SCHRES-06469; No of Pages 12

Schizophrenia Research xxx (2015) xxxxxx

Contents lists available at ScienceDirect

Schizophrenia Research
journal homepage: www.elsevier.com/locate/schres

Testosterone and reward prediction-errors in healthy men and men

with schizophrenia
R.W. Morris a,b,c,1, T.D. Purves-Tyson a,b,d, C. Shannon Weickert a,b,c, D. Rothmond a,b,c,
R. Lenroot a,b,c, T.W. Weickert a,b,c,
a

Neuroscience Research Australia, Barker St, Randwick, New South Wales 2031, Australia
Schizophrenia Research Institute, Liverpool St, Darlinghurst, New South Wales 2010, Australia
School of Psychiatry, University of New South Wales, Hospital Rd, New South Wales 2031, Australia
d
School of Medical Sciences, University of New South Wales, New South Wales 2031, Australia
b
c

a r t i c l e

i n f o

Article history:
Received 4 January 2015
Received in revised form 9 June 2015
Accepted 28 June 2015
Available online xxxx
Keywords:
Sex hormones
Reward
Ventral striatum
Associative learning
Tyrosine hydroxylase
Midbrain
Androgen receptor
Postmortem
Testosterone
Schizophrenia

a b s t r a c t
Sex hormones impact reward processing, which is dysfunctional in schizophrenia; however, the degree to which
testosterone levels relate to reward-related brain activity in healthy men and the extent to which this relationship may be altered in men with schizophrenia has not been determined. We used functional magnetic resonance
imaging (fMRI) to measure neural responses in the striatum during reward prediction-errors and hormone assays to measure testosterone and prolactin in serum. To determine if testosterone can have a direct effect on dopamine neurons, we also localized and measured androgen receptors in human midbrain with
immunohistochemistry and quantitative PCR. We found correlations between testosterone and predictionerror related activity in the ventral striatum of healthy men, but not in men with schizophrenia, such that testosterone increased the size of positive and negative prediction-error related activity in a valence-specic manner.
We also identied midbrain dopamine neurons that were androgen receptor immunoreactive, and found that
androgen receptor (AR) mRNA was positively correlated with tyrosine hydroxylase (TH) mRNA in human
male substantia nigra. The results suggest that sex steroid receptors can potentially inuence midbrain dopamine
biosynthesis, and higher levels of serum testosterone are linked to better discrimination of motivationallyrelevant signals in the ventral striatum, putatively by modulation of the dopamine biosynthesis pathway via
AR ligand binding. However, the normal relationship between serum testosterone and ventral striatum activity
during reward learning appears to be disrupted in schizophrenia.
Crown Copyright 2015 Published by Elsevier B.V. All rights reserved.

1. Introduction
Testosterone impacts male motivation, competitive drive and social
behavior (Spear, 2000; Archer, 2006; Coates and Herbert, 2008;
Richards et al., 2009; Morris et al., 2010). In schizophrenia, testosterone
levels inversely correlate with negative symptoms, such as lack of motivation, at affect and social withdrawal (Akhondzadeh et al., 2006; Ko
et al., 2007). The negative symptoms of schizophrenia relate to abnormal activity in a fronto-striatal circuit during reward processing
(Juckel et al., 2006; Schlagenhauf et al., 2008; Morris et al., 2012,
2014). However, the relationship between testosterone and neural activity during reward processing in schizophrenia is unknown.
Valence-specic changes in midbrain dopamine neuron ring are
known to underpin successful reward learning from a neurobiological

Corresponding author at: Neuroscience Research Australia, Barker Street, Randwick,

New South Wales 2031, Australia.
E-mail address: t.weickert@unsw.edu.au (T.W. Weickert).
1
Present address: School of Psychology, University of New South Wales, New South
Wales 2052, Australia.

perspective (Schultz, 2013). Midbrain dopamine neurons projecting to

the nucleus accumbens code for reward prediction-errors by increasing
ring after unexpected rewards (URs) and by decreasing ring after unexpected reward omissions (UOs) (Schultz et al., 1997). Consistent with
this, blood oxygenation level dependent (BOLD) responses in the
healthy human ventral striatum reliably show increased activity to
URs and decreased activity to UOs (D'Ardenne et al., 2008; Morris
et al., 2012). This, and other evidence (Pessiglione et al., 2006;
Knutson and Gibbs, 2007) suggests BOLD-related prediction-error signals in the ventral striatum reect regionally-specic and temporallyspecic input from midbrain dopamine neurons during rewardprocessing.
Testosterone may modulate the mesolimbic dopamine activity via
direct action on midbrain dopamine neurons. Testosterone can bind to
androgen receptors (AR) and after conversion by aromatase to estradiol,
to estrogen receptors (ER and ER). In rodents, most studies demonstrate that dopamine neurons express AR and both estrogen receptors
(Shughrue et al., 1997; Perez et al., 2003; de Souza Silva et al., 2009;
Purves-Tyson et al., 2012). Evidence from conditioned place preference
studies in rodents (de Beun et al., 1992; Alexander et al., 1994) and

http://dx.doi.org/10.1016/j.schres.2015.06.030
0920-9964/Crown Copyright 2015 Published by Elsevier B.V. All rights reserved.

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

anabolic steroid addiction in human males (Kanayama et al., 2009,

2010) suggest testosterone can act as a positive reinforcer. Furthermore,
oral administration of testosterone to women increases BOLD activation
in the ventral striatum during reward anticipation and has the greatest
effect in women with low intrinsic appetitive motivation (Hermans
et al., 2010). Thus, the effects of administering testosterone on the
mesolimbic path are most marked in people with low motivation; however, the relationship between endogenous testosterone and ventral
striatum BOLD activity in men, with or without low motivation, is
unknown.
Our aim was to determine the extent to which endogenous circulating testosterone levels were associated with ventral striatal BOLD activity during prediction error among healthy men and men with
schizophrenia, a group typically associated with anhedonia and motivation decits. We also identied a potential mechanism by which testosterone could be working by determining whether human midbrain
dopamine neurons have the capacity to directly respond to circulating
testosterone via sex steroid receptors. Our hypotheses were: 1) that testosterone will be positively related to ventral striatal BOLD activity in
healthy men during reward prediction-errors; 2) human nigral dopamine neurones will express AR and the level of expression will be related to dopamine synthesis capacity (as measured by tyrosine
hydroxylase mRNA); and 3) low levels of testosterone will co-occur
with abnormal BOLD ventral striatal activity in men with schizophrenia
during reward prediction-errors, and low BOLD activity will be related
to negative symptom severity.
2. Materials and methods

using the Positive and Negative Syndrome Scale (PANSS) (Kay et al.,
1987). All procedures involving humans or human tissue were approved by the University of New South Wales Human Research Ethics
Committees (HREC 07121, 09187, 07261 and 12435) and the South
Eastern Sydney and Illawarra Area Health Service HREC (07259,
09187), and written informed consent was obtained from all fMRI participants. See Table 1 for details on the fMRI participant demographics.
Six of the men with schizophrenia and eight of the healthy men were included in a previous report on the neural substrate of prediction-errors
(Morris et al., 2012). Thus, in the present study, an additional seven
healthy men and 12 men with schizophrenia were assessed.
2.2. Hormone samples
Fasting intravenous blood samples were collected between 9:45 am
and 11 am. Total testosterone was assayed using a solid-phase, competitive chemiluminescent immunometric assay (Siemens Healthcare Diagnostics Products Ltd, UK) and we also assayed prolactin since levels
of prolactin can indicate dopamine antagonism by antipsychotic drugs.
All hormonal assays were performed by the Prince of Wales Hospital
South Eastern Area Laboratory Services. For testosterone, reference
ranges were set at 7.2 to 25 nmol/L, sensitivity of assay was 0.7 nmol/
L, and the interassay coefcient of variation (CV) was 10.8%. For prolactin, reference ranges were set at 0 to 372 ml U/L, sensitivity of assay
was 11.0 ml U/L, the intra-assay CV was 6 and the interassay CV was
6.8%.
2.3. fMRI methods

2.1. Participants
Fifteen healthy men and 21 chronically ill men with schizophrenia or
schizoaffective disorder were recruited for this study. Three patients
were excluded for excessive head motion (N2 mm), inadequate task
performance (non-responding) or structural abnormalities leaving 18
men with schizophrenia, all of whom were receiving second generation
antipsychotic medication. All participants were native English speakers,
predominantly right-handed (as determined by the Edinburgh Handedness Inventory) and had no history of head injuries with loss of consciousness, seizures, central nervous system infection, uncontrolled
diabetes or hypertension or alcohol or drug abuse in the last ve
years. Trained clinicians administered the Structured Clinical Interview
for the Diagnostic and Statistical Manual IV (SCID) (First et al., 2007),
and obtained premorbid and current IQ estimates using the Wechsler
Test of Adult Reading (WTAR) and a short 4 subtest form of the
Wechsler Adult Intelligence Scale-third edition (WAIS-III), respectively
(Wechsler, 1997, 2001). Symptom severity ratings were obtained

2.3.1. fMRI acquisition

Imaging was acquired on a Phillips Achieva 3T scanner with an 8
channel bird cage head coil at Neuroscience Research Australia. We acquired 968 whole-brain T2* weighted echoplanar images (EPI) with a
gradient echo sequence. The slice thickness was 3 mm with a 1 mm
gap, consisting of 31 axial slices in ascending order, with a repetition
time (TR): 2000 ms, echo time (TE): 30 ms, ip angle: 90, matrix:
112 112, and eld of view (FOV): 240 mm. A T1-weighted high resolution anatomical scan was obtained for each participant for registration
and screening. The T1-weighted anatomical scan had a TR: 5.4 ms, TE:
2.4 ms, FOV: 256 mm, matrix: 256 256, sagittal plane, slice thickness
of 1 mm with no gap, and 180 slices.
2.3.2. Reward prediction task
A series of cue-outcome trials were presented in which participants
were instructed to predict a reward (an image of nine $50 dollar
Australian notes) that was contingent upon presentation of one of

Table 1
Mean (SD) clinical, demographic and hormone results.

Age
Years of education
WAIS-III IQ score
WTAR score
Handedness score
Testosterone (nmol/L)
Prolactin (mlU/L)

SZ (n = 18)a

HM (n = 15)

t (df = 31)

33.83 (8.92)
13.83 (2.73)
98.28 (12.75)
107.56 (10.67)
91.89 (14.13)
15.93 (5.49)
222.17 (194.23)

31.29 (8.34)
15.00 (1.92)
108.86 (13.61)
109.64 (7.47)
96.58 (9.15)
17.24 (5.19)
160.86 (134.73)

0.50
1.36
2.26
0.62
1.01
0.68
1.01

0.62
0.19
0.03
0.54
0.32
0.50
0.32

PANSS score
(General)
(Negative)
(Positive)

32.72 (10.13)
16.17 (3.90)
15.06 (7.46)

SZ: men with schizophrenia; HM: healthy males; WAIS-III: Weschler Adult Intelligence Scale, 3rd Edition; WTAR: Weschler Test of Adult Reading; PANSS: Positive and Negative Syndrome
Scale for Schizophrenia.
a
Antipsychotics (no. of patients): olanzapine: 6, clozapine: 3, risperidone: 2, amisulpride: 1, quetiapine: 1, clozapine & aripiprazole: 1, quetiapine & ziprasidone: 1, quetiapine &
zuclopenthixol: 1, risperidone & amisulpride: 1, and risperidone & quetiapine: 1.

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

four distinct playing-cards (the regularly winning trump card, see the
trial design in Fig. 1).
2.3.3. Pre-scan training
To establish expectations between the trump card and the money
stimulus each participant was initially trained before entering the scanner until a criterion of six consecutive correct responses occurred. Trials
were presented in a pseudo-randomized order to ensure each card was
presented at least once in every block of four trials. The trump card was
presented on half the trials and during training every trial containing
the trump card resulted in reward (i.e., no catch-trials in training).
2.3.4. Scanning task
To ensure prediction-errors occurred, catch-trials were introduced
during the scan and the 120 trials included 12 trials in which money
followed non-trump card hands (unexpected reward: UR) and 12 trials
in which no money followed a trump card hand (unexpected omissions
of reward: UO). Thus, during the scan the trump card was rewarded 80%
of the time. The identity of the trump card was counterbalanced between the card displaying diamonds, squares, circles and triangles
across participants. One catch-trial occurred in every block (4-trials) except for the rst three blocks in the session which were consistent with
training to ensure learned expectations from training were initially

reinforced in the scanner. Since the outcome was predetermined and

pre-training ensured all participants learned the contingency between
the trump card and reward, there was little variation in responses
among participants. The interval between trials was jittered and ranged
between 4051 to 6040 ms during which a cross hair was presented. Participants were required to predict the outcome before presentation of
the reward stimulus (money). Any prediction incongruent with the outcome was included as an unexpected event: thus, the obtained number
of trials in each condition differed slightly for each participant. Failure to
respond resulted in a missed response recorded and these trials were ignored in the analysis. Failure to respond to more than 25% of trials resulted in exclusion as a non-responder (1 patient was excluded for
non-responding, as mentioned previously). The duration of the task
was 32 min.
Stimulus presentation and timing of all stimuli and response events
were achieved using Presentation software (Neurobehavioral Systems,
CA) on a Dell computer running Microsoft Windows XP (SP3, 2002).
Stimulus presentation was automatically synchronized with the onset
of each EPI acquisition to ensure accuracy of event timing. Visual stimuli
were presented using an MRI-compatible mirrored screen placed at the
rear of the MRI scanner and viewed with a system of mirrors arranged in
the participants' direct line of sight while lying in the scanner. Response
times and accuracy were measured using a Lumina MRI-compatible

Fig. 1. Trial design. The order of events in each trial-type of the reward prediction task. Four different card stimuli were used (diamonds, squares, triangles and circles) in which one card
was trump (diamonds in this example). Money (reward stimulus) was either presented or omitted. Predictions were either correct or incorrect.

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

two-button response pad (Cedrus Corporation, CA). Vision correction

was provided as needed with a set of MRI-compatible glasses.

2.4. Immunohistochemistry
Fresh frozen post-mortem tissue provided by the New South Wales
Tissue Resource Centre (NSW TRC) from ve normal adult humans
(mean age = 52.8 years, mean pH = 6.47, mean postmortem
interval = 17.2 h, 3 males, 2 females) was blocked perpendicular to
the long axis of the CNS to include the quadrageminal plate dorsally
and the entire midbrain tegmentum ventrally. Brain slices were cut on
a cryostat (14 m) at the level of the exit of the oculomotor nerve, collected onto gelatin-subbed slides, and stored at 80 C. On the day of
immunolabeling, tissue was thawed, then xed (10 min) in 4% paraformaldehyde at 4 C. For the immunouorescent study, a donkey serum
(10% in PBS) block was applied prior to simultaneous incubation with
tyrosine hydroxylase (TH) mouse monoclonal antibody (MAB318,
Chemicon at 1:250) and androgen receptor (AR) rabbit polyclonal
(PA1-110 Afnity BioReagents 1:100) antibody. Secondary antibodies
were added together at a 1:1000 dilution (A21202 Donkey antimouse 488, A21207 Donkey anti-rabbit 594, Invitrogen). DAPI nuclear
stain was performed in a 1:1000 dilution followed by a copper solution
(5 mM CuSO4 in 50 mM AmmAc pH 5.0) to reduce autouorescence
prior to coverslipping.

2.5. Quantitative real-time PCR (qPCR) analysis

For mRNA quantication, frozen midbrain tissue at the level of the
oculomotor nerve from 19 male schizophrenia cases and 20 matched
control males were provided by the NSW TRC (post mortem midbrain
cohort; Table 2). Total RNA was extracted with Trizol (Life Technologies,
Scoresby, Australia) from midbrain tissue cryostat generated slices
(6 60 m thick) where the substantia nigra was excised based on TH
immunolabeling of adjacent slide-mounted sections. RNA quality was
determined using the Agilent Bioanalyzer 2100 (Agilent Technologies,
Palo Alto, CA) and cases with RIN below 5.0 were removed from analysis. cDNA was synthesized using the Superscript III First strand synthesis
kit (Life Technologies). qPCR analysis was conducted as reported previously (Weickert et al., 2010). Taqman gene expression assays (Applied
Biosystems, Forster, CA) were used for three house keeper control
mRNAs (-actin, Tata-binding protein, ubiquitin C), tyrosine hydroxylase and androgen receptor mRNAs. None of the house keeping genes,
nor the geometric mean of the three, varied between schizophrenia patients and controls (for geometric mean, t = 0.745; df = 55; p = 0.46).
Serial dilutions of pooled cDNA from all samples were included on
every qPCR plate for quantication of sample expression by the relative
standard curve method. All qPCR reactions were performed in triplicate.
Expression levels of the triplicate mean were then normalized to the
geometric mean of the three housekeepers.

Table 2
Demographic detail for the midbrain postmortem cohort.

N (male only)
Age at death
PH
Postmortem interval (PMI)
RNA integrity (RIN)
Duration of illness (yr) (range)
Daily chlorpromazine (CPZ) mean (mg)
Last recorded dose CPZ dose (mg)
Lifetime CPZ dose (g)

Control

Schizophrenia

20
49.35 12.19
6.63 0.26
30.63 9.29
5.55 1.14

19
49.0 11.93
6.51 0.19
35.76 18.25
5.61 1.48
23.75 (3.543)
669.32 273.12
659.26 550.57
6488.99 4581.86

Data are mean SD. Values in brackets indicate age range. There were no signicant differences in demographic detail between control and schizophrenia groups.

2.6. Data analysis

2.6.1. Task performance
The number of errors in the rst three blocks of the card game during
the scan (prior to the introduction of catch-trials) was used to calculate
the percentage of correct trials and conrm the success of training. The
remaining trials (including catch-trials) were also used to calculate the
percentage of correct trials to ensure group performance was maintained across the scan. Differences in the percentage of correct trials between groups were tested with independent two-tailed t-tests.
2.6.2. Hormones
Mean levels of serum testosterone and prolactin were calculated for
each group. Differences between each group were tested with independent two-tailed t-tests. To determine relationships between hormones
and symptoms, correlations between serum hormone levels and
PANSS positive and negative subscale scores were calculated using
Pearson r.
2.6.3. Imaging analyses
2.6.3.1. Preprocessing, rst-level and second-level analysis. Image processing and analysis was performed using SPM5. Images for each participant
were registered to the rst image in the sequence using a rigid-body
spatial transformation (6 parameters) and resliced using a 4th-degree
b-spline interpolation to correct for head motion. Slice time correction
to the rst slice was performed using SPM5's Fourier phase shift interpolation to adjust for differences in the time of slice acquisition. Images
were spatially normalized to the Montreal Neurological Institute
(ICBM452) template using a non-linear 12 parameter afne transformation. All images were smoothed to minimize noise and residual differences in gyral anatomy with a Gaussian kernel set at 8 mm full-width
at half-maximum (FWHM). Data sets were also manually screened for
scan stability (b2 mm head movement) and successful normalization.
The default high-pass lter (128 s) was used to remove slow signal
drift and serial correlations were accounted for using the default
autoregressive (AR1) autocorrelation model in SPM5.
The results from each individual subject were analyzed using a xed
effects model. Eight regressors were modeled in an event-related design
dened by the stimulus onset times and the participant's responses.
Four regressors of interest representing reward outcomes correctly
and incorrectly predicted, expected reward (ER) and unexpected reward (UR), and non-reward outcomes correctly and incorrectly predicted, expected omission of reward (EO) and unexpected omission of
reward (UO), and four regressors of no interest including two cue
regressors (reward cards and non-reward cards) and two response regressors (predicting reward and predicting non-reward). Six regressors
from the motion correction procedure were also included in the rstlevel analysis to account for slight variation due to head movement.
Events of interest were modeled as stick functions to represent the transient neural response to errors and each of the eight task regressors
were convolved with the canonical hemodynamic response function.
Individual beta-images for the four outcome regressors (UR, ER, UO
and EO) were taken to a second-level random-effects model to test for
group effects in separate ANOVAs.
We aimed to identify BOLD responses with a phasic increase to UR
(positive prediction-error), a phasic decrease after UO (a negative
prediction-error), and no change to predicted rewards in healthy men
(Niv and Schoenbaum, 2008; Morris et al., 2012). To achieve this, we
tested for regions with positive responses (activation) to UR over the
ER baseline (UR N ER), as well as negative responses (deactivation) to
UO relative to the EO baseline (UO b EO). This was performed in SPM
by testing the conjunction between UR N ER and UO b EO contrasts. To
ensure the BOLD response to predicted rewards did not change, we excluded regions according to an F-test of ER and EO (mask p b .05, uncorrected). Using a liberal threshold for the exclusion mask meant that only

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

voxels without signicant activation or deactivation during either the ER

or the EO baseline conditions were included in the analysis. After exclusion, the results were threshold at a whole-brain voxel-level FDR = .05
to correct for multiple comparisons and used to create a functional ROI
(fROI) for the covariate analyses described below. We also conrmed
our earlier results based on subsample of this cohort (Morris et al.,
2012) in the present analysis.

2.6.3.2. Conrmation analysis of diagnostic group differences in BOLD. We

tested for group differences in prediction-error signals within the
mesoaccumbens region (midbrain, caudate and putamen included in a
single ROI). As described in our previous report (Morris et al., 2012),
the three-way interaction: group reward surprise was tested to reveal the location of aberrant prediction-error signals.

2.6.3.3. fROI covariate analysis. Since we were specically interested in

the relationship between hormones and prediction-error signals, the
SPM of the conjunction analysis described above was saved as a binary
mask that was then used as an fROI in each covariate analysis (see
Fig. 2). In this way, we restricted any obtained effect of testosterone to
prediction-error signals, rather than non-specic effects of reward or
task stimuli. We performed a linear regression of testosterone and
prediction-error BOLD responses in each group (df = 13 and 16 for
healthy men and men with schizophrenia, respectively). To determine
testosterone's association with positive and negative prediction-error
signals, the differential BOLD responses (i.e., UR N ER and UO b EO)
were analyzed in separate regression models with testosterone as a covariate in each model (similarly, we tested symptom levels, prolactin
levels and drug dose as covariates as described below). After each test,
we conrmed signicant effects of testosterone were not due to changes in the ER or EO baseline by directly testing testosterone's association
with each ER and EO baseline. Signicant effects with the ER and EO
baseline in the opposite direction to the prediction-error effect would
suggest some (or all) of the effect of hormone with prediction-errors
may be due to a change in ER or EO baseline. We also directly compared
the correlation between groups by performing a covariate analysis including both groups with testosterone levels as separate covariates.
The group by testosterone interaction directly compared the linear effect of testosterone between groups to identify any potential regions
of aberrant response in men with schizophrenia relative to healthy men.

2.6.3.4. Covariate analysis with prolactin and CPZ. Since reward

prediction-error BOLD signals are thought to reect phasic dopamine signaling (Pessiglione et al., 2006; Knutson and Gibbs, 2007),
dopamine antagonism by antipsychotic drugs may attenuate the
prediction-error BOLD signal in patients with schizophrenia. To assess whether dopamine antagonism may be responsible for aberrant
prediction-error signals in men with schizophrenia, we tested
whether prolactin levels (which can be raised with increased dopamine antagonism) or the antipsychotic drug dose was inversely related to prediction-error BOLD signal in schizophrenia. The
antipsychotic drug dose was rst converted to a daily chlorpromazine equivalent dose (CPZ) to compare across different antipsychotics (Bollini et al., 2008). We included prolactin levels and CPZ
dose as covariates in separate linear-regression analyses of the differential BOLD responses (i.e., UR N ER and UO b EO) in the fROI
among men with schizophrenia to determine whether any relationship existed between prediction-error signals and these indirect
measures of dopamine antagonism.
2.6.3.5. Covariate analysis with PANSS positive and negative symptoms. To
determine whether aberrant ventral striatal activity was associated
with symptoms in schizophrenia, we tested whether positive or negative symptom scores were related to BOLD responses in the fROI
among men with schizophrenia. We included positive and negative
symptom scores in separate linear-regression analyses of the ER and
EO baseline.
For each covariate analysis above, we report signicant clusters larger than 25 contiguous voxels at p b .005 within the fROI, resulting in a
small volume corrected p = .05. The location of signicant peak voxels
was identied using the Anatomical Automatically Labeling (AAL) toolbox (Tzourio-Mazoyer et al., 2002) and was conrmed by reference to
the Atlas of the Human Brain (Mai et al., 2008).
2.6.4. Analysis of AR and TH gene expression in postmortem substantia
nigra
Independent sample 2-tailed t-tests were used to test for differences
in cohort demographics and housekeeper genes and/or geomean of the
housekeepers between control and schizophrenia individuals. For all
mRNA measures, Pearson's correlations were performed with demographic variables (pH, RIN, PMI, duration of illness) and Spearman's correlations were conducted with antipsychotic drug measurements. PMI
and pH correlated with AR and TH gene expression, as such they were
used as covariates when analyzing diagnostic differences in mRNA expression using analysis of covariance. Pearson's partial correlations
(controlling for pH and PMI) were conducted to determine the relationship between AR gene expression and TH gene expression. Statistical
signicance was set at p 0.05.
3. Results
3.1. Hormone levels were similar in healthy men and men with
schizophrenia
The mean levels of circulating testosterone and prolactin among
healthy men and men with schizophrenia were not signicantly different (see Table 1). The correlation analyses among hormone levels,
PANSS symptoms severity scores and daily antipsychotic dose (CPZ)
failed to reveal any strong, signicant correlations (see Table 3).
3.2. Healthy men and men with schizophrenia successfully learned to
predict the winning outcome in the reward prediction task

Fig. 2. Functional region-of-interest (fROI). The results of the conjunction analysis in

healthy men between positive and negative reward prediction-errors.

The mean (SEM) percent accuracy of predictions in the rst three

blocks before the introduction of catch-trials during the scan were
86.2 (4.7) and 88.1 (4.1) for healthy men and men with schizophrenia,
respectively, n.s. Introduction of the catch trials reduced accuracy in

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

Table 3
Pearson correlations (and p-values) among hormones, symptom severity and antipsychotic dose.

PANSS positive symptoms score

PANSS negative symptoms score
Chlorpromazine equivalent dose

Testosterone

Prolactin

0.06 (.81)
0.27 (.28)
0.32 (.20)

0.05 (.83)
0.11 (.67)
0.18 (.47)

conrming that the positive linear effect of testosterone was not due
to potential BOLD changes during expected reward. Thus, the results
are consistent with the neural response to unexpected rewards increasing as testosterone levels increased among healthy men. The largest relationship between testosterone and positive prediction-errors
occurred in the right ventral striatum of healthy men with signicant effects also occurring in the left putamen (see Table 5).

PANSS: Positive and Negative Syndrome Scale.

both groups during the remaining scanning trials to 68.0 (2.6) and 70.0
(3.4) percent, for healthy men and men with schizophrenia respectively, n.s.
3.3. Prediction-error signals occurred in the ventral striatum of healthy men
during reward prediction
Our conjunction analysis of positive activation to UR and negative
activation to UO revealed prediction-error signals throughout the
striatal region of healthy men, with the highest signicant peak voxel
in the left ventral striatum (see Table 4). These SPM results were used
to create the functional region-of-interest (fROI) to test the effects of
hormones as covariates in each group (Fig. 2).
3.4. Aberrant prediction-error signals in men with schizophrenia were related to negative symptoms
We replicated our earlier report of aberrant prediction-error signals
in schizophrenia by testing the three-way interaction: reward x surprise
x group (Morris et al., 2012). This revealed signicant aberrant
prediction-error signals occurred in the ventral striatum of men with
schizophrenia (Table 4). We also conrmed that aberrant activity in
the fROI was associated with symptom severity in schizophrenia:
PANSS negative symptom scores were signicantly, positively and
strongly correlated with right caudate (ventral striatal) activity during
the ER baseline (Fig. 3A and B). Examination of the peristimulus time
histogram (PSTH) extracted from the peak voxel in the right caudate
conrmed the aberrant activity occurred during the ER baseline in
men with schizophrenia (Fig. 3C). The increased response to expected
rewards, when there should be no change in BOLD signal, obscured positive prediction-error signals in schizophrenia. A scatterplot of the relationship between negative symptoms and aberrant activity in the right
caudate conrms the signicant correlation was not due to an outlier
(Fig. 3D).
3.5. Testosterone levels were related to positive prediction-error signals in
healthy men
The fROI covariate analysis revealed a signicant positive linear relationship between circulating testosterone levels and striatal activity
during positive (UR N ER) prediction-errors among healthy men
(Fig. 4A). A follow-up analysis testing if testosterone correlated with
the baseline BOLD striatal activity during ER was not signicant,

Table 4
Prediction-error signals in healthy men and men with schizophrenia.
Group

Region label

Cluster size

Prediction-errors in healthy men (UR N ER; UO b EO)

Left caudate (vStr)
1724
5.44
Right putamen (vStr)
222
3.42

MNI (x, y, z)

b0.001
0.001

10, 8, 4
26, 22, 2

Reward surprise group (Aberrant prediction-errors in schizophrenia)

Right putamen
178
2.92
0.002
26, 22, 2
Left caudate (vStr)
164
2.65
0.005
10, 10, 2
UR: Unexpected reward; ER: expected reward; UO: unexpected omission of reward; EO:
expected omission of reward; and vStr: ventral striatum.

3.6. Positive prediction-error signals did not vary with testosterone in men
with schizophrenia
Covariate analysis failed to reveal any signicant relationship between circulating testosterone levels and positive prediction-error related activity (UR N ER) in the fROI among men with schizophrenia.
We also directly compared the relationship between testosterone and
positive prediction-error signals in men with and without schizophrenia by including both groups in the same covariate analysis and testing
the group by testosterone interaction. A signicant group interaction
occurred in the midbrain and the ventral striatum of the fROI
(Table 6). Fig. 4B shows regions where the groups were signicantly different. Extraction of the parameter estimates at the highest signicant
peak voxel in the ventral striatum of healthy men conrmed that the
correlation with testosterone was strong and positive (Fig. 4C), while
a (weaker) inverse relationship existed in men with schizophrenia in
the same region (Fig. 4D). This indicates the source of the interaction
was due to a weaker or reversed effect in schizophrenia. There were
no group differences in the relationship between positive prediction
error signals and testosterone in the opposite direction
(i.e., schizophrenia N healthy controls).
3.7. Testosterone was inversely related to negative prediction-error signals
in healthy men
The fROI covariate analysis also revealed a signicant inverse linear
relationship between testosterone and negative prediction-error signals
(UO b EO) among healthy men. This indicates larger decreases in BOLD
responses (i.e., larger negative prediction-error signals) as testosterone
increased, suggesting that testosterone levels not only relate to
motivationally-relevant increases in BOLD, but also to motivationallyrelevant decreases in BOLD in the ventral striatum of healthy men.
Fig. 5A shows the peak voxel in the left putamen of healthy men and
Table 5 lists signicant peaks throughout the ventral striatum during
negative prediction error. There was no signicant positive relationship
of testosterone with the BOLD response to baseline EO among healthy
men; thus, changes in EO baseline do not explain the relationship of testosterone with negative prediction-error activity.
3.8. Negative prediction error signals were not inversely related to testosterone in schizophrenia
Among men with schizophrenia, a signicant positive linear relationship between testosterone and negative prediction-error signals
was shown in the fROI, specically in the right putamen (Table 5);
and this was not due to changes in the EO baseline with testosterone.
The positive relationship indicates that negative prediction-error signals (deactivation) were more attenuated with higher (normal) levels
of testosterone in men with schizophrenia. Overall, the pattern of results suggests that among men with schizophrenia with higher normal
levels of testosterone, motivationally-relevant neural changes in the
ventral striatum (increases and decreases) were blunted.
The group by testosterone interaction to directly compare healthy
men and men with schizophrenia revealed signicant group differences
in the linear effect of testosterone on left and right ventral striatal activity (Table 6 and Fig. 5B). Extraction of the parameter estimates conrmed the correlations with testosterone were in opposite directions

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

Fig. 3. Aberrant prediction-error signals in men with schizophrenia (SZ) correlate with negative symptom scores (PANSS). (A & B) Peak voxel in the right ventral striatum (8, 16, 6)
correlating with negative symptoms among SZ; (C) peristimulus time histogram (PSTH) at peak voxel showing mean aberrant response after ER is above zero. Shaded area represents
SEM; and (D) scatterplot of the signicant correlation between response at peak voxel during ER and negative symptom score in SZ. Thin red lines bounding the linear effect represent
95% condence intervals (SPM threshold p b .005).

in each group (Fig. 5C and D). There were no other signicant group differences from the interaction analysis.
3.9. No evidence that CPZ levels were related to prediction-error signals in
men with schizophrenia
The linear regression with CPZ revealed no signicant relationship
with prediction-error signals in the fROI or whole-brain analysis in
schizophrenia. We also failed to nd evidence that prolactin was inversely related to prediction-error signals in the fROI. However, prolactin was positively related to positive prediction-error signals (UR N ER)
in the ventral striatum (fROI) in men with schizophrenia (see Table 5).
3.10. Human midbrain dopamine neurons have potential to respond directly to testosterone
Many dopamine neuron tyrosine hydroxylase-positive (TH +)
immunopositive cells were identied in the substantia nigra pars
compacta (SNpc) in all ve human midbrains. Using double-label
immunouorescence we conrmed that human dopamine neurons
(TH + green cell in Fig. 6B) in the substantia nigra were
immunopositive for AR (red cell in Fig. 6C). Overlap of subcellular
distribution of TH and AR in human subtantia nigra was conrmed
in the cytoplasm, areas that are yellow due to overlap between TH
(green) and AR (red) in Fig. 6D, whereas only AR immunoreactivity
was found in the nucleus, purple at arrowhead due to overlap

between DAPI (blue) and AR (red), and TH immunolabeling was

more intense in the dendrite as compared to AR (thus, it appears
green, see arrow).
3.11. Positive relationship between AR and TH gene expression in the
substantia nigra
We identied no signicant differences in TH or AR gene expression
levels in substantia nigra between control and schizophrenia post
mortem male brains (For TH mRNA: F = 0.008; df = 38; p = 0.93, for
AR mRNA: F = 0.59, df = 38, p = 0.45). Neither TH mRNA nor AR
mRNA levels correlated with any of the three measures of chlorpromazine equivalents. For last dose CPZ vs. AR mRNA: N = 19, r = 0.27,
p = 0.251 and last dose CPZ vs. TH mRNA: N = 19, r = 0.32, p =
0.19, all comparisons between average daily CPZ or total lifetime CPZ
and TH or AR mRNA: r's b 0.2 and p's N 0.05.
Pearson's partial correlations (controlling for pH and RIN), including
data from both control and schizophrenia brains, revealed a signicant
positive relationship between AR and TH mRNA levels (r = 0.53, df =
33; p = 0.002) (Fig. 6E).
4. Discussion
The present study found testosterone levels were correlated with
the normal neural response to prediction-errors in the ventral striatum
of healthy men and this relationship was signicantly different from the

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

Fig. 4. Correlations between testosterone and positive prediction errors (UR N ER). (A) Regions of signicant correlations between striatal activity and testosterone in healthy men (HM);
(B) regions of a signicant two-way interaction between group, i.e., HM N men with schizophrenia (SZ), neural activity and testosterone levels in the striatum; (C) the signicant correlation between the BOLD parameter estimates (UR N ER) and testosterone at the peak voxel in HM; and (D) the negative (and non-signicant) correlation between BOLD parameter estimates (UR N ER) and testosterone in SZ, conrming the signicant two-way interaction in B was due to a weaker or reversed effect among SZ. Thin red lines bounding the linear effect
represent 95% condence intervals (SPM threshold p b .005).

correlation found in men with schizophrenia. The relationship with testosterone in the healthy men is consistent with the sex steroid modulation of reward prediction-error signals in the ventral striatum and may
reect changes in dopamine signaling. We also detected testosterone
receptors (AR) in the human midbrain consistent with the direct modulation of dopamine by sex steroids and thus, dopamine neurons could

be mediators of the testosterone-related changes we observed in the

ventral striatum BOLD signal which is also supported by ndings of testosterone directly regulating dopamine related gene expression in male
rodents (Purves-Tyson et al., 2012). In addition, in post mortem human
male substantia nigra we showed that tyrosine hydroxylase gene expression, an indicator of dopamine synthesis capacity, was positively

Table 5
Relationship of testosterone and prolactin to prediction-error signals in ventral striatum.
Hormone/group

Region label

Testosterone and positive prediction-errors (df = 29)

Healthy men
Right caudate (vStr)
Left putamen
Men with schizophrenia
n.s.
Testosterone and negative prediction-errors (df = 29)
Healthy Men
Left putamen (vStr)
Left putamen (vStr)
Right putamen (vStr)
Left putamen
Men with schizophrenia
Right putamen
Prolactin and positive prediction-errors (df = 16)
Men with schizophrenia
Left caudate (vStr)
Right putamen

Cluster
size

Pearson

99
107

4.70
4.24

0.76
0.73

b.001
b.001

10, 12, 12
12, 12, 6

34
37
30
37
93

3.88
3.86
3.83
3.69
3.42

0.70
0.69
0.69
0.68
0.65

0.001
0.001
0.001
0.001
0.002

26, 10, 4
14, 10, 8
34, 18, 4
34, 16, 6
30, 4, 10

398
279

4.69
4.34

0.76
0.74

b.001
b.001

2, 10, 8
24, 18, 2

MNI (x, y, z)
p

vStr: Ventral striatum; and n.s. non-signicant.

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

Table 6
Group differences in the interaction between testosterone and ventral striatum activity.
Group/hormone

MNI (x, y, z)

Healthy men N schizophrenia: testosterone and positive prediction-errors (df = 29)

Left putamen (vStr)
111
Midbrain (SN/VTA)
29
Right caudate (vStr)
54

Region label

Cluster size

3.91
3.59
3.48

b.001
0.001
0.001

14, 0, 0
10, 12, 4
10, 14, 2

Healthy men b schizophrenia: testosterone and negative prediction-errors (df = 29)

Left putamen
119
Right caudate (vStr)
118
Left putamen (vStr)
87

4.19
4.00
3.60

b.001
b.001
0.001

28, 10, 4
14, 14, 2
24, 12, 2

vStr: Ventral striatum; and SN/VTA: substantia nigra/ventral tegmental area.

correlated with androgen receptor gene expression, providing evidence

in the human male that testosterone signaling and dopamine production may be linked. However, if the BOLD signal was regulated via dopamine activity, the type of modulation by testosterone would not appear
to be a simple increase or decrease of dopamine neuronal ring in a unidirectional manner since testosterone was positively correlated with increasing BOLD response to unexpected reward (UR) and negatively
correlated to decreasing BOLD signal in response to unexpected omission of reward (UO) in healthy men. Thus, our results suggest that sex
steroids inuence the control of dopamine neurotransmission in a
valence-specic manner to augment or suppress release inversely
with reward expectations. This is consistent with evidence that testosterone modulates reward-learning to inuence motivated behavior,

rather than acting as a primary reinforcer (de Beun et al., 1992;

Alexander et al., 1994).
Given testosterone's amplifying effect on anticipatory rewardresponses in women with low intrinsic motivation (Hermans et al.,
2010), we also expected to see a benecial relationship between testosterone and ventral striatal activity in men with schizophrenia who generally have low intrinsic motivation. However, the correlation between
testosterone levels and neural responses in the ventral striatum of men
with schizophrenia was either nonexistent or in the opposite direction
to healthy men (e.g., Fig. 4C versus 4D). Thus, in contrast to our expectations, the present results suggest testosterone was having little effect
or even a detrimental effect on ventral striatal reward responses in men
with schizophrenia.

Fig. 5. Correlations between testosterone and negative prediction errors (UO b EO). (A) Regions of signicant correlations between striatal activity and testosterone in healthy men (HM);
(B) regions of a signicant two-way interaction between group, i.e., HM b men with schizophrenia (SZ), neural activity and testosterone levels in the striatum; (C) the signicant correlation between the BOLD parameter estimates (UO b EO) and testosterone at the peak voxel in HM; and (D) the positive (and non-signicant) correlation between BOLD parameter estimates (UO b EO) and testosterone in SZ, conrming the signicant two-way interaction in B was due to a weaker or reversed effect among SZ. Thin red lines bounding the linear effect
represent 95% condence intervals (SPM threshold p b .005).

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

10

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

Fig. 6. Tyrosine hydroxylase (TH) and androgen receptor (AR) immunoreactivity in human midbrain neurons. Immunouorescence of TH and AR in a human SN neuron; (A) DAPI nuclear
stain; (B) TH immunoreactivity; (C) AR immunoreactivity; and (D) localisation of TH in the cytoplasm and AR in the nucleus. The arrowhead indicates the nucleus with AR staining and the
arrow indicates an axon with TH staining and no AR. Scale bar = 100 m. (E) TH mRNA and AR mRNA were positively correlated in the substantia nigra from both control and schizophrenia brains (r = 0.53, df = 33, p = 0.002).

In contrast to our hypothesis, we did not nd a strong signicant correlation between testosterone levels and negative symptoms in schizophrenia (r = 0.27, n.s.). Some studies with larger sample sizes have
found increasing circulating testosterone levels are associated with
fewer negative symptoms (Akhondzadeh et al., 2006; Ko et al., 2007);
however, other studies of men with schizophrenia receiving maintenance treatment with antipsychotics as opposed to displaying acute
psychosis have also failed to show a relationship between negative
symptoms and circulating testosterone levels (Taherianfard and
Shariaty, 2004; Moore et al., 2013). While differences in sample size
makes it difcult to reach rm conclusions, the absence of an association
between serum testosterone levels and negative symptoms in our sample is at least consistent with the view that symptoms in schizophrenia
do not only vary with levels of circulating sex steroids, but also may be
due to sex steroid receptor dysfunction or abnormalities in downstream effectors in the presence of healthy hormone levels.
The positive relationship between testosterone levels and
mesolimbic function in the healthy males suggests that dopamine neurotransmission is normally sensitive to circulating testosterone. A potential mechanism by which testosterone exerts its effects is indicated
by our rodent and primate molecular studies, as well as the AR and TH
expression levels reported in the current study. In particular, we have
found that circulating testosterone levels in adolescent male rhesus macaques correlate with dopamine synthesis potential, as inferred from increased striatal tyrosine hydroxylase levels the rate-limiting step in
dopamine biosynthesis (Morris et al., 2010). Our studies in rodents
also nd that testosterone triggers a positive feedback loop to change
the level of dopamine receptors and breakdown enzymes and dopamine transporters via AR action (Purves-Tyson et al., 2012, 2014). Our
current study of human post mortem substantia nigra in patients and
controls also found a positive association between a potential functional
expression of AR and tyrosine hydroxylase. These converging lines of
evidence add support to the possibility that testosterone can directly
modify dopamine neurotransmission in the mesolimbic path (Sinclair
et al., 2014).
The enhancement of the mesolimbic response with testosterone did
not occur in men with schizophrenia, even though they appeared to
have healthy levels of testosterone and a separate cohort displayed normal midbrain TH/AR mRNA levels. The prediction-errors we found
among men with schizophrenia were obscured by aberrant responses
during expected rewards (baseline), which raises the problem of why

the prediction-error response was not enhanced above baseline in the

context of normal testosterone levels. Given the enhancement of dopamine transmission by testosterone we have observed elsewhere in
humans, macaques and rodents (as described above), we speculate
that sex steroid related dysfunction in schizophrenia may be responsible for the attenuated response to motivationally-relevant stimuli. The
site of dysfunction is likely downstream from the androgen receptor,
given that AR and TH mRNA was positively correlated in post-mortem
substantia nigra tissue from people with schizophrenia. Instead, we suggest that dysfunction at other points in the downstream pathway
(e.g., DAT or VMAT) could derail normal testosterone modulation of dopamine neurotransmission, even in the presence of normal sex steroid
receptor action. Thus, we speculate that the lack of relationship between
testosterone and ventral striatal responses in schizophrenia may be due
to a failure of testosterone to properly regulate other mRNAs encoding
proteins that control dopamine neurotransmission, i.e., release and reuptake. However, whatever the molecular mechanism, since goaldirected behavior is guided by reward prediction (Morris et al., 2014),
compromised prediction-error signaling may have far-reaching consequences on daily motivation and successfully achieving goals related
to education and employment for people with schizophrenia.
There were several limitations of the present study relating to the
measurement of testosterone and medication effects. The conrmation
we report here included data from 14 participants in the original study
of reward prediction-errors (SZ = 6, HA = 8) so it does not represent a
pure replication of our earlier study with an independent sample. We
did not estimate an a priori sample size for the new analysis and so
while the correlations we observed were substantial, the condence intervals around those estimates have a wide range. Our patient sample
was treated with second-generation antipsychotic medications, which
all have a common dopamine D2 receptor antagonistic effect (Seeman
and Tallerico, 1998, 1999), thus, we cannot entirely rule out the possibility that disruption to striatal function among our patient group was due
to dopamine antagonism. However, we found no relationship between
CPZ and ventral striatal BOLD responses in schizophrenia or to AR or TH
mRNA in the midbrain, which is consistent with our view that our results are due to pathology in the illness rather than a medication effect.
Another potential limitation is that we measured circulating serum testosterone levels, which are an indirect measure of the actual levels of
testosterone reaching the neurons within the brain. The majority of testosterone (N98%) is strongly bound to sex hormone-binding protein or

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

weakly bound to albumin, leaving only a small proportion free to diffuse

into the tissue and bind with sex steroid receptors. However, free testosterone levels correlate with total testosterone and circulating levels
are often used as an index of bioavailable testosterone (Vermeulen
et al., 1999).
In summary, the present results show for the rst time, that testosterone may modulate ventral striatal reward sensitivity in healthy
men, putatively by directly acting on midbrain dopamine neurons signaling prediction-errors. However, testosterone was uncoupled or negatively coupled with striatal activity in men with schizophrenia. While
the reasons for this are not known, it is possibly due to dysfunctional
sex steroid receptor signaling in the midbrain of people with schizophrenia (Weickert et al., 2008). This sex steroid receptor dysfunction
may be critical during periods of elevated hormone levels, such as during adolescence, when men in the prodromal phase prior to their rst
psychotic break are more likely to fail at school, work and social relationships. Further research on the effects of testosterone on the prodromal phase during reward processing would be important to understand
the genesis of motivational decits in schizophrenia and all stages of the
illness may benet from sex hormonal modulation which may improve
prediction error signaling and motivation.
Contributions
RW Morris designed the study in part, organized and analyzed the
data, and wrote and edited the manuscript.
TD Purves-Tyson managed qPCR, contributed to analysis, and edited
the manuscript.
CS Weickert designed the study in part, managed the study in part,
and edited the manuscript.
D Rothmond performed immunohistochemistry, managed RNA extraction and cDNA systhesis, and edited the manuscript.
R Lenroot assessed patients and edited the manuscript.
TW Weickert designed the study in part, wrote the protocol,
assessed patients, managed the study in part, and edited the
manuscript.
All authors contributed to and have approved the nal manuscript.
Funding body agreements and policies
This work was supported by the University of New South Wales
School of Psychiatry, National Health and Medical Research Council
(NHMRC) Project Grant #568807, Neuroscience Research Australia,
the Schizophrenia Research Institute (utilizing infrastructure funding
from the NSW Ministry of Health and the Macquarie Group Foundation), and the Australian Schizophrenia Research Bank (supported by
the NHMRC, the Pratt Foundation, Ramsay Health Care, and the Viertel
Charitable Foundation). CSW is a recipient of a NHMRC (Australia) Senior Research Fellowship (#1021970). RWM is presently supported by
NHMRC Project Grant #1069487, and the Australian Research Council
Centre of Excellence in Cognition and its Disorders.
Conict of interest
CSW is a recipient of a NHMRC (Australia) Senior Research Fellowship (#1021970).
RWM is presently supported by NHMRC Project Grant #1069487, and the Australian Research Council Centre of Excellence in Cognition and its Disorders. All authors declare no
relevant nancial interests or potential conicts of interest.
Acknowledgments
The authors thank Alice Rothwell for skilled laboratory support and editing assistance.
Midbrain postmortem tissues were received from the New South Wales Tissue Resource
Centre (NSW TRC) at the University of Sydney. NSW TRC is supported by the National Institute on Alcohol Abuse And Alcoholism of the National Institutes of Health under Award
Number R28AA012725 and the Schizophrenia Research Institute. The content is solely the
responsibility of the authors and does not necessarily represent the ofcial views of the
National Institutes of Health. The authors thank Toni McCrossin and Julia Stevens at the
New South Wales Brain Bank for the provision of midbrain tissue sections and collation
of clinical demographics for the post mortem cohort.

11

References
Akhondzadeh, S., Rezaei, F., Larijani, B., Nejatisafa, A.A., Kashani, L., Abbasi, S.H., 2006. Correlation between testosterone, gonadotropins and prolactin and severity of negative
symptoms in male patients with chronic schizophrenia. Schizophr. Res. 84, 405410.
Alexander, G.M., Packard, M.G., Hines, M., 1994. Testosterone has rewarding affective
properties in male rats: implications for the biological basis of sexual motivation.
Behav. Neurosci. 108, 424428.
Archer, J., 2006. Testosterone and human aggression: an evaluation of the challenge hypothesis. Neurosci. Biobehav. Rev. 30, 319345.
Bollini, P., Pampallona, S., Nieddu, S., Bianco, M., Tibaldi, G., Munizza, C., 2008. Indicators of
conformance with guidelines of schizophrenia treatment in mental health services.
Psychiatr. Serv. 59, 782791.
Coates, J.M., Herbert, J., 2008. Endogenous steroids and nancial risk taking on a London
trading oor. Proc. Natl. Acad. Sci. U. S. A. 105, 61676172.
D'Ardenne, K., McClure, S.M., Nystrom, L.E., Cohen, J.D., 2008. BOLD responses reecting
dopaminergic signals in the human ventral tegmental area. Science 319, 12641267.
de Beun, R., Jansen, E., Slangen, J.L., Van de Poll, N.E., 1992. Testosterone as appetitive and
discriminative stimulus in rats: sex- and dose-dependent effects. Physiol. Behav. 52,
629634.
de Souza Silva, M., Mattern, C., Topic, B., Buddenberg, T., Huston, J., 2009. Dopaminergic
and serotonergic activity in neostriatum and nucleus accumbens enhanced by intranasal administration of testosterone. Eur. Neuropsychopharmacol. 19, 5363 (Epub
2008 Sep 2024).
First, M.B., Spitzer, R.L., Gibbon, Miriam, Williams, J.B.W., 2007. Structured Clinical Interview for DSM-IV-TR Axis I Disorders Patient Edition. 1st Revision Edition. Biometrics Research Department, New York State Psychiatric Institute, New York, NY.
Hermans, E.J., Bos, P.A., Ossewaarde, L., Ramsey, N.F., Fernandez, G., van Honk, J., 2010. Effects of exogenous testosterone on the ventral striatal BOLD response during reward
anticipation in healthy women. Neuroimage 52, 277283.
Juckel, G., Schlagenhauf, F., Koslowski, M., Wustenberg, T., Villringer, A., Knutson, B.,
Wrase, J., Heinz, A., 2006. Dysfunction of ventral striatal reward prediction in schizophrenia. Neuroimage 29, 409416.
Kanayama, G., Brower, K.J., Wood, R.I., Hudson, J.I., Pope Jr., H.G., 2009. Anabolicandrogenic steroid dependence: an emerging disorder. Addiction 104, 19661978.
Kanayama, G., Hudson, J.I., Pope Jr., H.G., 2010. Illicit anabolicandrogenic steroid use.
Horm. Behav. 58, 111121.
Kay, S.R., Fiszbein, A., Opler, L.A., 1987. The Positive and Negative Syndrome Scale
(PANSS) for schizophrenia. Schizophr. Bull. 13, 261276.
Knutson, B., Gibbs, S.E., 2007. Linking nucleus accumbens dopamine and blood oxygenation. Psychopharmacology 191, 813822.
Ko, Y.H., Jung, S.W., Joe, S.H., Lee, C.H., Jung, H.G., Jung, I.K., Kim, S.H., Lee, M.S., 2007. Association between serum testosterone levels and the severity of negative symptoms in
male patients with chronic schizophrenia. Psychoneuroendocrinology 32, 385391.
Mai, J.K., Paxinos, G., Voss, T., 2008. Atlas of the Human Brain. 3rd ed. Elsevier, Oxford, UK.
Moore, L., Kyaw, M., Vercammen, A., Lenroot, R., Kulkarni, J., Curtis, J., O'Donnell, M., Carr,
V.J., Shannon Weickert, C., Weickert, T.W., 2013. Serum testosterone levels are related
to cognitive function in men with schizophrenia. Psychoneuroendocrinology 38,
17171728.
Morris, R.W., Fung, S.J., Rothmond, D.A., Richards, B., Ward, S., Noble, P.L., Woodward,
R.A., Weickert, C.S., Winslow, J.T., 2010. The effect of gonadectomy on prepulse
inhibition and fear-potentiated startle in adolescent rhesus macaques.
Psychoneuroendocrinology 35, 896905.
Morris, R.W., Vercammen, A., Lenroot, R., Langton, J.M., Short, B., Kulkarni, J., Curtis, J.,
O'Donnell, M., Shannon Weickert, C., Weickert, T.W., 2012. Disambiguating ventral
striatum fMRI related BOLD signal during reward prediction in schizophrenia. Mol.
Psychiatry 12, 280289.
Morris, R.W., Quail, S., Grifths, K.R., Green, M.J., Balleine, B.W., 2014. Corticostriatal control of goal-directed action is impaired in schizophrenia. Biol. Psychiatry http://dx.doi.
org/10.1016/j.biopsych.2014.06.005.
Niv, Y., Schoenbaum, G., 2008. Dialogues on prediction errors. Trends Cogn. Sci. 12, 265272.
Perez, S.E., Chen, E.Y., Mufson, E.J., 2003. Distribution of estrogen receptor alpha and beta
immunoreactive proles in the postnatal rat brain. Brain Res. Dev. Brain Res. 145,
117139.
Pessiglione, M., Seymour, B., Flandin, G., Dolan, R.J., Frith, C.D., 2006. Dopaminedependent prediction errors underpin reward-seeking behaviour in humans. Nature
442, 10421045.
Purves-Tyson, T.D., Handelsman, D.J., Double, K.L., Owens, S.J., Bustamante, S., Weickert,
C.S., 2012. Testosterone regulation of sex steroid-related mRNAs and dopaminerelated mRNAs in adolescent male rat substantia nigra. BMC Neurosci. 13, 95.
Purves-Tyson, T.D., Owens, S.J., Double, K.L., Desai, R., Handelsman, D.J., Weickert, C.S.,
2014. Testosterone induces molecular changes in dopamine signaling pathway molecules in the adolescent male rat nigrostriatal pathway. PLoS One 9, e91151.
Richards, A.B., Morris, R.W., Ward, S., Schmitz, S., Rothmond, D.A., Noble, P.L., Woodward,
R.A., Winslow, J.T., Weickert, C.S., 2009. Gonadectomy negatively impacts social behavior of adolescent male primates. Horm. Behav. 56, 140148.
Schlagenhauf, F., Juckel, G., Koslowski, M., Kahnt, T., Knutson, B., Dembler, T., Kienast, T.,
Gallinat, J., Wrase, J., Heinz, A., 2008. Reward system activation in schizophrenic patients switched from typical neuroleptics to olanzapine. Psychopharmacology 196,
673684.
Schultz, W., 2013. Updating dopamine reward signals. Curr. Opin. Neurobiol. 23, 229238.
Schultz, W., Dayan, P., Montague, P.R., 1997. A neural substrate of prediction and reward.
Science 275, 15931599.
Seeman, P., Tallerico, T., 1998. Antipsychotic drugs which elicit little or no parkinsonism
bind more loosely than dopamine to brain D2 receptors, yet occupy high levels of
these receptors. Mol. Psychiatry 3, 123134.

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030

12

R.W. Morris et al. / Schizophrenia Research xxx (2015) xxxxxx

Seeman, P., Tallerico, T., 1999. Rapid release of antipsychotic drugs from dopamine D2 receptors: an explanation for low receptor occupancy and early clinical relapse upon
withdrawal of clozapine or quetiapine. Am. J. Psychiatry 156, 876884.
Shughrue, P.J., Lane, M.V., Merchenthaler, I., 1997. Comparative distribution of estrogen
receptor-alpha and -beta mRNA in the rat central nervous system. J. Comp. Neurol.
388, 507525.
Sinclair, D., Purves-Tyson, T.D., Allen, K.M., Weickert, C.S., 2014. Impacts of stress and sex
hormones on dopamine neurotransmission in the adolescent brain. Psychopharmacology 231, 15811599.
Spear, L.P., 2000. The adolescent brain and age-related behavioral manifestations.
Neurosci. Biobehav. Rev. 24, 417463.
Taherianfard, M., Shariaty, M., 2004. Evaluation of serum steroid hormones in schizophrenic patients. Indian J. Med. Sci. 58, 39.
Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N.,
Mazoyer, B., Joliot, M., 2002. Automated anatomical labeling of activations in SPM
using a macroscopic anatomical parcellation of the MNI MRI single-subject brain.
Neuroimage 15, 273289.

Vermeulen, A., Verdonck, L., Kaufman, J.M., 1999. A critical evaluation of simple methods
for the estimation of free testosterone in serum. J. Clin. Endocrinol. Metab. 84,
36663672.
Wechsler, D., 1997. Wechsler Adult Intelligence Scale. 3rd ed. The Psychological Corporation, San Antonio, Texas.
Wechsler, D., 2001. Wechsler Test of Adult Reading. The Psychological Corporation, San
Antonio, Texas.
Weickert, C.S., Miranda-Angulo, A.L., Wong, J., Perlman, W.R., Ward, S.E., Radhakrishna, V.,
Straub, R.E., Weinberger, D.R., Kleinman, J.E., 2008. Variants in the estrogen receptor
alpha gene and its mRNA contribute to risk for schizophrenia. Hum. Mol. Genet. 17,
22932309.
Weickert, C.S., Sheedy, D., Rothmond, D.A., Dedova, I., Fung, S., Garrick, T., Wong, J.,
Harding, A.J., Sivagnanansundaram, S., Hunt, C., Duncan, C., Sundqvist, N., Tsai, S.Y.,
Anand, J., Draganic, D., Harper, C., 2010. Selection of reference gene expression in a
schizophrenia brain cohort. Aust. N. Z. J. Psychiatry 44, 5970.

Please cite this article as: Morris, R.W., et al., Testosterone and reward prediction-errors in healthy men and men with schizophrenia, Schizophr.
Res. (2015), http://dx.doi.org/10.1016/j.schres.2015.06.030