The neuromodulatory role of dopamine, serotonin, and
norepinephrine in working memory deficits in schizophrenia:
A comparison of task-related fMRI activity in dopaminergic, serotonergic, and noradrenargic brainstem regions between patients with schizophrenia and healthy controls.Nada Amekran, MSc Brain and Cognitive Sciences — University of Amsterdam
External examiner: Verónica Mäki-Marttunen, Leiden University Internal examiner: Tim Zimmermans, University of Amsterdam
Abstract
The dopaminergic system plays a crucial role in positive symptoms of schizophrenia, but its role in negative and cognitive symptoms is unclear. Cognitive symptoms are a key indicator of functional outcome in schizophrenia and often precede positive symptoms. The serotonergic and noradrenargic regions are likely to be implicated in the pathogenesis of cognitive symptoms. We investigated whether working memory (WM) impairment in schizophrenia is modulated by altered brainstem activity in dopaminergic, serotonergic, and noradrenargic regions. WM-related fMRI activity in the ventral tegmental area (VTA), substantia nigra (SN), locus coeruleus (LC), and raphe nuclei were compared between 17 patients with schizophrenia and 15 healthy controls during no, any, and high cognitive load. Patients showed reduced BOLD activity in the VTA, SN, and LC without cognitive load, but not when cognitive load was added. During high cognitive load, higher LC beta-coefficients corresponded to higher accuracy performance in healthy controls, but not in patients. This result is a promising shift of research focus from positive to cognitive symptoms and implicates norepinephrine in WM-deficits in schizophrenia. An important limitation of this study was that possible effects of antipsychotics on BOLD activity could not be taken into account.
Introduction
Schizophrenia is a psychiatric disorder that is mostly characterized by positive symptoms (e.g. hallucinations or delusions), negative (e.g. reduced emotion expression or decreased
motivation), and cognitive (e.g. decreased executive functioning) symptoms. One of the most accredited theory underlying part of the mechanism underlying schizophrenia is the dopamine hypothesis (Bear, Connors, & Paradiso, 2015). This hypothesis is derived from the observed
pharmacological links between the mesocorticolimbic dopamine system and schizophrenia (Meltzer & Stahl, 1976): Firstly, the use of amphetamines by healthy people increases the release of
dopamine (DA). Simultaneously, an overdose of amphetamines has been found to induce a psychotic episode with positive symptoms that are virtually the same as schizophrenic psychotic episodes (Bramness et al., 2012). A second implication of the dopaminergic system in schizophrenia is that antipsychotics used to treat symptoms of schizophrenia are mainly targeted at dopamine receptors (Uchida & Mamo, 2009). These antipsychotics can be divided into two kinds: typical (also referred to as classical) antipsychotics, such as haloperidol, and atypical antipsychotics, such as clozapine (Smieskova et al., 2010). Both types of antipsychotics affect the dopaminergic system mostly at the D2 receptor, with a difference in their affinity. Typical antipsychotics are known to be potent blockers of dopamine (D2) receptors in the mesocorticolimbic system (Dazzan et al., 2005). Atypical antipsychotics have the same antagonistic effect on D2 receptors but at a lower affinity. Altogether, these findings suggest hyperactivation of the dopaminergic system to be related to positive symptoms in schizophrenia.
While the dopamine hypothesis shows a clear relationship between the dopaminergic system and positive symptoms, the involvement in negative and cognitive symptoms is less understood. It is suggested for atypical antipsychotics to be more effective on negative and cognitive systems than typical antipsychotics (Wang et al., 2013). A key difference between these subtypes is that in
addition to D2 receptor binding, atypical antipsychotics also show high occupancy on
5-hydroxytryptamine (5-HT), or serotonin, receptors, implicating the role of serotonin in the pathway of schizophrenia (Meltzer, Matsubara, & Lee, 1989). Even though atypical antipsychotics have some effect on serotonin receptors, they have not yet been designed or approved for these effects. More specifically, the American Food and Drug Administration (FDA) currently requires 60-70% blockage of D2 receptors for all marketed antipsychotics (Kantrowitz, 2020), and the role of serotonin in the efficacy of antipsychotics remains less clear. Consequently, antipsychotics used to treat symptoms of schizophrenia are mainly targeted at dopamine receptors (Uchida & Mamo, 2009). With the dopamine hypothesis rising from the similarities that overdoses have with the
positive symptoms of schizophrenia, the treatment of negative and cognitive symptoms with antipsychotics remains deficient (Kishimoto, Hagi, Nitta, Kane, & Correll, 2019; Kantrowitz, 2020). Past research shows that blockage of the dopamine type 2 receptors is often not sufficient to improve negative and cognitive symptoms, and for some patients it also does not improve positive symptoms (McCutcheon, Krystal, & Howes,, 2020). Additionally, a recent meta analysis comparing previous research on efficacy of antipsychotics suggests effects of antipsychotics on negative and cognitive symptoms to be very limited (Huhn et al., 2019). While the past decades have shown the crucial role of dopamine in the pathogenesis of schizophrenia, these findings suggest that the dopamine hypothesis is not sufficient to explain all aspects of this complex disorder. Furthermore, negative and cognitive components of the disorder have been found to be strong indicators of functional outcome, and current research is slowly shifting towards treatment of schizophrenia that addresses these components as well as the positive symptoms (Kantrowitz, 2020).
As previously mentioned, the role of the serotonergic system in the pathogenesis of
schizophrenia could be of interest. Apart from the relationship between antipsychotics and serotonin (for an overview of binding affinity and efficacy, see Kantrowitz (2020)), serotonin also plays a role in cognition (Švob Štrac, Pivac, & Mück-Šeler, 2016). Serotonin has been found to play a major role in executive functions such as working memory, attention, and decision making (Švob Štrac et al., 2016; Cano-Colino et al., 2013; Williams, Rao, & Goldman-Rakic, 2002), all of which are cognitive symptoms that are often disrupted in patients with schizophrenia (Bowie & Harvey, 2006). These cognitive symptoms are mostly reported to precede positive symptoms in patients with schizophrenia and are a good indicator for future functional outcome (Bowie & Harvey, 2006). Whereas serotonergic regions are suspected to modulate cognitive symptoms of schizophrenia (Švob Štrac et al., 2016; Querdnow, Geyer, & Halberstadt, 2010; O’Connor, 1998), the exact neuromodulatory function in schizophrenia remain largely unknown. An interactive role between dopaminergic and serotonergic systems is suspected (Bear et al., 2015).
In addition to the involvement of the dopaminergic and serotonergic systems, mouse models have shown schizophrenia-like impairments in social behavior, spatial working memory, and
reference memory, to be related to hyper- and hypofunctionality of the glutamate and GABA pathways rather than dopaminergic pathways (Desbonnet, 2016). In accordance with this, another theory explaining part of the underlying mechanism of schizophrenia is the glutamate hypothesis. The glutamate hypothesis arose based on behavioral changes (such as hallucinations and paranoia) induced by drugs, namely phencyclidine and ketamine, and their similarities to symptoms of schizophrenia. Different from amphetamines, the neural mechanism underlying phencyclidine and
ketamine does not involve the dopaminergic system. Rather, these drugs work by blocking a type of glutamate receptors, namely N-methyl-D-aspartate (NMDA) receptors (Anis, Berry, Burton, & Lodge, 1983). The involvement of glutamate in schizophrenia is further supported by a post-mortem study that showed altered dendrites of glutamatergic neurons in schizophrenic cerebri (McCutcheon et al., 2020). Schizophrenia-like effects of phencyclidine and ketamine are not limited to positive symptoms, but have also been reported for both negative and cognitive changes similar to schizophrenic symptoms (Hu, MacDonald, Elswick, & Sweet, 2016). These drugs have also been reported to worsen positive, negative, and cognitive symptoms in schizophrenic patients, unlike amphetamines, which are only reported to worsen positive symptoms (McCutcheon et al., 2020). Both the dopamine and glutamate hypotheses have been well researched and make a compelling case for their involvement in the development of schizophrenia. It is important to note that the involvement of both neurotransmitters is not mutually exclusive, neither do they need to be. It is most likely for both neurotransmitters to be crucially involved in the complex development and maintenance of schizophrenia.
Whereas there is consensus on dopamine, serotonin, and glutamate modulating the course of schizophrenia, other neuromodulatory regions are much less extensively investigated. The role of norepinephrine (NE) and its interaction with the dopaminergic system is one of these
underinvestigated modulations (Mäki-Marttunen, Andreassen, & Espeseth, 2020). Implication of NE in schizophrenia has been suggested throughout the years. Decades ago, analysis of
cerebrospinal fluid showed higher concentrations of NE in individuals with schizophrenia than in healthy controls (Lake et al., 1980). Furthermore, some atypical antipsychotics (such as risperidone) were found to alter firing patterns of norepinephrine neurons in rodents (Grinchii & Dremencov, 2020). It has been difficult to investigate NE in schizophrenia, partly due to techniques using more conventional methodology, such as positron emission tomography (PET) scans, being
underdeveloped in measuring norepinephrine transporters (Takano, 2018). However, investigating the role of NE in schizophrenia is especially relevant since the NE is known to affect various cognitive processes that are impaired in individuals with schizophrenia, such as working memory (Borodovitsyna, Flamini, & Chandler, 2017).
Working memory, as defined by the Baddeley and Hitch (1974) model, is the cognitive capacity to temporarily store a limited amount of information over brief time periods. Previous research has found this cognitive process to be disrupted in individuals with schizophrenia (Nuechterlein et al., 2004). A meta-analysis conducted on a total of 36 working memory tasks comparing performance between healthy controls and patients with schizophrenia found working
memory to be impaired in all samples with schizophrenia and across all 36 tasks, showing a large and robust effect (Forbes, Carrick, McIntosh, & Lawrie, 2009). Furthermore, changes in functional connectivity in the dorsolateral prefrontal cortex (DLPFC) have been found in patients with
schizophrenia during working memory tasks (Meyer-Lindenberg et al., 2005). The DLPFC being an important area in the mesocortical dopamine pathway (Bear et al., 2015) again suggests the
involvement of the dopaminergic system in working memory impairment in schizophrenia. The importance of targeting cognitive symptoms in schizophrenia is increasing due to the opportunity it creates for early intervention (Bowie et al., 2006). However, research on understanding and
adequately treating cognitive symptoms in schizophrenia remains rather limited. The existence of cognitive impairment in schizophrenia is well represented in previous research, as well as research on possible neuromodulatory systems in schizophrenia. However, research that integrates these separate components involved in the complex pathogenesis of schizophrenia remains scarce, possibly partly due to the methodological challenges neuromodulatory research comes with.
Recent methods have allowed to look at the activity in NE and DA regions using brainstem fMRI as a measurement. A recent study by Köhler et al. (2019) investigated the role of dopamine and norepinephrine in working memory of patients with schizophrenia. DA activity was measured at the ventral tegmental area (VTA) and substantia nigra (SN), whereas NE activity was measured at the locus coeruleus (LC). Measurements were task-related, looking at BOLD activity during the STROOP task. Results showed reduced task-related BOLD activity in the VTA in schizophrenics compared to controls, whereas this difference was not found in the SN or LC. However, activation in the LC was found to significantly correlate with interference time in healthy controls, which ceased to exist in patients with schizophrenia. This finding suggests that the LC is more likely involved in conflict resolution rather than in task engagement in patients with schizophrenia (Köhler et al., 2019).
Taken altogether, there is compelling theoretical ground to study the role of multiple
neuromodulatory systems on cognitive symptoms in schizophrenia. In order to gain insight into the role of dopaminergic, noradrenargic, and serotonergic regions in modulating working memory in patients with schizophrenia, the current study used a similar method as Köhler et al. (2019) of brainstem fMRI. This method was used on already existing fMRI data of individuals with
schizophrenia and healthy controls while performing a WM task. More specifically, we investigated how neuromodulatory regions in the brainstem related to differences in working memory task difficulty, and whether task-related brainstem activity differs between patients with schizophrenia and healthy controls. We aimed to answer the question of whether working memory related
cognitive impairments in schizophrenia are modulated by altered brainstem activity patterns compared to healthy controls. In order to do this we first looked at the differences between healthy controls and patients with schizophrenia in performance on a working memory task. We expected healthy controls to perform better on the working memory task than individuals with schizophrenia. We then looked at brainstem activation during this task by computing differences in task-related BOLD activation in the VTA, SN, and LC in both patients with schizophrenia and healthy controls. In line with the findings from Köhler et al. (2019), we expected individuals with schizophrenia to show lower task-related BOLD activation in the VTA compared to healthy controls. We did not expect to find differences in BOLD activation in the locus coeruleus between individuals with schizophrenia and healthy controls. However, we expected BOLD-activation in the locus coeruleus to change with increased effort on the working memory task in healthy controls, whereas we did not expect this difference to occur in individuals with schizophrenia. Lastly, we looked at BOLD
activation in the raphe nuclei (which release serotonin) to see if there was a difference between patients and healthy controls, albeit task-related.
Method Participants
Participants were recruited and tested by Repovš and Barch (2012) and data were made available on OpenNeuro. The complete dataset consisted of data from patients with schizophrenia, healthy controls, and siblings of both groups. For this study we did not include data from siblings of patients with schizophrenia. We added an age requirement of minimum 18, and excluded all data of participants younger than 18. As we did not look at effects on siblings, we pooled the data of healthy controls and their siblings together into one control group to still have a balanced design after exclusion based on age.
Materials
N-back task. The n-back task was used as a measure of working memory performance. The n-back task is a working memory task that shows a series of letters to the participants. The task is to assess whether the shown letter is the same as a pre-specified letter (0-back) or the same as a
previously shown (1-back/2-back) letter in the sequence. With 1-back meaning the letter is the same as the one shown in the trial before, and 2-back meaning the letter is the same as the one shown 2 trials before. For the 0-back task the letter requiring a response is pre-specified for a couple of trials. Task-related BOLD scans were obtained during the n-back task, with separate runs for the 0-back, back, and 2-back condition. All participants did an n-back task in all conditions, i.e. 0-back,
1-back, and 2-back. A full description of the task done by this sample can be found in the article by Repovš & Barch (2012).
Behavioral results. A repeated measures analysis of variance (rANOVA) was conducted using IBM SPSS Version 27 (IBM corporation, 2020). These analyses were conducted separately for accuracy and reaction time on the n-back task. Cognitive load (0-back, 1-back, 2-back) was set as within-subject factor, while group (control or patient group) was set as between-subject factor. Helmert contrasts were used to look at load differences, comparing 0-back vs. 1-/2-back and 1-back vs. 2-back performance. Significance was tested at α = .05.
MRI dataset. Data were downloaded from OpenNeuro (https://openneuro.org/datasets/ ds000115/versions/00001). The original paper describing the full dataset can be found on https:// www.ncbi.nlm.nih.gov/pmc/articles/PMC3358772/ (Repovš & Barch, 2012). For each participant, data consisted of a T1-weighted structural image (TR = 2400 ms, TE= 3.16 ms, FOV = 256 mm, flip angle= 8°, voxel size = 1 mm x 1 mm x 1 mm) and three task-related functional T2* images (TR = 2500 ms, TE= 27 ms, FOV = 256 mm, flip angle = 90°, voxel size = 4 mm x 4 mm x 4 mm). Task-related scans were obtained during the n-back task. Each run consisted of two blocks of either the 0-back, 1-back, or 2-back task, resulting in three images for each participant for all runs. Each functional scan consisted of 137 volumes. All scans were acquired on a 3T Tim TRIO scanner. Full details of the data acquisition can be found in the previously mentioned original paper by Repovš and Barch (2012). Data also included event files for each n-back, specifying onset times for each stimulus throughout each functional scan.
Preprocessing. Preprocessing was done using FSL (Jenkinson, Beckmann, Behrens, Woolrich, & Smith, 2012). First, all T1 scans ran through the Brain Extraction Tool (BET;
Jenkinson, Pechaud, & Smith, 2005) in order to remove all non-brain tissue from the image. First level FSL-feat (Woolrich, Ripley, Brady, & Smith, 2001) was then used for preprocessing, using the T2* of each n-back task separately, with registration to the skull-stripped T1 output from BET. The five first volumes were deleted as advised by Repovš and Barch (2012). The high pass filter cut-off was set to 100. Motion correction occurred using MCFLIRT (Jenkinson, Bannister, Brady, & Smith, 2002), and slice time correction was done regular up. No spatial smoothing was done at this stage. This resulted in a preprocessed 4D fMRI image per n-back condition for each participant named filtered_func_data. The dimensions of this image were 64 x 64 x 36 x 132.
Brainstem segmentation. The Spatially Unbiased Infratentorial Template (SUIT) v3.4 toolbox (Diedrichsen, 2006; Diedrichsen, Balsters, Flavell, Cussans, & Ramnani, 2009; Diedrichsen et al., 2011; Diedrichsen & Zotow, 2015) was used for segmentation of the brainstem and
cerebellum. This toolbox is openly available on http://www.diedrichsenlab.org/imaging/
suit_download.htm. SUIT was used as a toolbox added to SPM 12 (Friston, Ashburner, Kiebel, Nichols, & Penny, 2007), which in turn ran on Matlab r2019b (Matlab, 2020).
All T1 scans were viewed in SPM to see whether the origin was set to the anterior
commissure. For participants where this was not the case, the origin of the T1 was manually set to the anterior commissure and the resulting matrix was applied to the task-related T2* scans.
The segmentation pipeline was divided into three steps. First, the function suit isolate was run on the T1 weighted scan of each participant, which resulted in a cerebellar isolation map including both the cerebellum and brainstem, and separate isolations of only gray and white matter respectively. Secondly, suit normalize dartel was run on the three isolation maps from the first step. The output of this step consisted of an ‘affine transformation matrix for the linear part of the
normalization' and a 4D nifti with the non-linear flowfield, that allowed to transform the individual brainstem and cerebellum into the template space. Lastly, suit reslice dartel was run. This step ran on the filtered functional image that was obtained from FSL-feat. For each n-back condition and each participant, this 4D image was first split into 132 3D volumes. Suit reslice dartel resampled each volume using the transformation matrix, the non-linear flowfield, and the isolation map from suit isolate. Following, all volumes were merged again into a resliced 4D image for each n-back separately.
All brainstem segmentations were visually inspected by overlaying the segmentation and T1 weighted images (see Figure 1). Following, some segmentations had to manually be corrected, while some images could not result in an acceptable segmentation resulting in exclusion of these runs. All segmentations were spatially smoothed in SPM with a Gaussian filter (fwhm = 3mm). We chose a small kernel for smoothing because of the small size of our ROIs.
Figure 1. Brainstem and cerebellum segmentation (in red) overlaid on the T1 weighted image (greyscale) of
fMRI analysis. First and second level analyses were run for each participant using SPM. A first level model was specified, estimated, and contrasted for each participant. Model estimation was done using the volumes of the 4-dimensional smoothed brainstem images of all n-back runs. The TR was 2.5 seconds, while microtime resolution and onset were left as default (16 and 8 respectively). Onset times (specified in seconds) and duration of each stimulus were imported from the original dataset. Individual motion correction regressors were added using the movement parameters from MCFLIRT. Additionally, the canonical HRF with its temporal derivatives was added for each functional scan. Altogether, this resulted in a total of eight regressors for each n-back condition. The aforementioned model was then estimated, and t-tests were run for three contrasts. The first contrast looked at BOLD activation during the 0-back task. This contrast was used as a measurement of no cognitive load. The second contrast compared BOLD activation during the 2-back to0-back, expecting higher activation during the 2-back task. This contrast was used as a measurement of high cognitive load. Lastly, the third contrast compared the difference between BOLD activation of the 2-back and 1-back tasks together with respect to BOLD activation during the 0-back task.This contrast was a measurement of any cognitive load, with low (1-back) and high (2-back) cognitive load taken together. The 0-back condition was always subtracted from the
cognitive load conditions to take baseline activity (without added cognitive load) into account and allows us to look at the change in activity when cognitive load is added.
Second level analyses were run for each of the specified first level contrast. T-tests were done for each contrast, comparing contrast-specific BOLD activation of patients with schizophrenia and healthy controls. The second level contrast was specified with the expectation of higher BOLD activation for healthy controls than for patients.
Regions of interests (ROIs). The Ventral Tegmental Area (VTA), Substantia Nigra (SN), Locus Coeruleus (LC), and Raphe Nuclei (RN) were defined as regions of interest. An indication of dopaminergic activation was inferred from the percentage in changed BOLD activation within a mask of the VTA (Murty et al., 2014) and SN (Murty et al., 2014). An indication of norepinephrine was inferred from the percentage in changed BOLD activation within the LC mask (Tona et al., 2017). All masks were registered to the same space as the brainstem segmentations using SPM and were thresholded in order to be binarized (see Figure 2 and Figure 3). The dorsal raphe nucleus was looked at using a 3 mm sphere around [0, -31, -9] in MNI305 space, which are coordinates for the dorsal raphe nucleus reported by Beliveau et al. (2015). These masks were then used for a small volume correction on the second level maps.
Beta-coëfficients fMRI. Beta coëfficients were extracted using the marsbal toolbox (Brett, Anton, Valabregue, & Poline, 2002; retrieved from http://marsbar.sourceforge.net). These parameters were extracted for each participant per ROI (4 ROIs in total) and separately for each first level contrast (3 contrasts), resulting in 12 parameters per participant. Accuracy and reaction times on the N-back task for the same contrasts (0-back, 2- and 1-back minus 0-back, and 2- minus 0-back) were calculated and Pearson correlations were computed in IBM SPSS to assess the relationship between task-related ROI activation and performance on the n-back task for healthy controls and patients with schizophrenia separately.
Figure 2. VTA (yellow/red) and SN (green/blue) masks overlaid on the brainstem segmentation of a single
participant.
Results Sample
Visual inspection of the brainstem segmentations revealed that segmentation was not successful for all participants. All steps of the segmentation process as described in the Method section were repeated manually and inspected after each step for all incomplete segmentations. Participants whose brainstem segmentation remained incomplete following manual correction were excluded from the final sample. Our final sample consisted of 15 participants in the control group and 17 participants in the patient group. Demographics of both groups can be found in Table 1. Groups did not differ significantly on gender and age. For standardized IQ scores and years of school attended, the means of the control group were significantly higher than of the patient group.
Table 1. Means (standard deviations) of demographic differences on gender, age, standardized IQ-scores and
years of school attended. An independent sample t-test was conducted to assess whether these demographics differed between patients with schizophrenia and healthy controls. Significance was tested at p < .05*.
Behavioral results
Accuracy. A repeated measures ANOVA was conducted in order to assess group differences on the three n-back loads. Accuracy on the n-back task for the three levels (0-back, 1-back, and 2-back, respectively) was set as within-subject factor, and group (control versus patient condition) was set as a between-subject factor. Mauchly’s test of sphericity showed that the assumption of sphericity was violated (Mauchly’s W = .561, X2(2) = 16.771, p < .001). In order to take this into account, univariate tests were looked at with the Greenhouse-Geisser correction (Fields, 2009). Univariate tests were still found to be significant following this correction for both the effect of accuracy (F(1.390, 41.691) = 10.758, p < .001) and the interaction effect of accuracy * group (F(1.390, 41.691) = 8.743, p = .002). Table 2 shows the mean percentage accuracy for each group. Within-subject tests using Helmert contrasts (see Table 3) show that the effect of accuracy is only found in 0-back versus 1/2-back, and not between 1-back and 2-back performance, whereas the interaction effect accuracy*group shows a significant effect for both contrasts. This interaction effect is illustrated in Figure 4.
Control (N=15) Patients (N=17) Significance
Gender 11 males 13 males t(30) = 1.98, p = .844
Age 22.54 (2.88) 24.89 (3.61) t(30) = -2.02, p = .052
IQ (standardized z-score) .12 (.75) -.77 (.85) t(30) = 3.027, p = .005* Years of school 14.40 (1.55) 12.35 (2.00) t(30) = 3.206, p = .003*
Between-subject analysis showed that, on average, the control group performed 12.2% more accurate on the n-back task than the patient group, but this difference was not significant (F(1,30) = 3.907, p = .057, 95% CI [-.004, .248]).
Table 2. Means (standard deviations) of percentage accuracy reaction times (in ms) per n-back for healthy
controls and patients with schizophrenia.
Table 3. Helmert contrasts of significant univariate tests. First, these contrasts show whether performance
between no cognitive load (0-back) and cognitive load (1-back and 2-back) are significantly different for the main effects (accuracy and RT) and interaction effect (accuracy*group). Secondly, this contrast looks at whether there is a significant difference between low (1-back) and high (2-back) cognitive load. This difference is not found for accuracy or reaction time, but is found for the interaction effect accuracy*group. Significance was tested at p < .05*.
Reaction time. A repeated measures ANOVA was conducted in order to assess group differences on reaction times (RT) of the 3 n-back loads. RT on the n-back task for the three levels (0-back, 1-back, and 2-back, respectively) was set as within-subject factor, and group (control versus patient condition) was set as a between-subject factor. Mauchly’s test of sphericity showed that the assumption of sphericity was violated (Mauchly’s W = .607, X2(2) = 14.5, p < .001). In order to take this into account, univariate tests were looked at with the Greenhouse-Geisser
Control (N=15) Patients (N=17) 0-back Accuracy 93.9 (13.2) 93.1 (17.0) Reaction time 690.23 (141.55) 726.53 (92.47) 1-back Accuracy 91.1 (16.5) 79.4 (18.9) Reaction time 753.33 (141.03) 815.06 (119.32) 2-back Accuracy 92.8 (16.0) 68.6 (30.3) Reaction time 753.57 (146.05) 852.79 (247.65)
Effect Helmert contrast Significance testing
Accuracy 0 - back vs. later F(1,30) = 19.972, p < .001*
1 - back vs 2 - back F(1,30) = 2.512, p = .123
Accuracy *
Group 0 - back vs. later F(1,30) = 13.278, p = .001*
1 - back vs 2 - back F(1,30) = 4.684, p = .039*
RT 0 - back vs. later F(1,30) = 13.30, p < .001*
correction (Fields, 2009). Univariate tests were found to be significant for the main (cognitive load) effect of RT (F(1.435, 43.058) = 15.179, p = .017), but the interaction effect of RT * group
remained absent (F(1.435, 43.058) = .566, p = .516). We therefore only looked at post-hoc contrasts for the found effect of RT. Within-subject tests using Helmert contrasts show that the load effect of RT was only found in 0-back versus 1/2-back, and not between 1-back and 2-back performance (see Table 2).
Between-subject analysis showed that, on average, the control group was 71.75 ms faster in performance on the n-back task than the patient group, but this difference was not significant (F(1,30) = 2.781, p = .106, 95%CI [-159.613, 16.113]).
Figure 4 (left). Estimated marginal means of accuracy performance on each n-back level. Accuracy
performance is plotted separately for the control group (blue) and schizophrenia group (green). The significant interaction effect that was found (see Table 3) becomes visible here: At the 0-back accuracy performance does not differ between the two groups, but with increased cognitive load, accuracy performance in the schizophrenia group drops significantly, whereas this does not happen in the control group. In the patient group, the accuracy is lower during 2-back than during 1-back, and lower during 1-back than during 0-back.
Figure 5 (right). Estimated marginal means of reaction time on each n-back level. Reaction time (ms) is
plotted separately for the control group (blue) and schizophrenia group (green). This plot shows the presence of a main effect (see Table 3), whereas an interaction effect of reaction time * group lacks. The plot shows an increase in reaction time with increased cognitive load, and this increase occurs approximately the same for the two separate groups (parallel).
fMRI analysis (voxel-wise)
Second level analyses showed significant differences in voxel activation between the control and patient group within the VTA, SN, and LC masks (see Table 4) during the 0-back task (no cognitive load). This means that during the 0-back task, BOLD activation in the VTA, SN, and LC was on average significantly higher in the control group than in the schizophrenia group. No significant clusters were found in the coordinates for the raphe nuclei. No significant group
differences were found when subtracting activation during no cognitive load (0-back) from low/ high (1/2-back) or high (2-back) cognitive load.
Table 4. Results of second level SPM analyses for the three contrasts specified in the first level analysis (see
Method section). A t-test contrast for controls > patients was computed for all second level analyses. All results were thresholded at p = .001. This table only shows significant activation differences within the relevant masks. Significant activation differences (for controls > patients) were only found during the 0-back task for the VTA, SN, and LC. Ke shows the cluster size of the found activation difference. Coordinates are reported in MNI-space.
Second level analysis of a t-test contrast for patients > controls showed no significant clusters for any of the three cognitive load conditions in the VTA, SN, LC, or RN.
ROI-based analysis
Pearson correlations between the beta-coefficients of each voxel activation mask and the corresponding performance on the n-back task showed a significant correlation between beta-coefficients of LC activation and accuracy in target trials with added cognitive load. These correlations were only found to be significant in the control group, but was non-existent in the patient group (see Table 5). Surprisingly, this significant correlations within the control group showed a negative relationship between LC activation and differences in accuracy when comparing any cognitive load (1- and 2-back) to no cognitive load (0-back) (p = .042; see Figure 7) while beta coefficients of LC activation and differences in accuracy between high cognitive load (2-back) and
First level contrast ROI T-test (df = (1,30)) Ke z-value puncorr MNI
coordinates
No cognitive load VTA 2.83 3 2.65 0.004 -1, -23, -6
SN 2.85 1 2.66 0.004 11, -21, -10
LC 3.29 8 3.01 0.001 8, -39, -29
RN No significant clusters
Any cognitive load VTA No significant clusters SN No significant clusters LC No significant clusters RN No significant clusters
High cognitive load VTA No significant clusters SN No significant clusters LC No significant clusters RN No significant clusters
no cognitive load (0-back) were positively correlated (p = .009; see Figure 6). It is important to note here that on average, the control group showed higher accuracy on the high cognitive load (2-back) trials than on the low cognitive load (1-back) trials. While this difference in accuracy was not significant (as can be seen in the Helmert contrast 1 vs. 2-back in Table 3), this difference does explain the paradoxical negative and positive correlation with the LC beta-coefficients (see Table 5). All Pearson correlations between beta coefficients and accuracy can be found in Table 5. No correlations were found to be significant for the SN, VTA, and RN masks. Reaction time correlations also showed no significant relationships for either group (see Table 6).
Table 5. Pearson correlations between the beta coefficients of the LC, SN, VTA, and RN (respectively), and
accuracy performance on the n-back task. Correlations were computed for the contrasts of no cognitive load (0-back), high cognitive load (2-0 back), and any cognitive load (2 and 1 - 0 back), and separately for the control group (N=15) and patients with schizophrenia (SCZ; N=17) (noted in the table as SCZ). * shows a significant correlation at p < .05. ** shows a significant correlation at p < .01.
Note. LC correlations of the SCZ group contain 16 participants due to the extracted beta coefficient of one participant being unreasonably high.
Table 6. Pearson correlations between the beta coefficients of the LC, SN, VTA, and RN (respectively), and
reaction time on the n-back task. Correlations were computed for the contrasts of no cognitive load (0-back), high cognitive load (2-0 back), and any cognitive load (2 and 1 - 0 back), and separately for the control group and patients with schizophrenia (noted in the table as SCZ). No significant correlations were found between the beta coefficients and reaction time on the n-back task.
Note. LC correlations of the SCZ group contain 16 participants due to the extracted beta coefficient of one participant being unreasonably high.
Accuracy LC SN VTA RN
Control: No cognitive load -.253 -.245 -.396 -..0139
Control: High cognitive load .531* -.189 -.153 -.151
Control: Any cognitive load -.645** -.132 .105 .167
SCZ: No cognitive load -.253 -.123 -.074 -.094
SCZ: High cognitive load .365 -.011 -.028 .022
SCZ: Any cognitive load .152 -.065 -.025 -.040
RT LC SN VTA RN
Control: No cognitive load -.364 .181 -.014 -.064
Control: High cognitive load .071 .266 -.107 -.080
Control: Any cognitive load .180 .296 .200 -.221
SCZ: No cognitive load .083 -0.123 .074 -.063
SCZ: High cognitive load .016 .018 .084 .183
Figure 6. High cognitive load contrast (2-back minus 0-back). Scatter plot visualizing the correlation
between accuracy performance and beta-coefficients of the LC when subtracting performance and activity during 0-back from performance and activity during the 2-back task. The plot on the left shows the data for the healthy control, showing a positive linear relationship (see Table 5). The plot on the right shows the data for patients with schizophrenia, where no relationship is visible (see Table 5).
Figure 7. Any cognitive load contrast (2- and 1-back minus 0-back). Scatter plot visualizing the
correlation between accuracy performance and beta-coefficients of the LC when subtracting performance and activity during 0-back from performance and activity during the 1- and 2-back task taken together. The plot on the left shows the data for the healthy control, showing a negative linear relationship (see Table 5). The plot on the right shows the data for patients with schizophrenia, where no relationship is visible (see Table 5).
An independent samples t-test comparing the means of the beta-coefficients for each mask between healthy controls and patients with schizophrenia showed significant mean differences in the VTA on all cognitive load contrasts, while no significant differences were found in the beta coefficients of the SN, LC, or raphe nuclei (see Table 7). Interestingly, the differences in the VTA show higher beta-coefficients for the control group than the patient group during the no cognitive
load contrast (0-back), whereas when looking at the any or high cognitive load contrasts this difference is reversed and the control group shows lower beta coefficients than the patient group (see Table 8).
Table 7. Independent samples t-test of control > schizophrenia. A positive t-value portrays a higher mean for
healthy controls than for patients with schizophrenia, whereas a negative t-value portrays a lower mean for healthy controls than for patients with schizophrenia. Significance was tested at p < .05, as marked by *.
Table 8. Means differences (control group - patient group) of the beta-coefficients of each mask, per
cognitive load contrast. Positive values refer to higher beta-coefficients for the control group compared to the patient group, whereas negative values refer to lower beta-coefficients for the control group compared to the patient group. Means differences that were found to be significantly different between groups (see Table 7) are reported in bold.
Discussion
The main purpose of this study was to investigate the involvement of dopaminergic,
noradrenargic, and serotonergic regions in working memory performance and whether this differs in patients with schizophrenia compared to healthy controls. We aimed to answer the question of whether cognitive impairments in schizophrenia that relate to working memory are possibly modulated by altered brainstem activity patterns compared to activity patterns in healthy controls. We first looked at differences in cognitive performance separately, which will be discussed under behavioral results. Following, we looked at the brainstem activity corresponding to cognitive performance. This was done for the VTA, SN, LC, and RN, and these results will be discussed under fMRI results.
Behavioral results
We looked at accuracy performance and reaction times on the n-back task as a measure of working memory. The n-back task consisted of three levels: No cognitive load (0-back), low
Contrast LC SN VTA RN
No cognitive load t(30) = 1.583
p = .124 t(30) = 1.164 p = .254 t(23.456) = 2.432 p = .023* t(30) = .694 p = .493
High cognitive load t(30) = 0.791
p = .435 t(30) = -.408 p = .686 t(30) = -2.222 p = .034* t(30) = -.347 p = .731
Any cognitive load t(30) = -.787
p = .437 t(30) = -1.135 p = .266 t(30) = -2.601 p = .014* t(30) = -.632 p = .532
Contrast LC SN VTA RN
No cognitive load 1.02 0.45 1.00 0.05
High cognitive load -0.60 -0.46 -1.13 -0.056
cognitive load (1-back), and high cognitive load (2-back). We expected healthy controls to perform better on the n-back task than patients with schizophrenia in both accuracy (higher accuracy) and reaction time (lower reaction time). We expected cognitive load to have a negative effect on performance, where added cognitive load resulted in lower accuracy performance and higher reaction times. We additionally expected these differences to be bigger in the patient group than in the control group.
In line with our hypothesis, we found a significant effect of cognitive load on accuracy performance and reaction time on the n-back task in both groups. For accuracy performance on the n-back task, we also found the expected interaction effect. This showed a significant decrease in accuracy performance in patients with schizophrenia with added cognitive load, whereas this
decrease was less evident in healthy controls. The same interaction effect was not found for reaction time. Our results are supported by previous literature. As previous research has shown, cognitive load has a bigger negative effect on patients with schizophrenia than on healthy controls (Keefe, 2000). A paper looking particularly at differences on the n-back performance in schizophrenia found the same main effects of cognitive load as our study (Krieger, Lis, Cetin, Callhofer, & Meyer-Lindenberg, 2005). A difference, however, is that this study also found an interaction effect for reaction time and group, as we originally expected. The sample of this study (Krieger et al., 2005; N=12) was similar in size to our sample, but they presented participants with shapes (squares and triangles) rather than letters, perhaps measuring a slightly different component of working memory than in our n-back task. Furthermore, this study took all trials on the n-back task into account, whereas we only looked at target trials. It is possible that reaction times increased more during non-target trials, but this analysis was not part of the current study as non-non-target trials do not necessarily measure working memory performance, as opposed to target trials (Meule, 2017). It should also be noted that we did not take differences in IQ score into account, even though IQ score differed significantly between groups. The meta-analysis by Forbes and colleagues (2009) showed that across 36 working memory tasks, IQ score did not modulate the cognitive impairment in this domain in patients with schizophrenia, suggesting a specific working memory deficit rather than a general cognitive deficit. However, none of working memory tasks taken into account in this meta-analysis were a variation of the n-back task, thus we cannot be sure that our results are equally robust to IQ score differences. Despite this, we do expect the cognitive impairment to be working memory specific in our study.
fMRI results
In order to investigate the modulating role of brainstem activity patterns on our behavioral results, we analyzed differences in BOLD activity in brainstem segmentations of a pre-existing fMRI dataset. We particularly looked at masks of the VTA, SN, LC, and the raphe nuclei. We compared changes in activity within these masks between patients with schizophrenia and healthy controls. These differences were looked at for three different contrasts: We looked at (1) differences in activation without added cognitive load (0-back), (2) differences between the change in
activation between the highest and no cognitive load (2-back minus 0-back), and the change in activation during any added cognitive load (2- and 1-back minus 0-back).
Dopaminergic region. Within the masks, we expected to find significantly lower task-related BOLD activation in patients with schizophrenia compared to healthy controls in the VTA and SN. This hypothesis was supported within the VTA mask, but not in the SN mask. Patients with schizophrenia showed significantly less BOLD activation in the VTA than healthy controls during the ‘no cognitive load’ conditions. This means that the difference in VTA BOLD activation was also present without cognitive load (at 0-back), which means the difference is not per se related to working memory performance. This is in line with previous literature suggesting the involvement of the dopaminergic pathway to correspond more to positive symptoms of schizophrenia than to cognitive symptoms (McCutcheon et al., 2020). However, our results are not in line with the findings of Köhler et al. (2019), who also looked at BOLD activation in the VTA and SN during a working memory task. It is unclear whether this study compared BOLD activation within the VTA and SN at baseline, as this is not reported, but baseline parameters were included in their model, suggesting their BOLD activation differences to be exclusively related cognitive performances. We additionally ran a t-test on the beta coefficients of VTA, which showed a significant difference between healthy controls and patients with schizophrenia during the cognitive load contrasts as well, which is different from our voxel-wise analysis possibly due to the two analyses having different sensitivities for detecting differences. However, the hyperactivity observed in patients in the any and high cognitive load conditions could also be related to antipsychotic
treatment rather than to the task. A key difference between our study and the study by Köhler et al. (2019) is that they added chlorpromazine equivalents to quantify antipsychotic treatment. As previously discussed, antipsychotics majorly affect the dopaminergic pathway, and it cannot be ruled out (in fact, it is highly likely that) any changes related to the dopaminergic region are affected by antipsychotic treatment if these effects are not corrected for. All patients used in our analysis were using atypical antipsychotics, and some (but not more than four) were possibly also
using atypical antipsychotics (Repovš & Barch, 2012). Unfortunately, the data on use of antipsychotics was not available to us. We have contacted the authors of the original paper that analyzed this data in order to try and get more information on this, but this data was not available anymore. Future work assessing working memory-related VTA and SN activity in patients using antipsychotic treatment would help in understanding our observed effect.
Noradrenargic region. The same difference in BOLD activation we expected for the dopaminergic region was not expected within the LC. In line with the findings of Köhler et al. (2019), we expected no (task-related) difference in BOLD activation between healthy controls and patients with schizophrenia. Instead, we expected activity in the LC to significantly change with increased cognitive load on the n-back task in healthy controls, whereas this effect was not expected in patients with schizophrenia. Our ROI analysis supports this hypothesis. We looked at the beta-coefficients of task-related BOLD activity within the LC mask as provided by Tona et al. (2017) and correlated this with performance on target trials of the n-back task (both accuracy and reaction time). We found a significant correlation between these beta-coefficients and accuracy performance within healthy controls during the high cognitive load contrast and during the any cognitive load contrast, but not when no cognitive load was added. For the high cognitive load, this relationship was positive, which means that increased beta coefficients in the LC (and with that, increased BOLD activation) correlates with higher performance accuracy on the 2-back task and vice versa. This significant correlation was completely absent in patients with schizophrenia. This finding is similar to the finding by Köhler et al. (2019), where they found a significant correlation between performance on the STROOP task (which also measures working memory) and BOLD activation within the LC in healthy controls, but not in patients with schizophrenia. An interesting finding in our study seemed to be that whereas the relationship between the LC beta coefficients and high cognitive load was positive, the relationship between the LC beta coefficients and any cognitive load was negative. Both of these relationships were strong (Cohen, 1988). These finding suggests that higher beta coefficients in the LC correlates with lower mean accuracy performance if we look at the 1 and 2-back taken together, but higher mean accuracy when we look at the n-back separately.
In order to better understand this difference, we looked at a low cognitive load (1-0) contrast separately and found that this the LC beta coefficients for this contrast negatively correlated with the LC beta coefficients of our other contrasts (no cognitive load and high cognitive load). This suggests that our unexpected finding in the ‘any cognitive load’ contrast was mainly driven by the effects during low cognitive load trials. The low and high cognitive load contrasts in this study do not seem to follow comparable trends. Retrospectively, pooling the data of these two loads in an
‘any cognitive load’ condition was not a reliable measure for ‘any cognitive load’, as it resulted in mixed results due to the different linearities of the cognitive loads. We therefore highlight that our results for the high cognitive load show a more reliable and valid effect in comparing patients with healthy controls.. Low cognitive load should be looked at separately in future research in order to understand whether this truly affects LC activity differently from high cognitive load.
So far, very little research has been towards the neuromodulatory role of norepinephrine (NE) in schizophrenia (Máki-Marttunen et al., 2020). It is possible that this is because NE is more related to cognitive symptoms rather than positive symptoms of schizophrenia, and neurobiological research in schizophrenia so far is more heavily focused on positive symptoms. To the best of our knowledge, this is the second study particularly looking at brainstem activation within the locus coeruleus in relation to working memory impairment in schizophrenia, and the results are promising for future research that focuses on cognitive symptoms in schizophrenia because of the clearer relationship between NE and cognitive rather than positive symptoms of schizophrenia
(Borodovitsyna et al., 2017). Furthermore, cognitive symptoms are a good indicator of functional outcome and often precede positive symptoms (Bowie et al., 2006), and insights in the
neurobiological pathways underlying these cognitive impairments hopefully contribute to better treatment methods of cognitive impairments in future research.
Serotonergic region. Due to the ambiguous role of the serotonergic region in
schizophrenia-related cognitive impairment, we did not have any confirmatory hypotheses for the BOLD activity patterns in the RN. However, due to the clear implication of the serotonergic region in the pathophysiology of schizophrenia we wanted to include this region in order to get a more complete overview of this complex mechanism. We found no significant differences in the
serotonergic region between healthy controls and patients with schizophrenia, in neither direction. This difference remained absent in all cognitive load contrasts. The lack of difference in our study does not per se mean that this difference is non-existent outside of this study. Firstly, as we could not find a mask for the brainstem segmentation to show the raphe nuclei within our segmentations, we used coordinates from a single study. This study reported coordinates specifically for the dorsal raphe nuclei, taking a very small part of very small nuclei, which decreases the accuracy of the location due to intersubject variability (Seghier & Price, 2018). Secondly, as discussed in relation to the dopaminergic system, we did not take possible effects of use of antipsychotics into account. As all patients were using atypical antipsychotics, and atypical antipsychotics are known to have high occupancy on serotonin receptors (Kantrowitz, 2020), it is also impossible to draw robust inferences any serotonergic findings without correcting for antipsychotic use.
Limitations
While our findings, albeit limitedly, contribute to the understanding of the neuromodulatory role of brainstem patterns on working memory performance in schizophrenia, some limitations should be taken into account. Even though the difference in age between our two groups was not technically significant, there is a very strong trend towards significance and therefore age
differences cannot be ruled out. Our healthy control group had younger participants than our patient group. Secondly, this study aimed to give a more complete overview of possible neuromodulatory systems in schizophrenia. Although we discussed the glutamate hypothesis of schizophrenia, due to the methodological limitations of measuring glutamate, including this was beyond the scope of this study. The involvement of glutamate in working memory impairment however should not be ruled out. Lastly, our study does not pose any findings for the dopaminergic and serotonergic systems due to the (unexpected) inability to take use of antipsychotics into account, and further research on cognitive symptoms in schizophrenia should still look at the neuromodulatory role of these regions after adjusting for dose-specific effects of antipsychotics (Gardner, Murphy, O'Donnell, Centorrino, & Baldessarini, 2010). The effects of antipsychotics should also not be ruled out in other
neuromodulatory regions, such as the LC. One previous study has found the use of Modafinil in schizophrenia to affect both VTA and LC activity, further highlighting the importance of taking the effects of medication into account (Minzenberg, Yoon, Soosman, & Carter, 2018).
However, our finding within the locus coeruleus are promising for future research looking at the involvement of noradrenaline in modulating working memory deficits in schizophrenia. While measuring fMRI activity in small brainstem nuclei is currently quite extended in human research and can attenuate the task-related involvement of brainstem nuclei, it is still an indirect measure that perhaps should be complemented in the future by e.g. PET studies. Due to the correlational nature of our study, no strong causal inferences can be drawn, but it is an interesting step for future
research to build on. With the slow shift towards addressing cognitive symptoms in schizophrenia in addition to positive symptoms, awareness of the possible role of noradrenaline is crucial in order to adequately treat cognitive symptoms in the future.
References
Anis, N. A., Berry, S. C., Burton, N. R., Lodge, D. (1983). The dissociative anaesthetics, ketamine and phencyclidine, selectively reduce excitation of central mammalian neurones by N-methyl-aspartate. British Journal of Pharmacology, 79(2), 565-575.
Baddeley, A. D., & Hitch, G. (1974). Working memory. Psychology of Learning and Motivation, 8, 47–89.
Bear, M. F., Connors, B. W., & Paradiso, M. A. (2015). Neuroscience: Exploring the brain (4th ed.). Lippincott Williams & Wilkins Publishers.
Beliveau, V., Svarer, C., Frokjaer, V. G., Knudsen, G. M., Greve, D. N., & Fisher, P. M. (2015). Functional connectivity of the dorsal and median raphe nuclei at rest. Neuroimage, 116, 187-195.
Borodovitsyna, O., Flamini, M., & Chandler, D. (2017). Noradrenergic modulation of cognition in health and disease. Neural Plasticity, 6031478.
Bowie, C. R., & Harvey, P. D. (2006). Cognitive deficits and functional outcome in schizophrenia. Neuropsychiatric Disease and Treatment, 2(4), 531-536.
Brett, M., Anton, J. L., Valabregue, R., & Poline, J. B. (2002). Region of interest analysis using an SPM toolbox. Neuroimage, 13(2), 210-217.
Bramness, J. G., Gundersen, Ø. H., Guterstam, J., Borger Rognli, E., Konstenius, M., Løberg, E. M., Medhus, S., Tanum, L., & Franck, J. (2012). Amphetamine-induced psychosis - a separate diagnostic entity or primary psychosis triggered in the vulnerable? BMC Psychiatry, 12, 221. Cano-Colino, M., Almeida, R., & Compte, A. (2013). Serotonergic modulation of spatial working
memory: Predictions from a computational network model. Frontiers in Integrative Neuroscience, 7, 71.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. New York, NY: Academic Press.
Dazzan, P., Morgan, K. D., Orr, K., Hutchinson, G., Chitnis, X., Suckling, J., Fearon, P., McGuire, P. K., Mallett, R. M., Jones, P. B., Leff, J., & Murrary, R. M. (2005). Different effects of typical and atypical antipsychotics on grey matter in first episode psychosis: The ÆSOP study. Neuropsychopharmacology, 30, 765-774.
Desbonnet, L. (2016). Mouse models of schizophrenia: Risk genes. Handbook of Behavioral Neuroscience, 23, 267-284.
Diedrichsen, J. (2006). A spatially unbiased atlas template of the human cerebellum. Neuroimage, 33(1), 127-138.
Diedrichsen, J., Balsters, J. H., Flavell, J., Cussans, E., & Ramnani, N. (2009). A probabilistic atlas of the human cerebellum. Neuroimage.
Diedrichsen, J., Maderwald, S., Kuper, M., Thurling, M., Rabe, K., Gizewski, E. R., et al. (2011). Imaging the deep cerebellar nuclei: A probabilistic atlas and normalization procedure. Neuroimage.
Diedrichsen, J., & Zotow, E. (2015). Surface-based display of volume-averaged cerebellar data. PLoS One, 7, e0133402.
Forbes, N. F., Carrick, L. A., McIntosh, A. M., & Lawrie, S. M. (2009). Working memory in schizophrenia: A meta-analysis. Psychological Medicine, 39(6), 889-905.
Fields, A. (2013). Discovering statistics using IBM SPSS Statistics. London, UK: Sage publications. Friston, K. J., Ashburner, J., Kiebel, S. J., Nichols, T. E., & Penny, W. D. (2007) Statistical
Parametric Mapping: The Analysis of Functional Brain Images. Academic Press.
Gardner, D. M., Murphy, A. L., O'Donnell, H., & Centorrino, F. (2010). International consensus study of antipsychotic dosing. American Journal of Psychiatry, 167(6), 686-693.
Grinchii, D., Dremencov, E. (2020). Mechanism of action of atypical antipsychotic drugs in mood disorders. International Journal of Molecular Sciences, 21(24), 9532.
Hu, W., MacDonald, M. L., Elswick, D. E., & Sweet, R. A. (2015). The glutamate hypothesis of schizophrenia: Evidence from human brain tissue studies. Annals of the New York Academy of Sciences, 1338(1), 38-57.
Huhn, M., Nikolakopoulou, A., Schneider-Thoma, J., Krause, M., Samara, M., Peter, N., Arndt, T., Bäckers, L., Rother, P., Cipriani, A., Davis, J., Salanti, G., & Leucht, S. (2019). Comparative efficacy and tolerability of 32 oral antipsychotics for the acute treatment of adults with multi-episode schizphrenia: A systematic review and network meta-analysis. Lancet, 394, 939-951. Jenkinson, M., Bannister, P., Brady, J. M., & Smith, S. M. (2002). Improved optimisation for the
robust and accurate linear registration and motion correction of brain images. NeuroImage, 17(2), 825-841.
Jenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W., & Smith, S. M. (2012). FSL. NeuroImage, 62(2), 782–790.
Jenkinson, M., Pechaud, M., & Smith, S. (2005). MR-based estimation of brain, skull and scalp surfaces. In Eleventh Annual Meeting of the Organization for Human Brain Mapping. Kantrowitz, J. T. (2020). Targeting serotonin 5-HT2A receptors to better treat schizophrenia:
Rationale and current approaches. CNS Drugs, 34, 947-959.
Functioning in Schizophrenia, ?, 16-49.
Kishimoto, T., Hagi, K., Nitta, M., Kane, J. M., & Correll, C. U. (2019). Long-term effectiveness of oral second-generation antipsychotics in patients with schizophrenia and related disorders: A systematic review and meta-analysis of direct head-to-head comparisons. World Psychiatry, 18(2), 208-224.
Köhler, S., Wagner, G., & Bär, K. (2019). Activation of brainstem and midbrain nuclei during cognitive control in medicated patients with schizophrenia. Human Brain Mapping, 40(1), 202-213.
Krieger, S., Lis, S., Janik, H., Cetin, T., Gallhofer, B., & Meyer-Lindenberg, A. Executive function and cognitive subprocesses in first-episode, drug-naive schizophrenia: An analysis of N-back performance. The American Journal of Psychiatry, 162(6), 1206-1208.
Lake, C. R., Sternberg, D. E., van Kammen, D. P., Ballenger, J. C., Ziegler, M. G., Post, R. M., Kopin, I. J., & Bunney, W. E. (1980). Schizophrenia: Elevated cerebrospinal fluid
norepinephrine. Science, 207, 331-333.
Mäki-Marttunen, V., Andreassen, O. A., & Espeseth, T. (2020). The role of norepinephrine in the pathophysiology of schizophrenia. Neuroscience & Biobehavioral Reviews, 118, 298-314. MATLAB. (2020). version 9.7.0 (R2019b). Natick, Massachusetts: The MathWorks Inc.
McCutcheon, R. A., Krystal, J. H., & Howes, O. D. (2020). Dopamine and glutamate in schizophrenia: Biology, symptoms and treatment. World Psychiatry, 19(1), 15-33. Meltzer, H. Y., Matsubara, S., & Lee, J. C. (1989). Classification of typical and atypical
antipsychotic drugs on the basis of dopamine D-1, D-2, and serotonin2 pKi values. Journal of Pharmacologicaly and Experimental Therapeutics, 251(1), 238-246.
Meltzer, H. Y., & Stahl, S. M. (1976). The dopamine hypothesis of schizophrenia: A review. Schizophrenia Bulletin, 2(1), 19–76.
Meule, A. (2017). Reporting and interpreting working memory performance in n-back tasks. Frontiers in Psychology, 8, 352.
Meyer-Lindenberg, A. S., Olsen, R. K., Kohn, P. D., Brown, T., Egan, M. F., Weinberger, D. R., & berman, K. F. (2005). Regionally specific disturbance of dorsolateral prefrontal-hippocampal functional connectivity in schizophrenia. AMA Archives of General Psychiatry, 62(4),
379-386.
Murty, V. P., Shermohammed, M., Smith, D. V., McKell Carter, R., Huettel, S. A., & Adcock, R. A. Resting state distinguish human ventral tegmental area from substantia nigra. Neuroimage, 100, 580-589.
Nuechterlein, K. H., Barch, D. M., Gold, J. M., Goldber, T. E., Green, M. F., & Heaton, R. K. (2004). Identification of separable cognitive factors in schizophrenia, 72(1), 29-39.
O'Connor, F. L. (1998). The role of serotonin and dopamine in schizophrenia. Journal of American Psychiatric Nurses Association, 4(4), S30-S34.
Quednow, B. B.,Geyer, M. A., & Halberstadt, A. L. (2009). Serotonin and Schizophrenia. In: Müller, C. R., & Jacobs, B. Handbook of the Behavioral Neurobiology of Serotonin. London, 585-620.
Repovš, G., & Barch, D. (2012). Working memory related brain network connectivity in individuals with schizophrenia and their siblings. Frontiers in Human Neuroscience, 6, 1-15.
Seghier, M. L., & Price, C. J. (2018). Interpreting and utilising intersubject variability in brain function. Trends in Cognitive Sciences, 22(6), 517–530.
Smieskova, R., Fusar-Poli, P., Allen, P., Bendfeldt, K., Stieglitz, R. D., Drewe, J., Radue, E. W., McGuire, P. K., Riecher-Rössler, A., & Borgwardt, S. J. (2010). Neuroimaging predictors of transition to psychosis -- A systematic review and meta-analysis. Neuroscientific &
Biobehavioral Reviews, 34(8), 1207-1222.
Švob Štrac, D., Pivac, N., & Mück-Šeler, D. (2016). The serotonergic system and cognitive function. Translational Neuroscience, 7(1), 35-49.
Takano, H. (2018). Cognitive function and monoamine neurotransmission in schizophrenia: Evidence from positron emission tomography studies. Frontiers in Psychiatry, 9, 228. Tona, K.D., Keuken, M.C., de Rover, M., Lakke, E., Forstmann, B.U., Nieuwenhuis, S., & van
Osch, M.J.P. (2017). In vivo visualization of the locus coeruleus in humans: Quantifying the test-retest reliability. Brain Structure and Function, 222, 4203-4217.
Uchida, H., & Mamo, D. (2009). Dosing of antipsychotics in schizophrenia across the life-spectrum. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 33(6), 917-920.
Wang, J., Hu, M., Guo, X., Wu, R., Li, L., & Zhao, J. (2013). Cognitive effects of atypical
antipsychotic drugs in first-episode drug-naïve schizophrenic patients. Neural Regeneration Research, 8(3), 277-286.
Williams, G. V., Rao, S. G., Goldman-Rakic, P. S. (2002). The physiological role of 5-HT2A receptors in working memory. The Journal of Neuroscience, 22(7), 2843-2854.
Woolrich, M. W., Ripley, B. D., Brady, M., & Smith, S. M. (2001). Temporal autocorrelation in univariate linear modeling of fMRI data. NeuroImage, 14(6), 1370–1386.
