females).A total of eight 'blocks' were acquired from each participant, comprising two rest MRS and two fMRS spectra for each of the brain regions, V1 and STS.Table 1 shows the quality metric for each analysis.To date, no consensus exists for a cut-off value for fMRS quality metrics, as fMRS often employs fewer transients to achieve higher temporal resolution compared with traditional MRS studies 21 and uses temporal modelling of the data, perhaps less sensitive to low SNR.Our reported quality metrics across various brain regions and analytical approaches are comparable, highlighting the robustness of our analysis method. 17Figure 3 shows example spectra from both static window analysis and sliding window analysis from both V1 and STS T A B L E 1 Quality metric for each analysis approach.

| Static window analysis
Based on static (block) level analysis (see Figure 4), we estimated GABA+/tCr and Glx/tCr ratios across the two stimulus types combined and found no significant changes between functional and rest blocks ( p > 0.1).Subsequent analyses based on stimulus type also found no significant change in GABA+/tCr or Glx/tCr for either face or object stimuli ( p > 0.1) (Figures S4 and S5).
We then further investigated event-related changes by comparing metabolite levels from transients corresponding to stimulus-ON and stimulus-OFF T R times separately, regardless of stimulus type. Figure 5 illustrates GABA+/tCr and Glx/tCr levels across stimulus-ON and stimulus-OFF in STS.No statistical significances in metabolite measures were found in the STS region ( p > 0.1). Figure 6 shows the GABA +/tCr and Glx/tCr levels across stimulus-ON and stimulus-OFF in V1.The fixed effects model of Glx/tCr in V1 shows that for stimulus-OFF Glx/tCr increased compared with baseline (OFF1) during the second half of the first functional block (FUNC1) when the stimulus was OFF (OFF2) and during the first half of FUNC2 when stimulus was off (OFF3) (OFF2, β ^= 0.005, SE = 0.002, t = 2.36, p = 0.02; OFF3, β ^= 0.005, SE = 0.002, t = 2.36, p = 0.0016).For the fixed effects model for GABA+/tCr in V1, the result showed a decrease during the second half of stimulus-ON in the second functional block (FUNC2, ON4) (β ^= À0.009, SE = 0.004, t = À2.095,p = 0.044).The supplementary Bayesian analysis for event-related analyses also supports the null results we observed in the STS.Given the significant but small effect observed in V1, Bayesian analysis further suggests that these effects are potentially negligible (see Supporting Information for full details).Table 2 displays the full results of fixed effect models.

| Sliding window analysis
The results obtained from the sliding window analysis indicate that there were fluctuating changes in Glx/tCr and GABA/tCr levels, with no obvious patterns observed irrespective of the type of stimulus presented in either V1 (Figure 7A, D) or STS (Figure 8A, D).However, it should be noted that the 95% confidence intervals (95% CIs) for metabolite levels were relatively large when considering the metabolite levels based on the type of stimulus presented for both brain regions.This may be due to the low number of spectra available for certain stimulus types in each functional block.
The results of the linear mixed model analysis indicated that the stimulus type had a significant effect on V1 GABA+/tCr levels during FUNC1 (Table 3).Specifically, V1 GABA+/tCr increased for object stimuli ( p < 0.001, β ^= 0.019, SE = 0.006) while decrease for faces stimuli ( p < 0.001, β ^= À0.019, SE = 0.005) when compared with REST1.There was also a weak but significant interaction between time and faces stimuli On the other hand, the linear mixed model showed no significant effect of social (face) stimuli on STS GABA+/tCr and STS Glx/tCr.The results showed increase STS GABA+/tCr levels during object stimulus presentation during FUNC2 compared with REST1 (p = 0.008, β ^= 0.041, SE = 0.015), while REST2 showed lower STS GABA+/tCr levels compared with REST1 ( p < 0.036, β ^= À0.000028, SE = 0.000013).The object stimuli during FUNC2 in STS also demonstrated a small negative interaction with time ( p < 0.001, β ^= À0.015, SE = 0.004), whereas REST2 in STS showed a positive interaction with time ( p < 0.001, β ^= 0.000038, SE = 0.000009).Similarly, the linear mixed model showed a significant positive relationship of object stimulus presentation during FUNC2 on STS Glx/tCr ( p = 0.003, β ^= À0.041, SE = 0.041), while REST2 demonstrated significantly lower STS Glx/tCr levels compared with REST1 ( p < 0.001, β ^= À0.028, SE = 0.006), with significant positive interaction with time (p < 0.001, β ^= 0.000036, SE = 0.000008).While complementary Bayesian analyses have shown that there is a possible effect of facial stimuli on STS Glx/tCr during FUNC1, and a different effect in REST2 compared with the baseline, the effect sizes are generally weak across brain regions (see Supporting Information).

| Changepoint analysis
Irrespective of stimulus types, the changepoint analysis demonstrated satisfactory fits with the provided models for both metabolites in V1 and STS (Rhat < 1.1) (Figure 9).The exploratory changepoint analysis, aiming to detect the initial shift in the trend of metabolites, demonstrated changes in both GABA+/tCr and Glx/tCr at the onset of the first rest block (REST1).The mean changepoint for V1 GABA+/tCr was determined to be 450.03s, with 95% CI (51.02, 1344.24), while for V1 Glx/tCr (Figure 9A) the average changepoint was found to be 211 s, with 95% CI (51.01, 769.05) (Figure 9B).A mean changepoint of 682.55 s, with 95% CI [108.87,1380.56]during the first functional block, was identified for STS GABA+/tCr (Figure 9A), and a changepoint of 967.23 s, with 95% CI [182.95,1380.91]during REST2, was identified for STS Glx/tCr (Figure 9B).The relatively large 95% CI intervals indicate uncertainty in estimating the changepoint locations, as demonstrated by the posterior probability distributions presented in Figure 9.

| DISCUSSION
While studies suggest a connection between differences in social processes and alterations in glutamate and GABA function in neurodevelopmental disorders, the neurophysiology underlying this relationship remains largely unexplored.In this present study, we examined glutamate and GABA changes in response to social stimuli in non-clinical participants, using MEGA-PRESS fMRS.While our data demonstrated some modulation of GABA and Glx in response to social stimuli, these changes were minimal and offered limited insights.The observed modulations of GABA and glutamate, given our current experimental design, were minimal and are more likely to reflect visual processing than social processing.
We have not found evidence that, at least in our paradigm, there are social-stimulus-specific changes in metabolite levels as measured with fMRS.
For the purpose of investigating E/I balance in, for example, autism, visual paradigms may indeed be useful to determine differences in the response of GABA and glutamate more generally, but the paradigm presented here does not appear to be a useful approach for studying socialspecific effects, relevant to autism.Initial block-level analyses provided preliminary evidence that there was no association between the social processing stimuli and changes in GABA+/tCr or Glx/tCr in either V1 or STS.Subsequent event-related analyses based on stimulus presentation events within each functional block (stimulus-ON or stimulus-OFF) indicated potential timing differences in how Glx/tCr and GABA+/tCr respond F I G U R E 5 Raincloud plots of GABA+/tCr and Glx/tCr in STS during stimulus-ON (24 s stimulus-ONs of 3 s movie clips of dynamic faces or dynamic objects) or stimulus-OFF (16 s rest-fixation cross) blocks.ON1-4 are the stimulus-on T R times (regardless of stimulus type) within each functional block (ON1, ON2 for FUNC1 and ON3, ON4 for FUNC2).OFF1-4 are the stimulus-off T R times within each functional block (OFF1, OFF2 for FUNC1 and OFF3, OFF4 for FUNC2).p values obtained from a linear mixed model are presented in Table 1.
to stimulus presentation.Specifically, changes in V1 GABA+/tCr were significantly affected during the stimulus-ON in the last functional block, while changes in Glx/tCr were significantly impacted during stimulus-OFFs in both functional blocks (FUNC1 and FUNC2).Utilizing the sliding window analysis as an exploratory technique to examine metabolite changes at a higher temporal resolution, we demonstrated alterations in V1 Glx/tCr and Glx/tCr levels early in the first functional block while STS Glx/tCr and GABA+/tCr levels changes happened toward the last two functional blocks of REST2 and FUNC2.
When considering GABA and Glx levels using a block-type paradigm (REST blocks versus FUNC blocks) we observed no significant changes between the rest and functional blocks for either V1 or STS.Subsequently, we also did not observe any significant differences between face and object stimuli.A possible explanation for this result is that a change in metabolite level is too marginal to be detected by fMRS with a typical number of transients ($200 transients).While a block level analysis holds the advantage of the presumably summative effect of repeated stimuli within the same block, small metabolite responses might be averaged out with background changes (i.e.brain energetic consumption). 32Another possible explanation for the absence of significant changes for faces versus objects is that these stimuli are simply not salient enough to elicit a metabolite response with sufficient SNR, given the relatively short stimulation period given for each presentation of 24 s.A previous fMRI-fMRS study demonstrated a lack of significant metabolite level changes at low contrast, but the response increased with increasing contrast levels. 413][44][45] These results suggest there is a possible sensitivity threshold for the functional stimulus to elicit a detectable response in fMRS.It is possible that this threshold will be strongly dependent on the stimulus type and paradigm design as well. 19However, this lack of significant change between conditions also suggests good reliability between REST1 and REST2 as the control condition, in agreement with studies showing a fixation cross has good reliability as a control condition. 46 additionally investigated the metabolite change in an event-related design, separating transients into stimulus-ONs and stimulus-OFFs regardless of stimulus type to investigate potential lag of metabolite response with stimulus blocks, FUNC1 and FUNC2.For V1, we found a relative decrease in GABA+/tCr during the last stimulus-ON block, which we hypothesized might be due to cumulative change with continuous stimulus-ON T R times.This agrees with studies suggesting that as stimulation continues more GABA becomes bound to receptors, leading to  2.
linewidth broadening, and thus making it less detectable with MRS. 47,48The negative relationship of GABA with time might suggest that the evoked inhibitory tone does not completely return to baseline after stimulation, which has been previously suggested in recent animal optogenetic-fMRS studies. 49Indeed, studies, including our recent meta-analysis, 21 suggest that GABA changes appear slowly.
With event-related analysis, we observed an increase in Glx/tCr relative to the start of stimulus-OFF (OFF1) in V1 toward the end of the first functional block (OFF2) and during the first half of the second functional block (OFF3).This increase in Glx/tCr levels generally agrees with several studies showing an increase in glutamate in response to visual stimulation. 19,29,44,45,50Interestingly, we found changes to be in stimulus-OFFs rather than stimulus-ON transients, suggesting a potential delay in the metabolic response of Glx/tCr in V1.This result is in agreement with previous studies, which observed 3%-6% glutamate increase after at least 16-20 s of stimulation as measured in a block paradigm. 29Other studies that used event-related paradigms showed 9%-12% glutamate changes after 300-1000 ms of stimulus onset. 32,51Here, we have illustrated how an event-related paradigm might capture task-related transient metabolic changes that might be hidden by potential habituation and homeostatic regulation in a block paradigm.
T A B L E 2 Summary of linear mixed model fit for stimulus-ON and stimulus-OFF measurements of GABA+/tCR and Glx/tCr across functional stimulus types.In an attempt to establish the metabolite dynamics of both GABA+ and Glx further, the spectra were analysed using a sliding window method.This analysis method allows for metabolite quantification with higher temporal resolution, in our case with an effective time course of 10 s (five transients) with the metabolite levels averaged over 1 min 40 s (50 transients).Our sliding window analysis results demonstrated fluctuations of Glx/tCr and GABA+/tCr with no discernible pattern in either V1 or STS, irrespective of stimulus type.
In this study, both face and object stimuli demonstrated a significant effect on both V1 Glx/tCr and V1 GABA+/tCr during FUNC1, and the interaction terms suggest that changes are dependent on time.Despite the significant changes detected with event-related analysis during FUNC2, no significant changes in response to stimulus were observed during FUNC2 with the sliding window approach.This suggests that the changes during FUNC2 might be more abrupt or a quick change in trend, as the sliding window analysis approach could potentially smooth out such alterations in metabolite dynamics.
For STS, the linear mixed model did not indicate any significant effect of the social stimulus presentation (face stimulus presentation on both STS Glx/tCr and STS GABA+/tCr).However, the results suggest a consistent effect of object stimulus presentation during FUNC2 on both Glx/tCr and GABA+/tCr, which is time independent for GABA+/tCr.The results also showed a significant effect of the second rest block on both STS Glx/tCr and STS GABA+/tCr, with a significant interaction between REST2 and time of measurement suggesting time dependence of the effect.It is possible that the significant effect of REST2, despite the absence of stimulus presentation, might be due to the nature of the sliding window analysis, which includes the neighbour signal in other stimulus blocks (i.e., the previous FUNC1 block).However, based on supplementary Bayesian analysis, there may be a potential effect of face stimuli on STS Glx/tCr levels, although this cannot be conclusively established in this current study (full details in the Supporting Information).
Changepoint analysis based on Bayesian inference approach reflected that GABA+/tCr and Glx/tCr levels acquired with sliding window analysis change in a gradual fashion for stimuli in both V1 and STS, with a broad estimation of 95% CI of the changepoint.In comparison to STS, the changepoint in V1 is suggested to be appear early in the first rest block with a changepoint of approximately 7 min for V1 GABA+/tCr and approximately 3 min V1 Glx/tCr.These results are in agreement with a recent study that investigated metabolite dynamics of GABA+ and Glx in V1 with high-resolution dynamic analysis and showed Glx and GABA drift despite the absence of stimulation. 20Another explanation for the detected changepoint without the stimulus presentation might be due to the fixation cross as the baseline, which may subtly influence metabolite dynamics.Indeed, there is some evidence of eyes open versus eyes closed. 45Nonetheless, a fixation cross was demonstrated to induce the least variation in glutamate levels and serves as the most behavioural constrained task compared with other rest conditions (i.e., flashing checkerboard or finger tapping), 46 so should have minimal effect on the results.Other possible explanations for Glx change could be other brain processes and/or that the detected Glx is influenced by glutamine rather than glutamate. 20 the other hand, the changepoint analysis identified the changepoint of STS GABA/tCtr at around 3 min after the start of FUNC1 (around 11 min from the start of the experiment), while a changepoint of about 2 min after the start of REST was identified for Glx/tCr (around 17 min from the start of the experiment).These results agree with the linear mixed model analysis, which found a significant effect of REST2 on STS Glx/tCr levels that depends on the time of measurement, suggesting glutamate might take roughly 2-3 min to reach its steady state through the synaptic reorganization process after the stimulus onset. 18,42,50The different changepoints identified between STS and V1 could be explained by timing differences between metabolite types and between brain regions, as previously shown in several studies. 19,29 did not observe significant changes in metabolite responses evoked by social stimuli (i.e., face stimuli) in the STS region with the sliding window analysis approach.While STS is a cortical region responsible for social perception from visual cues, in this current study only V1 showed significant changes to stimulation.These results are consistent regardless of analytical approach.Therefore, our findings probably reflect neurochemical responses to 'basic' visual processing rather than social processing per se.While previous studies have shown the sensitivity of fMRS in detecting glutamate changes that differ in the response to the presentation of either objects or abstract pictures, 51 this current study suggests that the Glx response is not specific to the type of visual stimulation.Another possible explanation for the changes observed in V1 but not STS is that V1 has closer proximity to the receiver coils, thus higher SNR, thereby increasing the statistical power to detect changes. 52e current study does have some limitations.It should be noted that the sample size of this study is small and thus the effects observed should be interpreted as a preliminary finding that requires future exploration with larger sample size.Additional post hoc power analysis was conducted to guide further research and suggests that a relatively large number of subjects are required to detect the small effect size indicated in our study (full details in Supporting Information).However, it is likely that the effect size would vary across different brain regions, stimulus designs, and analysis approaches. 53Furthermore, power calculations could vary for studies focusing on autistic individuals or those with other neurodevelopmental conditions, as these groups may exhibit differences in both effect size and latency that have yet to be elucidated.The order of V1 and STS was not randomized, as our primary interest was in STS given the social context.However, fatigue and possible examiner bias could have influenced these results.We recognize that counterbalancing is important in future studies.T A B L E 3 Summary of linear mixed model fit for sliding window analysis of GABA+/tCR and Glx/tCr in STS and V1.Only parameters with significant effect are shown here; for full parameters for each model please refer to Table S1.
MRI and MRS data were acquired using a Siemens Trio 3 T MRI scanner (Siemens, Erlangen, Germany) with 32-channel head coil at Swinburne University of Technology, Melbourne, Australia.T 1 -weighted structural images for MRS voxel localization were acquired using a magnetization pre-prepared rapid gradient echo (MPRAGE) pulse sequence with an inversion recovery (176 slices, slice thickness = 1.0 mm, voxel resolution = 1 mm 3 , repetition time (T R ) = 1900 ms, echo time (T E ) = 2.52 ms, inversion time (TI) = 900 ms, flip angle = 9 , field of view 350 mm Â 263 mm Â 350 mm, acquisition time = 5 min).

F
I G U R E 4 GABA+/tCr and Glx/tCr levels in V1 (A, B) and STS (C, D) regions based on fMRS blocks across both functional stimulus types (i.e.faces and objects).FUNC1, FUNC2, participants passively viewed social (faces) or non-social (objects) stimuli, counterbalanced; REST1, REST2, participants passively viewed fixation cross.
Raincloud plots of GABA+/tCr and Glx/tCr in V1 during stimulus-ON (24 s stimulus-ONs of 3 s movie clips of dynamic faces or dynamic objects) or stimulus-OFF (16 s rest-fixation cross) blocks.p values (<0.01**, and <0.05*) were obtained from a linear mixed model, indicating the significant difference between the marked condition with reference condition (the first block of ONs and OFFs), and the full results are presented in Table

F
I G U R E 7 GABA+/tCr and Glx/tCr dynamics from sliding window analysis in V1 (window width = 50 transients).A, B, GABA+/tCr (A) and Glx/tCr (B) levels irrespective of stimulus type during FUNC1 and FUNC2.C, E, V1 GABA+/tCr levels in response to face (C) or object (E) stimuli during FUNC1 and FUNC2.D, F, V1 Glx/tCr levels in response to face (D) or object (F) stimuli during FUNC1 and FUNC2.The time represents the time at the centre of each window.The line fit represents the mean value of metabolite levels in each window; the purple ribbon indicates the 95% CI.

F
I G U R E 8 GABA+/tCr and Glx/tCr dynamics from sliding window analysis in STS (window width = 50 transients).A, B, GABA+/tCr (A) and Glx/tCr (B) levels irrespective of stimulus type during FUNC1 and FUNC2.C, E, STS GABA+/tCr levels in response to face (C) or object (E) stimuli during FUNC1 and FUNC2.D, F, STS Glx/tCr levels in response to face (D) or object (F) stimuli during FUNC1 and FUNC2.The time represents the time at the centre of each window.The line fit represents the mean value of metabolite levels in each window; the purple ribbon indicates the 95% CI.

F I G U R E 9
Results of changepoint analysis for V1 (A, B) and STS (C, D) regions.The grey/black line represents the posterior fit derived from the fitted models, the blue/black lines below each plot illustrates the posterior density of changepoints of each iteration chain at various time points.The height of the posterior density of changepoints signifies the potential values for changepoint locations, with the y axis denoting the density (i.e.posterior probability) at each specific time point.