Introduction
Treatment-response heterogeneity (TRH) remains a central challenge in schizophrenia. Approximately 20%–35% of patients do not respond adequately to first-line antipsychotics and are classified as having treatment-resistant schizophrenia (TRS) (Hietala et al., Reference Hietala, Syvälahti, Vuorio, Räkköläinen, Bergman, Haaparanta, Solin, Kuoppamäki, Kirvelä, Ruotsalainen and Salokangas1995). Clozapine is the only established treatment for TRS, yet 40%–70% of patients remain symptomatic despite clozapine treatment and may be considered to have ultra-resistant schizophrenia (URS) (Shah et al., Reference Shah, Iwata, Brown, Kim, Sanches, Takeuchi, Nakajima, Hahn, Remington, Gerretsen and Graff-Guerrero2020). TRS and URS are associated with substantial clinical and functional burden, but biologically informed markers that distinguish different forms of treatment resistance remain limited.
Dopaminergic dysfunction remains central to pathophysiological models of schizophrenia. Positron emission tomography (PET) studies have consistently demonstrated elevated presynaptic dopamine function in schizophrenia, particularly within the dorsal and associative striatum (Abi-Dargham et al., Reference Abi-Dargham, Rodenhiser, Printz, Zea-Ponce, Gil, Kegeles, Weiss, Cooper, Mann, van Heertum, Gorman and Laruelle2000; Caravaggio, Borlido, Wilson, & Graff-Guerrero, Reference Caravaggio, Borlido, Wilson and Graff-Guerrero2015; Howes et al., Reference Howes, Montgomery, Asselin, Murray, Grasby and McGuire2007; Howes & Kapur, Reference Howes and Kapur2009). However, PET findings also suggest that dopamine abnormalities may differ according to treatment response. Increased striatal dopamine synthesis capacity has been reported in treatment-responsive patients, whereas TRS does not consistently show the same dopaminergic pattern (Demjaha et al., Reference Demjaha, Murray, McGuire, Kapur and Howes2012, Reference Demjaha, Egerton, Murray, Kapur, Howes, Stone and McGuire2014; Jauhar et al., Reference Jauhar, Veronese, Nour, Rogdaki, Hathway, Turkheimer, Stone, Egerton, McGuire, Kapur and Howes2019). Although PET provides important mechanistic information, it is invasive, costly, and not widely available in psychiatric practice, which limits its broader use for characterizing treatment-response heterogeneity.
Neuromelanin-sensitive MRI (NM-MRI) offers a noninvasive way to examine monoaminergic-related signal in vivo. Neuromelanin (NM), a byproduct of dopamine and noradrenaline metabolism, accumulates over years in monoaminergic nuclei, including the substantia nigra (SN), ventral tegmental area (VTA), and locus coeruleus (LC). Because NM is resistant to degradation, NM-MRI is thought to reflect cumulative rather than transient monoaminergic-related processes (Sasaki et al., Reference Sasaki, Shibata, Tohyama, Takahashi, Otsuka, Tsuchiya, Takahashi, Ehara, Terayama and Sakai2006). Meta-analyses suggest that NM-MRI signal in the SN is modestly increased in schizophrenia relative to controls (Ueno et al., Reference Ueno, Iwata, Nakajima, Caravaggio, Rubio, Horga, Cassidy, Torres-Carmona, de Luca, Tsugawa, Honda, Moriguchi, Noda, Gerretsen and Graff-Guerrero2021; Wieland et al., Reference Wieland, Fromm, Hetzer, Schlagenhauf and Kaminski2021). These findings suggest that NM-MRI may help characterize biological heterogeneity within schizophrenia, particularly in relation to treatment responsiveness.
To date, only two NM-MRI studies have examined treatment responsiveness in schizophrenia, and their findings were inconsistent (Tarumi et al., Reference Tarumi, Honda, Cassidy, Etani, Homma, Sekihara, Kaneko, Moriyama, Tobari, Takahashi, Ueno, Horga, Tsugawa, Rubio, Matsui, Fujii, Graff-Guerrero, Noda, Uchida and Nakajima2026; van der Pluijm et al., Reference van der Pluijm, Wengler, Reijers, Cassidy, Tjong Tjin Joe, de Peuter, Horga, Booij, de Haan and van de Giessen2024). One study, conducted in early-phase schizophrenia, reported lower SN NM-MRI signal in antipsychotic nonresponders than in responders, with nonresponders resembling healthy controls (HCs) (van der Pluijm et al., Reference van der Pluijm, Wengler, Reijers, Cassidy, Tjong Tjin Joe, de Peuter, Horga, Booij, de Haan and van de Giessen2024). Another study reported higher SN NM-MRI signal in TRS than in HCs, whereas responders did not differ clearly from either group (Tarumi et al., Reference Tarumi, Honda, Cassidy, Etani, Homma, Sekihara, Kaneko, Moriyama, Tobari, Takahashi, Ueno, Horga, Tsugawa, Rubio, Matsui, Fujii, Graff-Guerrero, Noda, Uchida and Nakajima2026). These divergent findings suggest that illness stage, treatment exposure, and the clinical definition of treatment resistance may influence observed NM-MRI patterns.
Against this background, we examined NM-MRI signal in the SN-VTA across four groups: patients with URS, patients with clozapine-responsive TRS (non-URS), patients with first-line antipsychotic response (FLR), and HCs. Based on prior PET findings, we expected FLR to show higher SN-VTA NM-MRI signal than HCs. Given the limited and inconsistent NM-MRI literature, analyses involving URS and non-URS were considered exploratory and aimed to examine whether distinct monoaminergic profiles might characterize different forms of treatment resistance. Clarifying these subgroup differences may help refine biologically informed models of TRH in schizophrenia. In secondary analyses, we examined NM-MRI signal in the LC and its associations with clinical and cognitive measures because noradrenergic dysfunction may also contribute to schizophrenia heterogeneity (Klimek et al., Reference Klimek, Rajkowska, Luker, Dilley, Meltzer, Overholser, Stockmeier and Ordway1999; Pelegrino et al., Reference Pelegrino, Guimaraes, Sena, Emele, Scoriels and Panizzutti2023; Sternberg et al., Reference Sternberg, Charney, Heninger, Leckman, Hafstad and Landis1982).
Methods
Participants
This single-center cross-sectional trial was conducted at the Centre for Addiction and Mental Health (CAMH) between 2014 and 2022. The study was approved by the Research Ethics Board at CAMH. Each participant provided written informed consent to participate in this study. All participants aside from HCs were patients receiving clinical care at CAMH. These participants met the criteria for schizophrenia or schizoaffective disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV)/Structured Clinical Interview for DSM-IV. Antipsychotic treatment response or resistance was defined by the modified Treatment Response and Resistance in the Psychosis Working Group Consensus criteria (Howes et al., Reference Howes, McCutcheon, Agid, de Bartolomeis, van Beveren, Birnbaum, Bloomfield, Bressan, Buchanan, Carpenter, Castle, Citrome, Daskalakis, Davidson, Drake, Dursun, Ebdrup, Elkis, Falkai and Correll2017). We defined standard antipsychotic treatment, and antipsychotic treatment response and failure based on the previous study (Supplementary Material 1) (Iwata et al., Reference Iwata, Nakajima, Plitman, Caravaggio, Kim, Shah, Mar, Chavez, de Luca, Mimura, Remington, Gerretsen and Graff-Guerrero2018). Detailed inclusion/exclusion criteria for URS (clozapine-resistant TRS), non-URS (clozapine-responsive TRS), FLR (first-line antipsychotic-responsive schizophrenia), and HCs were described in Supplementary Material 2. The patient and HC groups were matched for age and sex. The sample size was calculated based on an effect size reported from a previous study that compared NM levels in patients with schizophrenia who presented severe positive symptoms (Positive and Negative Symptom Scale [PANSS] [Kay, Fiszbein, & Opler, Reference Kay, Fiszbein and Opler1987] Positive subscale scores ≥19) to HCs (Cohen’s d = 0.92) (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019), which resulted in the estimated number of 24 participants in each group (α = 0.05, 1 − β = 0.85). We aimed to enroll 24 participants per group; however, due to the COVID-19 pandemic, the study was terminated, and we did not reach the target enrollment for the patient groups.
Magnetic resonance imaging
Participants were scanned at CAMH in a 3T GE Discovery MR750 scanner equipped with an eight-channel head coil. Participants underwent a three-dimensional inversion recovery–prepared T1-weighted MRI scan (BRAVO; GE Healthcare, Wauwatosa, WI) (echo time = 3 ms, repetition time = 6.74 ms, inversion time = 650 ms, flip angle = 8°, field of view = 230 mm, 256 × 256 matrix, and slice thickness = 0.9 mm) in support of the preprocessing of NM-MRI images.
NM-MRI acquisition and data processing
NM-MRI imaging was acquired using a two-dimensional gradient echo sequence with magnetization transfer contrast (TR = 260 ms; TE = 2.68 ms; flip angle = 40°; in-plane resolution = 0.39 × 0.39 mm2; partial brain coverage with FOV = 162 × 200; matrix 416 × 512; number of slices = 10; slice thickness = 3 mm; slice gap = 0 mm; magnetization transfer frequency offset = 1200 Hz; number of excitations = 8; and acquisition time = 8.04 min). The protocol for slice prescription was to align the image stack along the line between the anterior and posterior commissures and to place the top slice 3 mm above the floor of the third ventricle. This protocol covered the SN-containing portions of the midbrain and surrounding structures. The quality of the NM-MRI images was inspected visually.
The preprocessing steps were performed in accordance with the methodologies outlined in previous studies (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019; Wengler, He, Abi-Dargham, & Horga, Reference Wengler, He, Abi-Dargham and Horga2020). NM-MRI scans were aligned to each participant’s T1-weighted images and then normalized to the MNI space using Advanced Normalization Tools (ANTs). Visual inspection was conducted after each preprocessing stage. Quality control procedures included visual inspection of raw NM-MRI images, image alignment to the T1-weighted image, spatial normalization to MNI space, and anatomical coverage of the SN-VTA and LC regions. Scans were excluded when image quality was insufficient for reliable contrast ratio (CR) estimation, including incomplete anatomical coverage or suboptimal registration or normalization. Subsequently, the SN-VTA and LC underwent different processing methods. For the SN-VTA, intensity normalization and spatial smoothing were performed sequentially using custom MATLAB scripts. We calculated CR as the relative difference in NM-MRI signal intensity to the white matter tracts of crus cerebri, which is known to have minimal NM content, in the following way:
where, I is NM-MRI signal intensity, V is a voxel, and RR is a reference region (i.e. crus cerebri). The reference region mask for the SN was derived from a previous study (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019). For each participant, the mode (IRR) was calculated by applying a kernel-smoothing function applied to the histogram of the voxel-intensity distribution within the mask. The masks overlaid on the CR mean image for all participants are shown in Figure 1. The resulting NM-MRI CR maps were then further smoothed spatially using a Gaussian kernel with a full width at half maximum of 1 mm. For ROI analyses, NM-MRI signal was extracted from the fully processed image using an SN-VTA mask adapted from prior work (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019).
Voxels masks. SN-VTA and LC masks were derived from prior research: (a) SN-VTA and (b) LC in red; reference regions ((a) crus cerebri, (b) central pons) in blue. Note: LC, locus coeruleus; SN, substantia nigra; VTA, ventral tegmental area.
Figure 1. Long description
A multi-panel anatomical diagram showing brain regions in grayscale with colored overlays.
Panel a, labeled S N dash V T A, contains three views.
* Top left: Coronal y = minus 23 view showing two diagonal red regions representing the S N dash V T A, flanked by blue reference regions representing the crus cerebri.
* Top right: Sagittal x = minus 8 view showing a red cluster positioned above a blue reference region.
* Bottom: A large Axial z = minus 13 view showing two large red central regions flanked by blue regions on the left and right sides.
Panel b, labeled L C, contains two views.
* Top: Coronal y = minus 23 view showing two small vertical red strips at the center.
* Bottom: Axial z = minus 23 view showing a large central blue reference region representing the central pons, with two small red dots located posteriorly representing the L C.
For the LC, that signal was measured using a semi-automated algorithm called the ‘funnel tip method’ described in the previous article (Cassidy et al., Reference Cassidy, Therriault, Pascoal, Cheung, Savard, Tuominen, Chamoun, McCall, Celebi, Lussier, Massarweh, Soucy, Weinshenker, Tardif, Ismail, Gauthier and Rosa-Neto2022; Sibahi et al., Reference Sibahi, Gandhi, Al-Haddad, Therriault, Pascoal, Chamoun, Rosa-Neto and Cassidy2023). This method conducts an intensity-threshold-free cluster search within an overinclusive LC mask in native space. The mask was derived from prior research (McCall et al., Reference McCall, Forouhandehpour, Celebi, Richard-Malenfant, Hamati, Guimond, Tuominen, Weinshenker, Jaworska, McQuaid, Shlik, Robillard, Kaminsky and Cassidy2024). A cluster-forming algorithm was employed to segment the LC within this space, defining it as the eight adjacent voxels per side and axial slice with the highest mean signal in each of the right and left LCs. The automated segmentation was visually inspected and, in some cases, required manual correction to address suboptimal performance. The slices that required manual correction accounted for 2.03% of all slices across all participants.
CR for each voxel v in a given axial slice was calculated as the relative difference in NM-MRI signal intensity from a reference region in the same slice, similar to the process for the SN-VTA. The reference region mask was derived from previous research (Cassidy et al., Reference Cassidy, Therriault, Pascoal, Cheung, Savard, Tuominen, Chamoun, McCall, Celebi, Lussier, Massarweh, Soucy, Weinshenker, Tardif, Ismail, Gauthier and Rosa-Neto2022). LC signal was calculated by averaging the CR values from the three brightest voxels on each side and slice within the LC mask.
Clinical assessment
Clinical assessments included the following: the PANSS, Clinical Global Impression Severity (CGI-S) (Guy, Reference Guy1976), and Global Assessment of Functioning (GAF) (Jones, Thornicroft, Coffey, & Dunn, Reference Jones, Thornicroft, Coffey and Dunn1995). Cognitive functioning was assessed using the following scales: the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) (Randolph, Tierney, Mohr, & Chase, Reference Randolph, Tierney, Mohr and Chase1998), Executive Interview (EXIT) (Royall, Mahurin, & Gray, Reference Royall, Mahurin and Gray1992), and Mini-Mental State Examination (MMSE) (Folstein, Folstein, & McHugh, Reference Folstein, Folstein and McHugh1975) for cognitive function.
Statistical analysis
Statistical analyses were performed using R 4.1.1 (Ihaka & Gentleman, Reference Ihaka and Gentleman1996) and JMP® 12.2.0 (SAS Institute Inc., Cary, NC). The Shapiro–Wilk test was performed to confirm the normality of data distribution. Clinico-demographic characteristics were compared between the groups using analyses of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. Levene’s test was conducted to confirm if variances of continuous variables were equally distributed among the groups. ROI analyses of SN–VTA CR were considered the primary analyses. LC analyses, voxel-wise analyses, and correlation analyses were considered secondary or exploratory.
Group differences in NM levels, as assessed with average CR values within the SN-VTA or LC, were examined. Analyses of covariance (ANCOVA) were performed controlling for age and sex because they were reported to be associated with NM accumulation (Wengler et al., Reference Wengler, Baker, Velikovskaya, Fogelson, Girgis, Reyes-Madrigal, Lee, de la Fuente-Sandoval, Ojeil and Horga2024; Zecca et al., Reference Zecca, Fariello, Riederer, Sulzer, Gatti and Tampellini2002). ANCOVA were also carried out controlling for smoking status and education in addition to age and sex. To address potential confounding by antipsychotic exposure and illness chronicity, additional patient-only sensitivity analyses were conducted. These models examined group differences among URS, non-URS, and FLR groups after further adjustment for chlorpromazine (CPZ)-equivalent daily dose and duration of illness, in addition to age, sex, smoking status, and education. Comparisons in CR used a significance threshold of p < 0.05, and others were corrected for multiple comparisons using the Benjamini–Hochberg method, which employed the conservative false discovery rate (FDR) of 5%. FDR was also applied for planned post-hoc tests such as ANCOVA.
In addition, the voxel-wise analysis of CR values within the SN-VTA mask was conducted. This analysis was repeated for two sets of comparisons: (1) URS versus HCs and (2) FLR versus HCs. After excluding participants not relevant to the test (e.g. non-URS and URS for the FLR vs HC comparison), robust linear regression was applied to predict CR at each voxel within the SN-VTA mask with a binary group variable as fixed factor and age as a nuisance covariate. Following previous work (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019), we censored data points with incomplete SN-VTA coverage or extreme values. Multiple comparisons were corrected by defining the spatial extent of the effect (number of voxels showing group difference at p < 0.05) using a permutation test with 10,000 iterations. To obtain an effect size less biased by voxel selection, we performed a leave-one-subject-out analysis. For each participant, suprathreshold voxels were identified after excluding that participant, and mean CR was extracted from those voxels for the left-out participant. Cohen’s d was then calculated from these extracted values.
For exploratory analyses, Spearman correlations were used to examine the association between CR and participants’ characteristics or symptom severity scales (i.e. PANSS total and subscale scores, RBANS total scale and subscale scores, and EXIT scores) in each group and the whole patient group. These analyses were conducted for CR in both the SN-VTA and LC.
Results
Participants
This study included 78 participants, consisting of 16 patients with URS, 16 with non-URS, 20 with FLR, and 26 HCs. Table 1 contains participants’ clinical and demographic characteristics. The URS group had a higher ratio of smokers compared with the non-URS and HCs groups, and the FLR group also had a higher ratio of smokers compared with HCs. Compared with each patient group, HCs had more years of education. Compared with the URS group, the non-URS and FLR groups had more years of education. The duration of illness was longer in URS than in non-URS. While CPZ dose was higher in URS than in FLR, there were no differences in either CPZ or clozapine doses between the URS and non-URS groups. The URS group showed greater symptom severity compared with the non-URS and FLR groups in general. Cognitive assessment scores were generally better in the HCs than in the patient groups. Two participants (1 URS and 1 FLR) were excluded from subsequent analyses due to insufficient quality. The final analytic sample therefore included 76 participants: 15 patients with URS, 16 with non-URS, 19 with FLR, and 26 HCs.
Characteristics of participants
Table 1. Long description
The table is divided into two sections.
Section (a) Clinico-demographic characteristics:
- Age, years: U R S 49.3, non-U R S 39.5, F L R 46.9, H C s 43.6. F(3,74) = 1.39, p = 0.25.
- Sex, female: U R S 2 (12.5%), non-U R S 5 (31.3%), F L R 4 (20.0%), H C s 7 (26.9%). Chi-square(3) = 2.03, p = 0.57.
- Smoker: U R S 8 (50.0%), non-U R S 2 (12.5%), F L R 6 (30.0%), H C s 1 (3.9%). Chi-square(3) = 14.6, p = 0.002.
- Education, years: U R S 11.7, non-U R S 14.0, F L R 14.3, H C s 16.7. F(3,74) = 9.65, p < 0.0001.
- Schizophrenia: U R S 12 (75.0%), non-U R S 11 (68.8%), F L R 20 (100.0%). Chi-square(2) = 10.04, p = 0.01.
- Chlorpromazine dose, mg/day: U R S 698.4, non-U R S 603.6, F L R 408.4. F(2,49) = 4.92, p = 0.01.
Section (b) Clinical severity:
- P A N S S total score: U R S 81.9, non-U R S 51.5, F L R 51.9. F(2,49) = 53.62, p < 0.0001.
- C G I-S: U R S 4.8, non-U R S 2.6, F L R 2.4. F(2,49) = 50.27, p < 0.0001.
- G A F: U R S 46.9, non-U R S 67.5, F L R 67.0. F(2,49) = 13.07, p < 0.0001.
- M M S E: U R S 27.9, non-U R S 29.2, F L R 28.6, H C s 29.4. F(3,72) = 3.42, p = 0.02.
- R B A N S Total Score: U R S 73.3, non-U R S 80.9, F L R 79.1, H C s 91.5. F(3,72) = 8.12, p < 0.0001.
- R B A N S Attention index: U R S 84.3, non-U R S 87.4, F L R 85.7, H C s 106.0. F(3,72) = 8.54, p < 0.0001.
Note: Values are mean (SD) or n (%). Bold values mean statistically significant results (p < 0.05). The Benjamini–Hochberg method was used for post-hoc tests (FDR < 0.05). Antipsychotics in the FLR group: Paliperidone, n = 6; Olanzapine, n = 4; Risperidone, n = 4; Quetiapine, n = 2; Flupentixol, n = 1; Lurasidone, n = 1; Perphenazine, n = 1; Trifluoperazine, n = 1.
† Data was missing for two participants (1 URS and 1 HC).
‡ Data was missing for three participants (2 URS and 1 HC).
a Smoker: non-URS < URS, p = 0.04; HCs < URS, p = 0.002; HCs < FLR, p = 0.04.
b Education: URS < HCs, p < 0.0001; non-URS < HCs, p = 0.02, URS < FLR, p = 0.02; FLR < HCs, p = 0.02; URS < non-URS, p = 0.04.
c URS < FLR, p = 0.003; non-URS < FLR, p = 0.02.
d Chlorpromazine Dose: FLR < URS, p = 0.01.
e PANSS Total score, Positive subscale, General subscale, and CGI-S: non-URS < URS, p < 0.0001; FLR < URS, p < 0.0001.
f Negative subscale: non-URS < URS, p < 0.0001; FLR < URS, p = 0.001.
g GAF: URS < FLR, p < 0.0001; URS < non-URS, p < 0.0001.
h MMSE: URS < HCs, p = 0.02.
i RBANS Total score, URS < HCs, p < 0.0001; FLR < HCs, p = 0.001; non-URS < HCs, p = 0.01.
j RBANS Immediate memory index, URS < HCs, p < 0.0001; non-URS < HCs, p = 0.002; FLR < HCs, p = 0.003; URS < FLR, p = 0.04.
k RBANS Attention index, URS < HCs, p = 0.0003; non-URS < HCs, p = 0.001; FLR < HCs, p < 0.0001.
l URS < HCs, p = 0.005; non-URS < HCs, p = 0.04.
Abbreviation: ANOVA, analysis of variance; CGI-S, clinical global impression severity scale; EXIT, executive interview; FDR, false discovery rate; FLR, first-line responders; GAF, global assessment of functioning; HCs, healthy controls; MMSE, mini mental state examination; PANSS, positive and negative syndromes scale; RBANS, repeatable battery for the assessment of neuropsychological status; SD, standard deviation; SUD, substance use disorder; URS, ultra-resistant schizophrenia.
ROI analyses in the SN-VTA
Group comparisons in CR in the SN-VTA are shown in Table 2 and Figure 2a. Mean SN-VTA CR was 12.89 (1.10) in URS, 12.44 (1.26) in non-URS, 12.84 (1.40) in FLR, and 11.71 (1.51) in HCs. ANCOVA found group differences in average CR levels of the whole SN-VTA after controlling for age and sex, F(3,70) = 4.45, p = 0.01. Post-hoc analyses revealed higher CR levels in the URS (p = 0.01, d = 1.01) and FLR groups (p = 0.01, d = 0.94) compared with HCs, while no significant differences were found in the other comparisons. There were also group differences in CR in the SN-VTA in comparison among the TRS (URS + non-URS), FLR, and HC groups, and between the entire patient (URS + non-URS + FLR) and HC groups. Post-hoc analyses showed higher SN-VTA CR in the TRS, 12.66 (1.19) versus 11.71 (1.51), p = 0.02, d = 0.72, and FLR group, 12.84 (1.40) versus 11.71 (1.51), p = 0.02, d = 0.91, as well as the entire patient group, 12.73 (1.26) versus 11.71 (1.51), p = 0.002, d = 0.80, respectively, in comparison with HCs. Results survived after controlling for age, sex, smoking status, and education history (Supplementary Table 1). In patient-only sensitivity analyses, SN-VTA CR did not differ across URS, non-URS, and FLR groups after adjustment for age and sex or after additional adjustment for smoking status and education, or after further adjustment for CPZ-equivalent dose and duration of illness (Supplementary Table 4). Effect sizes and 95% confidence intervals for the main pairwise comparisons are reported in Table 2 and Supplementary Table 5.
Comparisons of CR values between the groups
Table 2. Long description
The table is divided into two sections.
Section a, C R, lists mean values and standard deviations for two brain regions across six columns: Entire patients n = 50, T R S n = 31, U R S n = 15, non-U R S n = 16, F L R n = 19, and H Cs n = 26.
* For S N V T A: Entire patients 12.73 1.26, T R S 12.66 1.19, U R S 12.89 1.10, non-U R S 12.44 1.26, F L R 12.84 1.40, and H Cs 11.71 1.51.
* For L C: Entire patients 26.40 3.70, T R S 27.36 3.37, U R S 28.19 3.47, non-U R S 26.59 3.19, F L R 24.84 3.76, and H Cs 25.97 3.37.
Section b, A N C O V A with age and sex covariates, provides F values and p values for three comparison sets.
* Patients versus H Cs: S N V T A shows F 1,72 = 10.70, p = 0.002. L C shows F 1,72 = 0.39, p = 0.53.
* T R S, F L R, and H Cs: S N V T A shows F 2,71 = 5.52, p = 0.01. L C shows F 2,71 = 3.11, p = 0.05.
* U R S, non-U R S, F L R, and H Cs: S N V T A shows F 3,70 = 4.45, p = 0.01. L C shows F 3,70 = 2.87, p = 0.04.
Significant p values are bolded. Footnotes indicate H Cs have lower C R than patients in S N V T A, and F L R has lower C R than U R S in the L C region.
Note: Values are mean (SD). Bold values mean statistically significant results (p < 0.05). (Post-hoc test) FDR < 0.05.
a HCs < Patients (Standardized adjusted mean difference, d = 0.80, 95% CI = 0.31–1.28).
b HCs < FLR, p = 0.02 (d = 0.91, 95% CI = 0.30–1.52), HCs < TRS, p = 0.02 (d = 0.72, 95% CI = 0.19–1.25).
c HCs < URS, p = 0.01 (d = 1.01, 95% CI = 0.35–1.67), HCs < FLR, p = 0.01 (d = 0.94, 95% CI = 0.33–1.55).
d FLR < URS, p = 0.04 (d = 0.99, 95% CI = 0.30–1.68).
Abbreviation: ANCOVA, analysis of covariance; CI, confidential interval; CR, contrast ratio; FDR, false discovery rate; FLR, first-line antipsychotic response; HCs, healthy controls; LC, locus coeruleus; SD, standard deviation; SN, substantia nigra; TRS, treatment-resistant schizophrenia; URS, ultra-resistant schizophrenia; VTA, ventral tegmental area.
Comparisons of CR in the SN-VTA and in the LC between the groups. CR in the SN-VTA (a) and LC (b) were compared among patients with URS, non-URS, FLR, and HCs. Group differences were observed after controlling sex and age. The plotted CR values represent mean CR extracted from voxels that showed significant group differences in the voxel-wise analysis. Note: CR, contrast ratio; FLR, first-line antipsychotic response; HCs, healthy controls; LC, locus coeruleus; SN, substantia nigra; URS, ultra-resistant schizophrenia; VTA, ventral tegmental area.
Figure 2. Long description
A two-panel figure showing box plots with overlaid individual data points.
Panel a is titled C R in the S N dash V T A. The y-axis is labeled Adjusted C R in the S N dash V T A with a scale from 10 to 14. The x-axis lists four groups: U R S (red dots), non dash U R S (green dots), F L R (teal dots), and H C (purple dots). The U R S and F L R groups show higher median C R values (approximately 13.8 and 13.6 respectively) compared to non dash U R S (approximately 12.8) and H C (approximately 12.4). Horizontal bars at the top indicate significant differences between U R S and H C (asterisk p equals 0.01) and between F L R and H C (asterisk p equals 0.01).
Panel b is titled C R in the L C. The y-axis is labeled Adjusted C R in the L C with a scale from 20 to 35. The x-axis lists the same four groups: U R S, non dash U R S, F L R, and H C. The U R S group shows the highest median C R (approximately 29), while the other three groups have lower medians ranging between 26 and 28. A horizontal bar at the top indicates a significant difference between U R S and H C (asterisk p equals 0.04).
Voxel-wise analysis in the SN-VTA
We conducted voxel-wise comparisons to identify the specific areas driving these effects. We observed increased signals in the dorsal area of the SN-VTA in both URS and FLR groups compared with HCs (URS, 549 out of 1807 voxels at p < 0.05, corrected p = 0.0003), permutation test, (peak t = 5.09, peak MNI coordinates [x, y, z] = −6, −23, −14, unbiased leave-one-out Cohen’s d = 1.15); FLR, (556 out of 1807 voxels at p < 0.05, corrected p = 0.0006), permutation test, (peak t = 5.20, peak MNI coordinates [x, y, z] = 6, −21, −17), and unbiased leave-one-out Cohen’s (d = 1.30). The areas of increased signal are shown in Figure 3.
A map of significant voxels in the SN-VTA in the URS and FLR. Voxels are highlighted where significant group differences were observed in SN-VTA CR. Blue indicates regions where URS exhibited higher SN-VTA CR than HCs, red represents regions where FLR showed higher SN-VTA CR than HCs, and pink denotes overlapping areas where both URS and FLR had higher SN-VTA CR compared to HCs. Note: CR, contrast ratio; FLR, first-line antipsychotic response; HCs, healthy controls; SN, substantia nigra; URS, ultra-resistant schizophrenia; VTA, ventral tegmental area.
Figure 3. Long description
A collection of nine brain scan panels showing significant group differences in S N V T A contrast ratio.
At the top right, a legend identifies three color-coded categories. Blue indicates H Cs less than U R S. Red indicates H Cs less than F L R. Pink indicates an overlap where both U R S and F L R are greater than H Cs.
The top two rows contain five small panels showing coronal and sagittal views of the midbrain. Significant voxels are clustered in the bilateral S N V T A regions, showing a mixture of blue, red, and pink pixels.
The center of the image features a large axial slice providing a detailed view of the midbrain. The colored voxels form two distinct clusters on either side of the midline. Pink overlap areas are concentrated in the center of these clusters, surrounded by red voxels on the lateral edges and blue voxels scattered primarily on the superior and medial borders.
The bottom row consists of three axial slices labeled with z-coordinates from left to right.
- At z equals minus 13, the clusters are broad with a high density of pink and red.
- At z equals minus 17, the clusters appear more crescent-shaped and move slightly more medial.
- At z equals minus 19, the clusters are smaller and more concentrated, showing a high proportion of pink overlap.
ROI analyses in the LC
Group differences in CR were observed in the LC (Table 2 and Figure 2b). Mean LC CR was 28.19 (3.47) in URS, 26.59 (3.19) in non-URS, 24.84 (3.76) in FLR, and 25.97 (3.37) in HCs. ANCOVA showed a group difference in LC CR after controlling for age and sex, F(3,70) = 2.87, p = 0.04. Post-hoc tests demonstrated that the URS group had higher CR in the LC than the FLR group (p = 0.04, d = 0.99). Other group comparisons showed no significant differences. The results were maintained after controlling for age, sex, smoking status, and education history (Supplementary Table 1). In patient-only sensitivity analyses, LC CR differed across patient groups after adjustment for age and sex, with post-hoc analyses showing higher LC CR in URS than in FLR (Supplementary Table 4). This group effect was no longer significant after additional adjustment for smoking status and education, or after further adjustment for CPZ-equivalent dose and duration of illness.
Relationships between CR values and participant characteristics
There were no relationships between CR in the SN-VTA or LC and clinico-demographic characteristics (Supplementary Table 2).
Relationships between CR values and clinical variables
Exploratory correlation analyses examining relationships between NM-MRI CR values and clinical or cognitive measures are shown in Supplementary Table 3. These exploratory analyses did not identify significant associations between NM-MRI signals and clinical or cognitive measures in either individual subgroups or the overall patient sample.
Discussion
This cross-sectional study suggested dissociable monoaminergic signatures associated with antipsychotic treatment response in schizophrenia. Elevated NM-MRI signal in the dorsal SN–VTA was observed in both first-line responders and individuals with clozapine nonresponse, consistent with altered long-term dopaminergic-related NM-MRI signal as a common feature across these groups. In contrast, increased LC signal was specific to clozapine nonresponse, implicating additional noradrenergic system involvement in extreme treatment resistance. Together, these results supported a model in which dopaminergic alterations may be necessary but not sufficient to account for treatment resistance, and where interactions between dopaminergic and noradrenergic systems might contribute to variability in treatment response.
SN-VTA findings and antipsychotic responsiveness
Our SN-VTA findings were partly consistent with our a priori expectation and exploratory aims. Consistent with prior PET studies, FLR showed higher SN-VTA NM-MRI signal than HCs (Demjaha et al., Reference Demjaha, Egerton, Murray, Kapur, Howes, Stone and McGuire2014, Reference Demjaha, Murray, McGuire, Kapur and Howes2012; Jauhar et al., Reference Jauhar, Veronese, Nour, Rogdaki, Hathway, Turkheimer, Stone, Egerton, McGuire, Kapur and Howes2019). Analyses involving URS and non-URS were exploratory because prior NM-MRI findings on treatment responsiveness have been limited and inconsistent. In this context, the lack of a significant non-URS versus HC difference was broadly consistent with prior PET evidence suggesting that TRS does not consistently show the same increase in striatal dopamine synthesis capacity observed in treatment-responsive psychosis. By contrast, the elevation in URS was less expected and requires cautious interpretation.
One interpretation is that NM-MRI and PET capture related but distinct aspects of monoaminergic function. PET measures more dynamic aspects of presynaptic dopamine synthesis over a relatively short time scale, whereas NM-MRI is thought to reflect cumulative monoaminergic activity over years (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019; Sasaki et al., Reference Sasaki, Shibata, Tohyama, Takahashi, Otsuka, Tsuchiya, Takahashi, Ehara, Terayama and Sakai2006). The two methods should therefore not be expected to produce identical subgroup patterns, especially in chronic and extensively treated samples. This distinction may help explain why FLR in our study showed elevated SN-VTA signal. Increased presynaptic dopamine function is one of the most consistent findings in treatment-responsive schizophrenia, particularly in the dorsal and associative striatum (Abi-Dargham et al., Reference Abi-Dargham, Rodenhiser, Printz, Zea-Ponce, Gil, Kegeles, Weiss, Cooper, Mann, van Heertum, Gorman and Laruelle2000; Caravaggio et al., Reference Caravaggio, Borlido, Wilson and Graff-Guerrero2015; Howes et al., Reference Howes, Montgomery, Asselin, Murray, Grasby and McGuire2007; Howes & Kapur, Reference Howes and Kapur2009). Our data fit that literature and suggest that long-term dopaminergic burden may also be increased in this subgroup.
The voxel-wise localization to the dorsal SN-VTA is also relevant (Tarumi et al., Reference Tarumi, Honda, Cassidy, Etani, Homma, Sekihara, Kaneko, Moriyama, Tobari, Takahashi, Ueno, Horga, Tsugawa, Rubio, Matsui, Fujii, Graff-Guerrero, Noda, Uchida and Nakajima2026). This region contributes to both nigrostriatal and mesolimbic dopamine pathways, so the present signal increase may not correspond directly to PET findings focused on the striatum. NM-MRI topography may therefore provide information that is not captured by striatal PET measures alone. At minimum, the present findings support the view that alterations linked to treatment response are not confined to a single dopaminergic projection field.
The URS finding requires more cautious interpretation. In our sample, URS also showed elevated SN-VTA signal relative to HCs, despite our expectation that severe treatment resistance might be associated with lower NM-MRI signal, based on prior work linking ultra-resistance with glutamatergic alterations (Iwata et al., Reference Iwata, Nakajima, Plitman, Caravaggio, Kim, Shah, Mar, Chavez, de Luca, Mimura, Remington, Gerretsen and Graff-Guerrero2018; Jauhar et al., Reference Jauhar, McCutcheon, Borgan, Veronese, Nour, Pepper, Rogdaki, Stone, Egerton, Turkheimer, McGuire and Howes2018). One possibility is that URS is not simply a more severe extension of clozapine-responsive TRS, but a biologically distinct subgroup in which persistent dopaminergic abnormality remains relevant. This would fit the broader view that treatment resistance is heterogeneous and may reflect different combinations of dopaminergic, glutamatergic, and noradrenergic dysfunction (Demjaha et al., Reference Demjaha, Egerton, Murray, Kapur, Howes, Stone and McGuire2014, Reference Demjaha, Murray, McGuire, Kapur and Howes2012; Iwata et al., Reference Iwata, Nakajima, Plitman, Caravaggio, Kim, Shah, Mar, Chavez, de Luca, Mimura, Remington, Gerretsen and Graff-Guerrero2018; Jauhar et al., Reference Jauhar, McCutcheon, Borgan, Veronese, Nour, Pepper, Rogdaki, Stone, Egerton, Turkheimer, McGuire and Howes2018, Reference Jauhar, Veronese, Nour, Rogdaki, Hathway, Turkheimer, Stone, Egerton, McGuire, Kapur and Howes2019).
Our results also help place recent NM-MRI studies in context. The first study of treatment responsiveness using NM-MRI, conducted in relatively early-phase schizophrenia, found lower SN NM signal in antipsychotic nonresponders than in responders, with nonresponders resembling HCs (van der Pluijm et al., Reference van der Pluijm, Wengler, Reijers, Cassidy, Tjong Tjin Joe, de Peuter, Horga, Booij, de Haan and van de Giessen2024). In contrast, the more recent study cited reported higher SN NM signal in TRS than in HCs, whereas responders did not differ clearly from either group (Tarumi et al., Reference Tarumi, Honda, Cassidy, Etani, Homma, Sekihara, Kaneko, Moriyama, Tobari, Takahashi, Ueno, Horga, Tsugawa, Rubio, Matsui, Fujii, Graff-Guerrero, Noda, Uchida and Nakajima2026). Our findings align more closely with the latter pattern for URS, while extending it by separating treatment-resistant patients according to clozapine response. This distinction may be important. If non-URS and URS are pooled together, biologically meaningful variation within TRS may be obscured. The lack of a clear difference between non-URS and HCs in our study is consistent with the possibility that clozapine-responsive TRS differs from URS in ways that are lost when all TRS cases are analyzed as one category. These findings support the view that TRS should not be conceptualized as a biologically uniform category.
Differences in illness stage and treatment exposure may also contribute to discrepancies across studies. The earlier NM-MRI study focused largely on first-episode or minimally exposed patients (van der Pluijm et al., Reference van der Pluijm, Wengler, Reijers, Cassidy, Tjong Tjin Joe, de Peuter, Horga, Booij, de Haan and van de Giessen2024), whereas our sample had chronic illness and substantial treatment histories. Because NM accumulates gradually, NM-MRI may reflect the long-term balance of dopamine metabolism, illness course, and medication exposure rather than acute presynaptic state alone (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019; Sasaki et al., Reference Sasaki, Shibata, Tohyama, Takahashi, Otsuka, Tsuchiya, Takahashi, Ehara, Terayama and Sakai2006). This may be especially relevant when comparing FLR, non-URS, and URS, because these groups differ not only in treatment responsiveness but also in cumulative pharmacological exposure and chronicity. Longitudinal multimodal studies that combine NM-MRI with PET will be needed to clarify how these measures relate across illness stages and treatment histories. At the same time, differences in cumulative antipsychotic exposure and illness chronicity across groups may also have contributed to the observed NM-MRI patterns. The present findings should therefore not be interpreted as reflecting disease-related monoaminergic alterations alone.
LC findings and possible noradrenergic involvement in URS
The LC findings provide preliminary evidence that noradrenergic-related NM-MRI signal may differ across treatment-response subgroups. Compared with FLR, the URS group showed higher LC CR, which may suggest altered long-term noradrenergic-related NM-MRI signal in the subgroup with the poorest treatment response. Although the URS-HC difference was not statistically significant, the overall pattern raises the possibility that URS involves alterations beyond the dopaminergic system. This interpretation is biologically plausible, because noradrenergic dysfunction has long been implicated in schizophrenia, particularly in relation to arousal, salience processing, stress responsivity, and positive symptoms (Klimek et al., Reference Klimek, Rajkowska, Luker, Dilley, Meltzer, Overholser, Stockmeier and Ordway1999; Pelegrino et al., Reference Pelegrino, Guimaraes, Sena, Emele, Scoriels and Panizzutti2023; Sternberg et al., Reference Sternberg, Charney, Heninger, Leckman, Hafstad and Landis1982). The LC plays a central role in noradrenergic regulation of arousal and stress responses, and has been implicated in modulating dopaminergic function through stress-related pathways. The elevated LC NM-MRI signal observed in URS may therefore reflect sustained noradrenergic tone that interacts with dopaminergic systems, potentially contributing to persistent symptoms despite clozapine treatment.
This interpretation also suggests a possible pharmacological link to clozapine response. Clozapine differs from other antipsychotics in several respects, including effects on serotonergic, glutamatergic, and adrenergic systems (Svensson, Reference Svensson2003). Its antagonism at α2-adrenergic receptors may contribute to its efficacy in some forms of TRS (Kalkman & Loetscher, Reference Kalkman and Loetscher2003; Svensson, Reference Svensson2003). At the same time, the role of α2-adrenergic mechanisms in schizophrenia remains complex. A meta-analysis suggested that α2-antagonist augmentation does not consistently improve core psychotic symptoms and may worsen some symptom domains in TRS (Hecht & Landy, Reference Hecht and Landy2012). The present findings therefore should not be interpreted as establishing a specific noradrenergic mechanism of clozapine nonresponse, but rather as supporting the possibility that noradrenergic involvement may contribute to heterogeneity within severe treatment resistance.
Although these findings should be interpreted cautiously, they are consistent with emerging evidence suggesting noradrenergic involvement in treatment-resistant phenotypes and highlight the need to consider multisystem monoaminergic interactions when investigating pharmacological targets. The LC finding was based on a between-patient difference between URS and FLR, not on a robust separation between URS and HCs. Accordingly, the present data do not establish a definitive noradrenergic marker of URS. Rather, they support the more limited conclusion that LC involvement deserves further study when severe treatment resistance persists despite clozapine. In patient-only sensitivity analyses, the LC group effect was significant in the age- and sex-adjusted model, but was attenuated after additional adjustment for smoking status and education and after further adjustment for CPZ-equivalent dose and duration of illness. This pattern suggests that the LC finding should be interpreted cautiously and may partly reflect differences in clinical characteristics, illness chronicity, or cumulative treatment exposure rather than a noradrenergic-related signal specific to clozapine nonresponse alone. Nevertheless, the direction of the URS–FLR difference was unchanged, supporting the more limited conclusion that LC involvement remains a hypothesis-generating finding requiring replication in larger samples. Replication in larger samples will be necessary to determine the robustness and clinical relevance of these LC findings.
Lack of association with symptoms and cognition
We did not find significant associations between NM-MRI signals and clinical or cognitive measures. This absence of correlation should be interpreted cautiously. First, NM-MRI likely reflects cumulative monoaminergic activity over years rather than current symptom state (Sasaki et al., Reference Sasaki, Shibata, Tohyama, Takahashi, Otsuka, Tsuchiya, Takahashi, Ehara, Terayama and Sakai2006). A marker with this temporal profile may be better suited to distinguishing subgroups than to tracking cross-sectional severity. Second, symptom ratings and cognitive performance are influenced by many factors beyond midbrain monoaminergic function, including medication effects, illness chronicity, psychosocial factors, and heterogeneity in underlying circuitry. Third, restricted variability within the patient groups may have reduced power to detect associations.
The lack of significant correlations is nevertheless broadly consistent with part of the existing NM-MRI literature. Some NM-MRI studies have examined associations between SN NM signal and symptom severity, but findings have been inconsistent, with positive associations reported mainly in unmedicated samples (Tarumi et al., Reference Tarumi, Honda, Cassidy, Etani, Homma, Sekihara, Kaneko, Moriyama, Tobari, Takahashi, Ueno, Horga, Tsugawa, Rubio, Matsui, Fujii, Graff-Guerrero, Noda, Uchida and Nakajima2026; Ueno et al., Reference Ueno, Iwata, Nakajima, Caravaggio, Rubio, Horga, Cassidy, Torres-Carmona, de Luca, Tsugawa, Honda, Moriguchi, Noda, Gerretsen and Graff-Guerrero2021; van der Pluijm et al., Reference van der Pluijm, Wengler, Reijers, Cassidy, Tjong Tjin Joe, de Peuter, Horga, Booij, de Haan and van de Giessen2024; Wieland et al., Reference Wieland, Fromm, Hetzer, Schlagenhauf and Kaminski2021). In a chronic, medicated sample such as ours, subgroup differences may therefore be more informative than symptom correlations. The same logic may apply to cognition. Although dopamine is relevant to cognitive function, cognitive performance in schizophrenia is unlikely to be explained by a single midbrain marker. Our negative findings therefore do not argue against monoaminergic contributions to symptoms or cognition; rather, they suggest that SN-VTA or LC NM-MRI signal alone is unlikely to serve as a simple cross-sectional marker for either domain.
Potential implications for stratification
Although preliminary, the present pattern suggests that NM-MRI may help refine biological stratification in schizophrenia. FLR and URS both showed elevated SN-VTA signal relative to HCs, whereas non-URS did not show the same clear elevation. URS also showed higher LC signal than FLR. One possible framework is that FLR and URS may share altered long-term monoaminergic-related NM-MRI signal, but URS may additionally involve altered noradrenergic function. In contrast, non-URS may represent a subgroup in which clozapine-responsive mechanisms are less tightly linked to elevated NM accumulation in either region.
This framework should not be overstated, but it may help generate testable hypotheses about treatment response. A practical implication is that treatment resistance may be biologically heterogeneous even within patients who meet similar clinical definitions. Separating clozapine-responsive and clozapine-unresponsive cases may therefore be important in biomarker studies, because combining them could dilute subgroup-specific signals. Future stratification approach and treatment trials may therefore benefit from distinguishing clozapine-responsive and clozapine-unresponsive forms of TRS rather than treating TRS as a single entity.
Limitations
Several limitations should be considered. First, the cross-sectional design precludes causal inference and cannot determine whether the observed NM-MRI differences preceded treatment exposure or emerged during illness progression. Second, the sample size, particularly in the URS and non-URS groups, was modest, which may have limited power to detect between-patient differences. Post hoc power estimates indicated adequate power for several large observed effects, but limited power to exclude smaller or moderate subgroup differences. Accordingly, nonsignificant comparisons, particularly those involving URS and non-URS groups, should be interpreted cautiously. Relatedly, we did not have a quantitative measure of head motion during NM-MRI acquisition, such as framewise displacement. Although visual quality control was performed and scans with insufficient image quality were excluded, residual motion-related effects cannot be fully ruled out. Such effects would generally be expected to reduce image sharpness and attenuate CR estimates rather than produce higher CR values, but future studies should include quantitative motion metrics where available. Third, because NM accumulates gradually over time, cumulative antipsychotic exposure and chronic illness course may have influenced the observed group differences. The patient-only sensitivity analyses further suggested that the LC finding was attenuated after adjustment for smoking status, education, CPZ-equivalent dose, and duration of illness, indicating that this secondary result should be interpreted as exploratory. Smoking status also differed across groups and may have affected monoaminergic measures, although additional analyses adjusting for smoking did not materially change the main findings. Fourth, patients in the FLR group were treated with different antipsychotic medications, and medication-specific effects could not be separated. In addition, although clozapine treatment in TRS was clinically characterized, adherence in the FLR group was not monitored with equivalent precision. Fifth, treatment response and resistance were defined mainly by positive symptom outcomes and may not capture other clinically relevant dimensions of illness. In addition, the NM-MRI acquisition protocol used partial brain coverage focused on the brainstem monoaminergic nuclei and therefore did not permit the assessment of a broader cortical or subcortical networks potentially relevant to treatment resistance. Sixth, the biological determinants of NM-MRI contrast are not fully understood, and NM-MRI should therefore still be regarded as an indirect marker of long-term monoaminergic-related signal rather than a direct measure of monoaminergic activity (Cassidy et al., Reference Cassidy, Zucca, Girgis, Baker, Weinstein, Sharp, Bellei, Valmadre, Vanegas, Kegeles, Brucato, Kang, Sulzer, Zecca, Abi-Dargham and Horga2019; Sasaki et al., Reference Sasaki, Shibata, Tohyama, Takahashi, Otsuka, Tsuchiya, Takahashi, Ehara, Terayama and Sakai2006). Finally, this study did not include complementary molecular or spectroscopic measures. As a result, we could not determine how the present NM-MRI findings relate directly to dopamine synthesis capacity, glutamatergic alterations, or other neurochemical processes implicated in treatment resistance. Longitudinal multimodal studies will be needed to clarify whether the subgroup differences observed here correspond to distinct neurobiological profiles.
Conclusion
In summary, the present findings support the view that treatment-response subgroups in schizophrenia may differ in long-term monoaminergic-related NM-MRI signal patterns. NM-MRI revealed elevated SN-VTA signal in both URS and FLR relative to HCs, with voxel-wise differences localized mainly to the dorsal SN-VTA and higher LC signal in URS than in FLR. These findings suggest that altered long-term monoaminergic-related NM-MRI signal may characterize both URS and FLR, whereas URS may also involve additional noradrenergic alterations. The recent NM-MRI study further supports the relevance of this interpretation by showing that TRS can be associated with higher SN NM signal than HCs (Tarumi et al., Reference Tarumi, Honda, Cassidy, Etani, Homma, Sekihara, Kaneko, Moriyama, Tobari, Takahashi, Ueno, Horga, Tsugawa, Rubio, Matsui, Fujii, Graff-Guerrero, Noda, Uchida and Nakajima2026). Overall, the present results support the view that schizophrenia treatment response is biologically heterogeneous and that NM-MRI may help characterize that heterogeneity. However, these findings remain preliminary and require replication in larger longitudinal studies that combine NM-MRI with complementary modalities such as PET and magnetic resonance spectroscopy.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726106576.
Acknowledgments
Support for the title page creation and format were provided by a tool developed at the National Cancer Institute. The authors used OpenAI’s ChatGPT to assist with English language editing and refinement of the manuscript text. All scientific content and interpretations are the authors’ own.
Author contributions
Analyzed data: F.U., S.H., S.T.; Designed research: F.U., S.N., Y.I., A.G.-G.; Performed research: F.U., S.N., Y.I.; Supervised the work: S.N., C.C., G.H., A.G-G.; Wrote the article: F.U., S.N. All authors reviewed and approved the final manuscript.
Funding statement
BBRF Young Investigator Grant (ID: 31084 to F.U.), Japan Agency for Medical Research and Development (AMED: JP24wm0625302 to S.N.). Canadian Institutes of Health Research: MFE-171222 (F.U.), MOP-141968 (A.G.-G.), and MOP-142493 (A.G.-G.). Ministry of Economic Development and Innovation, Ontario (A.G.-G.). Ontario Mental Health Foundation Type A grant (A.G.-G.) and NARSAD Independent Investigator (A.G.-G.).
Competing interests
F.U. has received grants from Canadian Institutes of Health Research (CIHR), and Brain & Behavior Research Foundation (BBRF); and consultant fees from WCG Clinical and Uchiyama Underwriting within the past 3 years. F.U. is also supported by Blended Research Chair from the University of Ottawa. S.H. has received the JSPS Research Fellowship for Young Scientists (DC1) and The Keio University Doctorate Student Grant-in-Aid Program from Ushioda Memorial Fund. S.N. has received grants from Japan Society for the Promotion of Science (18H02755, 22H03002), Japan Agency for Medical Research and Development (AMED: JP24wm0625302, JP24wm0625307), Japan Research Foundation for Clinical Pharmacology, Naito Foundation, Takeda Science Foundation, Watanabe Foundation, Osakeno-Kagaku Foundation, and Astellas Foundation within the past 3 years; S.N. has also received research support, manuscript fees, or speaker’s honoraria from Asahi Quality & Innovations, Ltd., Teijin Pharma, Sumitomo Pharma, Meiji Seika Pharma, Otsuka, PDR pharma, and MSD within the past 3 years. Y.I. has received fellowship grants from CIHR, research support from Japan Society for the Promotion of Science, SENSHIN Medical Research Foundation, Japanese Society of Clinical Neuropsychopharmacology, Inokashira Hospital Grants for Psychiatry Research, manuscript fees from Dainippon Sumitomo Pharma, and speaker’s fees from Eli Lilly, Meiji-Seika Pharma, Mochida Pharmaceutical, and Yoshitomi Yakuhin within the past 3 years. G.H. received funding from the NIMH (R01 MH117323 and R01 MH114965), and G.H. and C.M.C have a patent for analysis and use of NM imaging in central nervous system disorders, licensed to Terran Biosciences, but have received no royalties. G.R. has received research support from the Canadian Institutes of Health Research (CIHR), University of Toronto, and HLS Therapeutics Inc. P.G. reports receiving research support from CIHR, OMHF, CAMH, CAMH Foundation, and an Academic Scholars Award from the Department of Psychiatry, University of Toronto. E.T.-C., J.S., T.E., S.H., S.S., S.T., V.D.L., S.M.A., and A.G.-G. do not have any conflict of interest to declare.
Open access
