Schizophrenia Bulletin vol. 45 no. 5 pp.

1092–1100, 2019
doi:10.1093/schbul/sby154
Advance Access publication 2 November 2018

Towards a Unifying Cognitive, Neurophysiological, and Computational

Neuroscience Account of Schizophrenia

Andreas Heinz1, Graham K. Murray2, Florian Schlagenhauf1,3, Philipp Sterzer1, Anthony A. Grace4–6, and

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

James A. Waltz*,7,
1
Department of Psychiatry and Psychotherapy, Charité Campus Mitte, Charité—Universitätsmedizin Berlin, Berlin, Germany;
2
Department of Psychiatry, University of Cambridge, Cambridgeshire, UK; 3Max Planck Institute for Human Cognitive and Brain
Sciences, Leipzig, Germany; 4Department of Neuroscience, University of Pittsburgh, Pittsburgh, PA; 5Department of Psychiatry,
University of Pittsburgh, Pittsburgh, PA; 6Department of Psychology, University of Pittsburgh, Pittsburgh, PA; 7Maryland Psychiatric
Research Center, Department of Psychiatry, University of Maryland School of Medicine, Baltimore, MD 21228
*To whom correspondence should be addressed;  tel: 410-402-6044, fax: 410-402-7198, e-mail: [email protected]

Psychotic experiences may be understood as altered infor- Introduction

mation processing due to aberrant neural computations.
Biological accounts of schizophrenia and related psy-
A  prominent example of such neural computations is the
chotic states have long focused on dopamine dysfunction,
computation of prediction errors (PEs), which signal the
based on the effects of antipsychotic medication and neu-
difference between expected and experienced events. Among
roimaging studies showing increased striatal dopamine
other areas showing PE coding, hippocampal-prefrontal-
synthesis capacity and release in unmedicated psychotic
striatal neurocircuits play a prominent role in information
patients.1–4 Decisive steps toward computational accounts
processing. Dysregulation of dopaminergic signaling, often
of dopamine function were taken when Schultz and
secondary to psychosocial stress, is thought to interfere with
coworkers5 showed that phasic dopamine release reflects
the processing of biologically important events (such as re-
the valence and magnitude of mismatches between ex-
ward prediction errors) and result in the aberrant attribution
pected and obtained reward outcomes (called “reward
of salience to irrelevant sensory stimuli and internal rep-
prediction errors,” or RPEs), and when Robinson and
resentations. Bayesian hierarchical predictive coding offers
Berridge6 postulated that phasic dopamine release reflects
a promising framework for the identification of dysfunc-
the attribution of incentive salience to sensory stimuli
tional neurocomputational processes and the development
and internal representations (for further early theories
of a mechanistic understanding of psychotic experience.
on altered learning and inference in schizophrenia, see
According to this framework, mismatches between prior
supplementary section 1). We and others7–9 proposed
beliefs encoded at higher levels of the cortical hierarchy and
that, in schizophrenia, delusion formation is promoted
lower-level (sensory) information can also be thought of as
by dysregulated, and therefore noisy, firing of dopamin-
PEs, with important consequences for belief updating. Low
ergic neurons, thus imbuing otherwise irrelevant stimuli
levels of precision in the representation of prior beliefs rel-
with meaning and motivating subjects to focus their
ative to sensory data, as well as dysfunctional interactions
attention on these cues—a phenomenon termed “aber-
between prior beliefs and sensory data in an ever-changing
rant salience attribution.” Increased dopamine tone may
environment, have been suggested as a general mechanism
“drown” the encoding of actually relevant stimuli, such
underlying psychotic experiences. Translating the promise
as primary reinforcers and conditioned cues associated
of the Bayesian hierarchical predictive coding into patient
with rewards.10
benefit will come from integrating this framework with ex-
Dopaminergic prediction error (PE) signaling and its
isting knowledge of the etiology and pathophysiology of
postulated cognitive correlate of salience attribution can
psychosis, especially regarding hippocampal-prefrontal-
be grounded in a more general computational framework
striatal network function and neural mechanisms of infor-
of how individuals make inferences about their environ-
mation processing and belief updating.
ment and thus shape their subjective model of the world.11
According to the theory of Bayesian inference, beliefs
Keywords:  schizophrenia/computational modeling/ are continuously updated by testing prior beliefs against
dopamine/reward/prediction error/delusions/hallucinations novel incoming information, resulting in posterior beliefs.
© The Author(s) 2018. Published by Oxford University Press on behalf of the Maryland Psychiatric Research Center.
All rights reserved. For permissions, please email: [email protected]

1092
 Computational Neurophysiology of Schizophrenia

It has been proposed that the brain uses this mechanism experience and discuss the neural circuits related to altered
of belief updating to infer the hidden states of the world information processing in psychosis.
from incomplete and noisy sensory data.12 In a Bayesian
sense, the term belief is not exclusively used for higher-
level cognition but also refers to a probability distribution Computational Models of Reinforcement Learning
over some unknown environmental or internal state and Alterations in Psychosis
may or may not be consciously accessible.13 The mean of Computational models of reinforcement learning (RL)
this distribution denotes the most expected state, whereas have been frequently applied to data from individuals
its variance denotes the uncertainty of the belief (and the performing tasks reliant on the ability to process and
inverse variance its precision). Critically, if a belief is held learn from feedback, and to decide based on representa-
with high uncertainty (ie, low precision), mismatching tions of value.11,24 While the in-the-moment experience

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

sensory information will be very effective in updating the of rewarding stimuli appears to be largely intact in the
belief. In contrast, if a belief has low uncertainty (ie, high majority of schizophrenia patients, there is solid ev-
precision), mismatching information can be ignored and idence for deficits in value-based decision making and
will have little effect on belief updating.13 altered neuronal signals during aspects of reward-based
The “predictive coding” framework has been suggested learning.25–27 In animal studies, RPEs have been shown
as one biologically plausible algorithmic implementa- to evoke phasic dopamine bursts that act as teaching
tion of Bayesian inference in the brain.14,15 This frame- signals during associative learning about rewards,28–30 as
work describes a hierarchical organization of beliefs, well as salient nonrewarding events.31 Importantly, alter-
where mismatches between prior beliefs and incoming ations in the (largely dopaminergic) signaling of RPEs
signals result in PEs, which serve as learning signals for have been proposed to underlie both the increased ten-
the updating of posterior beliefs and are propagated dency to attribute aberrant salience to irrelevant stimuli
upwards in the hierarchy from low to high levels.16 The (thought to contribute to positive symptoms7,8,10) and the
updating of lower-level beliefs by PEs is weighted by reduced ability to adaptively learn about reward value
higher-level beliefs regarding the certainty attributed to and attribute incentive salience to biologically important
a lower-level belief. The predictive coding hierarchy is stimuli and events (thought to contribute to motivational
thought to reflect the hierarchical structure of informa- impairments32,33).
tion in  the world,17 whereby different sources of uncer- Several groups have used functional magnetic reso-
tainty are computed at different levels of the hierarchy: nance imaging (fMRI) in conjunction with computa-
low-level uncertainty results from limited or noisy sen- tional modeling in a RL framework, point to a disruption
sory information, whereas high-level uncertainty results in midbrain RPE signaling in acutely psychotic individu-
from unpredictable state changes over time, ie, environ- als,34–38 which has also been observed in conjunction with
mental volatility.13,18 This hierarchical structure in un- the encoding of informative errors in the ventral striatum
certainty computation is supposedly represented in the during reversal learning.39 Given the evidence for dopa-
sensory cortices for lower-level processing and in associa- minergic dysfunction in psychosis,2,3,40 a link between
tive cortices for higher-level processing.18,19 alterations in dopamine function and RPE signaling in
In schizophrenia, an altered balance between the encod- psychotic illness is very plausible. Indeed, one study in
ing of prior beliefs and sensory information processing healthy volunteers found an inverse association between
may result in maladaptive PE signaling, not only in dopa- dopamine synthesis capacity and the strength of stri-
mine-dependent subcortical circuits, but also in cortical atal RPE signals.41 Additional fMRI studies have dem-
brain circuits.16,20,21 In a Bayesian framework,12 this im- onstrated that individuals with schizophrenia (especially
pairment can be formalized as a relative imprecision in the those with more severe negative symptoms and/or deficits
signaling of prior beliefs vis-a-vis input signals emanating in intellectual function)42–44 exhibit attenuations in ven-
from low-level sensory cortices, leading to a kind of aber- tral striatal activation during reward anticipation, relative
rant PE signaling, apart from dopamine-dependent sub- to healthy controls. Such alterations in neuronal learning
cortical RPE signaling. Such erratic PE signaling could signals can help to explain previously observed RL defi-
impair the ability to distinguish between relevant and ir- cits in schizophrenia patients.45–47
relevant sensory information, resulting in maladaptive Recent Bayesian accounts of learning and inference
inferences and belief-updating. Maladaptive inferences provide a computational account of dopamine that goes
may arise from aberrant encoding of the precision of prior beyond signaling the magnitude and valence of RPEs,
beliefs at different levels of the predictive coding hierarchy. postulating that dopamine plays an important role in
However, such accounts need to be integrated with existing belief updating by encoding the precision of PEs.48–51
knowledge regarding hippocampal-prefrontal-striatal net- In short, the effects of PEs on learning are weighted by
work dysfunction in schizophrenia.22,23 In this review, we the belief about the current precision of the learning
focus on Bayesian predictive coding accounts of psychotic signal.52,53 This probabilistic belief is thought to reflect

1093
A. Heinz et al

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

Fig. 1.  Circuit model of frontal and hippocampal control of dopamine neuron firing. The ventral hippocampus exerts potent control
over dopamine neurons firing spontaneously. The number firing determines the amplitude, and hence salience, of the signal. Dopamine
neurons are normally inhibited by the ventral pallidum, which in turn is inhibited by the nucleus accumbens. When the ventral
hippocampus is activated, it activates the nucleus accumbens, which in turn inhibits the ventral pallidum and releases dopamine neurons
from inhibition, allowing them to initiate firing.55 The ventral hippocampus is potently regulated by the PFC via the thalamus. The
(infralimbic) PFC normally holds the ventral hippocampus in a less-active state. However, when infralimbic PFC activity is decreased,
the primary effect is deactivation of the reticular nucleus of the thalamus, which in turn disinhibits the thalamic nucleus reuniens. This
increases the tonic excitatory drive of the nucleus reuniens on the ventral hippocampus, disinhibiting dopamine firing, which impacts
cognitive control via the associative striatum.56 The human analogue of the infralimbic cortex is the subgenual cingulate area 25. The
arrows from the VTA to cortical and subcortical regions denote modulatory dopaminergic projections.

prior assumptions about informational uncertainty and exploration have been found to contribute to impairments
is, in turn, modulated by higher-level beliefs regarding in learning and motivation in schizophrenia patients.25,47
the volatility of the environment, encoded at higher Further evidence points to roles for glutamater-
levels of neural computation including the prefrontal gic neurotransmission (eg, via NMDA receptors) and
cortex (PFC) and hippocampus (figure  1).54 A  computa- GABAergic inhibition in encoding the precision of learn-
tional framework that models this hierarchical structure ing signals in hippocampal and cortical networks (fig-
of belief updating is the Hierarchical Gaussian Filter13 ure 1). While the signaling of precision-weighted PEs has
(figure  2). This approach extends RL accounts and pro- not been investigated in schizophrenia patients thus far,
vides a broader framework that goes beyond learning of the administration of the NMDA receptor antagonist
beliefs about reward-related value expectation. Using this ketamine, to healthy controls, was found to be associated
framework, a recent model-based fMRI study of associ- with both a reduced ability to use confidence (precision)
ative audiovisual learning found that high-level beliefs estimates to regulate RL parameters and altered fronto-
regarding the strength of dynamic cue-target contingencies parietal activity.62
relied on hippocampus and orbitofrontal cortex, whereas Finally, dopamine has also been implicated in hippo-
the representations of low-level conditional target prob- campus-dependent novelty detection and novelty seeking
abilities was associated with early visual cortex activity.19 behavior in both humans and rodents.54,63–65 Of course,
The idea that dopaminergic transmission contributes merely because a stimulus is “novel” does not necessarily
to the signaling of the precision of learning signals also mean that it would be involved in belief updating. Rather,
suggests a mechanism by which dopamine may influence it has been proposed that a stimulus must be both novel
higher-level decision making. Specifically, representations and salient to impact belief, and that prefrontal-hippo-
of uncertainty have been shown to drive exploratory be- campal dynamics thought to underlie novelty detection
havior.58–60 It is therefore noteworthy that dopamine interact with dopaminergic salience signals16,54 to bring
synthesis capacity in the striatum has been found to be about relevant memory incorporation.54
associated with the extent to which participants used a Regarding neurobiological correlates of the symptoms
goal-directed learning strategy, in the context of a com- of psychotic illness, positive symptoms are thought to be
plex cognitive task,61 and that deficits in goal-directed mediated by dopamine hyper-responsivity, with cognitive
1094
 Computational Neurophysiology of Schizophrenia

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

Fig. 2.  (A) Schematic model of altered hierarchical inference in the visual system. Sensory input represents processing in early visual
cortex. Low-level “sensory” beliefs are encoded at the next higher hierarchical level, eg, mid- or high-level visual areas, and high-
level “conceptual” beliefs at the highest cortical levels, eg, PFC. Arrows represent top-down signaling of prior beliefs and bottom-up
signaling of prediction errors (PEs), with arrow thickness representing their respective precisions. The putative decrease in precision
of low-level beliefs may lead to increased weighting of the sensory input, thus enhancing PEs, potentially compensated by increased
precision of conceptual high-level beliefs.57 Brain image courtesy of Flickr/IsaacMao. (B) Schematic representation of the Hierarchical
Gaussian Filter (adapted from Mathys et al13). Levels x1(k) , x2(k) , x3(k) represent hidden environmental states at time k. They depend on their
immediately preceding values x2(k −1) , x3(k −1) and on the parameters κ (coupling of levels 2 and 3), ω (step size at level 2), and ϑ (learning
speed about environmental volatility). The probability at each level is determined by the variables and parameters at the level above. The
levels relate to each other by determining the step size of a random walk. Note that the levels of this model are not equivalent to those
in (A). However, reduced learning speed about environmental volatility might contribute to a stronger top-down influence of high-level
beliefs.

symptoms arising via frontal cortical dysfunction, and symptoms—possibly via elevated striatal dopamine syn-
negative symptoms via impaired function of fronto- thesis72—especially in those with elevated genetic liability
striatal circuits, cingulate cortex, and the amygdalae.66,67 for, or social vulnerability to, psychotic illness.73–75 As yet,
While decades of research have revealed that dopamine no studies of individuals at elevated genetic or environ-
can have different actions depending on its projection mental risk for psychosis have examined the precision in
site,54,68,69 one central aspect of dopamine function seems encoding of prior beliefs, or influence of priors on deci-
to be the signaling of salience or, in Bayesian terms, of sion making. Regardless of whether they are driven by
the precision during information processing. psychosocial or primary neurobiological factors, mal-
adaptive inferences related to sensory and high-level
processing may ultimately implicate the same circuits in-
Sensory Information Processing and Its Association volved in the comparison of prior beliefs and new sensory
With the Formation and Maintenance of Delusions information. Such neural circuits include temporo-limbic
Insufficient precision in the encoding of low-level prior brain regions involved in executive control and stress
beliefs relative to sensory input could occur either due to reactivity.23,54
the faulty acquisition of prior beliefs, or due to a reduced Importantly, different levels of neural processing may
ability to appropriately use prior beliefs in perceptual be differentially involved in false inferences underlying
inference. The acquisition of prior beliefs could be faulty psychotic experience; Schmack and coworkers76 observed
if sensory information is misleading, if the detection and/ that delusional ideation in both healthy volunteers and
or encoding of relevant sensory information are per- schizophrenia patients is associated with reduced per-
turbed (or excessively stochastic),54 or if the detection ceptual stability during the viewing of ambiguous visual
and/or computation of higher-level PEs are perturbed or stimuli that represent informational uncertainty,76,77 which
insufficiently precise.16,21 A  reduced ability to appropri- points to a weaker influence of perceptual priors built up
ately use prior beliefs in perceptual inference could occur at previous encounters with the same visual information.
if high-level beliefs include false assumptions regarding This observation is in line with the well-known finding of
the volatility of a novel environment. reduced susceptibility to some visual illusions in schizo-
These considerations are consistent with the view that phrenia, pointing to reduced precision of prior beliefs at
both genetic and psychosocial factors contribute to the low hierarchical levels.21,78 Interestingly, the finding of a
development of psychosis. While it is still necessary to decreased influence of such lower-level perceptual beliefs
differentiate among the various forms of stress, a mecha- was accompanied by an increased influence of higher-level
nistic understanding of stress effects on neural circuits is cognitive beliefs on sensory processing. Belief-related con-
emerging, indicating, eg, that stress can elevate microglial nectivity between regions encoding high-level beliefs in
activity in hippocampus70 and increase striatal dopamine the orbitofrontal cortex and visual brain areas encoding
release.71 Experiences of both acute and early-life stress visual motion perception was increased in delusion-prone
have been shown to engender and exacerbate psychotic individuals and schizophrenia patients.76,79 Accordingly,
1095
A. Heinz et al

reduced precision of perceptual beliefs encoded at low the cell body of the pyramidal neurons and have greater
levels (eg, in sensory cortices) may be compensated by influence on overall activity.91 Compared with the total
increased precision of more abstract conceptual beliefs GABAergic interneuron population, the number of
encoded in higher-level brain circuits (figure 2).79 In this PV interneurons is small, so that measuring overall
case, such a disambiguating top-down signal may reflect GABA concentrations will not give a precise picture of
a belief in low volatility of the environment, as previ- excitability.
ously suggested.16 In the context of social interactions, Findings of systematic relationships between gluta-
an individual may falsely categorize ambivalent social matergic concentrations in brain regions of interest and
interactions as threatening if informational uncertainty PET measures of dopamine synthesis92 suggest that per-
regarding social cues is wrongly disambiguated by delu- turbations of excitatory glutamatergic neurotransmission
sion-congruent top-down signals (thereby contributing contribute to dopamine system abnormalities observed in

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

to the stability of delusions). psychosis patients, in accordance with preclinical studies
At this point, our understanding of the role of dopa- showing that activation of hippocampal pyramidal neu-
minergic modulation of (and by) fronto-striatal circuits rons drive dopamine neuron activity states.93,94 Within a
comes back into view.22,80 In the following section, we Bayesian framework, alterations in glutamatergic neu-
review a series of studies suggesting that aberrations in rotransmission as well as the balance between excitatory
glutamatergic and dopaminergic signaling emerge from glutamatergic and inhibitory GABAergic signal trans-
dysfunctional interactions among temporo-limbic, pre- duction in the temporo-limbic and PFC may be just spe-
frontal, and striatal brain areas, and consider how these cific correlates of a more widespread alteration in the
empirical findings relate to the theoretical framework of precision of information processing. In this context, vol-
Bayesian predictive coding. atility may be underestimated at high hierarchical levels
(see supplementary material; Jardri et al95).
We argue that genetic- and stress-dependent vulnera-
Reverberating Circuits and Temporal Limbic-Cortical bility factors can contribute to psychosis by shifting the
Dysfunction in Psychosis balance between the respective precisions of prior beliefs
Despite genetic associations between dopamine receptor and sensory evidence. In people with these risk factors,
variants and schizophrenia risk,81 even the most ardent slight alterations in the balance between excitatory and
proponents of the dopamine hypothesis of schizophrenia inhibitory neurotransmission may result in overall nonse-
would not suggest that schizophrenia involves a primary lective information processing and reduced precision of
deficit in dopamine dysfunction akin to that characteriz- prior beliefs, thus increasing the relative effects of sen-
ing Parkinson’s disease. Rather, it has been proposed that sory inputs, which in turn results in a shift of posteriors
fronto-striatal circuits may be disrupted in schizophre- toward the sensory evidence and enhanced precision-
nia, thus putting a focus on dopamine–glutamate inter- weighted PEs. As a result, irrelevant information could
actions1,22,23 (see supplementary section 2.1). have a greater tendency to be perceived as overly salient,
Preclinical studies suggest that interactions of tem- leading to an impaired distinction between relevant and
poro-limbic brain areas with PFC control regions second- irrelevant stimuli. This could be tested by examining rele-
arily disinhibit subcortical dopamine release, potentially vant neurocognitive and computational measures in indi-
in a “compensatory effort” to increase the signal-to-noise viduals at elevated risk for schizophrenia. These processes
ratio when confronted with noisy information process- can strongly involve brain areas associated with novelty
ing.10,82–84 Furthermore, a series of studies85,86 of temporo- detection including the hippocampus,23,54 which is critical
limbic-prefrontal interactions has focused on the balance in regulating dopamine neuronal activity.94 Furthermore,
of excitatory glutamatergic and inhibitory GABAergic all of the regions implicated in the expression of the
neurotransmission. Complementary fMRI and magnetic symptoms of psychosis (eg, fronto-striatal circuits, cingu-
resonance spectroscopy (MRS) studies86–88 suggest that late cortex, and the amygdalae) receive innervation from
abnormal GABAergic transmission can lead to elevated the limbic hippocampus.96 Thus, dysfunctional informa-
hippocampal and frontocortical glutamate concentra- tion processing in the limbic hippocampus has far-reach-
tions as well as hippocampal hyperactivity in schizophre- ing consequences, necessarily disrupting related circuits
nia patients. The most recent meta-analysis of glutamate in the cerebral cortex and influencing volatility estimates
MRS studies89 reported elevations in glutamatergic at the highest levels.79,97,98
metabolites in the medial frontal cortex in individuals at Stress-dependent and/or chaotic dopamine firing can
high risk for schizophrenia, and in the medial temporal lead to noisy signaling of common primary and sec-
lobes and basal ganglia of schizophrenia patients (but ondary reinforcers by augmenting otherwise irrelevant
see supplementary section 2.2; Schür et al90). In preclin- signals: in the frontal cortex and ventral striatum, dys-
ical models, and postmortem studies, of schizophrenia, regulated dopamine release may blunt signals elicited
a consistent loss of parvalbumin (PV) GABAergic inter- by the surprising manifestation of cues associated with
neurons is found in schizophrenia, which are located at positive reinforcers and thus contribute to motivational
1096
 Computational Neurophysiology of Schizophrenia

deficits99 (see supplementary section 2.3). On the other corresponding to the magnitude, valence, and precision
hand, when phasic dopamine release is associated with of predictions and violations thereof. This would allow
stressful or chaotically coinciding stimuli, salience can us to assess relationships among molecular measures of
be attributed to these otherwise random co-occurrences, neurochemical concentration and function and measures
thus contributing to delusional perceptions and delusion of neural circuit function.
formation.26,100 Bayesian posteriors may be further shifted Thirdly, Bayesian models of decision making should
toward sensory input by circular inference, thus amplify- make increasing use of more formalized approaches
ing certain information aspects at the expense of others to dynamic interactions between brain areas, such as
(supplementary section 3). dynamic causal modeling and laminar neuroimaging
methods. Dynamic causal modeling has already helped
to identify subgroups of schizophrenia patients based on

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

Summary and Consequences for Future Research
the severity of cognitive impairments and negative symp-
In sum, these findings suggest that, beyond simple toms,111,112 while laminar neuroimaging methods may
effects of sensory priors on the processing of sensory provide more precise assays of top-down and bottom-up
input, there are complex interactions in psychotic illness information flow.113,114
between low-level information processing in sensory cor- Finally, more studies should attempt to integrate com-
tices and information processing at higher levels of the putational neuropsychiatry methods investigating the
predictive coding hierarchy. We argue that such interac- pathophysiology of psychosis with advances in identi-
tions of higher-level with lower-level beliefs might be a fication of genetic and environmental risk factors for
general and important mechanism underlying psychotic schizophrenia to better understand how, why, and when
experiences in a volatile environment. psychopathology emerges.115 In the larger picture, compu-
The framework of Bayesian inference offers a promis- tational modeling—specifically Bayesian models of pre-
ing approach for developing computational models that dictive coding—offer the possibility to link various, thus
account for psychotic experiences by capturing exam- far largely separate, strains of research focused on dif-
ples of nonselective information processing and failed ferent aspects of psychotic experience, such as aberrant
attempts to deal with a lack of precision. Mathematical salience attribution in delusion formation, the tenacity of
models, particularly when constrained by biological plau- delusions, and the development of impaired self-ascrip-
sibility, offer the possibility of identifying neurocomputa- tion of thoughts and actions.
tional steps that the individual has to take when solving
a certain task (eg, the computation of PEs101) and the Supplementary Material
neural signals associate these steps. Importantly, the idea
of nonselective information processing within a Bayesian Supplementary data are available at Schizophrenia
framework, as a consequence of disrupted temporo-lim- Bulletin online.
bic and prefrontal-striatal interactions, might account
for a wide range of psychotic experiences, including self-
Funding
disorders related to the initiation and performance of
action as well as the self-attribution of intentions and This work was supported by the German Research
thoughts.102–104 Beyond RL, Bayesian approaches can Foundation (SCHL 1969/1-2/3-1 and 4-1 to F.S.) and the
be applied to a range of cognitive and perceptual pro- US National Institutes of Health (5R01MH094460 to
cesses, such as visual information processing,78,105,106 reli- J.A.W. and MH057440 to A.A.G.).
ant on frontal-occipital connectivity.76 To further develop
a mechanistic understanding of the emergence of the
symptoms of psychotic illness, 4 steps are recommended. References
First, future research should place greater emphasis on
auditory information processing. Even though auditory 1. Carlsson A, Waters N, Carlsson ML. Neurotransmitter
interactions in schizophrenia—therapeutic implications. Biol
hallucinations clearly occur outside of psychotic illness, Psychiatry. 1999;46:1388–1395.
complex auditory hallucinations figure prominently in 2. Laruelle M, Abi-Dargham A, van Dyck CH, et  al. Single
psychotic illness,107–109 and substantial evidence points to photon emission computerized tomography imaging of
predictive coding alterations in relation to auditory proc- amphetamine-induced dopamine release in drug-free schizo-
essing in schizophrenia.98,110 phrenic subjects. Proc Natl Acad Sci USA. 1996;93:9235–9240.
Secondly, we strongly recommend employing multi- 3. Howes OD, Kambeitz J, Kim E, et al. The nature of dopa-
modal imaging approaches, including the combination mine dysfunction in schizophrenia and what this means for
treatment. Arch Gen Psychiatry. 2012;69:776–786.
of MRS assessment of excitatory and inhibitory neu-
4. Jauhar S, Nour MM, Veronese M, et al. A test of the trans-
rotransmitters and their metabolites with PET studies diagnostic dopamine hypothesis of psychosis using positron
of monoaminergic neurotransmission and fMRI, EEG, emission tomographic imaging in bipolar affective disorder
or magnetoencephalographic measures of neural signals and schizophrenia. JAMA Psychiatry. 2017;74:1206–1213.

1097
A. Heinz et al

5. Schultz W, Dayan P, Montague PR. A neural substrate of 27. Cohen AS, Minor KS. Emotional experience in patients with
prediction and reward. Science. 1997;275:1593–1599. schizophrenia revisited: meta-analysis of laboratory studies.
6. Robinson TE, Berridge KC. The neural basis of drug crav- Schizophr Bull. 2010;36:143–150.
ing: an incentive-sensitization theory of addiction. Brain Res 28. Schultz W. Updating dopamine reward signals. Curr Opin
Brain Res Rev. 1993;18:247–291. Neurobiol. 2013;23:229–238.
7. Heinz A. Dopaminergic dysfunction in alcoholism and 29. Steinberg EE, Keiflin R, Boivin JR, Witten IB, Deisseroth K,
schizophrenia—psychopathological and behavioral corre- Janak PH. A causal link between prediction errors, dopamine
lates. Eur Psychiatry. 2002;17:9–16. neurons and learning. Nat Neurosci. 2013;16:966–973.
8. Kapur S. Psychosis as a state of aberrant salience: a frame- 30. Eshel N, Tian J, Bukwich M, Uchida N. Dopamine neurons
work linking biology, phenomenology, and pharmacology in share common response function for reward prediction error.
schizophrenia. Am J Psychiatry. 2003;160:13–23. Nat Neurosci. 2016;19:479–486.
9. Miller R. Schizophrenic psychology, associative learn- 31. Sharpe MJ, Chang CY, Liu MA, et al. Dopamine transients

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

ing and the role of forebrain dopamine. Med Hypotheses. are sufficient and necessary for acquisition of model-based
1976;2:203–211. associations. Nat Neurosci. 2017;20:735–742.
10. Heinz A, Schlagenhauf F. Dopaminergic dysfunction in 32. Waltz JA, Gold JM. Motivational deficits in schizophrenia
schizophrenia: salience attribution revisited. Schizophr Bull. and the representation of expected value. Curr Top Behav
2010;36:472–485. Neurosci. 2016;27:375–410.
11. Maia TV, Frank MJ. An integrative perspective on the role of 33. McClure SM, Daw ND, Montague PR. A computa-
dopamine in schizophrenia. Biol Psychiatry. 2017;81:52–66. tional substrate for incentive salience. Trends Neurosci.
12. Friston KJ, Stephan KE, Montague R, Dolan RJ. 2003;26:423–428.
Computational psychiatry: the brain as a phantastic organ. 34. Murray GK, Corlett PR, Clark L, et al. Substantia nigra/ven-
Lancet Psychiatry. 2014;1:148–158. tral tegmental reward prediction error disruption in psych-
13. Mathys CD, Lomakina EI, Daunizeau J, et al. Uncertainty in osis. Mol Psychiatry. 2008;13:239, 267–276.
perception and the Hierarchical Gaussian Filter. Front Hum
35. Corlett PR, Murray GK, Honey GD, et  al. Disrupted pre-
Neurosci. 2014;8:825.
diction-error signal in psychosis: evidence for an associative
14. Aitchison L, Lengyel M. With or without you: predictive cod- account of delusions. Brain. 2007;130:2387–2400.
ing and Bayesian inference in the brain. Curr Opin Neurobiol.
2017;46:219–227. 36. Schlagenhauf F, Sterzer P, Schmack K, et  al. Reward feed-
back alterations in unmedicated schizophrenia patients: rele-
15. Valton V, Romaniuk L, Douglas Steele J, Lawrie S, Seriès P. vance for delusions. Biol Psychiatry. 2009;65:1032–1039.
Comprehensive review: computational modelling of schizo-
phrenia. Neurosci Biobehav Rev. 2017;83:631–646. 37. Romaniuk L, Honey GD, King JR, et al. Midbrain acti-
vation during Pavlovian conditioning and delusional
16. Adams RA, Huys QJ, Roiser JP. Computational psychiatry:
symptoms in schizophrenia. Arch Gen Psychiatry.
towards a mathematically informed understanding of mental
2010;67:1246–1254.
illness. J Neurol Neurosurg Psychiatry. 2016;87:53–63.
17. Friston K. A theory of cortical responses. Philos Trans R Soc 38. Ermakova AO, Knolle F, Justicia A, et al. Abnormal reward
Lond B Biol Sci. 2005;360:815–836. prediction-error signalling in antipsychotic naive individu-
als with first-episode psychosis or clinical risk for psychosis.
18. Payzan-LeNestour E, Bossaerts P. Risk, unexpected uncer- Neuropsychopharmacology. 2018;43:1691–1699.
tainty, and estimation uncertainty: Bayesian learning in un-
stable settings. PLoS Comput Biol. 2011;7:e1001048. 39. Schlagenhauf F, Huys QJ, Deserno L, et al. Striatal dysfunc-
tion during reversal learning in unmedicated schizophrenia
19. Weilnhammer VA, Stuke H, Sterzer P, Schmack K. The
patients. Neuroimage. 2014;89:171–180.
neural correlates of hierarchical predictions for perceptual
decisions. J Neurosci. 2018;38:5008–5021. 40. Howes OD, McCutcheon R, Owen MJ, Murray RM. The
20. Fletcher PC, Frith CD. Perceiving is believing: a Bayesian role of genes, stress, and dopamine in the development of
approach to explaining the positive symptoms of schizo- schizophrenia. Biol Psychiatry. 2017;81:9–20.
phrenia. Nat Rev Neurosci. 2009;10:48–58. 41. Schlagenhauf F, Rapp MA, Huys QJ, et  al. Ventral striatal
21. Adams RA, Stephan KE, Brown HR, Frith CD, Friston KJ. prediction error signaling is associated with dopamine syn-
The computational anatomy of psychosis. Front Psychiatry. thesis capacity and fluid intelligence. Hum Brain Mapp.
2013;4:47. 2013;34:1490–1499.
22. Weinberger DR. Implications of normal brain development 42. Reddy LF, Waltz JA, Green MF, Wynn JK, Horan WP.
for the pathogenesis of schizophrenia. Arch Gen Psychiatry. Probabilistic reversal learning in schizophrenia: stability of
1987;44:660–669. deficits and potential causal mechanisms. Schizophr Bull.
23. Meyer-Lindenberg A. From maps to mechanisms through 2016;42:942–951.
neuroimaging of schizophrenia. Nature. 2010;468:194–202. 43. Culbreth AJ, Westbrook A, Xu Z, Barch DM, Waltz JA.
24. Deserno L, Boehme R, Heinz A, Schlagenhauf F. Intact ventral striatal prediction error signaling in medi-
Reinforcement learning and dopamine in schizophrenia: cated schizophrenia patients. Biol Psychiatry Cogn Neurosci
dimensions of symptoms or specific features of a disease Neuroimaging. 2016;1:474–483.
group? Front Psychiatry. 2013;4:172. 44. Dowd EC, Frank MJ, Collins A, Gold JM, Barch DM.
25. Strauss GP, Cohen AS. A transdiagnostic review of nega- Probabilistic reinforcement learning in patients with schizo-
tive symptom phenomenology and etiology. Schizophr Bull. phrenia: relationships to anhedonia and avolition. Biol
2017;43:712–719. Psychiatry Cogn Neurosci Neuroimaging. 2016;1:460–473.
26. Pankow A, Friedel E, Sterzer P, et al. Altered amygdala acti- 45. Waltz JA, Gold JM. Probabilistic reversal learning impair-
vation in schizophrenia patients during emotion processing. ments in schizophrenia: further evidence of orbitofrontal
Schizophr Res. 2013;150:101–106. dysfunction. Schizophr Res. 2007;93:296–303.

1098
 Computational Neurophysiology of Schizophrenia

46. Murray GK, Cheng F, Clark L, et  al. Reinforcement and 64. Wittmann BC, Daw ND, Seymour B, Dolan RJ. Striatal ac-
reversal learning in first-episode psychosis. Schizophr Bull. tivity underlies novelty-based choice in humans. Neuron.
2008;34:848–855. 2008;58:967–973.
47. Waltz JA. The neural underpinnings of cognitive flexi- 65. Tobler PN, O’Doherty JP, Dolan RJ, Schultz W. Reward value
bility and their disruption in psychotic illness. Neuroscience. coding distinct from risk attitude-related uncertainty coding
2017;345:203–217. in human reward systems. J Neurophysiol. 2007;97:1621–1632.
48. Schwartenbeck P, FitzGerald TH, Mathys C, et al. Optimal 66. O’Donnell P, Grace AA. Dysfunctions in multiple interre-
inference with suboptimal models: addiction and active lated systems as the neurobiological bases of schizophrenic
Bayesian inference. Med Hypotheses. 2015;84:109–117. symptom clusters. Schizophr Bull. 1998;24:267–283.
49. Friston K, Schwartenbeck P, FitzGerald T, Moutoussis M, 67. O’Donnell P, Grace AA. Synaptic interactions among ex-
Behrens T, Dolan RJ. The anatomy of choice: dopamine citatory afferents to nucleus accumbens neurons: hip-
and decision-making. Philos Trans R Soc Lond B Biol Sci. pocampal gating of prefrontal cortical input. J Neurosci.

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

2014;369: 20130481. 1995;15:3622–3639.
50. Iglesias S, Tomiello S, Schneebeli M, Stephan KE. Models 68. Morales M, Margolis EB. Ventral tegmental area: cellular
of neuromodulation for computational psychiatry. Wiley heterogeneity, connectivity and behaviour. Nat Rev Neurosci.
Interdiscip Rev Cogn Sci. 2016. 2017;18:73–85.
51. Haarsma J, Fletcher PC, Ziauddeen H, Spencer TJ, Diederen 69. Schultz W. Dopamine reward prediction-error signalling: a
KMJ, Murray GK. Precision weighting of cortical unsigned two-component response. Nat Rev Neurosci. 2016;17:183–195.
prediction errors is mediated by dopamine and benefits learn- 70. Calcia MA, Bonsall DR, Bloomfield PS, Selvaraj S,
ing. bioRxiv. 2018. Barichello T, Howes OD. Stress and neuroinflammation: a
52. Mathys C, Daunizeau J, Friston KJ, Stephan KE. A Bayesian systematic review of the effects of stress on microglia and the
foundation for individual learning under uncertainty. Front implications for mental illness. Psychopharmacology (Berl).
Hum Neurosci. 2011;5:39. 2016;233:1637–1650.
53. Iglesias S, Mathys C, Brodersen KH, et al. Hierarchical pre- 71. Vaessen T, Hernaus D, Myin-Germeys I, van Amelsvoort T.
diction errors in midbrain and basal forebrain during sensory The dopaminergic response to acute stress in health and psy-
learning. Neuron. 2013;80:519–530. chopathology: a systematic review. Neurosci Biobehav Rev.
54. Lisman JE, Grace AA. The hippocampal-VTA loop: control- 2015;56:241–251.
ling the entry of information into long-term memory. Neuron. 72. Egerton A, Howes OD, Houle S, et al. Elevated striatal dopa-
2005;46:703–713. mine function in immigrants and their children: a risk mech-
55. Floresco SB, West AR, Ash B, Moore H, Grace AA. Afferent anism for psychosis. Schizophr Bull. 2017;43:293–301.
modulation of dopamine neuron firing differentially regu- 73. Myin-Germeys I, van Os J. Stress-reactivity in psychosis: evi-
lates tonic and phasic dopamine transmission. Nat Neurosci. dence for an affective pathway to psychosis. Clin Psychol Rev.
2003;6:968–973. 2007;27:409–424.
56. Zimmerman EC, Grace AA. The nucleus reuniens of the 74. Veling W. Ethnic minority position and risk for psychotic dis-
midline thalamus gates prefrontal-hippocampal modula- orders. Curr Opin Psychiatry. 2013;26:166–171.
tion of ventral tegmental area dopamine neuron activity. J 75. Cantor-Graae E, Selten JP. Schizophrenia and migration: a
Neurosci. 2016;36:8977–8984. meta-analysis and review. Am J Psychiatry. 2005;162:12–24.
57. Jardri R, Hugdahl K, Hughes M, et  al. Are hallucinations 76. Schmack K, Gòmez-Carrillo de Castro A, Rothkirch M,
due to an imbalance between excitatory and inhibitory influ- et al. Delusions and the role of beliefs in perceptual inference.
ences on the brain? Schizophr Bull. 2016;42:1124–1134. J Neurosci. 2013;33:13701–13712.
58. Behrens TE, Woolrich MW, Walton ME, Rushworth MF. 77. Schmack K, Schnack A, Priller J, Sterzer P. Perceptual in-
Learning the value of information in an uncertain world. Nat stability in schizophrenia: probing predictive coding accounts
Neurosci. 2007;10:1214–1221. of delusions with ambiguous stimuli. Schizophr Res Cogn.
59. Krugel LK, Biele G, Mohr PN, Li SC, Heekeren HR. Genetic 2015;2:72–77.
variation in dopaminergic neuromodulation influences the 78. Notredame CE, Pins D, Deneve S, Jardri R. What visual illu-
ability to rapidly and flexibly adapt decisions. Proc Natl Acad sions teach us about schizophrenia. Front Integr Neurosci.
Sci USA. 2009;106:17951–17956. 2014;8:63.
60. Friston KJ, Shiner T, FitzGerald T, et  al. Dopamine, 79. Schmack K, Rothkirch M, Priller J, Sterzer P. Enhanced
affordance and active inference. PLoS Comput Biol. predictive signalling in schizophrenia. Hum Brain Mapp.
2012;8:e1002327. 2017;38:1767–1779.
61. Deserno L, Huys QJ, Boehme R, et al. Ventral striatal dopa- 80. Alexander GE, DeLong MR, Strick PL. Parallel organiza-
mine reflects behavioral and neural signatures of model- tion of functionally segregated circuits linking basal ganglia
based control during sequential decision making. Proc Natl and cortex. Annu Rev Neurosci. 1986;9:357–381.
Acad Sci USA. 2015;112:1595–1600. 81. Ripke S, O’Dushlaine C, Chambert K, et  al.; Multicenter
62. Vinckier F, Gaillard R, Palminteri S, et  al. Confidence and Genetic Studies of Schizophrenia Consortium; Psychosis
psychosis: a neuro-computational account of contingency Endophenotypes International Consortium; Wellcome Trust
learning disruption by NMDA blockade. Mol Psychiatry. Case Control Consortium 2. Genome-wide association ana-
2016;21:946–955. lysis identifies 13 new risk loci for schizophrenia. Nat Genet.
63. Salamone JD, Correa M, Farrar A, Mingote SM. Effort- 2013;45:1150–1159.
related functions of nucleus accumbens dopamine and 82. Winterer G, Ziller M, Dorn H, et al. Schizophrenia: reduced
associated forebrain circuits. Psychopharmacology (Berl). signal-to-noise ratio and impaired phase-locking during in-
2007;191:461–482. formation processing. Clin Neurophysiol. 2000;111:837–849.

1099
A. Heinz et al

83. Winterer G, Coppola R, Goldberg TE, et  al. Prefrontal 98. Powers AR, Mathys C, Corlett PR. Pavlovian conditioning-
broadband noise, working memory, and genetic risk for induced hallucinations result from overweighting of percep-
schizophrenia. Am J Psychiatry. 2004;161:490–500. tual priors. Science. 2017;357:596–600.
84. Abi-Dargham A, Mawlawi O, Lombardo I, et al. Prefrontal 99. Juckel G, Schlagenhauf F, Koslowski M, et al. Dysfunction
dopamine D1 receptors and working memory in schizo- of ventral striatal reward prediction in schizophrenia.
phrenia. J Neurosci. 2002;22:3708–3719. Neuroimage. 2006;29:409–416.
85. Gonzalez-Burgos G, Lewis DA. NMDA receptor hypofunc- 100. Kienast T, Hariri AR, Schlagenhauf F, et  al. Dopamine
tion, parvalbumin-positive neurons, and cortical gamma oscil- in amygdala gates limbic processing of aversive stimuli in
lations in schizophrenia. Schizophr Bull. 2012;38:950–957. humans. Nat Neurosci. 2008;11:1381–1382.
86. Heckers S, Konradi C. GABAergic mechanisms of hip- 101. Montague PR, Hyman SE, Cohen JD. Computational roles for
pocampal hyperactivity in schizophrenia. Schizophr Res. dopamine in behavioural control. Nature. 2004;431:760–767.
2015;167:4–11. 102. Frith CD. The Cognitive Neuropsychology of Schizophrenia.

Downloaded from https://academic.oup.com/schizophreniabulletin/article/45/5/1092/5155538 by guest on 03 July 2023

87. Olbrich HM, Valerius G, Rüsch N, et al. Frontolimbic glu- Hillsdale, NJ: Lawrence Erlbaum Associates, Inc; 1992.
tamate alterations in first episode schizophrenia: evidence 103. Sterzer P, Mishara AL, Voss M, Heinz A. Thought insertion as
from a magnetic resonance spectroscopy study. World J Biol a self-disturbance: an integration of predictive coding and phe-
Psychiatry. 2008;9:59–63. nomenological approaches. Front Hum Neurosci. 2016;10:502.
88. van Elst LT, Valerius G, Büchert M, et al. Increased prefrontal 104. Voss M, Moore J, Hauser M, Gallinat J, Heinz A, Haggard P.
and hippocampal glutamate concentration in schizophrenia: Altered awareness of action in schizophrenia: a specific deficit
evidence from a magnetic resonance spectroscopy study. Biol in predicting action consequences. Brain. 2010;133:3104–3112.
Psychiatry. 2005;58:724–730. 105. Seymour K, Stein T, Sanders LL, Guggenmos M, Theophil
89. Merritt K, Egerton A, Kempton MJ, Taylor MJ, McGuire I, Sterzer P. Altered contextual modulation of primary visual
PK. Nature of glutamate alterations in schizophrenia: a cortex responses in schizophrenia. Neuropsychopharmacology.
meta-analysis of proton magnetic resonance spectroscopy 2013;38:2607–2612.
studies. JAMA Psychiatry. 2016;73:665–674. 106. Anderson EJ, Tibber MS, Schwarzkopf DS, et  al. Visual
90. Schür RR, Draisma LW, Wijnen JP, et al. Brain GABA lev- population receptive fields in people with schizophrenia have
els across psychiatric disorders: a systematic literature review reduced inhibitory surrounds. J Neurosci. 2017;37:1546–1556.
and meta-analysis of (1) H-MRS studies. Hum Brain Mapp.
107. Schneider K. Psychischer Befund und Psychiatrische Diagnose.
2016;37:3337–3352.
Leipzig: Georg Thieme; 1942.
91. Modinos G, Simsek F, Azis M, et al. Prefrontal GABA lev-
108. Heinz A, Voss M, Lawrie SM, et al. Shall we really say goodbye
els, hippocampal resting perfusion and the risk of psychosis.
to first rank symptoms? Eur Psychiatry. 2016;37:8–13.
Neuropsychopharmacology. 2018.
92. Gleich T, Deserno L, Lorenz RC, et al. Prefrontal and stri- 109. Waters F, Blom JD, Jardri R, Hugdahl K, Sommer IEC.
atal glutamate differently relate to striatal dopamine: poten- Auditory hallucinations, not necessarily a hallmark of psych-
tial regulatory mechanisms of striatal presynaptic dopamine otic disorder. Psychol Med. 2018;48:529–536.
function? J Neurosci. 2015;35:9615–9621. 110. Horga G, Schatz KC, Abi-Dargham A, Peterson BS. Deficits
93. Lodge DJ, Grace AA. The hippocampus modulates dopa- in predictive coding underlie hallucinations in schizophrenia.
mine neuron responsivity by regulating the intensity of J Neurosci. 2014;34:8072–8082.
phasic neuron activation. Neuropsychopharmacology. 111. Deserno L, Sterzer P, Wüstenberg T, Heinz A, Schlagenhauf
2006;31:1356–1361. F. Reduced prefrontal-parietal effective connectivity and
94. Grace AA, Floresco SB, Goto Y, Lodge DJ. Regulation of working memory deficits in schizophrenia. J Neurosci.
firing of dopaminergic neurons and control of goal-directed 2012;32:12–20.
behaviors. Trends Neurosci. 2007;30:220–227. 112. Brodersen KH, Deserno L, Schlagenhauf F, et al. Dissecting
95. Jardri R, Duverne S, Litvinova AS, Denève S. Experimental psychiatric spectrum disorders by generative embedding.
evidence for circular inference in schizophrenia. Nat Commun. Neuroimage Clin. 2014;4:98–111.
2017;8:14218. 113. Lawrence SJD, Formisano E, Muckli L, de Lange FP.
96. Grace AA, Gomes FV. The circuitry of dopamine system Laminar fMRI: applications for cognitive neuroscience.
regulation and its disruption in schizophrenia: insights into Neuroimage. 2017.
treatment and prevention. Schizophr Bull. 2018. 114. Stephan KE, Petzschner FH, Kasper L, et al. Laminar fMRI and
97. Sekutowicz M, Schmack K, Steimke R, et al. Striatal activa- computational theories of brain function. Neuroimage. 2017.
tion as a neural link between cognitive and perceptual flexi- 115. van Os J, Kenis G, Rutten BP. The environment and schizo-
bility. Neuroimage. 2016;141:393–398. phrenia. Nature. 2010;468:203–212.

1100