Dopamine and Glutamate in Antipsychotic-Responsive Compared With Antipsychotic-Nonresponsive Psychosis: A Multicenter Positron Emission Tomography and Magnetic Resonance Spectroscopy Study (STRATA)

Abstract The variability in the response to antipsychotic medication in schizophrenia may reflect between-patient differences in neurobiology. Recent cross-sectional neuroimaging studies suggest that a poorer therapeutic response is associated with relatively normal striatal dopamine synthesis capacity but elevated anterior cingulate cortex (ACC) glutamate levels. We sought to test whether these measures can differentiate patients with psychosis who are antipsychotic responsive from those who are antipsychotic nonresponsive in a multicenter cross-sectional study. 1H-magnetic resonance spectroscopy (1H-MRS) was used to measure glutamate levels (Glucorr) in the ACC and in the right striatum in 92 patients across 4 sites (48 responders [R] and 44 nonresponders [NR]). In 54 patients at 2 sites (25 R and 29 NR), we additionally acquired 3,4-dihydroxy-6-[18F]fluoro-l-phenylalanine (18F-DOPA) positron emission tomography (PET) to index striatal dopamine function (Kicer, min−1). The mean ACC Glucorr was higher in the NR than the R group after adjustment for age and sex (F1,80 = 4.27; P = .04). This was associated with an area under the curve for the group discrimination of 0.59. There were no group differences in striatal dopamine function or striatal Glucorr. The results provide partial further support for a role of ACC glutamate, but not striatal dopamine synthesis, in determining the nature of the response to antipsychotic medication. The low discriminative accuracy might be improved in groups with greater clinical separation or increased in future studies that focus on the antipsychotic response at an earlier stage of the disorder and integrate other candidate predictive biomarkers. Greater harmonization of multicenter PET and 1H-MRS may also improve sensitivity.

http://adni.loni.usc.edu/methods/documents/mri-protocols/). T1-weighted images were automatically reformatted to provide axial, sagittal and coronal views. where M is the uncorrected metabolite concentration/ level, and WM, GM and CSF indicate the percentages of tissue content in the voxel. 4,5 The terms in the numerator are the ratios of the water content for each tissue type to the water content of white matter, since by default voxels are assumed to be pure white matter in the LCModel analysis. Since a relatively short TE and long TR were used, no corrections were applied for metabolite and tissue water T1 and T2 relaxation times, except for assuming T2 of tissue water = 80 ms. seconds, 5x120 seconds, 16x300 seconds) over the 95-minute period immediately postinjection.
Raw data were transferred to KCL for analysis. Head movement was corrected for by frameby-frame realignment using mutual information image registration. 8,9 An 18 F-DOPA template, 10 together with a striatal brain atlas, 11 and cerebellum 12 were normalized to each 18 Table 3).

Supplementary Discussion
As with many other imaging modalities, both PET and 1 H-MRS suffer from between-scanner variation that impacts on comparisons of images across imaging sites, scanners and over time. There are several factors that are responsible for this variability. Some of these will be related directly to patient preparation and execution of the PET or MRI examination while others relate technical factors.
In our 1 H-MRS data, although we took steps to harmonize the acquisitions, by using the same to Philips or Siemens scanners. 17 This effect is likely due to the use of outer volume saturation (OVS) using very selective saturation (VSS) pulses on the GE platform, which is not available on Philips or Siemens, leading to underestimation of the unsuppressed water peak. 18 In future multicenter studies, similarity of metabolite estimates across sites might be improved by avoiding VSS-OVS on GE platforms, although there are differences in other acquisition parameters (e.g. RF pulse shapes and timings) that may also need to be considered.
Nonetheless, across sites and metabolites, CRLB were <10%. This indicates that site differences in spectral quality were sufficiently small, and that spectra were of sufficient quality, to allow good fit.
To standardize our analysis pipeline and avoid the potential of introducing additional inter-site variance when acquiring manufacturer or scanner-specific basis sets, we chose to use a single basis set to analyze all 1 H-MRS spectra. This is the recommended approach within LCModel, 1 in situations where the data to be analyzed match an existing basis set in terms of localization sequence (i.e. PRESS) and (to within a few percent), B0 field strength, and TE, as was the case in our study. At short echo times as applied in our protocol, TE1 varies for the different manufacturers by ~ 1 ms, compared to the much larger variations that occur at longer TEs, which would affect the signals from strongly coupled spins. The limited number of other crossplatform multisite spectroscopy studies in psychiatry have also used single basis sets. 16,[19][20][21] Nonetheless Povazan et al., 22 recently compared single voxel PRESS at TE=35ms in healthy volunteers across multiple sites and manufacturers, using manufacturer-specific basis sets, and found no effects of manufacturer for the ratio to creatine of Glx, NAA or mI. Site contributed more markedly than vendor to variance in metabolite estimates 22 although another multicenter study at 3 Tesla using a single manufacturer have found high inter-site reproducibility at a similar TE. 23 Overall, there are very few multisite / multi-manufacturer reproducibility studies at 3 Tesla and there currently a lack of consensus on the optimal parameters including the use a single or customized basis set when combining data. These will be topical issues as further multicenter 1 H-MRS consortia emerge.
Effects of PET scanner on Ki cer have also been reported on previously. 24 This relates to the intrinsic sensitivity of the PET scanner as well as the reconstruction method and related parameters. As a result, PET images acquired in different sites can differ quite significantly in term on spatial image resolution, partial volume effects and noise content. Even with harmonization procedures to ensure comparable data quality are in place, residual differences in measured PET image may still be present. 25 In this project, despite having implemented a standardised PET acquisition protocol and analysis pipeline, the intrinsic differences of the KCL (Siemens Biograph 6 HiRez PET/CT) and Manchester (Siemens Biograph TruePoint TrueV PET/CT) scanners showed differences in PET data quality (Supplement Figure 4), which ultimately lead to differences in Ki estimates.
In addition to the main analysis investigating metabolite estimates corrected for voxel tissue composition (Mcorr), for further comparison with previous reports we additionally analysed ACC glutamate and Glx in ratio to creatine. There were no significant effects of age or gender on ACC Glu/Cr or ACC Glx/Cr z-scores (P>0.05). The group effect on ACC Glu/Cr was not significant in an unadjusted model (F1,84 = 0.31; P = 0.58), and was borderline significant for Glx/Cr (F1,84 = 3.86; P = 0.05) Inspection of water-referenced creatine z-scores found no significant effects of age (n = 85; r = -0.12; P = 0.28), gender (T84 = 1.60; P = 0.11) nor group (T84 = 1.24; P = 0.22, Supplement Table 4). However, while creatine is often used as an internal reference in 1 H-MRS studies 26 and did not significantly differ between the patient groups in our study, there are also reports of creatine differences in schizophrenia, 27, 28 and as a ratio denominator, small group differences in creatine may also influence results.