Neuroimaging as a Window Into the Pathophysiological Mechanisms of Schizophrenia

Schizophrenia is a complex neuropsychiatric disorder with a diverse clinical phenotype that has a substantial personal and public health burden. To advance the mechanistic understanding of the illness, neuroimaging can be utilized to capture different aspects of brain pathology in vivo, including brain structural integrity deficits, functional dysconnectivity, and altered neurotransmitter systems. In this review, we consider a number of key scientific questions relevant in the context of neuroimaging studies aimed at unraveling the pathophysiology of schizophrenia and take the opportunity to reflect on our progress toward advancing the mechanistic understanding of the illness. Our data is congruent with the idea that the brain is fundamentally affected in the illness, where widespread structural gray and white matter involvement, functionally abnormal cortical and subcortical information processing, and neurometabolic dysregulation are present in patients. Importantly, certain brain circuits appear preferentially affected and subtle abnormalities are already evident in first episode psychosis patients. We also demonstrated that brain circuitry alterations are clinically relevant by showing that these pathological signatures can be leveraged for predicting subsequent response to antipsychotic treatment. Interestingly, dopamine D2 receptor blockers alleviate neural abnormalities to some extent. Taken together, it is highly unlikely that the pathogenesis of schizophrenia is uniform, it is more plausible that there may be multiple different etiologies that converge to the behavioral phenotype of schizophrenia. Our data underscore that mechanistically oriented neuroimaging studies must take non-specific factors such as antipsychotic drug exposure or illness chronicity into consideration when interpreting disease signatures, as a clear characterization of primary pathophysiological processes is an imperative prerequisite for rational drug development and for alleviating disease burden in our patients.

Schizophrenia is a complex neuropsychiatric disorder with a diverse clinical phenotype that has a substantial personal and public health burden. To advance the mechanistic understanding of the illness, neuroimaging can be utilized to capture different aspects of brain pathology in vivo, including brain structural integrity deficits, functional dysconnectivity, and altered neurotransmitter systems. In this review, we consider a number of key scientific questions relevant in the context of neuroimaging studies aimed at unraveling the pathophysiology of schizophrenia and take the opportunity to reflect on our progress toward advancing the mechanistic understanding of the illness. Our data is congruent with the idea that the brain is fundamentally affected in the illness, where widespread structural gray and white matter involvement, functionally abnormal cortical and subcortical information processing, and neurometabolic dysregulation are present in patients. Importantly, certain brain circuits appear preferentially affected and subtle abnormalities are already evident in first episode psychosis patients. We also demonstrated that brain circuitry alterations are clinically relevant by showing that these pathological signatures can be leveraged for predicting subsequent response to antipsychotic treatment. Interestingly, dopamine D2 receptor blockers alleviate neural abnormalities to some extent. Taken together, it is highly unlikely that the pathogenesis of schizophrenia is uniform, it is more plausible that there may be multiple different etiologies that converge to the behavioral phenotype of schizophrenia. Our data underscore that mechanistically oriented neuroimaging studies must take non-specific factors such as antipsychotic drug exposure or illness chronicity into consideration when interpreting disease signatures, as a clear characterization of primary pathophysiological processes is an imperative prerequisite for rational drug development and for alleviating disease burden in our patients.

INTRODUCTION
Schizophrenia is a complex neuropsychiatric disorder with a diverse clinical phenotype that manifests in variable levels of positive and negative symptoms and cognitive impairment. Even though the course of the illness can be variable, for many patients the disease is chronic and debilitating, resulting in a substantial personal and public health burden (1). Available pharmacological interventions can alleviate positive symptoms, but effective treatments across symptom dimensions are lacking and a cure remains elusive. This can in part be attributed to the gaps in the knowledge of the underlying brain pathology. To close these gaps, tremendous efforts geared toward elucidating key pathophysiological brain signatures and at advancing our mechanistic understanding of the illness are undertaken with the ultimate goal to lower disease burden and improve long-term outcomes for patients.
Neuroimaging offers versatility in terms of capturing different aspects of brain pathology in vivo, including brain structural integrity deficits, functional dysconnectivity, and altered neurotransmitter systems, positioning the field well to contribute to the discovery of clinically relevant biological processes in schizophrenia. The first mechanistic discovery studies utilizing neuroimaging date back to the 1970s and 80s, when ventricular enlargement was described as a potential diagnostic marker (2)(3)(4). Since these early studies, the field has grown exponentially; more than 16,000 neuroimaging manuscripts in schizophrenia spectrum disorder patients have been published (Figure 1). Even though major strides toward delineating relevant pathophysiological processes have been made, progress in the field has been impeded by a number of confounding factors that make interpretation of data difficult. This ultimately hampers the development of a comprehensive explanatory model capturing the complex underlying pathophysiology and integrating findings from diverse imaging modalities. Two principal confounding variables need to be considered when interpreting neuroimaging studies. These are antipsychotic medication exposure and illness chronicity, which impact virtually all imaging measures. This has resulted in a debate to what extent brain imaging alterations reported in schizophrenia studies reflect primary disease pathology (5). To make further progress, it is imperative to disentangle confounding effects from illness specific brain signatures. Investigating the impact of antipsychotic medications in longitudinal brain imaging studies and studying patients in the early illness stages who have had no prior antipsychotic medication exposure affords the opportunity to characterize antipsychotic drug action, mitigate confounds and gain a more in depth understanding of the nature of the illness.
Our group's work has been dedicated to contribute to the efforts in delineating the schizophrenia pathophysiology and in disentangling disease signatures from those of antipsychotic medications and disease chronicity. In the following paragraphs, we discuss a number of key scientific questions and take the opportunity to reflect on our progress toward advancing the mechanistic understanding of the illness. Antipsychotic medications are the cornerstone of treating patients with schizophrenia spectrum disorders. They principally act on the dopamine D2 receptors (6), but also modulate a number of other neurotransmitter systems (7,8), affect receptor expression profiles, have downstream effects on protein synthesis (9), and impact brain plasticity (9). Given the complexity of antipsychotic drug action, it is plausible that molecular imaging geared at investigating the dopamine system and other imaging markers are impacted by these medications. Longitudinal brain imaging studies that contrast neural signatures in the same patients before and after they are treated can shed light on modulatory effects of antipsychotic medications on brain structure, function and neurochemistry [for reviews of the larger field, see (10)(11)(12)(13)(14)(15)]. We have conducted several studies utilizing different imaging techniques to broadly characterize antipsychotic drug effects on neural signatures ( Table 1).
Our structural imaging studies show mixed results in terms of antipsychotic medication effects. We observed antipsychotic medication related changes in cortical thickness (17), but not in white matter microstructural integrity or radial fiber trophy (a surrogate marker for white matter volume) (18,19). The disparity in effects across tissue classes is perhaps not surprising, given the different anatomical, functional and neurochemical makeup of these tissues. For example, brain derived neurotrophic factor (BDNF) is a neurotrophin that regulates synaptic plasticity (37,38) and spine density (39). The 66met allele has been associated with diminished activity-dependent BDNF secretion (40) and decreased cortical volumes (41) but not with white matter integrity in healthy subjects (42). BDNF also is found to be abnormally low in schizophrenia (43) and can be increased by second generation antipsychotics (44). It is therefore possible that antipsychotic medication related modulation of BDNF results in alleviation of dendritic deficits in the outer layers of the cortex (45), which would be reflected in an increase in cortical thickness, but not white matter integrity. Alternatively, it is possible that the acute gray matter tissue response can be captured at the time scale at which we investigated structural changes, but the duration is not sufficient to demonstrate subtle changes in white matter architecture and microstructure that may become more evident with longer duration of antipsychotic treatment.
Our longitudinal task based and resting state functional neuroimaging studies consistently show changes associated with antipsychotic drugs at different time scales, ranging from a single dose drug administration to several weeks of treatment. Our findings are in agreement with drug challenge studies in healthy subjects demonstrating that dopamine antagonism affects brain function (46) and animal studies showing that antipsychotics can be used to restore cortical synchronization and functional connectivity after disruption with hallucinogenic agents (47,48). When interpreting our functional data in context of differences between patients and healthy controls, we show that antipsychotic treatment appears to, at least partially, restore abnormal brain function to levels that are more similar to those of healthy controls. We also observed a recurring theme of brain regions that are functionally modulated by antipsychotic treatment. Relevant areas include the anterior cingulate cortex, medial prefrontal cortex, hippocampus, caudate and putamen. This suggests that cortico-striatal circuitry may be functionally remodeled by antipsychotic medications. Interestingly, anterior cingulate and medial frontal cortex function (35,(49)(50)(51)(52) and cortico-striatal connectivity have also been identified as critical for antipsychotic drug action by other groups (53,54). When examining brain function on a more global level, we found that antipsychotic medications affect brain network topology, i.e., global efficiency and clustering, but not dynamic connectivity states (26,29), which is consistent with the idea that antipsychotic medications partially but not fully restore abnormal brain function.
In our magnetic resonance spectroscopy studies, we unexpectedly did not find changes with antipsychotic medications on glutamate levels. Several lines of evidence describe complex interactions between the dopamine and glutamate systems and identify glutamate receptor complexes as potentially important indirect targets for dopamine D2 receptor blockers (55)(56)(57), suggesting antipsychotic medications may affect glutamate levels. A number of prior studies report a decrease of glutamate in the temporal lobe (58), striatum, and the anterior cingulate cortex (59), though the group failed to replicate the latter in a subsequent report (60). It is possible that this discrepancy is due to differential effects of various antipsychotic medications on the glutamate system. Alternatively, risperidone may affect glutamate levels only in a subset of patients, or the shifts in glutamatergic neurotransmission may be too subtle to be captured with magnetic resonance spectroscopy, which measures tissue metabolites rather than compartment specific levels.
Our data also suggest that heterogeneity in the patient population may need to be taken into account when examining  We demonstrated in several studies that changes in brain structure and function following antipsychotic treatment may be more prominent in patients with a favorable clinical response, compared to those who show little improvement with medications (17,26,27). It is tempting to speculate that the variability in modulatory effects of antipsychotics is reflective of differences in brain pathology. It is possible that patients who respond favorably to medications do so because they still have actively destructive, yet reversible, neurobiological changes that can be ameliorated with antipsychotic treatment. In contrast, poor responders may be beyond the period of active deficit process formation in their brain and the pathological state becomes more permanent, rendering antipsychotic medications ineffective both on a clinical level and in reversing structural and functional brain injury. If that is the case, it should be possible to leverage neuroimaging data to aid in the prediction of response to antipsychotic treatment. Much of the work described here has been performed in patients who had prior antipsychotic medication exposure, and in mixed groups of antipsychotic medication-naïve patients previously medicated patients.
Adequately powered longitudinal studies conducted in exclusively medication-naïve patients are direly needed to reduce the possible effects of illness chronicity and prior antipsychotic medication exposure on changes related to antipsychotic medication treatment.
Can Neuroimaging Data Aid in Predicting Subsequent Clinical Response to Antipsychotic Treatment?
The current strategy for management of psychosis spectrum disorders consists of sequential treatment with different antipsychotic medications based on trial and error. Unfortunately, it is not possible to predict how a patient will respond to treatment based on clinical assessments alone.
To make progress in this context, we have conducted a number of studies geared toward characterizing neural signatures in unmedicated and antipsychotic medication-naïve patients that predict a subsequent favorable response to antipsychotic treatment ( Table 2). Here, the ultimate goal is to realize the potential of neuroimaging in guiding treatment decisions based on the underlying brain pathology [for further review of the topic, see (63,64)].
In structural imaging studies, we found that both gray and white matter integrity at baseline are predictive of subsequent response to antipsychotic treatment. Lower cortical thickness and greater basal ganglia volumes [basal ganglia volumes are increased in patients, possibly as a result of antipsychotic drug exposure (65)] are associated with better treatment response (17,61). This suggests that greater baseline alterations in gray matter are indicative of a favorable clinical outcome. In contrast, greater alterations in whole brain fractional anisotropy, a nonspecific marker of white matter integrity, and in whole brain orientation dispersion, a measure of fiber complexity, were associated with subsequent poor treatment response, suggesting that greater alterations in white matter are indicative of poor clinical outcomes (18,19,62).
Our functional neuroimaging studies, both positron emission tomography (PET) and functional MRI, found consistent associations between more intact brain function at baseline and a favorable subsequent response to antipsychotic medications. Here again we found a recurring theme of brain regions and networks that were relevant for clinical outcomes, specifically anterior cingulate cortex, hippocampus, thalamus, striatum and caudate, as well as the default mode network and dorsal attention network. Interestingly, we also found that the degree of normalization in brain function after 1 week of antipsychotic medication was predictive of clinical outcomes after 6 weeks of treatment (35,36). Our data suggest that both baseline function and early functional changes may be important predictors of treatment response.
Our magnetic resonance spectroscopy studies investigated neurometabolite levels in the anterior cingulate cortex and the hippocampus (30). We found that N-acetyl-aspartate (NAA), a marker of neuronal health, in the anterior cingulate cortex and glutamate in the hippocampus at baseline were associated with response to risperidone after 6 weeks of treatment. At first glance, the association between low cortical NAA and good response to treatment appears counter-intuitive. It is striking, however, that greater alterations in both NAA and glutamate are linked to better treatment response. One could speculate that these features point toward an excitation/inhibition imbalance and a deprived neuronal state that is reversible by successful antipsychotic treatment.
Taken together, our studies suggest two groups of neuroimaging markers that are relevant for subsequent antipsychotic treatment response. The first is a set of markers for which greater abnormalities predict subsequent poor response to antipsychotic treatment. These markers are largely centered around the theme of brain connectivity, where greater abnormal functional connectivity and white matter integrity deficits are indicative of worse response to antipsychotic treatment. The second is a set of markers where greater abnormalities predict a favorable response to antipsychotic treatment. These markers include cortical thickness, subcortical brain volumes and neurometabolites levels. Abnormalities in these imaging features may signify a deprived neuronal state or a disequilibrium in the excitation/inhibition balance that may be reversible with successful antipsychotic treatment. Overall, our findings suggest that the brain is structurally and functionally "wired" in a way that does or does not favor response to antipsychotic medication treatment. However, based on these studies, it remains unclear if the neural signatures relevant for antipsychotic treatment response are also core features of the schizophrenia pathophysiology or not. To disentangle core illness features from confounds of antipsychotic medication exposure as well as illness chronicity, it is imperative to characterize neuroimaging signatures in antipsychotic medication-naïve first-episode psychosis patients.  There is a growing sense that progressive brain changes occur beyond the first psychotic episode and that illness chronicity may fundamentally affect the brain (66). It is difficult to pinpoint the extent to which abnormalities can be attributed to antipsychotic medication effects or to ongoing disease progression. Studying patients at later illness stages makes it problematic to discern downstream effects of pathology and prior treatment attempts from primary pathology. Even though this patient population is difficult to recruit (67), studying antipsychotic medication-naïve first episode psychosis patients can mitigate these issues. Our group has conducted studies in this population using different imaging modalities in an effort to characterize relevant neural signatures ( Table 3).
Using structural neuroimaging techniques, we found that a number of alterations are already present in medicationnaive patients. Alterations include decreased gyrification in the pre-and post-central gyri, decreased total hippocampus and hippocampus subfield volumes, a trend level decrease in cortical thickness, as well as a decrease in whole brain fractional anisotropy and axial diffusivity (17,18,62,69). Additionally, an increase in mean diffusivity in the hippocampal part of the cingulum, which is often interpreted as evidence of neuroinflammation, is already evident in first episode patients (18). Importantly, the duration of untreated psychosis, which is the time between first onset of psychotic symptoms and the first antipsychotic treatment, seems to be an important modulatory factor for structural integrity, as a longer duration of untreated psychosis is associated with greater gray and white matter alterations (62,68,69).
We saw similar results in our functional imaging studies where greater duration of untreated psychosis was associated with greater disruptions in functional connectivity in largescale brain networks supporting higher order cognition (68). In the overall group of antipsychotic medication-naïve patients, we note that hippocampus connectivity is disrupted. Spatial patterns of dysconnectivity resemble those in unmedicated, chronic schizophrenia patients (71). When examining dynamic connectivity, we found that quasi-periodic patterns had a greater impact of fronto-parietal control network connectivity in first episode patients compared to controls, suggesting that brain network dynamics are already altered in this patient group (70). Taken together, connectivity alterations are clearly already present in medication-naïve patients, which is consistent with the hypothesis that brain network dysconnectivity is a core feature of the illness (73)(74)(75).
In magnetic resonance spectroscopy studies, we found elevated choline and a disruption of the relationship between NAA and glutamate in the left striatum, indicating possible mitochondrial, membrane, and glial dysfunction (72). However, we reported no neurometabolite abnormalities in the medial prefrontal cortex or the hippocampus when contrasting antipsychotic medication-naïve first episode patients and controls in the overall group (31,69). It is important to note that glutamate levels in both areas showed a greater variance in patients compared to controls, suggesting heterogeneity in glutamatergic metabolism in first episode psychosis. When the duration of untreated psychosis is taken into consideration, those with a shorter but not those with a longer duration of untreated psychosis did have an elevation in hippocampal glutamate (69).
Collectively, it is evident that functional and structural abnormalities are already present in patients who are in their early illness stages who have no prior exposure to antipsychotic medications. However, wee did not observe group level alterations in NAA or glutamate which was somewhat unexpected. Interestingly, the variance in measurements was greater in patients compared to controls, suggesting that neurometabolite alterations may only be present in a subset of patients, or become more pronounced as the illness progresses. Importantly, we found the duration of untreated psychosis to be an important mediator, such that a longer interval adversely impacts brain structure and function, underscoring the importance of early intervention efforts in this syndrome.

Can Multimodal Brain Imaging Help Us Better Characterize the Complex Pathophysiology in Schizophrenia?
The field is moving from traditional models of schizophrenia focused on the disruption of a single molecule such as dopamine or glutamate, to models of pathway dysregulation as a means of integrating findings from diverse imaging modalities. An individual component of a pathway in this context can be best understood as part of a highly interactive network that may influence other parts of the network and in turn be regulated by a number of other factors (76). An abnormality may represent a primary etiological factor, but could also be present because of a disruption in modulatory inputs and thus reflect a secondary consequence of a pathophysiological process. It is critical to develop disease models with a degree of complexity that accurately reflect these highly interactive patterns. Multimodal imaging allows investigation of brain dysfunction using a range of techniques within the same individual and to test hypotheses about relationships between biological mechanisms. This approach may reveal details about the pathophysiology that would not be detectable when using one modality alone.
For example, glutamatergic hyperactivity is hypothesized to be a key pathological feature in schizophrenia (77, 78). Glutamate neurotransmitter flux, neuronal firing rate, and the blood oxygen level dependent (BOLD) response are tightly coupled (79,80), and glutamate plays a role in long-range functional connections (81). In preclinical studies, excess glutamate has been shown to be associated with disorganized neuronal activity (82) and may result in increased synapse turnover as well as axonal or glial injury (83)(84)(85). If this premise holds true, multimodal neuroimaging studies should detect relationships between glutamate levels and brain function in healthy subjects that are disrupted in patients with schizophrenia. They also should reveal an association between glutamate excess and structural integrity deficits in patients. Our group has conducted a number of such studies to empirically test the relationship between glutamate and brain function the impact of glutamate excess on brain structure.
We used several task paradigms and resting state functional MRI to characterize the relationship between glutamate levels and brain function in healthy subjects and schizophrenia spectrum patients. We reported a positive correlation between hippocampal glutamate and inferior frontal activation during a memory retrieval task in controls that was absent in medicated chronic schizophrenia patients (86). The relationship between hippocampal glutamate and the BOLD response in default mode network regions during memory encoding also differed between unmedicated patients and controls (24). Using a Stroop task, we observed a negative correlation between anterior cingulate cortex glutamate and the BOLD response in the posterior cingulate cortex, precuneus, occipital cortex and cerebellum in controls; this correlation was inverted in medicated firstepisode psychosis patients (87). Similarly, we noted a correlation between anterior cingulate cortex glutamate and the BOLD response in the salience network and posterior default network during a Stroop task in healthy controls but not unmedicated patients with schizophrenia (22). Using a reward task, we observed an association between the prediction error related BOLD response in the midbrain and substantia nigra glutamate in healthy controls but not medicated patients with chronic schizophrenia. Interestingly, glutamate levels were elevated in patients, suggesting that glutamatergic dysfunction might contribute to abnormal neural prediction error coding (88). At rest, we initially did not detect a relationship between hippocampal glutamate and aberrant hippocampal connectivity to the precuneus in a small group of unmedicated schizophrenia patients (89). However, after expanding the sample size, we did observe that higher glutamate levels were correlated with higher hippocampus resting state connectivity to the anterior default mode network in healthy controls, but the relationship between measures was inverse in first episode psychosis patients (71). Taken together, our studies consistently demonstrated a link between glutamate levels and brain function, regardless of task condition and even in absence of a task in healthy subjects. Importantly, this relationship was consistently found to be altered or absent in patients across paradigms tapping into various aspects of brain function, which implicates that the disruption in this coupling may be a fundamental feature of the illness. Interestingly, a detectable alteration in glutamate levels does not seem to be a necessary element to reach the threshold of a disruption in brain function, suggesting the presence of subtle abnormalities in glutamatergic neurotransmission that may be below the level of detection with magnetic resonance spectroscopy. Our longitudinal studies further demonstrate that the relationship between glutamate and brain function in patients is modulated by antipsychotic drug treatment, suggesting a potential mechanism of antipsychotic drug action (22,24).
We also conducted studies to examine the impact of glutamate (excess) on brain structures. In unmedicated chronic schizophrenia patients, we found that higher hippocampal glutamate was associated with lower hippocampal volumes, suggesting that glutamate related excitotoxicity (neurotransmitter excess related increased synapse turnover) might affect brain structure (90). Our finding is in agreement with a later study that reported a negative relationship between glutamate excess and brain volumes in the caudate nucleus in first episode psychosis patients (91). In contrast, we did not observe a linear relationship between hippocampus subfield volumes and glutamate levels in antipsychotic medication-naïve first episode psychosis patients (69). Interestingly though, those with a longer duration of untreated psychosis showed lower hippocampus subfield volumes. While this discrepancy between our studies at first glance does not support the hypothesis that glutamate impacts brain structures as a fundamental mechanism of the illness, it is possible that the amount of time the brain is exposed to altered glutamate is the key factor in adversely affecting brain structures rather than the amount of glutamate present in the brain at any given time.
As demonstrated with the above examples, multimodal neuroimaging clearly can give important insights into the pathophysiology of the illness, and allows testing of hypotheses about the relationships between biological mechanisms which is not possible with a single imaging modality. However, because different aspects of the pathophysiology are typically assessed at the same time in multimodal imaging, interpretations are limited to discovery of associations but not causality. In other words, it is possible to detect a relationship between two different imaging markers, but it is not feasible to discern if one abnormality is caused by the other or vice versa.

Can Pharmacological Challenge Studies Provide a Framework for the Interpretation of Neuroimaging Findings in Schizophrenia?
A complementary line of research, pharmacological challenge studies, which are designed to model aspects of the pathophysiology, can inform inferences drawn about causality for neuroimaging alterations seen in schizophrenia spectrum patients. Our group has done several pharmacological challenge studies using subanesthetic ketamine, a non-competitive Nmethyl-d-aspartate (NMDA) receptor blocker that transiently induces a behavioral phenotype similar to that seen in the illness (92)(93)(94)(95)(96), to test the hypothesis that experimentally induced NMDA receptor hypofunction causes changes in brain function and neurochemistry that are comparable to neural signatures observed in schizophrenia.
Early imaging studies testing the effects of a pharmacological challenge found that sub-anesthetic ketamine affects regional cerebral blood flow (rCBF). In response to the drug, an increase in psychosis symptom severity was paralleled by an increase in rCBF in the anterior cingulate cortex, both in healthy controls and medicated patients with schizophrenia (97). In addition to a ketamine related increase in rCBF in the anterior cingulate cortex, a reduction in rCBF was found in the hippocampus, lingual gyrus and fusiform gyrus in a different study of healthy controls (98). Interestingly, kinetic analyses suggest that the ketamine induced rCBF response differs across brain areas (99). Using a MRI based rather than PET based technique to quantify blood flow, our group later reported increased rCBF in the prefrontal cortex, cingulate cortex caudate, putamen, thalamus and hippocampus, as well as interregional connectivity alterations in areas of the salience network (100). This again suggests systems level, but not uniform, effects of experimentally induced NMDA receptor hypofunction.
In a combined magnetic resonance spectroscopy and resting state functional connectivity study of the hippocampus (101), we reported that experimentally induced NMDA receptor hypofunction resulted in an increase in glutamate that was similar in magnitude to that we saw in unmedicated schizophrenia patients (90). In parallel, we also found a reduction in hippocampal resting state connectivity to the anterior cingulate cortex, medial prefrontal cortex, insula, hippocampus, precuneus, posterior cingulate cortex and lingual gyrus. Spatial patterns of abnormalities here closely resemble those we observed in both unmedicated schizophrenia patients (27,71) and antipsychotic medication-naïve first episode psychosis patients (71).
Taken together, our findings from ketamine challenge studies lend empirical support to the putative link between NMDA receptor hypofunction, disruption in brain function and glutamate excess and provide a theoretical framework for the interpretation of abnormal brain signatures in schizophrenia in the context of NMDA receptor hypofunction. It is important to note that this type of framework can also be leveraged to test target engagement for putative novel pharmacological agents as recently demonstrated a multi-center proof of mechanism study of pomaglumetad (102), which will ideally translate into accelerated development of novel drugs.

How Can Postmortem Work Provide Context for the Neuroimaging Findings in Schizophrenia?
Postmortem studies reporting abnormalities in γ-Aminobutyricacid (GABA)ergic interneurons (103,104) and glutamatergic signaling (105) provided the impetus for the measurement of glutamate and GABA using MR spectroscopy, as well as the measurement of brain oscillations using EEG/MEG. The fast-spiking GABA-interneurons play a fundamental role in controlling of the synchrony of cortical pyramidal neurons by producing rhythmic inhibitory postsynaptic potentials. Consequently, these interneurons appear to be key to the generation of gamma oscillations (106). In a group of first episode psychosis patients scanned using an ultra-high field magnet [7 Tesla (T)] we reported a decrease in glutamate, but not GABA levels, in the anterior cingulate cortex (107). Interestingly, in a larger group of first episode psychosis patients also scanned at 7T, Wang reported both a decrease in anterior cingulate glutamate and GABA levels (108). In the same group of first episode psychosis patients in which we found a glutamate decrease (107), we did not find abnormalities in gamma range oscillations using MEG (109), although others, also in first episode patients, have reported such alterations (110)(111)(112). All three of these studies used EEG, and it has been shown that there can be differences in the cortical auditory evoked response between MEG and EEG (113). Inspired by imaging studies reporting decrease in N-acetyl-aspartate, a marker of neuronal integrity, in the anterior cingulate cortex in schizophrenia (114), Roberts (115) used electron microscopy in anterior cingulate cortex postmortem brain samples and identified a decrease in the number of excitatory synaptic connections, as well as a decrease in the number of mitochondria per neuronal somata, suggesting a decrease in cortical efficiency in schizophrenia.
How Does Neuroimaging Inform Our Mechanistic Understanding of the Illness?
Our work has contributed to the extensive efforts in studying the structural and functional neuroanatomy of schizophrenia by mapping relevant neural signatures in psychosis spectrum patients. Our data is congruent with the idea that the brain is fundamentally affected in the illness, where widespread structural gray and white matter involvement, functionally abnormal cortical and subcortical information processing, and neurometabolic dysregulation are present in patients. Importantly, data indicate that pathology is not merely diffusely distributed across the entire brain, rather it appears that certain brain circuits are preferentially affected. The evidence is compelling that many of the subtle abnormalities described in chronic schizophrenia are already evident in first episode psychosis patients, highlighting that these brain signatures are likely to be relevant to the core pathology and not a just be a consequence of other non-specific factors associated with the illness (116). We also demonstrated that these brain circuitry alterations are clinically relevant by showing that these pathological signatures can be leveraged for predicting subsequent response to antipsychotic treatment, further underscoring that they are key features of the illness.
Despite these successes in delineating disease signatures, it remains challenging to identify casual factors leading to these alterations through cross-sectional unimodal mapping alone (117). To gain further insights, our group has leveraged different approaches. First, we used various pharmacological challenges to characterize the effects of active modulation of major neurotransmitter systems on neural signatures. We clearly demonstrated that dopamine D2 receptor blockers alleviate neural abnormalities to some extent, and that experimentally induced NMDA receptor blockage results in alterations that resemble those seen schizophrenia, underscoring the pathophysiological relevance of these neurotransmitter systems. Second, we combined findings from different imaging modalities to gain additional insights into the role of glutamatergic neurotransmission for the modulation of functional brain networks. Our studies consistently demonstrated a disruption in the link between glutamate levels and brain function in patients across paradigms tapping into various aspects of brain function, which implicates that the disruption in this coupling may be a fundamental feature of the illness. Third, postmortem studies can also inform mechanistic understanding. For example, decrease in mitochondrial function observed in schizophrenia postmortem samples (115) suggest abnormalities in brain bioenergetics.
Taken together, it is highly unlikely that the pathogenesis of schizophrenia is uniform (118), it is more plausible that there may be multiple different etiologies that converge to the behavioral phenotype of schizophrenia (119). Our data underscore that mechanistically oriented neuroimaging studies must take non-specific factors such as antipsychotic drug exposure or illness chronicity into consideration when interpreting disease signatures, as a clear characterization of primary pathophysiological processes is an imperative prerequisite for rational drug development and for alleviating disease burden in our patients.

AUTHOR CONTRIBUTIONS
AL and NK wrote the manuscript. Both authors contributed to and approved the final version of this manuscript.