Auditory cortex 1H-MRS data was included from three groups of participants: (1) participants meeting CHR-P criteria; (2) participants who did not meet CHR-P criteria but met criteria for non-psychotic disorders, such as mood and anxiety disorders, eating disorders and substance abuse (CHR-N), indicated by the Mini-International Neuropsychiatric Interview (44); and (3) healthy control individuals (HC) without an axis I diagnosis or family history of psychotic disorders. CHR-P status was confirmed by meeting ultra-high-risk (UHR) criteria according to the Comprehensive Assessment of At-Risk Mental States (CAARMS) interview (45) and the Cognitive Disturbances and Cognitive-Perceptive Basic Symptoms criteria according to the Schizophrenia Proneness Instrument, Adult version [SPI-A: (46)]. In total, 25 CHR-Ps met only basic symptom criteria, n = 23 met only UHR, and 51 CHR-Ps met both basic symptom and UHR criteria. Individuals classified as CHR-N did not meet threshold for UHR and basic symptoms, but met criteria for other non-psychotic disorders, excluding autism, ADHD, or first-degree relatives with ScZ. Cognition was assessed with the Brief Assessment of Cognition in Schizophrenia [BACS: (47)].

Participants meeting CHR-P criteria were reassessed at 3-, 6-, 9-, 12-, 18-, 24-, 30-, and 36-month intervals to examine persistence of APS and functional outcomes, using with the CAARMS interview.

Data from bilateral auditory cortex were acquired on a Siemens 3T-Tim-Trio scanner, using a 32-channel head coil for reception. First, T1-weighted anatomical MRI images were collected using 3D-MPRAGE sequences (192 slices, voxel size 1 mm³, FOV = 256 × 256 × 176 mm, TR = 2,250 ms, TE = 2.6 ms, FA = 9°), and resliced into axial and coronal views to allow more precise and consistent placement of the MRS voxels of interest. Voxels of 2 cm³ were then placed in right (RAUD) or left auditory cortex (LAUD), positioned over Heschl’s gyrus, horizontally aligned with the lower bank of the supra-temporal plane on the coronal slide (Figure 1, top panel), and optimized in location to a final position with minimal Cerebral Spinal Fluid (CSF) contamination. FASTMAP (48) semi-automatic shimming of the voxel was used to improve local field homogeneity in the area of interest. A total of three scans were acquired, including a full spectrum acquisition, a GABA-edited MEGA-PRESS (WIP: VB-17A) scan (128 trials), and an unsuppressed water scan to provide an absolute concentration reference (64 trials). MEGA-PRESS scanning parameters included: TR/TE 1,500/68 ms, 1.9 ppm ON- and 1.5 ppm OFF-resonance editing pulse frequencies (symmetric editing to suppress macromolecule contribution), 44 Hz editing Gaussian pulse bandwidth, 50 Hz water suppression, 90^° flip angle, acquisition bandwidth of 1,200 Hz, duration 426 ms, number of points 512.

The number of available samples differed for LAUD and RAUD voxels. In total across the three groups, 135 participants had recordings from both voxels, 62 only from right, and 1 only from left auditory cortex. As a result, analyses were run separately for both voxels. After removal of participants with poor quality data, the final sample included 92 CHR-P, 30 CHR-N, and 44 HC subjects for the RAUD voxel, and 60 CHR-P, 20 CHR-N, and 31 HC subjects for the LAUD voxel.

Cleaning involved removing data with incomplete/missing metabolite concentrations, and/or data that were more than 2 standard deviations away from the mean across groups (within each voxel) for linewidth of N-acetyl-aspartate [NAA] and/or unsuppressed water, and/or absolute Cramér-Rao Lower Bound (CRLB) values. Absolute CRLB values were favored over percentages following most recent recommendations for clinical situations of potential low concentrations of metabolites (49, 50). A full overview of data quality measures is presented in Table 1.

MEGA-OFF, MEGA-ON, and unsuppressed water spectra were processed in jMRUI v6.0 (51). Firstly, edited spectra were generated by subtracting MEGA-OFF from MEGA-ON and GABA, Glx and MEGA edited-NAA were quantified using the AMARES routine (52). Based on previous work (53), we used QUEST/QUASAR (54) to estimate glutamate and glutamine separately from edited MEGA-PRESS spectra (see Figure 1 for AMARES and QUEST example spectra). This approach has been shown to estimate levels of glutamate and glutamine that are consistent with literature values, provided spectroscopic quality is good, with the principal determinant being the linewidth of NAA (<10.6 Hz: means were around 6 Hz, see Table 1). As only glutamate spectral quality measures were met (%CRLB for LAUD 8.2 ± 3.1%, RAUD 7.1 ± 3.0%), but not those for glutamine (%CRLB for LAUD 40.9 ± 41.1%, RAUD 18.2 ± 26.7%), only glutamate levels are reported.

The edited MEGA-PRESS ON spectra were fitted to a basis set consisting of NAA, GABA, and glutamate in which the full edited spectra were represented. The basic model sets were generated from a simulated MEGA-PRESS sequence using the routine NMRSCOPE-B. GABA was fitted as a single Gaussian line with the linewidth constrained to be between 20 and 25 Hz at full width, half maximum height with the frequency constrained to be +1.00 ppm away from NAA. Glx was fitted as a symmetric doublet centered at 3.76 ppm with a Lorentzian line shape, a 10.4 Hz separation, with the same linewidth as NAA. NAA was fitted as a Lorentzian line, 180° out of phase to GABA and Glx. The MEGA-OFF spectra were also fitted by AMARES to give estimates of the unedited NAA (peak at 2.02 ppm), of total creatine (Crn: peak at 3.03 ppm) and of choline-containing compounds (Cho: peak at 3.21 ppm; glycerol-phosphorylcholine, phosphorylcholine and free choline).

Finally, data was referenced to the separately collected unsuppressed water concentration and corrected for contamination of cerebrospinal fluid (CSF) fractions (metabolite concentration * 1/1-CSF fraction).

Group differences in demographic data, clinical assessment scores, as well as cognition (BACS scores) were assessed with non-parametric Kruskal-Wallis tests, using an alpha-level of 0.05 (2-sided tests). Post-hoc comparisons were Games-Howell corrected for multiple comparisons and corrected for ties. BACS data were first z-normalized to HC data, separately for females and males. Gender differences were tested with Chi-square tests.

As the number of available data were different for the RAUD and LAUD voxels, the main analyses described below were run for each voxel separately, using an alpha-level of 0.05 (2-sided).

First, potential group differences in voxel segmentation, NAA and Creatine levels were investigated using one-way Welch’s ANOVAs with the dependent-variables NAA/water, Creatine/water, gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), and GM/WM ratio, and the fixed-factor GROUP (HC, CHR-N, CHR-P). Both NAA and Creatine concentrations were obtained from the MEGA-OFF sequence, while water concentration was estimated from the unsuppressed water scan.

Subsequent analyses included General Linear Model ANCOVAs, using the dependent-variables GABA, Glx, glutamate (Glu), Glx/GABA ratio, and Glu/GABA ratio, and the fixed-factor GROUP (HC, CHR-N, and CHR-P), with AGE, GENDER and the ratio between GM and WM as covariates. Subsequent post-hoc pairwise comparisons for significant metabolites were Games-Howell corrected to control for multiple comparisons.

In addition, post-hoc analysis included two sets of follow-up CHR-P subgroups: 1) APS-persistent vs. APS non-persistent group, based on APS-criteria both at baseline and during 1-year follow-up, and 2) CHR-P participants with either good or poor functional outcomes (GAF scores good ≥ 65, poor < 65) at their latest follow-up assessment.

Finally, correlations between significant LAUD Glx concentrations and clinical, social and cognitive functioning, as well as age, known to correlate with Glx values (22, 55), were tested across the CHR-N and CHR-P groups, using linear regression with backward selection to identify the strongest predictors of abnormal metabolite concentrations. Potential collinearity issues were checked in the final model with VIF statistics.

There were no differences in age, or gender distribution between groups (Table 2). CHR-P and CHR-N groups, however, differed from HC in GAF scores (CHR-N: p < 0.001; CHR-P: p < 0.001) as well as in GF-social functioning (CHR-N: p = 0.002; CHR-P: p < 0.001), but GF-role was different from HC only in the CHR-P group (p < 0.001). Furthermore, the CHR-P group had less years of education, compared to controls (p = 0.017), and general cognition was affected only in CHR-Ps, who performed worse than controls on the BACS Token Motor task (p < 0.001), Symbol Coding task (p = 0.046), and had lower BACS composite scores (p = 0.018).

After removing data with partially missing metabolite concentrations or violations of the other quality control measures (Line width NAA and water, and CRLB values), group means and standard deviations for NAA and Creatine concentration, as well as GM, WM, CSF, or GM/WM ratios were not different between groups (HC, CHR-N, and CHR-P), neither for the LAUD nor for the RAUD voxel (Table 1).

ANCOVA analyses revealed no GROUP effects in any metabolites in the RAUD voxel (all p-values > 0.34). In contrast, the LAUD voxel showed a significant GROUP difference in Glx [F(2, 107) = 3.3, p = 0.039, uncorrected, partial η² = 0.060] that was affected by the AGE [F(1, 107) = 10.3, p = 0.002, partial η² = 0.089], but not by GENDER or GM-WM ratio. Exploratory post-hoc tests, including only AGE as covariate, revealed that this was driven mostly by the difference between the CHR-N and CHR-P groups (p = 0.029 uncorrected, p = 0.072 Games-Howell [GH] corrected), while the difference between either CHR-N vs. HC group (p = 0.603 uncorrected) nor CHR-P vs. HC group contrasts was significant (p = 0.058 uncorrected). Although there were two outliers in the CHR-P group (see Supplementary Material), these did not significantly influence the results. In addition, none of the other LAUD metabolites (Figure 2) or ratio scores were significant (Glx/GABA p = 0.65, or Glu/GABA p = 0.85).

Post-hoc analyses of LAUD Glx concentrations in the CHR-P group included ANCOVAs with age as covariate for two different contrasts, one for clinical outcome (APS persistent vs. APS non-persistent) as well as for functional outcome (poor vs. good functional outcome).

A total of 17 CHR-P participants with LAUD cortex data were assigned to the APS-P group. For n = 20 CHR-Ps, their baseline-recorded APS symptoms remitted (non-persistent APS-NP group). These two groups did not significantly differ from each other in LAUD Glx concentrations (APS-P vs. APS-NP: p = 0.400).

In a second contrast, CHR-P participants with good functional outcome in 12 months (GAF scores ≥ 65, n = 23) were compared in their LAUD Glx concentrations to a group with poor functional outcome (GAF scores < 65, n = 32). These groups also did not differ from each other in their LAUD Glx concentrations (GAF high vs. GAF low: p = 0.286).

A total of 11 CHR-Ps developed a psychotic episode during the follow-up period (mean/standard deviation of 17.6 ± 8.3 months; 3 < 12 month, 2 at 12 months, 6 > 12 month). For 7 of these CHR-Ps, LAUD cortex MRS data was available. Compared to healthy controls, their Glx levels were not significantly affected (p = 0.84).

Left auditory cortex Glx data from CHR-N and CHR-P group were entered together into a linear regression analysis with backward selection of potential predictive variables, including AGE, GAF, GF-role, GF-social, SPI-A total severity, CAARMS total severity, CAARMS subscales, BACS composite scores and scores on six included tests (verbal memory, digit coding, token motor, verbal fluency, symbol coding, and Tower of London task). A significant model was found [F(3, 76) = 4.9, p = 0.003], including the predictors AGE (Beta = −0.321, t = −3.0, p = 0.003, VIF = 1.010), CAARMS Perceptual Abnormalities (Beta = −0.198, t = −1.8, p = 0.081, VIF = 1.134), and CAARMS Disorganized Speech (Beta = 0.226, t = 2.0, p = 0.047, VIF = 1.133) items, that showed the same trend for both groups (Figure 2). In summary, LAUD cortex Glx levels were lower in CHR-N/P individuals with higher age, higher severity of perceptual abnormality, but lower disorganized speech symptoms.