Does Theta-Burst Stimulation Change Brain GABA in Methamphetamine Use Disorder? The 50-Person Interaction Test

Direct answer: A randomized sham-controlled study involving 50 adults with methamphetamine use disorder reported a within-group decrease in a left-prefrontal GABA/NAA magnetic-resonance-spectroscopy signal after four weeks of active intermittent theta-burst stimulation. But the crucial active-versus-sham group-by-time interaction for GABA/NAA was not statistically significant (p = .695). Glx/NAA likewise lacked a significant treatment interaction (p = .176). In the same trial, cue-induced craving and a verbal-learning measure did show statistically significant group-by-time interactions. Therefore the study supports selective behavioral treatment signals, but does not establish a stimulation-specific brain-GABA effect, prove a mechanism of improved cognition, measure methamphetamine in blood, or establish an individual legal impairment threshold.

The original research was conducted by Hang Su, Tianzhen Chen, Na Zhong, Haifeng Jiang, Jiang Du, Ke Xiao, Ding Xu, Zheng Wang, and Min Zhao. Their 2020 report, γ-aminobutyric acid and glutamate/glutamine alterations of the left prefrontal cortex in individuals with methamphetamine use disorder: a combined transcranial magnetic stimulation-magnetic resonance spectroscopy study, appeared in Annals of Translational Medicine 8(6):347. Read the original empirical study for its Methods, main figures, supplementary numerical tables, and authors’ complete analysis. The distinctions below depend on the printed comparisons rather than a general claim that brain stimulation improves addiction.

What question does this experiment actually answer?

The experiment asked whether four weeks of active magnetic stimulation over a specifically selected left-prefrontal scalp target would change regional proton-spectroscopy metabolite signals and selected cognitive or craving outcomes in adults already diagnosed with severe methamphetamine use disorder. That is a question about a stimulation intervention within a rehabilitation setting. It is not a controlled experiment assigning an acute methamphetamine dose, nor a field sobriety validation or a blood-test calibration study.

At least three distinct outcomes have to remain separate. First, a spectroscopy ratio summarizes resonances attributed to brain metabolites in one small region. Second, a self-reported visual-analog craving rating describes a subjective response to drug cues. Third, computerized learning and maze tasks summarize performance on particular standardized tasks. Finding change in one does not demonstrate an equal change in the others, still less that each has become an independent forensic marker of intoxication or actual conduct. This article examines exactly where the randomized evidence is stronger and where it is weaker.

How many independent participants were studied?

Investigators recruited 50 adults meeting DSM-5 criteria for severe methamphetamine use disorder, including 20 women. Randomization assigned 25 participants to active stimulation and 25 to sham treatment. Both groups had previously used methamphetamine, and the study did not include a healthy comparison arm or deliberately administer methamphetamine during outcome testing. Each person underwent preintervention and postintervention assessments; two study visits do not double the number of independent people.

Likewise, twenty stimulation sessions per participant, multiple nuclear-magnetic-resonance signal averages, repeated cognitive trials and individual survey items do not produce hundreds of independent patients. The central biological N is 50, not 1,000 sessions or the total number of spectral data points. Baseline group comparisons did not show the authors a statistically significant difference on the assessed participant characteristics, but absence of detected differences in this small trial does not establish perfect exchangeability on all possible prior exposures, health conditions, or concurrent influences.

What treatment did active and sham groups receive?

Active sessions used intermittent theta-burst stimulation directed toward the left dorsolateral prefrontal cortex, located by the Beam F3 scalp method. The source describes short bursts of three 50-Hz magnetic pulses repeated at 5 Hz, with two-second trains separated by eight-second pauses. Each session delivered 900 pulses over approximately five minutes, and treatment was repeated on weekdays for four weeks—twenty planned sessions. The nominal intensity was set relative to each person’s measured resting motor threshold.

The sham procedure used the same timetable but turned the stimulation coil away from the skull. Participants and outcome assessors were described as blinded. A rotated coil provides a comparison for apparatus exposure and expectations; it should not be silently equated with an ideal physically identical placebo for every scalp sensation or auditory experience. No experimental methamphetamine milligram dose was supplied in either arm. Here “900” is a count of magnetic pulses, not drug molecules, blood concentrations or independent volunteers.

What does proton magnetic resonance spectroscopy measure here?

A Siemens 3-tesla MRI system obtained proton magnetic resonance spectra from a 20 × 20 × 20 millimeter voxel in the left dorsolateral prefrontal cortex. The researchers used an edited MEGA-PRESS acquisition for the GABA-associated signal and a shorter-echo PRESS acquisition for major resonances including combined glutamate/glutamine, commonly labeled Glx. The spectral signal was fitted computationally rather than obtained by removing brain tissue or running chromatography on a blood specimen.

The investigators compared GABA and Glx signals with a resonance attributed to N-acetylaspartate, abbreviated NAA. Their principal outcomes therefore were GABA/NAA and Glx/NAA ratios. A ratio is not a direct count of neurotransmitter molecules, not a chemical confirmation of methamphetamine and not a validated conversion into legal venous whole-blood drug concentration. The same signal fitting and tissue composition can be influenced by spectral quality, voxel placement, metabolite overlap, normalization, and the assumptions embedded in the fitting method.

Why was NAA used as the reference?

The researchers reported no significant baseline-to-postintervention change in the NAA signal and normalized the other resonances to that reference. Normalization can make measurements across scans more interpretable by reducing some variation in overall scale, but a stable average reference is still a methodological choice. It does not turn a relative signal into an independently calibrated absolute brain concentration.

Moreover, apparent stability of an average reference signal should not be presumed to hold perfectly for each participant or to exclude measurement uncertainty. A ratio can change because its numerator changes, its denominator changes slightly, or both contribute. The original protocol performed tissue-segmentation work and recorded signal-quality indicators; those steps add methodological discipline, yet they do not establish a forensic assay for circulating stimulants or an individual chemical impairment threshold.

Did active stimulation significantly lower GABA/NAA?

Within the active arm, yes: the authors reported mean GABA/NAA decreasing from 0.21 ± 0.03 to 0.18 ± 0.03, with a within-group p value of .040. The sham mean changed from 0.20 ± 0.02 to 0.19 ± 0.03, with p = .143. These two p values do not, however, answer the randomized comparison. The directly relevant analysis asked whether the amount of change differed between treatment groups.

The printed group-by-time interaction was F = 0.158, p = .695—not significant. The authors also reported a time main effect F = 4.221, p = .046 and a post-treatment group main effect F = 0.270, p = .608. The appropriate conclusion is a nominal active-arm before/after signal alongside no demonstrated active-specific difference in change. “Significant in one group but not the other” is not itself evidence that the two groups differed significantly. This statistical distinction is central to evaluating an asserted stimulation-specific neurochemical mechanism.

Did Glx/NAA change specifically because of active stimulation?

No significant active-versus-sham Glx/NAA interaction was reported. In the active arm, mean Glx/NAA went from 1.51 ± 0.24 to 1.45 ± 0.21, with within-arm p = .458. In sham it went from 1.58 ± 0.24 to 1.38 ± 0.20, with p = .012. The reported group-by-time interaction was F = 1.926, p = .176, so the apparent within-sham decrease does not establish a significant difference in treatment response compared with the active arm.

The authors reported a significant overall time effect F = 8.878, p = .005 and a nonsignificant post-treatment group comparison F = 1.072, p = .309. This pattern illustrates why individual-arm change tests, main effects and interactions must be kept separate. The combined Glx resonance also does not resolve glutamate from glutamine as two independently verified concentrations. Assigning all movement to one neurotransmitter or deriving an individual stimulant-induced chemical imbalance from this composite signal would exceed the study’s reported measurement.

Evidence table 1: main original spectroscopy comparisons

The table deliberately places significant within-arm results next to the nonsignificant treatment interactions. Its entries are regional spectroscopy ratios rather than drug levels in blood or brain-tissue chromatography. The numerical signs and precision are retained as printed. No missing interaction confidence interval, baseline-adjusted effect size, or individual threshold has been calculated or invented.

Did participants report different changes in drug-cue craving?

Yes. Cue-induced craving was rated on a visual-analog scale after exposure to methamphetamine-associated stimuli, and the group-by-time interaction was F = 9.706, p = .003. The active group’s mean decreased from 30.47 ± 28.59 to 5.03 ± 7.82, whereas the sham group changed from 28.97 ± 25.07 to 26.52 ± 28.59. Within the active group the change was p < .001; the sham within-group comparison was p = .745.

This is a stronger randomized between-group signal than the one for GABA/NAA. It remains a subjective cue-triggered rating in the study context, not a chemical assay and not a documented individual abstinence or relapse outcome. A participant’s rating can be affected by context, expectations, time, intervention experience, recovery process, and how the cues were presented. Neither a low postintervention group mean nor a significant interaction independently determines whether an identified person used a stimulant, was intoxicated, or engaged in conduct alleged in a legal dispute.

Did the verbal-learning test respond differently between groups?

The International Shopping List task, abbreviated ISL, showed a significant group-by-time interaction F = 14.917, p < .001. Correct scores increased from 20.60 ± 4.74 to 23.83 ± 4.97 in the active group (within-group p = .006). Sham scores changed from 20.38 ± 4.74 to 18.69 ± 5.12 (within-group p = .092, not significant). This supports a treatment-related difference in performance on a particular verbal-learning test under the trial’s analysis.

The test is not a forensic validation of general intelligence, witness reliability, prospective real-world memory, a driver’s situational awareness, or the reason for any event. Its performance variable is a correct-item score on a specified learning protocol. Importantly, the investigators did not show that the spectroscopy GABA ratio statistically mediated the randomized verbal-learning difference; a group difference in ISL and a separate GABA within-arm finding should not be joined into a causal molecular pathway without direct evidence.

What happened on the Groton Maze Learning task?

The investigators also presented performance on the Groton Maze Learning task, a computerized measure in which fewer errors generally means better maze learning and problem solving. The active-arm mean error measure declined from 71.93 ± 27.14 to 59.80 ± 23.35, within-arm p = .003. The sham mean declined from 79.97 ± 32.10 to 73.86 ± 29.48, within-arm p = .205. The source reports an overall time effect F = 6.166, p = .016.

Crucially, the main-text passage does not provide the same clearly documented significant treatment-by-time interaction that it provides for craving and ISL. It would therefore be an overstatement to convert the active-arm p = .003 into proof of a randomized differential treatment effect on maze errors. Repeated tests can change with learning or practice. These are scores in a structured laboratory task; the researchers did not validate them as a cutoff for crime-scene reconstruction, a measured measure of vigilance during driving, or a biomarker of current methamphetamine exposure.

Evidence table 2: behavioral outcomes and metabolite associations

As with the first table, the p values address different hypotheses: a randomized treatment interaction, a main effect over time, or a within-group association. Combining them as interchangeable evidence of a “confirmed GABA mechanism” would disregard the analysis structure. Changes in cognitive task scores and survey ratings do not verify chemically measured illicit drug levels in a legal biological specimen.

Did GABA changes track cognitive performance?

The study reported a correlation between the change in its GABA-associated signal and the change in Groton Maze Learning errors within active stimulation participants: r = −.499, p = .011. In sham participants the corresponding correlation was r = −.190, p = .374, which was nonsignificant. The source’s Figure 4 shows individual points and fitted lines for these two associations. A correlation is a relationship between observed changes, not demonstration that altering GABA causes the maze error score to improve.

Interpretation also requires care about the direction of the plotted variables and whether larger GML scores reflect worse performance. The authors themselves described their direction-of-change interpretation as confusing. It would be misleading to reproduce the visually appealing slope as if it confirmed a mechanistic chain from stimulation to less GABA to fewer errors. The randomized treatment interaction for GABA was nonsignificant, and this exploratory correlation involved multiple possible neurochemical and behavioral comparisons. At most it motivates more focused mechanistic research.

Was the brain GABA change correlated with craving reduction?

Not convincingly in this reported analysis. Supplementary Table S3 gives the active-arm change in GABA versus change in craving as r = .049, p = .817, and the sham-arm association as r = −.321, p = .126. Both were nonsignificant. The combined observations therefore include a clear group difference in self-reported cue-craving trajectories but no significant within-arm correspondence between that craving change and the GABA change in the supplementary correlation test.

This discrepancy matters scientifically. If the claim is that stimulation changed craving by changing this particular prefrontal metabolite ratio, one would seek appropriately powered, prespecified causal or mediation evidence. Separate p values for the intervention and the brain ratio, plus a nonsignificant correlation with craving, do not establish that pathway. Nor does a single voxel capture the whole executive-control network; other cortical and subcortical changes could be relevant but were not comprehensively measured.

Were other metabolites or scanner quality indicators reported?

Yes. The supplementary material contains additional neurometabolites and quality measures. For example, sham-group inositol changed from 1.79 ± 0.46 to 1.48 ± 0.37 (within-arm p = .026) and sham creatine plus phosphocreatine from 0.86 ± 0.12 to 0.76 ± 0.11 (p = .018). Those changes were reported separately from the main GABA/Glx outcomes and do not themselves demonstrate beneficial sham effects. Several other listed measurements, including active-arm inositol and creatine-plus-phosphocreatine comparisons, were nonsignificant.

The investigators documented spectral linewidth, signal-to-noise ratios, fitting uncertainty, and tissue-fraction indices. The preintervention GABA spectrum signal-to-noise ratio, for instance, was 14.74 ± 3.15 in active participants and 13.78 ± 3.49 in sham (baseline p = .335). Such quality indices help a reader assess comparability, but a good fit or clean scanner signal is not chemical chain-of-custody validation. It does not identify circulating methamphetamine, rule out alternative drug use at a legal event, or establish the specificity of a metabolite-to-behavior causal interpretation.

Why does a significant within-group p value not prove a treatment effect?

The wrong comparison can change a scientifically cautious finding into an overclaim. Imagine two groups: one moves slightly more than the other on a repeated assessment. It is possible for a pre-versus-post calculation to cross an arbitrary significance boundary in one group while the other does not, even when the uncertainty about the difference between their changes is large. A proper randomized differential-effect test addresses that latter uncertainty directly.

In this trial, GABA/NAA illustrates the issue concretely. Active within-arm p was .040 and sham p was .143, yet the treatment-by-time p was .695. Glx/NAA supplies the mirror image: sham within-arm p was .012, active within-arm p was .458, but their interaction p was .176. Neither pattern proves the treatment selectively caused the observed metabolite decrease. This matters especially for an expert addressing a jury: statistical significance in a subcomparison does not license an unreported or unsupported cross-group conclusion.

What are the trial’s most important methodological limitations?

The treatment groups contained only 25 people each, and they were selected from one rehabilitation setting rather than randomly sampled from all people who use methamphetamine. Baseline clinical and prior-use histories are not the same as standardized toxicological assays. The original measurements covered one left-prefrontal voxel and selected tests at baseline and after four weeks, not continuous spectroscopy throughout recovery or direct sampling from connected brain regions. A spectral ratio in one site is not an unbiased survey of every neural structure.

The investigators evaluated multiple metabolite signals, psychiatric scales, cognitive tasks, quality measures, and exploratory correlations. Reported nominal p values may be sensitive to model specification, sample size, and multiplicity; the article should not be converted into a personalized therapy-response predictor without external validation. Other potentially influential issues include imperfect sensory matching between active and rotated-coil sham, task familiarity, treatment-setting effects, past nicotine and methamphetamine exposure, and uncertainty in metabolite normalization. The study cannot resolve all of those factors through group randomization alone.

How does this differ from other iTBS methamphetamine experiments?

This original study’s distinctive feature is proton MRS measurement of left-prefrontal GABA/NAA and Glx/NAA in 50 randomized adults. Other research may investigate scalp event-related potentials during methamphetamine-word Stroop tasks, visual working memory with patterned stimuli, or self-rated executive-function domains after different pulse schedules and in different cohorts. Such findings should not be pooled as though the same participants generated brain metabolite spectra, EEG components, executive ratings, and error-trial outcomes.

Even studies that place a coil over the same broad scalp target may differ in number of sessions, pulse count, sham method, rehabilitation conditions, completion rates, outcome definitions, statistical models and reporting quality. A shared label such as “theta-burst stimulation” does not make every result interchangeable. For SEO and scientific accuracy, the relevant precise question here is the contrast between within-arm MRS changes and randomized GABA/Glx interaction tests alongside craving and ISL interactions—not whether every stimulation paper generally demonstrates reduced drug craving.

Can an MRS ratio identify methamphetamine intoxication in a specific person?

No. The investigators did not chemically quantify parent methamphetamine or amphetamine in a legally collected venous whole-blood specimen. They did not report validated paired circulating plasma/serum concentrations from which a sample-specific exposure could be reconstructed. There was no quantitative methamphetamine assay in urine, oral fluid, hair, sweat, vitreous, milk or real exhaled breath, and no validated one-to-one relationship between an MRS brain metabolite ratio and any such specimen was established.

This distinction is more than terminology. A brain spectroscopy resonance, a fitted concentration ratio, a self-report score and a confirmation test for a parent drug are different measurements with different reference standards. See the published explanatory guides on biological specimen matrices and confirmatory chemical drug testing. Neither scientific approach can simply substitute for the other without validated calibration and sampling history. Even a scientifically sound group effect would not establish an identified person’s exact last-use time.

What would be an appropriate forensic statement of the evidence?

A careful expert can describe the randomized population, 25-person arms, 20-session stimulation schedule, voxel dimensions, metabolite normalization, and actual behavioral endpoints. The report should say explicitly that active-arm GABA/NAA decreased at nominal within-group p = .040, but there was no significant active-versus-sham change interaction (p = .695). It should likewise preserve the Glx/NAA null interaction (p = .176) and distinguish significant craving (p = .003) and ISL (p < .001) treatment interactions from the more exploratory or incomplete GML comparison.

It would not be justified to present the source as demonstrating a specific legal whole-blood drug level, a current psychoactive dose, a patient’s reliable abstinence status, individualized prediction of treatment response, witness truthfulness, incident fault, or the cause of an actual collision. Any such case-specific conclusion would require separate authenticated physical specimens, chain of custody, fit-for-purpose analytical confirmation, medical and behavioral chronology, and validated linkage to the legal question. The legitimate strength of the trial is a bounded scientific observation, not a universal forensic rule.

Seven focused questions about iTBS, brain spectroscopy and methamphetamine

Was brain GABA significantly reduced compared with sham?

A nominal within-active before/after GABA/NAA decline was reported (p=.040), but the randomized active-versus-sham group-by-time test was nonsignificant (p=.695). The latter does not show a specific treatment effect on this metabolite ratio.

Did Glx/NAA rise after stimulation?

No consistent active-arm rise was reported. Active mean Glx/NAA declined numerically from 1.51 to 1.45 with p=.458, while sham declined from 1.58 to 1.38 with p=.012; the group-by-time contrast was nonsignificant, p=.176.

Were 50 people or 1,000 treatment sessions the independent sample?

Fifty independent adults were randomized, 25 per group. Repeated sessions, spectral averages, cue ratings and computerized trials were nested within those participants, not additional independently recruited people.

Was methamphetamine administered to the participants during this trial?

No controlled methamphetamine dose was assigned. The manipulated exposure was active versus sham magnetic stimulation in adults with an existing history and diagnosis of methamphetamine use disorder.

Did craving and verbal learning change differently by treatment?

Yes under the authors’ models: cue-craving group-by-time p=.003 and International Shopping List correct-score interaction p<.001. Neither is an assay of circulating methamphetamine or proof of a biological GABA mediation mechanism.

Does the GABA–maze correlation establish causal cognitive improvement?

No. In active participants the within-arm change correlation was r=−.499, p=.011, whereas sham was r=−.190, p=.374. Correlation, multiple exploratory tests and unclear direction coding do not demonstrate that changing the GABA/NAA ratio caused a maze improvement.

Can the findings establish an individual DUI, last-use time or crash cause?

No. The study did not quantify methamphetamine or amphetamine in legally collected human venous whole blood and did not validate any MRS, craving or cognitive measure against individual roadside intoxication, incident causation, driving errors or witness credibility.

Primary research and measured conclusion

The evidence rests on one randomized fifty-person original empirical article, not a pooled meta-analysis or a rephrasing of another stimulation cohort. The source should be consulted for its original participant table, printed GABA and Glx figure captions, the ΔGABA–ΔGML scatterplot and supplementary metabolite correlations. The original primary DOI provides the authoritative underlying study; the present explanation is independently written and does not reproduce the publisher’s figures or original tabular layouts.

In summary, active treatment was accompanied by within-arm GABA/NAA change, but neither GABA/NAA nor Glx/NAA showed a statistically established differential active-versus-sham interaction. Craving and verbal-learning outcomes did show reported treatment interactions, while a maze-task association was exploratory and did not demonstrate a GABA-mediated causal mechanism. The paper is informative about one brain-stimulation protocol and a narrowly defined population. It cannot substitute for quantitative forensic drug chemistry, individual impairment evaluation, or fact-specific legal causation analysis.