For decades the search for novel biomarkers has driven neurobiological research on schizophrenia. Although there is no comprehensive mechanistic pathophysiological model to date, excessive dopaminergic signaling in the midbrain has been hypothesized to be central in the etiology of psychotic symptoms (1–3). The primarily used methods in molecular imaging, such as positron emission tomography (PET) or single photon emission computed tomography (SPECT) reliably found dopaminergic aberrations in patients with schizophrenia (4). Similarly, dopaminergic aberrations in patients with a diagnosis of bipolar disorder and psychotic symptoms were found (5). Therefore, dopaminergic signaling might function as a transdiagnostic marker for the symptomatically uniform category of psychosis instead of being limited to the diagnostic group of schizophrenia. However, obtaining these in vivo proxies of dopamine function is challenging in patient populations, since the needed radiopharmaceuticals render these measures highly expensive, invasive and time-consuming. Technological advances, such as improved signal intensity and increased signal-to-noise ratio recently enabled neuromelanin-sensitive magnetic resonance imaging (NM-MRI) as a possible readily available window into molecular alterations in psychiatric disorders in vivo (6, 7).

Neuromelanin (NM) is a byproduct of dopamine and catecholamine synthesis, primarily found in the dopaminergic neurons of the substantia nigra (SN) and the noradrenergic neurons of the locus ceruleus (LC) (8). It appears following several reactions, which begin with the oxidation of cytosolic dopamine, advanced by ferric iron (Fe⁺³). These oxidized dopamine-quinones react with cytosolic proteins and are polymerized, which results in an undegradable iron-melanin-protein complex. Possibly, because they are too large for proteasome degradation, macroautophagy is initiated and the complex is taken up into an autophagic vacuole. Fusion with lysosomes enables this vacuole to merge with other vacuoles, containing lipids and proteins and eventually result in neuromelanin organelles, that accumulate with increasing age of the organism (9, 10) (see Figure 1).

The specific features from the interaction between neuromelanin and iron in the iron-neuromelanin complexes make them highly paramagnetic and give them their unique signature in magnetic resonance imaging. A combination of reduced magnetization transfer and shortened T1 relaxation times results in high signal intensities for NM-containing brain regions (11). Although too small for conventional MRI techniques, novel NM-MR sequences acquired at a higher field-strength are sensitive enough to detect and quantify tissue containing neuromelanin. Initially, NM-MRI was developed as a candidate biomarker for Parkinson's disease, which is marked by the depletion of dopaminergic neurons in the SN and noradrenergic neurons in the LC (12). Typically, anatomical reference regions, which do not contain neuromelanin (such as crus cerebri) are used to calculate so-called contrast-to-noise ratios (CNR). In order to obtain these CNR, the signal intensity in the SN is set in ratio to the added signal intensities of the SN and a region without neuromelanin [see e.g., (12–15)], as the following exemplary formula shows:

After this initial breakthrough as a biomarker for neurodegenerative illness, a multitude of studies employed NM-MRI in patients with Parkinson's disease. Integrating these results in a recent meta-analysis, NM-MRI shows a pooled sensitivity of 89% and a pooled specificity of 83% in differentiating healthy individuals from those with Parkinson's disease (16). However, its usefulness as a biomarker beyond Parkinson's disease has been unclear thus far. As described above, dopaminergic alterations in the nigrostriatal and mesolimbic pathway have been hypothesized as a potential cause for psychosis (17), making the SN and ventral tegmental area (VTA) ideal candidate targets for NM-MRI. Due to the technical difficulties in detecting VTA with early NM-MRI sequences [e.g., (15)] the focus of the present review will be the SN.

In an extensive validation study, by Cassidy et al. (9), NM-MRI signal intensity was associated with regional NM concentration in post-mortem human brain tissue of the SN. In contrast to previous studies, a semiautomated voxelwise approach to identify SN and reference regions was used to avoid high inter-rater variability, present in manual tracing approaches. When investigating the relation to dopaminergic alterations, NM-MRI CNR were positively related to dopamine release in the dorsal striatum and resting blood flow in the SN in patients with schizophrenia. According to these results, NM-MRI in the SN could serve as a proxy of individual differences in presynaptic dopaminergic function.

Several studies have established excellent test-retest reliability with voxelwise intraclass coefficients (ICC) above 0.9 and test-retest intervals of several weeks in healthy individuals (18–20) and a few hours in a small sample of schizophrenic patients (9). This provides further evidence that NM-MRI might be a suitable marker with clinical value.

In order to elucidate the use of NM-MRI as a biomarker for schizophrenia, a quantitative synthesis of studies investigating NM-MR in case-control design is needed. To this end, all available case-control studies using NM-MRI in schizophrenia spectrum disorders are reviewed. We hypothesize that CNR of the NM-MRI signal in the SN will be significantly elevated in patients compared to controls, whereas CNR in the LC will not be significantly different between patients and controls. Furthermore, a meta-regression with year of publication and age as predictors will be calculated to control for further sources of variance.

We could show that neuromelanin-sensitive CNR in the SN were significantly higher in patients from the schizophrenia spectrum disorder, compared to healthy controls. Potential effects of age were not detectable in this meta-analysis. These results contribute to the development of NM-MR imaging of the SN as a promising biomarker for patients with schizophrenia.

However, some open questions remain regarding the relation to transdiagnostic features of the biomarker. The specificity of the signal, and the potential impact of confounds such as medication and age need to be addressed in future studies. To begin with, NM-MRI CNR were only related to positive symptom severity in clinical-high risk individuals as well as clinically manifest patient(s) with schizophrenia in one study (9). The absence of this association in previous studies (22, 23) might be explained by differential effects of antipsychotic medications. As suggested with regard to pharmacological imaging (5), a transdiagnostic perspective on biomarkers of psychosis is more useful than strict adherence to diagnostic categories. This assumption aligns with the positive association between symptom scores and NM-MRI signal intensity, reported in one study on a syndromal and prodromal sample (9). It also demonstrates the importance of assessing subjects, regarding a transdiagnostic psychotic phenotype in future studies.

Results regarding the anatomical specificity of this biomarker must be interpreted with caution, considering that CNR values for the LC were only available in three studies. Due to poor detection of VTA signals, findings are limited to the SN. However, given the more prominent findings concerning hyperdopaminergic alterations in associative regions (i.e., associative and sensorimotor striatum) (17), that mainly receive input from SN (31), this does not limit our findings. In a further study (22), which could not be included in the present meta-analysis due to different reporting of group differences as absolute instead of relative values, VTA values were significantly decreased in patients with schizophrenia. This supports the recently posed hypothesis of mesocortical deficits and nigrostriatal excess in the dopaminergic aberrations observed in schizophrenia (17). With regard to SN, one study used a different approach than the others and used a smaller subset of so-called psychosis overlap voxels instead of the whole SN (9). Furthermore, most studies included employed a manual tracing approach to determine regions of interest (ROI), which is less reliable compared to semiautomated thresholding techniques according to a recent systematic evaluation (19). Manual tracing is not only prone to human bias but also to circularity, when high-signal regions in the SN are determined to be ROI and subsequently used to calculate CNR (20). The voxelwise approach provides another clear advantage, in that it allows to investigate anatomical subregions of the midbrain and striatum, which are hypothesized to correspond to symptom groups of schizophrenia (32). Recently improved 3-D NM-MRI sequences also allow investigation of the VTA but were only investigated in patients with schizophrenia in a single study (22). The use of absolute values instead of CNR to determine signal intensity excluded the study from the present synthesis.

The present meta-analysis is limited by the following factors: The number of studies included is relatively low. However, NM-MR sequences have emerged within the last decades and were only recently suggested as a specific biomarker for schizophrenia (9). This is reflected in the relatively low number of studies in the field. Due to the low number of studies extracted, power was not sufficient for subgroup analyses, for example by clinical status. Another important possible confound is a potential effect of antipsychotic medication. Since there was only data on medication available in three out of six studies, there was too little data to provide a reliable statement. If the moderate effect sizes found in our meta-analysis are confirmed in further studies, usefulness of NM-MR sequences as a biomarker will have to be examined critically. One sample only includes patients with a comorbid substance use disorder (13), which has been shown to impact the dopaminergic system (33, 34) but also reflects clinical reality with a twofold increased use of substances in patients with first-episode psychosis (35) and therefore increases ecological validity of our meta-analysis.

Although test-retest reliability has been shown in 16 participants (9), larger samples investigating test-retest reliability and inter-site reliability of NM-MRI, particularly in psychotic samples, are warranted. Furthermore, as age effects are well-established in NM-MRI (11), potential interaction between diagnosis and age have to be addressed in future studies. Crucially, the translational application requires optimization of neuromelanin-sensitive sequences with short acquisition times (36, 37) and harmonization across sites.

Publicly available datasets were analyzed in this study. This data can be found at: https://github.com/agschlagenhauf/qMRI.

LW, SF, SH, FS, and JK: conceptualization, writing—review, and editing. LW and SF: data curation. LW, SF, and JK: formal analysis. LW: writing—original draft. All authors contributed to the article and approved the submitted version.

This work was supported by the Einstein Center for Neurosciences (LW and SF), the German Research Foundation (Grant No. SCHL 1969/1-2/3-1/5-1) (FS) and the BIH-Charité Clinician Scientist Program funded by the Charité–Universitätsmedizin Berlin and the Berlin Institute of Health (JK).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.