At 12 weeks, 27 patients were classified as responders and 15 were classified as non-responders, approximately 64% response rate in keeping with the STRATA study data [6]. One non-responder was excluded as CRLB exceeded 20% for glutamate at pACC voxel. As shown in Table 1, in the final dataset, the three groups were matched in demographic characteristics. Responders and Non-Responders did not differ in symptom burden at baseline, duration of psychosis, or risperidone dosage at 12 weeks. While both Responders and Non-Responders showed an improvement in symptom burden over 12 weeks, non-Responders had significantly higher symptom burden than Responders at 12 weeks across all symptom domains. In particular, non-responders showed no notable improvement in negative symptoms between the time points (t= -0.41, p=0.688).
As shown in Table S1, dropouts and included FES in the study haven’t significantly difference in age, education, gender, and symptom burden.
Baseline glutamate metabolites and prediction of response
As shown in Table 2 and Figure 1, a repeated measure ANOVA showed the main effect of group (F1,79=3.20, p=0.046, partial η2 = 0.075), site (F1,79=78.68, p<0.001, partial η2 = 0.499) and for metabolites (F2,79=422.83, p<0.001, partial η2 = 0.843). Non-responders had higher neurometabolite levels compared to responders, and pACC voxel had higher levels compared to dACC voxel. There were no group by site (F2,79=0.82, p=0.446, partial η2 = 0.020) or group by metabolite (F2,79=0.49, p=0.613, partial η2 = 0.012) interactions. Both higher Glu in pACC (B = -0.18; S.E. = 0.08; Wald = 4.59; df = 1; P = 0.032; Exp(B) = 0.84; 95%CI (0.71,0.99)) and higher Glx in dACC (B = -0.24; S.E. = 0.11; Wald = 4.30; df = 1; P = 0.038; Exp(B) = 0.79; 95%CI (0.64,0.99) at baseline significantly predicted non-Responder status at 12 weeks. Higher Glx in pACC (B = -0.11; S.E. = 0.06; Wald = 3.28; df = 1; P = 0.070; Exp(B) = 0.90; 95%CI (0.80,1.00)) and higher Glu in dACC(B = -0.26; S.E. = 0.13; Wald = 3.82; df = 1; P = 0.051; Exp(B) = 0.77; 95%CI (0.62,1.02) had associations with non-Responder status in the same direction but these relationships were not statistically significant.
ROC analysis returned comparable area under curve for both Glu and Glx at both voxels (0.71 for pACC Glu; 0.68 for PACC Glx; 0.70 for dACC Glu; 0.72 for dACC Glx) for non-responders. Glu in pACC was associated with an overall accuracy of 68.3% (proportion of correct predictions), sensitivity of 71.4 % (proportion of non-responders correctly predicted based on baseline Glu), specificity 77.8% (proportion of responders correctly predicted to respond). Glx in dACC was associated with an overall accuracy of 61.9%, sensitivity of 85.7%, specificity 63%.
Baseline glutamate metabolites and continuous measures of response
Glu and Glx in dACC and pACC at baseline were not significantly associated with the continuous measure of change in total PANSS scores over 12 weeks (P = 0.617 to 0.792). There was a significant relationship between baseline dACC Glu (Beta = -0.38; t = -2.61; P = 0.013) and Glx (Beta = -0.43; t = -3.03; P = 0.004) and change in PANSS negative subscale (Figure 2), while other PANSS subscale changes were not significantly associated with Glu and Glx at baseline in dACC (P = 0.386 to 0.723). In the pACC, Glu and Glx at baseline were not significantly associated with the change PANSS subscale over 12 weeks (P = 0.061 to 0.565). Thus, only the dACC metabolite levels, when higher, indicated poor improvement (even worsening burden), but this association was restricted to the negative symptoms domain.
Longitudinal changes in glutamate metabolites over 12 weeks
A repeated measure ANOVA showed that there were no significant changes in glutamate metabolite concentrations over time (F1,31=1.26, p=0.270, partial η2 = 0.039), and the effect of time on glutamate metabolites also did not differ according to response status over 12 weeks (F1,31=0.60, p=0.444, partial η2 = 0.019), (Table 2, Figure 3). The main effects of the voxel (F1,31=50.68, p<0.001, partial η2 = 0.620; pACC>dACC) and group (F1,31=5.92, p=0.021, partial η2 = 0.160; Non-Responder>Responder) were still seen after including this longitudinal MRS data, indicating that the pattern of a high glutamatergic metabolite level in non-responders compared to responders continued during the course of the 12 weeks of treatment, driven largely by the baseline difference.