Although they make an important contribution to the burden of schizophrenia (e.g. Fervaha, Foussias, Agid, & Remington, Reference Fervaha, Foussias, Agid and Remington2014; Kirkpatrick, Reference Kirkpatrick2014), the negative symptoms of the disorder – lack of volition, poverty of speech and flattening of affect – remain poorly understood. One of two major current approaches is that they are due to impaired prefrontal cortex function. Stemming ultimately from the resemblance between schizophrenic lack of volition and the apathy of neurological patients with the frontal lobe syndrome (Frith, Reference Frith1992; Liddle, Reference Liddle1987; Weinberger, Reference Weinberger1988), this theory is supported principally by neuropsychological studies, which have found significant associations between negative symptom scores and poor performance on a range of executive tests (de Gracia Dominguez, Viechtbauer, Simons, van Os, & Krabbendam, Reference de Gracia Dominguez, Viechtbauer, Simons, van Os and Krabbendam2009; Dibben, Rice, Laws, & McKenna, Reference Dibben, Rice, Laws and McKenna2009), although it should be noted that similar associations have been also found with non-executive tests (de Gracia Dominguez et al., Reference de Gracia Dominguez, Viechtbauer, Simons, van Os and Krabbendam2009).
The other main explanatory proposal for negative symptoms is that they are due to reduced responsiveness to rewarding stimuli in the environment (for a review see Deserno, Boehme, Heinz, & Schlagenhauf, Reference Deserno, Boehme, Heinz and Schlagenhauf2013). This theory grew out of the dopamine hypothesis of schizophrenia, in particular the so-called revised or extended version of this (Davis, Kahn, Ko, & Davidson, Reference Davis, Kahn, Ko and Davidson1991; Howes & Kapur, Reference Howes and Kapur2009), which proposes that negative symptoms reflect a functional deficiency of this transmitter. It became the subject of significantly increased interest after Schultz, Dayan, and Montague (Reference Schultz, Dayan and Montague1997) demonstrated conclusively that phasic release of dopamine from midbrain neurons acts as a reward prediction error (RPE) signal in animals. In humans, functional imaging studies have mapped a series of brain regions that activate in response to RPE (Garrison, Erdeniz, & Done, Reference Garrison, Erdeniz and Done2013), as well as to other aspects of reward processing, such as reward anticipation (i.e. when stimuli are viewed that are predictive of reward) and reward feedback (i.e. when information that reward has been won is provided) (Diekhof, Kaps, Falkai, & Gruber, Reference Diekhof, Kaps, Falkai and Gruber2012; Jauhar et al., Reference Jauhar, Fortea, Solanes, Albajes-Eizagirre, McKenna and Radua2021; Liu, Hairston, Schrier, & Fan, Reference Liu, Hairston, Schrier and Fan2011; Oldham et al., Reference Oldham, Murawski, Fornito, Youssef, Yucel and Lorenzetti2018). Prominent among these regions are the striatum, especially its ventral sector, and the orbitofrontal and adjacent ventromedial frontal cortex, as well as parts of the anterior cingulate cortex and the midbrain.
In schizophrenia, a substantial number of fMRI studies have found that patients show reduced ventral striatal activation during reward anticipation (for meta-analyses see Radua et al., Reference Radua, Schmidt, Borgwardt, Heinz, Schlagenhauf, McGuire and Fusar-Poli2015; Zeng et al., Reference Zeng, Yan, Cao, Su, Song, Luo and Yang2022). Whether this is also the case outside the ventral striatum is currently the subject of conflicting findings (for meta-analyses see Chase et al., Reference Chase, Loriemi, Wensing, Eickhoff and Nickl-Jockschat2018, Leroy et al., Reference Leroy, Amad, D'Hondt, Pins, Jaafari, Thomas and Jardri2020; Zeng et al., Reference Zeng, Yan, Cao, Su, Song, Luo and Yang2022). At the stage of receiving feedback about reward, findings point to a mixed pattern, with increased activation in schizophreniain the striatum, the amygdala and the hippocampus, plus various cortical regions, coupled with reduced activation in the medial frontal cortex and the dorsolateral frontal cortex (Zeng et al., Reference Zeng, Yan, Cao, Su, Song, Luo and Yang2022). The finding of increased activation at this stage is of interest because the RPE signal is generated when feedback about reward gain or loss is given. RPE itself, however, has not been found to be altered in schizophrenia according to a recent meta-analysis (Yaple, Tolomeo, & Yu, Reference Yaple, Tolomeo and Yu2021), although at 10 the number of studies is so far relatively small.
Although RPE is the aspect of reward processing most closely associated with dopamine, and hence with the ‘reduced responsiveness to reward’ theory of negative symptoms, only a few studies to date have empirically examined the relationship between RPE and these symptoms. Katthagen, Kaminski, Heinz, Buchert, and Schlagenhauf (Reference Katthagen, Kaminski, Heinz, Buchert and Schlagenhauf2020) found that RPE-associated activation within a right ventral striatal region of interest (ROI) was inversely correlated with negative symptoms in 19 patients with schizophrenia. Culbreth, Westbrook, Xu, Barch, and Waltz (Reference Culbreth, Westbrook, Xu, Barch and Waltz2016) found a significant inverse correlation between negative symptoms and RPE-associated activations in a ventral striatal ROI in one of two patient samples (N = 28) but not in the other (N = 30). Finally, in 38 patients, Dowd, Frank, Collins, Gold, and Barch (Reference Dowd, Frank, Collins, Gold and Barch2016) found that RPE-related activations were not associated with either clinician- or self-rated anhedonia in an ROI in the striatum.
The aim of the present study was to further investigate the reduced responsiveness to reward theory of negative symptoms. We focused on RPE, on the basis that this is the aspect of reward processing that is most clearly linked to dopamine function. Because correlational analysis has been argued to have limited power to detect brain:behaviour correlations in fMRI studies (Yarkoni, Reference Yarkoni2009), we used a design comparing groups of schizophrenia patients with and without negative symptoms. A matched healthy control group was also examined.
Methods
Participants
The patient sample consisted of two groups of right-handed patients meeting criteria for either high or absent negative symptoms (HNS and ANS, see below). The patients were selected from a larger group of patients recruited from four hospitals in the Barcelona area (Benito Menni CASM, Hospital de Sant Rafael, Hospital Sagrat Cor de Martorell, Hospital Mare de Déu de la Mercè). All patients had a DSM-5 diagnosis of schizophrenia or schizoaffective disorder, made on the basis of clinical interview and review of casenotes.
Patients were excluded if they (a) were younger than 18 or older than 65 years, (b) had a history of brain trauma or neurological disease, or (c) had shown alcohol/substance abuse within 12 months prior to participation. They were also required to have a premorbid IQ in the normal range (⩾70), as estimated using the Word Accentuation Test (Test de Acentuación de Palabras, TAP) (Del Ser, Gonzalez-Montalvo, Martinez-Espinosa, Delgado-Villapalos, & Bermejo, Reference Del Ser, Gonzalez-Montalvo, Martinez-Espinosa, Delgado-Villapalos and Bermejo1997; Gomar et al., Reference Gomar, Ortiz-Gil, McKenna, Salvador, Sans-Sansa, Sarro and Pomarol-Clotet2011), which requires pronunciation of low-frequency Spanish words whose accents have been removed. Patients with a current IQ < 70 based on four subtests from the WAIS-III (Vocabulary, Similarities, Matrix Reasoning and Block Design) were also excluded. All patients were taking antipsychotic medication.
Healthy controls (HC) were recruited from non-medical hospital staff, their relatives and acquaintances, plus independent sources in the community. The HC met the same exclusion criteria as the patients. They were also excluded if they reported a history of mental illness and/or treatment with psychotropic medication, or a history of major mental illness in a first-degree relative. The Structured Clinical Interview for DSM-IV (First, Spitzer, Gibbon, & Williams, Reference First, Spitzer, Gibbon and Williams2002) (modified to make it compatible with DSM-5) was administered to exclude current or past psychiatric disorders.
All participants gave written informed consent prior to participation. The study was approved by the ethics committee for the relevant hospitals (Comité d’Ètica i Investigació Clínica de Germanes Hospitalàries). Healthy controls received a gift card as a compensation for their participation.
Clinical assessment
Negative symptoms were rated using the Positive and Negative Syndrome Scale (PANSS). On the basis of a review of factor analytic studies by Wallwork, Fortgang, Hashimoto, Weinberger, and Dickinson (Reference Wallwork, Fortgang, Hashimoto, Weinberger and Dickinson2012) the following items were considered negative symptoms: blunted affect (N1), emotional withdrawal (N2), poor rapport (N3), apathetic social withdrawal (N4), lack of spontaneity (N6), and motor retardation (G7). All these items are scored in a scale ranging from 1 (absent) to 7 (extreme), indicating the severity of the symptom. Negative symptoms were additionally assessed with the Clinical Assessment Interview for Negative Symptoms (CAINS) (Kring, Gur, Blanchard, Horan, & Reise, Reference Kring, Gur, Blanchard, Horan and Reise2013; Valiente-Gomez et al., Reference Valiente-Gomez, Mezquida, Romaguera, Vilardebo, Andres, Granados and Bernardo2015). Scores on individual CAINS items are summed to give an overall score, and also yield two subscale scores: motivation and pleasure (CAINS-MAP, 9 items) focuses on lack of motivation and anhedonia, whereas expressivity (CAINS-EXP, 4 items) rates lack of facial expression and expressive gestures, and prosody and amount of speech.
Patients were assigned to high negative symptoms (HNS) and absent negative symptoms (ANS) groups based on PANSS negative symptom scores, according to criteria devised by Bucci and Galderisi (Reference Bucci and Galderisi2017). Patients in the HNS group had scores of at least moderate (i.e. score of 4 out of a maximum of 7) on three or more PANSS negative symptom items, or moderately severe (i.e. 5 or higher) on two or more items. In contrast, patients in the ANS group had scores of either absent or minimal (i.e. a maximum of 2) on all PANSS negative symptom items.
Task description
During fMRI scanning, subjects performed a probabilistic monetary learning task adapted from Pessiglione, Seymour, Flandin, Dolan, and Frith (Reference Pessiglione, Seymour, Flandin, Dolan and Frith2006). In each trial, two stimuli were shown (one at each side of the screen) for 2.5s. The participant had to select one of them by pressing a button on the left or the right. The selected stimulus was then highlighted and feedback was displayed in the form of either a 10-eurocent coin (gain) or a yellow circle (no gain) (see Fig. 1). After a further 1s a fixation cross appeared with variable duration (from 1 to 6.5s, mean 2s) until a new trial began.
Fig. 1. The probabilistic reinforcement learning task. The participant was required to choose through button press one of the two abstract stimuli shown on the screen for 2.5s. The selected stimulus was highlighted and feedback indicating gain (10-cent coin) or no gain (yellow circle) appeared (1s). The participant had to learn by trial and error which of the two stimuli was more likely to yield a reward in order to win as much money as possible. In half of the trials, one of the stimuli had a reward probability of 80% and the other of 20% (rewarded condition). In the other half, both stimuli had the same probability (bivalent condition).
Two types of stimuli were presented. In rewarded pairs, one of the two stimuli led to reward in 80% of trials, whereas the other stimulus was rewarded on 20% of trials. In bivalent pairs, both stimuli were rewarded on 50% of trials. The same pair of stimuli were presented for 16 successive trials, during which (in the case of the rewarded pairs) the participant had the opportunity to learn, through trial and error, which of the two stimuli was more likely to lead to reward, and earn as much money as possible. After 16 trials, the pair of stimuli changed and the reward contingencies had to be learned from scratch. The task included 10 different pairs of stimuli (i.e. 160 trials), half of them corresponding to rewarded pairs (1st, 3rd, 4th, 5th and 7th) and the other half to bivalent pairs. Subjects were paid the money they won during the experiment. Participants underwent a training session with 48 trials before the scanning session.
We excluded from the analysis participants who chose the optimal stimulus less than 50% of the time during the last eight trials of each rewarded pair (i.e. they had not learnt to choose the rewarded stimulus after 16 trials). Subjects who showed no response or a premature response (reaction time < 100 ms) in more than 10% of the trials were also excluded.
Computational model
RPE, defined as the difference between the expected and the actual outcome, was estimated using the Q-learning model (Watkins & Dayan, Reference Watkins and Dayan1992). For each pair of stimuli, the model estimates the expected value of choosing stimulus A (EVA) and stimulus B (EVB) based on the sequence of choices of subjects. Expected values were set to zero at the beginning of each 16-trial block, and the expected value of the chosen stimulus was updated after each trial as a function of RPE (for further details see online Supplementary Information).
To estimate RPE, the model generates two parameters for each subject, α and β. α is the rate of learning for the chosen stimulus; it ranges between zero and one and weights the influence of RPE on the updating of expected value. A high α value indicates a strong influence of recent outcomes, whereas a low value indicates slow learning. β is the (inverse) temperature, a parameter that adjusts the trade-off between exploration and exploitation, i.e. how consistently the model chooses the higher-value stimulus, especially when the values of both stimuli are similar. The higher the β value, the more exploratory is the way of choosing; the lower the β value, the more deterministic (for further details see online Supplementary Information). The free parameters of the model, α and β, were estimated for each subject through the maximum likelihood technique to maximize the probability of the actual choices. In the imaging analysis, the learning model was calculated using the same parameters across subjects, through the median of the individual fitted parameters (Daw, Reference Daw, Delgado, Phelps and Robbins2011).
We carried out two additional analyses, (a) employing RPEs extracted from individually-fitted models instead of using group parameters; and (b) examining brain activation patterns separately for positive and negative RPEs. The latter was with the aim of determining whether negative symptoms were associated with abnormal responses to RPEs of one or both valences (for details of both analyses see online Supplementary Information).
Image acquisition and pre-processing
Images were acquired with a 3T Philips Ingenia scanner (Philips Medical Systems, Best, The Netherlands). Functional data were acquired using a T2*-weighted echo-planar imaging (EPI) sequence with 443 volumes and the following acquisition parameters: TR = 2000 ms, TE = 30 ms, flip angle = 70°, in-plane resolution = 3.5 × 3.5 mm, FOV = 238 × 245 mm, slice thickness = 3.5 mm, inter-slice gap = 0.75 mm. Slices (32 per volume) were acquired with an interleaved order parallel to the AC-PC plane. In addition, a high-resolution anatomical volume was acquired using an FFE (Fast Field Echo) sequence for anatomical reference and inspection (TR = 9.90 ms; TE = 4.60 ms; Flip angle = 8°; voxel size = 1 × 1 mm; slice thickness = 1 mm; slice number = 180; FOV = 240 mm).
Pre-processing was carried out with the FEAT module included in the FSL (FMRIB Software Library) software (Smith et al., Reference Smith, Jenkinson, Woolrich, Beckmann, Behrens, Johansen-Berg and Matthews2004). The first 6 s (3 volumes) of the sequence, corresponding to signal stabilization, were discarded. Pre-processing included motion correction (using the MCFLIRT algorithm), co-registration and normalization to a common stereotactic space (MNI, Montreal Neurological Institute template). For accurate registration, a two-step process was used. First, brain extraction was applied to the structural image, and the functional sequence was registered to it. Then the structural image was registered to the standard template. These two transformations were used to register the functional sequence to the standard space. Before group analyses, normalized images were spatially filtered with a Gaussian filter (FWHM = 5 mm). To minimize unwanted movement-related effects, individuals with an estimated maximum absolute movement >3.0 mm or an average absolute movement >0.3 mm were excluded from the study.
Statistical analysis
Statistical analysis was performed by means of a General Linear Model (GLM) approach using the FEAT module in FSL. At the first level (within-subject), the following regressors were created: onset of stimuli as 2.5s events in bivalent and reward pairs separately (2 regressors) and onset of feedback as 1s events in bivalent and reward pairs separately (2 regressors). Finally, RPE derived from the computational model was introduced as a parametric modulator at feedback onset as a 1s event across all conditions (one regressor that contained all feedback events across the task, parametrically modulated by the RPE value at each trial). RPE values were signed, meaning that the parametric regressor included both positive and negative RPE values. All regressors were convolved with a gamma response function. We added temporal derivates and motion parameters to the model. Our contrast of interest was the activation associated with the RPE regressor.
GLMs were fitted to generate whole brain individual activation maps for the contrast of interest and second level (group) analyses were performed by means of mixed-effects GLMs (Beckmann, Jenkinson, & Smith, Reference Beckmann, Jenkinson and Smith2003), to obtain mean activation maps for each group. Then, we performed two-sample t tests to compare brain activation between the different groups. In all these analyses, whole-brain voxel-wise statistical tests were carried out with a p < 0.05, with cluster correction for multiple comparisons, with a cluster-forming threshold of z > 3.1 (p < 0.001), using Gaussian Random Field correction (Worsley, Marrett, Neelin, & Evans, Reference Worsley, Marrett, Neelin, Evans, Myers, Cunningham, Bailey and Jones1996). In all analyses, age, sex and premorbid IQ were included as covariates in order to minimize the effects of non-significant differences (and in one pairwise comparison a significant difference) among the groups on these variables.
Given the established role of the ventral striatum, orbitofrontal cortex and midbrain in reward processing, we carried out a complementary ROI analysis. Separate right and left ROIs were anatomically defined using the Harvard-Oxford Cortical and Subcortical Atlases (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases) for the ventral striatum and orbitofrontal cortex. The midbrain ROI contained the bilateral ventral tegmental area and substantia nigra as defined by Murty et al. (Reference Murty, Shermohammed, Smith, Carter, Huettel and Adcock2014).
Results
From an initial sample of 36 ANS patients, 3 did not complete the task, 3 were excluded due to excessive head motion during the fMRI session, and 3 more due to poor behavioural performance, leaving a final sample of 27 patients. From the initially included 36 HNS patients, 1 did not complete the task and 8 more were excluded, for excessive head motion (3), current IQ < 70 (1), an incidental MRI finding (1), loss of behavioural data due to technical problems (1) and poor behavioural performance (2), leaving 27 patients. Thirty HC were included who were recruited to be similar to the two patient groups for age, sex and TAP-estimated premorbid IQ.
As shown in Table 1, the three groups did not significantly differ in age and sex or TAP-estimated premorbid IQ, although premorbid IQ was significantly lower in the HNS group compared to the HC (p = 0.03). As expected, current IQ was significantly lower in both patient groups than in the HC (ANS v. HC: p = 0.003; HNS v. HC: p < 0.001), but the two patient groups did not differ significantly on this variable (p = 0.56). The patient groups did not significantly differ in the severity of positive (reality distortion) symptoms, illness duration and antipsychotic dose. Both groups were also similar in the type of antipsychotic used: two patients were taking first-generation antipsychotics (one in the ANS and one in the HNS group), 46 were taking second-generation antipsychotics (23 in the ANS and 23 in the HNS group), and 5 were taking both (2 in the ANS group and 3 in the HNS group). Data from pharmacological treatment was missing for one patient in the ANS group. The HNS patients showed higher levels of PANSS disorganization symptoms than the ANS patients, and lower levels of functioning, as measured by the Global Assessment of Functioning scale (GAF).
Table 1. Socio-demographic and clinical data in the three groups
HC, Healthy controls; ANS, Absent negative symptoms; HNS, High negative symptoms; M/F, Male/Female; PANSS, Positive and Negative Symptoms Scale; CAINS, Clinical Assessment Interview for Negative Symptoms; CAINS-MAP, CAINS motivation and pleasure subscale; CAINS-EXP, CAINS expressivity subscale; GAF, Global Assessment of Functioning; CPZ, chlorpromazine.
Values are means (s.d.). For clinical data, non-parametric tests (Wilcoxon rank sum test) were used, given the non-normal distribution of the data.
a The ANS and HNS patients were a priori selected to differ on PANSS negative symptom scores.
Behavioural results
The patients showed a lower proportion of correct choices in rewarded pairs (defined as the proportion in which the stimulus with the highest reward probability was chosen, irrespective of the actual outcome) than the HC (HC: 0.80 ± 0.09; ANS: 0.75 ± 0.12; HNS 0.73 ± 0.12, F = 3.04, p = 0.053), which was significant in the comparison between HC and HNS (p = 0.02). For bivalent pairs, there were no group differences in the proportion of choices (HC: 0.46 ± 0.11; ANS: 0.50 ± 0.10; HNS: 0.49 ± 0.08, F = 0.83, p = 0.44). The three groups showed evidence of learning in the rewarded condition. In the bivalent condition, choice proportions remained around 50% throughout the whole trial sequence for the three groups. The findings are shown in Fig. 2.
Fig. 2. Evolution of responding throughout the task. Upper row plots show the proportion of trials where each group selected the more rewarded stimulus in the rewarded condition (blue line) or one of the two stimuli in the bivalent condition (orange line), in each of the 16 trials that formed each trial block. Lower row plots show the choices predicted by the reinforcement learning model for each group (i.e. the probability of choosing the rewarded stimulus in the rewarded condition, or one of the two stimuli in the bivalent condition). Coloured regions around the lines indicate 95% confidence intervals. The three groups showed evidence of learning in the rewarded condition, since the proportion of choices for the rewarded stimulus progressively increased throughout the trial sequence, although patients (especially in the HNS group) showed a slower increase and did not reach the same proportion of choices than controls. In the bivalent condition, choice proportion remained around 50% throughout the whole trial sequence, as predicted.
The three groups did not differ in the estimates of the Q-learning model parameters α (rate of learning) (HC: 0.44 ± 0.33; ANS: 0.43 ± 0.41; HNS: 0.42 ± 0.37, χ2 = 0.78, p = 0.68) and β (temperature) (HC 0.22 ± 0.21; ANS 0.21 ± 0.24; HNS 0.28 ± 0.34, χ2 = 1.06, p = 0.59). Model fit, measured as the root mean square error (RMSE) of the regression between actual and predicted choices, was similar for HC and ANS patients but was slightly reduced in the HNS patients (p < 0.05, see online Supplementary Information for details).
RPE-associated activations
In the HC, RPE value was positively correlated with activation in a wide range of brain regions. These included the ventral striatum and putamen, and portions of the prefrontal cortex, including the bilateral orbitofrontal and ventrolateral prefrontal cortex, the DLPFC (more pronounced on the left) extending into the precentral and medial superior frontal cortex. RPE was also correlated with activation in the posterior cingulate cortex, the bilateral inferior parietal cortex (more pronounced on the left), the bilateral occipital cortex extending into the fusiform gyri, the bilateral amygdala and hippocampus, and the cerebellum (see Fig. 3 and online Supplementary Table S1).
Fig. 3. Mean activation maps for positive correlation with RPE in the healthy controls (HC) and the ANS and HNS patients. Coronal slices show ventral striatal activation, present in all three groups. The right side of the image corresponds to the right side of the brain. Colour bar depicts z values.
The ANS patients showed a broadly similar pattern of RPE-associated activations, although this appeared less marked in the dorsolateral, inferior frontal, and inferior parietal cortices, particularly on the right. In the HNS patients, regions positively correlated with RPE were limited to the ventral striatum, the bilateral inferior parietal cortex, the posterior cingulate cortex, and the visual cortex extending into the cerebellum.
Group comparisons are shown in Fig. 4a. There were no clusters of significant difference between the HC and the ANS patients. Compared to the HC, the HNS patients showed clusters of significantly reduced RPE-associated activation in the superior medial frontal cortex, the left ventrolateral and dorsolateral prefrontal cortex, the left middle temporal cortex and the bilateral occipital cortex. Compared to the ANS patients, the HNS patients showed reduced activation in the left inferior ventrolateral prefrontal cortex and the DLPFC, and in the left lingual and fusiform gyrus.
Fig. 4. (a) Maps of group comparisons for RPE-associated activations. Top row shows areas of reduced association with RPE for HNS patients relative to heathy controls (HC). Bottom row shows areas of reduced association with RPE for HNS patients relative to ANS patients. The right side of the image corresponds to the right side of the brain. Colour bar depicts z values. (b) Scatter plots of RPE-associated activations in ventral striatal, orbitofrontal and midbrain ROIs. Plots show mean parameter estimates (beta values) for the correlation between RPE values and activation within the ROI. The ventral striatum was defined as the nucleus accumbens. The midbrain comprised the ventral tegmental area and substantia nigra bilaterally. OFC, orbitofrontal cortex. * p < 0.05.
Since disorganization scores differed significantly between the HNS and ANS patients, we repeated the whole-brain analysis comparing the two patient groups after covarying for this variable. This caused an additional small cluster of differences to appear in the right inferior parietal cortex (MNI coordinates x = 64, y = −38, z = 22; Z value = 4.66; cluster size = 132 voxels; p = 0.01).
Examination of RPE-associated activation in the predetermined anatomical ROIs revealed no group differences in the left or right ventral striatum (left: F = 0.64, p = 0.53; right: F = 0.46, p = 0.63) (see Fig. 4b). In the left orbitofrontal ROI, there were significant group differences (F = 3.28, p = 0.04) that reflected a reduced RPE-associated activation in the HNS group relative to the ANS patients (p = 0.02) and the HC (p = 0.03). The right orbitofrontal ROI showed a similar pattern, but only reaching a trend level (F = 2.96, p = 0.06). This reflected reduced RPE-related activation in HNS relative to ANS (p = 0.02) and HC (p = 0.08), with no differences between HC and ANS. The midbrain ROI showed no significant group differences (F = 2.00, p = 0.14).
Whole brain findings using other analytic models
Findings employing RPEs extracted from individually-fitted models instead of using group parameters, and examining positive and negative RPEs separately are reported in detail in the online Supplementary Information. Briefly, repeating the whole brain analysis with RPEs extracted from individually-fitted models instead of using group parameters yielded closely similar results to the original analysis. The findings for positive RPEs were also broadly similar to those in the original analysis: the healthy subjects showed clusters of activation in the bilateral ventral striatum and left putamen, the left ventrolateral and dorsolateral prefrontal cortex, the posterior cingulate cortex, the bilateral inferior parietal cortex, the occipital cortex and cerebellum. The pattern was closely similar in the ANS patients, whereas in the HNS patients, activation was seen in the bilateral ventral striatum, the left inferior parietal cortex, the bilateral occipital cortex, and the cerebellum only. Group comparisons showed reduced activation in the HNS patients compared to the ANS patients in the left DLPFC (MNI coordinates: x = −42, y = 4, z = 32; Z = 3.94; cluster size = 130 voxels; p = 0.01) and left occipital cortex/cerebellum (MNI coordinates: x = −38, y = −72, z = −18; Z = 4.53; cluster size = 153 voxels; p = 0.006), as shown in online Supplementary Fig. S4. There were no significant group differences in any comparisons for negative RPE associated activations. (For full details and results concerning negative RPEs, see online Supplementary information).
Discussion
The main finding of this study was that both healthy subjects and schizophrenia patients without negative symptoms showed a broadly similar pattern of activations associated with RPE signalling (albeit slightly less extensive in the latter), whereas activation was reduced in patients with high levels of negative symptoms. Importantly, the areas showing reductions in high negative symptom patients were limited to the dorsolateral and ventrolateral sectors of the prefrontal cortex and did not include the ventral striatum. A further reward-processing- associated region, the orbitofrontal cortex, showed equivocal evidence of reduced activation.
Our findings in healthy subjects show some similarities to those of Garrison, Erdeniz, and Done (Reference Garrison, Erdeniz and Done2013) in their meta-analysis of studies measuring RPE across a wide variety of functional imaging paradigms. Like us, they found activations affecting large areas of the basal ganglia and other subcortical nuclei such as the thalamus, amygdala and hypothalamus. On the other hand, they found only scattered cortical activations, which took the form principally of a series of discrete clusters in the anterior and middle zones of the medial frontal cortex. More recently, Corlett, Mollick, and Kober (Reference Corlett, Mollick and Kober2022) meta-analysed studies of RPE in response to secondary rewards (i.e. money) and found more areas in common with those that we found, including the ventrolateral (and to a small extent the dorsolateral) prefrontal cortex, a large expanse of the medial frontal cortex and the middle/posterior cingulate cortex. Neither meta-analysis found much in the way of occipital cortex activation, but one possible reason for the extensive activations we found in our study might be that our task employed visual stimuli that changed at regular intervals.
Our study failed to find RPE-associated activation differences between patients with high and absent negative symptoms (or between either group or the healthy controls) in the ventral striatum, one of the key target areas for the RPE signal. In this respect, it is noteworthy that three other studies which examined RPE in relation to negative symptoms have not had strong findings in this brain region. One (Katthagen et al., Reference Katthagen, Kaminski, Heinz, Buchert and Schlagenhauf2020) found that right ventral striatal activation was inversely correlated with negative symptoms; however, the authors appear to have only detected this using a small-volume-corrected voxel-based analysis carried out within a ventral striatal ROI. Another study (Culbreth et al., Reference Culbreth, Westbrook, Xu, Barch and Waltz2016) found no significant correlation between RPE-related activation and negative symptoms in ventral striatal ROIs in one sample of patients, but a significant inverse correlation on the left in another. The third study (Dowd et al., Reference Dowd, Frank, Collins, Gold and Barch2016) found no significant associations between RPE-related activation in striatal ROIs and clinician- or self-rated anhedonia/amotivation.
Our findings for another area importantly involved in reward processing, the orbitofrontal cortex, were equivocal. In the whole brain analysis, there were no differences between patients with high and absent negative symptoms in this region (or between the high negative symptom patients and healthy controls). However, examination of an ROI in the orbitofrontal cortex revealed significantly lower activation in the HNS group on the left and a difference in the same direction on the right. Clearly, a positive finding in an ROI analysis against a background of negative findings in a whole brain analysis needs to be interpreted with caution.
Where we did find evidence of altered RPE-related activation in high negative symptom patients was in the prefrontal cortex, specifically its ventrolateral and dorsolateral regions. Such a finding is of some theoretical interest, as it points to disrupted processing of RPE information in a brain region implicated by the other, ‘impaired frontal function’ theory of negative symptoms. The meaning of such a finding, if replicated in other studies, is uncertain, but one possible interpretation is that altered RPE-related activation in the prefrontal cortex is secondary to a broader prefrontal dysfunctionality underlying negative symptoms, one that would show itself on other tasks as well. To date, functional imaging studies have not consistently found associations between negative symptoms and reduced prefrontal activation during executive (and non-executive) task performance [for a review see Goghari, Sponheim, & MacDonald (Reference Goghari, Sponheim and MacDonald2010)]. Recently, however, our group found support for reduced activation in parts of the lateral prefrontal cortex in patients with negative symptoms compared to those without, in a study using a novel paradigm putatively more closely linked to negative symptoms than standard executive tasks (Fuentes-Claramonte et al., Reference Fuentes-Claramonte, Ramiro, Torres, Argila-Plaza, Salgado-Pineda, Soler-Vidal and Pomarol-Clotet2022).
It is important to point out that our results do not necessarily exclude a role for other aspects of reward processing than RPE in negative symptoms. Probably the main candidate here is reward anticipation, i.e. the motivational properties shown by stimuli that are predictive of reward, a role for dopamine in which has been for argued for persuasively by Berridge (Reference Berridge2007). Radua et al. (Reference Radua, Schmidt, Borgwardt, Heinz, Schlagenhauf, McGuire and Fusar-Poli2015), in their meta-analysis of reward anticipation in the ventral striatum in schizophrenia, found that reduced activation was associated with negative symptoms, although the number of studies was small (N = 6) and the association was only seen on the left. Subsequent studies have had inconsistent findings: apathy/amotivation was found to be inversely correlated with ventral striatal activation during reward anticipation in three studies (Kirschner et al., Reference Kirschner, Hager, Bischof, Hartmann, Kluge, Seifritz and Kaiser2016; Kluge et al., Reference Kluge, Kirschner, Hager, Bischof, Habermeyer, Seifritz and Kaiser2018; Simon et al., Reference Simon, Cordeiro, Weber, Friederich, Wolf, Weisbrod and Kaiser2015), but this was only equivocally the case in a fourth (Wolf et al., Reference Wolf, Satterthwaite, Kantrowitz, Katchmar, Vandekar, Elliott and Ruparel2014). Another study (Moran, Culbreth, Kandala, & Barch, Reference Moran, Culbreth, Kandala and Barch2019) found a significant inverse correlation between the ‘motivation and pleasure’ subscale of the CAINS and activation in the anterior cingulate cortex and insula.
In conclusion, our study does not suggest that a general dysfunction in RPE signalling underlies negative symptoms, but does find evidence for one that is restricted to the lateral prefrontal cortex. Some limitations of our study need to be acknowledged. At 54, the numbers of patients included was larger than some other studies of RPE in relation to negative symptoms, but the high and absent negative symptom groups were still individually relatively small. All the patients in the study were taking antipsychotic medication: while this may have influenced the findings in patients compared to healthy controls, it seems unlikely that this factor would have played a role in the differences found between patients with and without negative symptoms. Finally, the results in the online Supplementary analysis of positive and negative RPEs found evidence of an abnormality in relation to positive RPEs, but were weaker than in the main analysis.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723000521
Acknowledgements
This work was supported by CIBER (Consorcio Centro de Investigación Biomédica en Red), Instituto de Salud Carlos III, Ministerio de Ciencia y Innovación and the Catalonian Government (2014-SGR-1573 and 2017-SGR-1271 to FIDMAG), and by the Instituto de Salud Carlos III, co-funded by European Union (ERDF/ESF, ‘Investing in your future’): Sara Borrell research contracts (CD19/00149 to PFC and CD22/00106 to MAG-L), Miguel Servet research contracts (CPII13/00018 to RS and MS10/00596 to EP-C), and Research Project Grants (PI14/01151 to RS, PI14/01148, PI21/00416 to EP-C and PI14/01691 to PM, and PI21/00823 to SS). MAG-L, JS-V and SS were supported by Júlia Gil Pineda Research Fellowships.
Open access
