Shift toward prior knowledge confers a perceptual

advantage in early psychosis and psychosis-prone

healthy individuals
Christoph Teufela,b,1, Naresh Subramaniamb, Veronika Doblerc,d, Jesus Perezc,d, Johanna Finnemannb,e, Puja R. Mehtab,
Ian M. Goodyerc,d, and Paul C. Fletcherb,d
a
School of Psychology, Cardiff University, Cardiff CF10 3AT, United Kingdom; bBrain Mapping Unit, Department of Psychiatry, University of Cambridge,
Cambridge CB2 3EB, United Kingdom; cDevelopmental Psychiatry Section, Department of Psychiatry, University of Cambridge, Cambridge CB2 8AH, United
Kingdom; dCambridgeshire and Peterborough National Health Service Foundation Trust, Cambridge CB21 5EE, United Kingdom; and eDepartment of
Psychology, University of Cambridge, Cambridge CB2 3EB, United Kingdom

Edited by Thomas D. Albright, The Salk Institute for Biological Studies, La Jolla, CA, and approved May 28, 2015 (received for review February 26, 2015)

Many neuropsychiatric illnesses are associated with psychosis, i.e., Psychosis—a loss of contact with external reality—is charac-
hallucinations (perceptions in the absence of causative stimuli) and terized by delusions (irrational, often bizarre beliefs) and hal-
delusions (irrational, often bizarre beliefs). Current models of brain lucinations (perceptions in the absence of causative stimuli).
function view perception as a combination of two distinct sources Conceptual and computational models of psychosis have hy-
of information: bottom-up sensory input and top-down influences pothesized that an imbalance in the combination of bottom-up
from prior knowledge. This framework may explain hallucinations sensory evidence and top-down prior knowledge is at the core of
this altered state of mind (8–12). According to such models, at

PSYCHOLOGICAL AND
COGNITIVE SCIENCES
and delusions. Here, we characterized the balance between visual
bottom-up and top-down processing in people with early psychosis
the perceptual level, an undue reliance on prior knowledge in
perception may lead to the emergence of aberrant perceptions
(study 1) and in psychosis-prone, healthy individuals (study 2) to
such as hallucinations. The current study tests this hypothesis in
elucidate the mechanisms that might contribute to the emergence of the visual domain by characterizing the impact of prior knowl-
psychotic experiences. Through a specialized mental-health service, edge on the perception of ambiguous stimuli in two groups of
we identified unmedicated individuals who experience early psy- people: a clinical group with early psychotic experiences (study
chotic symptoms but fall below the threshold for a categorical diagnosis. 1) and healthy volunteers showing differing levels of proneness
We observed that, in early psychosis, there was a shift in information to such experiences (study 2). Although the conventional view
processing favoring prior knowledge over incoming sensory evidence. focuses preferentially on auditory hallucinations in psychosis,
In the complementary study, we capitalized on subtle variations in epidemiological evidence indicates that hallucinations in the vi-
perception and belief in the general population that exhibit graded sual domain are very common in, for example, schizophrenia
similarity with psychotic experiences (schizotypy). We observed (13). In fact, vision seems to play a prominent role in the de-
that the degree of psychosis proneness in healthy individuals, and, velopment of psychosis given that basic visual symptoms identi-
specifically, the presence of subtle perceptual alterations, is also fied before illness onset are one of the most powerful predictors
associated with stronger reliance on prior knowledge. Although, in of the emergence of later psychotic disorders (14).
the current experimental studies, this shift conferred a performance To determine mechanisms for the emergence of perceptual psy-
benefit, under most natural viewing situations, it may provoke chotic symptoms as purely as possible, we conducted two comple-
anomalous perceptual experiences. Overall, we show that early mentary studies. First, using a case-control study design, we
psychosis and psychosis proneness both entail a basic shift in visual
characterized the balance between visual bottom-up and top-down
information processing, favoring prior knowledge over incoming
sensory evidence. The studies provide complementary insights to a Significance
mechanism by which psychotic symptoms may emerge.
Perceiving things that are not there and holding unfounded,
visual perception | psychosis | top-down processing | predictive coding | bizarre beliefs (hallucinations and delusions, respectively) are
schizophrenia psychotic symptoms that occur in particular syndromes including
affective psychoses, paranoid states, and schizophrenia. We
studied the emergence of this loss of contact with reality based
T o interact successfully with our physical and social environ-
ment, we need appropriate information about relevant states
of the world, such as the size, location, or distance of an object.
on current models of normal brain function. Working with
clinical individuals experiencing early psychosis and nonclinical
individuals with high levels of psychosis proneness, we show
However, there is no direct access to this information, only to
that their visual perception is characterized by a shift that favors
sensory stimulation caused by the environment. This sensory in-
prior knowledge over incoming sensory evidence. Given that
formation is inherently ambiguous and, on its own, rarely suffices
these alterations in information processing are evident early on
to uniquely specify our surroundings (1). The human visual system
in psychosis and even in association with subtle perceptual
overcomes this challenge by combining ambiguous sensory in-
changes indicating psychosis proneness, they may be important
formation with prior knowledge of the environment to generate a
factors contributing to the emergence of severe mental illnesses.
robust and unambiguous representation of the world around us (1–
7). This insight has been formalized under the tenets of Bayesian Author contributions: C.T. conceived of the study; C.T. and P.C.F. designed research; C.T.,
decision theory and is typically modeled within a predictive coding N.S., V.D., J.F., and P.R.M. performed research; C.T. and P.C.F. analyzed data; C.T., N.S., V.D.,
framework. Here, the notion is that expectations based on prior J.P., J.F., I.M.G., and P.C.F. wrote the paper; and J.P. and I.M.G. managed patient pool.
knowledge are fed back from higher to lower levels of information The authors declare no conflict of interest.
processing, thereby shaping the way incoming signals are treated This article is a PNAS Direct Submission.
by lower-level mechanisms. This influence is labeled top-down Freely available online through the PNAS open access option.
processing. The present study tests the hypothesis that psychotic 1
To whom correspondence should be addressed. Email: teufelc@cardiff.ac.uk.
experiences arise from an increased use of prior knowledge in This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
constructing meaningful percepts from ambiguous sensory inputs. 1073/pnas.1503916112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1503916112 PNAS Early Edition | 1 of 6

 processing in a group of patients with early psychotic experiences
and matched healthy controls (SI Materials and Methods and Table
S1). Individuals in our clinical group were recruited from a dedi-
cated mental health service identifying help-seeking people who
have low-level but measurable psychotic experiences. Although, at
the time of testing, these individuals fell below the threshold for a
categorical diagnosis, they already showed symptoms and have an
increased risk for transitioning to a severe mental illness such as
schizophrenia or an affective disorder (15). Importantly, working
with such a group of patients and comparing them to controls
enabled us to focus on the features of early psychosis before any
formal categorical diagnosis. Moreover, and also critically, this
comparison is not confounded by the effects of antipsychotic
medication or the impact of chronic illness, allowing us, as purely
as possible, to explore the mechanisms of early psychosis.
In a second study, we explored psychosis proneness in healthy
participants characterized according to the presence of perceptual
(16) and belief-related schizotypal features (17). Schizotypy refers
to a personality measure that has established predictive value for
psychotic and other mental illnesses (18). Although it has been
traditionally considered a specific risk measure for schizophrenia,
more recently it has been proposed to reflect a general psychosis
proneness. A number of schizotypy scales have been devised to
characterize various dimensions of psychosis. In the current study,
we focused on individual variation in measures relating to per-
ception and belief (16, 17) because they most clearly relate to the
key features of psychosis. These measures provided us with a fine-
grained index for relevant perceptual experience and beliefs,
allowing us to characterize the bottom-up/top-down balance in
relation to subtle, nonclinical but specific and measurable markers
associated with psychosis proneness.
Characterizing these two situations enabled us to pursue our
central aim of exploring information-processing mechanisms that are
altered in association with the occurrence of early symptoms (study
1) and also identifiable even before such symptoms arise (study 2).
As well as offering a purer assessment of the emergence of psychotic
experiences, this approach is inspired by growing evidence suggesting
that psychosis lies on a continuum with normality (19, 20) and is
associated with a range of different psychiatric disorders (15, 21).
According to this perspective, existing diagnostic categories group
biologically heterogeneous syndromes with potentially different
pathophysiological mechanisms into one disorder (22); this may
obfuscate our attempts to understand the neurobiological under-
pinnings of mental illness. In keeping with a broader move within the
field, the aim of this approach is therefore to characterize deeper
dimensions in their own right, such as psychosis as in the current
study, irrespective of diagnostic categorization to advance our
mechanistic understanding of specific symptom clusters.
In summary, we explored how the use of prior knowledge in visual
information processing is related to early psychosis and to psychosis
proneness. Importantly, given our hypothesis, we predicted that the
putative mechanism associated with the emergence of psychosis
would confer a relative advantage in this task, given that successful
performance required the use of prior knowledge to discriminate
ambiguous stimuli. Together, the two studies provide evidence to

of the different experimental parts in seconds; var. indicates a part of vari-

able length. In total, every observer participated in 12 sessions, separated by
Fig. 1. The case-control study (study 1). (A) Illustration of a test (Left) and a self-paced breaks. (C) Illustration of one individual trial. After a brief pre-
control stimulus (Right). Test and control images had similar low-level statis- sentation of either a test or a control image, observers were asked to in-
tical properties (SI Materials and Methods) but differed in whether or not they dicate whether the image contained a person or not. Observers had as
contained an embedded person. (B) Illustration of one session. Initially, ob- much time as needed to respond but in case they had not responded within
servers viewed a block of ten two-tone images (Before Block). Across blocks, 1,250 ms, a reminder to respond was shown on the screen. (D) Results of
half of the stimuli were test images that contained an embedded person, the discrimination task. (Left) Discrimination sensitivity in terms of d′ (mean ± SEM)
other half were similar-looking control images without a person. The pre- in the Before and the After Block, slightly jittered for both groups to correctly
sentation of two-tone images was followed by three blocks of ten color im- display error bars; the higher the d′ values, the better observers were able to
ages, presented for 2 s in random order and back-to-back (Template Blocks). discriminate between test and control images. (Right) Response bias in terms
For every test image viewed in the Before Block a matching color template of the observers’ criterion (mean ± SEM); a lower criterion indicates a response
was presented, showing the clear version of the same image. Finally, in the bias to report the presence of a person in a given stimulus (independent of
After Block, observers saw the same two-tone images as in the Before Block, whether a test or control image is shown). Healthy controls are plotted in red;
presented in random order. Numbers along the timeline indicate the length the clinical group is shown in blue.

2 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1503916112 Teufel et al.

suggest that early psychosis and psychosis proneness is associated Additional analyses showed that, on average, observers in both
with a shift in visual processing that favors prior knowledge over groups had a stronger response bias (as measured by a criterion
incoming sensory evidence. We also demonstrate that this relation shift) to report seeing an object after being given relevant image
is specific to atypical perceptual experiences rather than being information compared with before [F(1,32) = 37.77, P < 0.001;
linked to psychotic experiences more generally. Fig. 1D, Right]. This finding did not differ between groups [F(1,32) <
1, not significant) and is an expected change given that observers
Results received additional information to solve the task and were therefore
To quantify top-down processing, we used two-tone images as more confident in the After than in the Before Block.
stimuli (3, 23–26). Without appropriate knowledge, these images In sum, study 1 shows that, in comparison with healthy controls,
appear to depict meaningless 2D black-and-white patches (Fig. observers with early psychosis can call more strongly on prior knowl-
1A). Once an observer is exposed to relevant image information, edge at the expense of sensory evidence when faced with a visual
however, grouping mechanisms in the visual system are able to signal. This group comparison result thus establishes the clinical
integrate patches in such a way that a robust and coherent percept importance of this basic information-processing shift but, as is typical
of a 3D object forms. In the present study, we provided the prior for such studies, our clinical group differs from the control group on
knowledge necessary to bind a two-tone image into a coherent a range of psychiatric measures (Table S1). It is therefore difficult to
percept by exposing observers to the color templates that were determine the specificity of the effect with respect to the two primary
used to generate the two-tone images (Fig. 2). Importantly, rather psychotic symptom clusters: hallucinations and delusions. In study 2,
than seeing an individual two-tone and the corresponding template we sought to gain a more specific and nuanced understanding of the
image back-to-back, observers freely viewed the stimuli in blocks of relationship between psychotic symptoms and the use of knowledge
10 (SI Materials and Methods). This procedure ensured that purely in perception. We capitalized on subtle alterations in perception and
bottom-up priming was minimized in the disambiguation of two- belief across the healthy population that show a graded similarity
tone stimuli after template exposure. In combination with the fact
with psychotic experiences. This variability can be characterized in-
that sensory stimulation is identical before and after template
exposure, the perceptual change of the two-tone image in this dependently for perception and belief by two schizotypy scales,

PSYCHOLOGICAL AND
the Cardiff Anomalous Perception Scale (CAPS) (16) and the

COGNITIVE SCIENCES
experimental design thus provides an ideal index of the impact
of knowledge on perception. This notion is supported by pre- Peters Delusion Inventory (PDI) (17). Note that, although other
vious psychophysical and neuroimaging literature, which indi- schizotypy scales such as Oxford-Liverpool Inventory of Feel-
cates that disambiguation of two-tone images is due to top-down ings and Experiences (O-LIFE) (27) and Schizotypal Personality
influences from high-level processes onto low-level visual func- Questionnaire (SPQ) (28) characterize a number of schizotypal
tion (3, 23–26). In particular, two-tone image perception recruits features, CAPS and PDI are specific to clinical features relating
memory processes and object knowledge associated with cortical to hallucinations and delusions respectively. In study 2, we thus
areas such as precuneus and dorsolateral prefrontal cortex; these recruited a larger sample of individuals from the general population
processes activate stored, high-level visual representations of the and related their tendency to rely on prior knowledge in visual
template images and feed back information to lower-level areas perception to their schizotypy scores. The task was identical to that
to shape perceptual processing, even as early as the primary visual used in the previous experiment except for two aspects: Given that
cortex V1. study 2 involved healthy observers only, we extended testing to 16
In study 1, observers performed a yes/no discrimination task, rather than 12 sessions to obtain a more precise estimate of every
in which they viewed briefly presented two-tone images of people observer’s d′; second, an improved stimulus set was used that con-
and similar-looking control images without an embedded object tained two image categories (people and animals) rather than one
(Fig. 1A). Within each session, these images were presented (people) as in study 1.
before and after the observer received prior knowledge by The results of study 2 are plotted in Fig. 3A. As expected, prior
viewing template images (Figs. 1B and 2). On every trial ob- knowledge resulted in a significant improvement in the ability to
servers were required to indicate whether a given image con- distinguish between test and control images (Fig. 3A left plot; paired
tained a person or not (Fig. 1C and SI Materials and Methods). t test, df = 39, t = 7.08, P < 0.001). Similar to the previous experi-
Based on signal detection theory, we derived two measures from ment, we also observed a criterion shift (Fig. 3A, Right; paired t test,
the observers’ performance: (i) d′, a pure index of an observer’s df = 39, t = 4.11, P < 0.001). Due to the differences in experimental
perceptual mechanisms independent of response bias; and (ii) design, we cannot directly compare performance across experiments.
criterion, which captures the general bias of observers to report Nevertheless, as expected, it is evident that inclusion of a second
the presence of an object. As an objective performance measure image category in study 2 made the task more difficult, explaining
of the influence of knowledge on perception, we were primarily the performance differences between the two experiments in healthy
interested in the change in d′ after having been exposed to rel- observers. Importantly and in line with the results of study 1, Fig. 3B
evant image information. indicates that the performance benefit due to prior knowledge was
The results support our hypothesis (Fig. 1D). As expected, a related to both the extent to which an observer experienced
mixed 2 × 2 ANOVA with the between-subjects factor group and anomalous perceptual phenomena similar to those of psychotic
the within-subjects factor block found an overall increased ability
patients as measured by CAPS (Fig. 3B, Left; product-moment,
to distinguish between test and control images measured in terms
r = 0.42, df = 38, t = 2.98, P < 0.01) and the extent to which they
of d′ after having viewed template images in comparison with
before having seen them [F(1,32) = 29.27, P < 0.001]. Critically, showed a delusion-like style of thinking as measured by PDI
this main effect was qualified by an interaction between group (Fig. 3B, Right; r = 0.33, df = 38, t = 2.14, P < 0.05). As was
and block, indicating that the improvement in performance dif- expected, however, the schizotypy scales for perception and belief
fered between the groups [F(1,32) = 5.02, P < 0.05]. As can be were positively correlated (r = 0.40, df = 38, t = 2.67, P < 0.05),
seen in the left plot in Fig. 1D and supported by post hoc tests, and we therefore conducted a first-order partial correlation
both groups showed an improvement in performance (paired analyses to determine the amount of variability in performance
t test, controls: df = 15, t = 2.17, P < 0.05; clinical group: df = 17, that can be explained by schizotypal features related to perception
t = 5.60, P < 0.001), but this was on average more than twice as or belief alone. This analysis indicated a very specific effect. The
large in the clinical group (mean ± SE: 0.35 ± 0.06) compared relation between a delusion-like style of thinking and performance
with controls (0.14 ± 0.07). It is noteworthy that the increased benefit due to prior knowledge disappeared once an observer’s
benefit in the clinical group was observed in the absence of an anomalous perceptual experiences were taken into account (first-
overall difference between groups [F(1,32) < 1, not significant] order partial correlation, r = 0.19, df = 37, t = 1.17, not signifi-
and was not due to a difference in baseline performance in the cant). By contrast, a significant relation between CAPS and
Before block (Welch’s two-sample t test, before clinical vs. be- performance benefit remained even when PDI was controlled for
fore controls: df = 31.16, t = 0.19, not significant). (r = 0.35, df = 37, t = 2.28, P < 0.05).

Teufel et al. PNAS Early Edition | 3 of 6

Fig. 2. Illustration of a template image. This template image was used to
generate the two-tone image shown in Fig. 1A, Left. To experience the
effect of knowledge on perception, the reader should look back at Fig. 1A
after having viewed this template for some time. The perceptual experi-
ence of the two-tone image in Fig. 1A, Left, should have drastically changed
from the initial viewing: rather than looking like meaningless patches, a
coherent percept of the child and parts of the puzzle should now be ex-
perienced. The control image in Fig. 1A, Right, should still appear as
meaningless patches. Fig. 3. The individual-differences study (study 2). (A) Results of discrimination
task. (Left) Discrimination sensitivity in terms of d′ (mean ± SEM) in the Before
and After Block; again, the higher the d′ values, the better observers were able
Discussion to discriminate between test and control images. (Right) Response bias in terms
of the observers’ criterion (mean ± SEM); as in Fig. 1, a lower criterion indicates
Our studies were designed to characterize, in complementary a response bias to report the presence of a person in a given stimulus (in-
ways, the balance between visual bottom-up and top-down pro- dependent of whether a test or control image is shown). (B) Relation of task
cessing in clinical individuals with early psychosis and healthy performance with schizotypy. (Left) Performance benefit due to prior knowl-
people prone to developing psychotic symptoms. A relative ad- edge against the global CAPS score, which captures perceptual alterations that
vantage in using prior knowledge to discriminate between am- show graded similarity with perceptual atypicalities experienced during psy-
biguous images was observed in both situations. This finding is chosis. Performance benefit was calculated as the difference in d′ between
especially striking in the clinical group in study 1 given that After and Before Blocks. (Right) Relation between performance benefit and
performance in this group (as in psychiatrically ill individuals global PDI scores of the participants. PDI captures belief-related alterations
similar to those seen during psychosis.
more generally) is typically impaired. Such a result is rare and
revealing in that it highlights a specific information-processing
atypicality rather than a general performance deficit. Study 2 These findings fit neatly with and support current conceptual
allowed us to characterize these alterations in visual function and computational models of psychotic symptoms (8–12). For in-
more completely by adopting an individual differences approach stance, it has been hypothesized that a single core disturbance
with healthy participants and by capitalizing on subtle variations relating to the balance between bottom-up and top-down pro-
in perception and belief that exhibit graded similarity with psy- cessing can explain both the hallucinatory experiences and the
chotic experiences. In line with our clinical findings, we uncov- bizarre delusional beliefs of psychotic patients (8, 11). Importantly,
ered a relation between an individual’s visual performance we show that, on the perceptual level, a shift in this balance toward
benefit due to prior knowledge and their scores on two scales of prior knowledge is present both in a clinical group of individuals
psychosis proneness. Importantly, also, our data suggest that this with early psychosis and even associated with psychosis proneness
relation is primarily driven by perceptual alterations rather than in the general population. Although schizotypy is a marker for
unusual beliefs. Taken together, these results indicate that visual psychosis proneness as ascertained by previous longitudinal studies
function in early psychosis and in healthy people who are prone (18), it is important to acknowledge that individuals in study 2 were
to such experiences is characterized by a basic information- not suffering from psychosis or even a diagnosed mental illness.
processing shift that favors existing knowledge over incoming Rather, those individuals scoring high on the scales identified a
sensory evidence. Although, in the current experimental task, number of unusual perceptual experiences. It is therefore striking
this shift conferred a performance benefit, under most natural that the same information-processing shift was observed as was
viewing situations, it may provoke anomalous perceptual expe- found in early psychosis. Indeed, even in the early psychosis group,
riences. Specifically, it might impose prior expectations on inputs no formal, categorical diagnosis was applicable (although it is known
to the extent that, ultimately, formed percepts are generated that that such groups have a high risk of transition to full psychiatric ill-
have no direct sensory cause: hallucinations. ness) (15). The findings may therefore suggest that the altered balance

4 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1503916112 Teufel et al.

is a fundamental trait that contributes to the emergence of psychosis pathological trajectories associated with different diagnostic cate-
rather than a reflection or consequence of the psychotic state. gories within such a unifying approach.
The specificity of the relation between performance on our task Some authors argue that, on a neuronal level, the bottom-up/
and perceptual aspects of schizotypy is of particular interest. It has top-down balance is implemented by neuromodulatory gain
long been known that altered perceptual experiences form a key control of feedforward and feedback neuronal circuitries;
part of the emergence of psychosis (29). Given that the CAPS is adopting this approach, psychosis may thus be understood in
selective for measuring schizotypal perceptual phenomena rather terms of atypicalities in these control mechanisms (35). Although
than targeting schizotypy in general (16), our findings indicate that a some models have focused on the emergence of delusions as a
shift in visual information processing that favors prior knowledge consequence of enhanced gain of feed-forward connections (and
over sensory evidence might be a marker for the mechanisms un- thus of the bottom-up signal) (8, 35, 36), the current findings
derlying this observation. The finding that healthy individuals that suggest that a shift in the opposite direction—i.e., either en-
score high on this scale share this marker with our clinical group is in hanced gain of feedback connections or reduced gain of feed-
line with the growing belief that psychotic mental illnesses are part forward connections—may account for perceptual changes
of a continuum with normality (19, 20). It supports the idea that the associated with psychosis. Given that our study did not directly
putative atypicality underlying the emergence of perceptual psy- measure neuromodulatory gain control, this assertion is specu-
chotic experiences relates directly to normal function of the system. lative but it might provide a fruitful avenue for future research.
In other words, the potential for psychotic experiences such as In schizophrenia, one important diagnostic category associated
hallucinations might be a logical consequence of the way in which with psychosis, both increased and decreased susceptibility to var-
our brain deals with the inherent ambiguity of sensory information ious visual illusions is observed (37–41). Although most studies
by incorporating prior knowledge into our perceptual processing. implicate atypicalities in early visual function as the source of these
The current study uncovered an imbalance of this processing type differences, some authors argue that certain illusions rely on some
that shows its effects at the perceptual level. However, within a hi- form of top-down processing (40, 41), a claim that has been con-
erarchical and recurrent information-processing system such as the tested (42, 43). Critically, however, most researchers agree that
human brain, an imbalance at any level will, in time, propagate up these visual illusions can be explained by processes within the visual

PSYCHOLOGICAL AND
COGNITIVE SCIENCES
and down the hierarchy and affects the whole system (8, 30), a system, rather than by influences on visual function from higher-
notion that might ultimately account for atypicalities in both lower- level knowledge areas outside the visual system. Our findings are
level perceptual processing and higher-level belief formation in se- therefore not at odds with decreased susceptibility to some visual
vere mental illnesses and psychosis proneness (30). illusions in schizophrenic patients (37, 38, 40, 41). In fact, we argue
As previously discussed, our aim in conducting these studies was that our findings dovetail with and complement these previous re-
specifically to try to understand the processes that contribute to sults in a surprising way: independently of the underlying mecha-
the emergence of psychotic symptoms rather than to examine a nisms, atypicalities within the visual system result in the outputs of
particular categorical diagnosis within which psychosis may occur. the perceptual analysis being less well structured and less adaptive.
This approach is part of a more general move in translational This could lead to an increased reliance on sources of information
research aimed at developing our understanding of the mecha- from outside the visual system to structure visual percepts and
nisms by which symptoms arise (22); this is seen as a necessary might make psychotic patients “hungry” for prior object knowledge.
prelude to a successful classificatory system in psychiatry. We do It is this latter effect that was demonstrated in the current study.
not advocate that we can do without diagnostic systems in psy- We should add that an explanation of these findings in terms
chiatry but rather selected our approach to fit the specific question of better memory for the information provided by the templates
we aimed to address. Ultimately, it will be extremely interesting to is implausible in both studies. Existing evidence suggests that
determine whether the shift in the bottom-up/top-down balance both early psychosis and high schizotypal characteristics may be
identified here is a transdiagnostic effect, occurring generally in associated with poorer working memory (44, 45).
early psychosis and psychosis proneness, or whether it is specific to Although performance improvements due to prior knowledge
a particular diagnostic category such as schizophrenia. This future were increased in the clinical group in study 1, these individuals
work will be important and we suggest that having identified such were no worse (than controls) in their ability to discriminate am-
a cognitive biomarker will prove very useful in examining, and biguous two-tone images without prior knowledge. This finding is
validating or challenging existing taxonomies of mental illness. not directly relevant to the main question addressed in this study
Computational explanations of the balance between sensory but it is interesting because it contrasts with previous studies in
evidence and prior knowledge as highlighted by our study are patients with established schizophrenia who show reduced ability
typically cast in terms of Bayesian models of perception and for- to spontaneously disambiguate two-tone images of faces without
malized within a predictive coding framework (5, 31, 32). In this prior knowledge (46, 47). This previous effect is most likely related
setting, the integration of bottom-up and top-down signals is me- to well-established schizophrenic deficits in early and midlevel vi-
diated by the relative amount of information each of these com- sion that affect perceptual organization, context processing and
ponents provides: the stronger the sensory evidence is relative to integration (37, 38, 48–53) rather than to top-down influences from
prior knowledge, the more it will impact on the final processing high-level visual cognition as in the current study. We did not di-
output; conversely, if prior knowledge provides a relatively greater rectly probe early and midlevel visual function in our participants,
amount of information it will be weighted more strongly. Our but it seems most likely that the absence of impairments in spon-
findings fit comfortably with such a computational account: a taneous disambiguation of two-tone images in the clinical group
stronger reliance on top-down processing in psychosis-prone in- might be due to the specific nature of our stimulus material, which
dividuals as identified in both of our experiments suggests that was extensively piloted to be difficult to disambiguate without prior
prior knowledge provides a large amount of information relative to knowledge (for details, see SI Materials and Methods).
the amount of sensory evidence. Importantly, this could either be To conclude, if we are to make progress in understanding the
due to atypically strong knowledge representations or, alterna- nature of psychotic experiences and how they relate to cognitive
tively, to an unusually noisy sensory system. Given well-established and neural markers, we have to identify candidate mechanisms for
psychotic deficits in early visual processing as discussed below, the how they may arise based on a growing understanding of the
latter explanation might be more likely. However, the findings relevant perceptual and cognitive systems. Here, we identified a
cannot distinguish between these hypotheses. In this context, it is shift in visual information-processing in early psychosis and in
interesting to note that the same theme of an imbalance between psychosis-prone healthy individuals. In both cases, top-down prior
bottom-up and top-down processing figures prominently in current knowledge was favored over bottom-up sensory evidence. These
debates about autism (33, 34). This similarity highlights the notion alterations directly relate to visual function in healthy individuals
that predictive coding might provide a common framework within and our findings support the notion of a continuum between
which to understand mental illness more generally. An important normality and psychotic experiences. The changes in information
task for the future will be to explain the specificities of the different processing we report here might be among the influences that, in

Teufel et al. PNAS Early Edition | 5 of 6

concert with other factors, contribute to the emergence of major template, the two-tone images should give the strong experience of a coherent
mental illnesses such as schizophrenia and affective disorders. object. Very few studies using two-tone images use stimuli that have both of
these properties. In the current study, we ran an extensive piloting phase to
Materials and Methods generate appropriate images. For details, see SI Materials and Methods.
Observers. In study 1, 34 individuals from the database of the Cambridge and
Peterborough Assessing, Managing, and Enhancing Outcomes (CAMEO) early Experimental Procedure and Analysis. As described here and in Fig. 1, we used
intervention in psychosis service (www.cameo.nhs.uk) based in Cambridge a yes/no task with an analysis derived from signal detection theory to
consented to take part in the study: 16 healthy volunteers (9 males) and 18 characterize the balance between bottom-up and top-down processing in
at-risk mental state (ARMS) individuals (13 males). The groups were matched our observers. A complete description of experimental procedures and
on age and IQ, using Cattell’s Culture Fair Intelligence Test. For full de- analyses is presented in SI Materials and Methods.
scription of the recruitment procedure and the clinical and demographic
profile of participants, see SI Materials and Methods and Table S1. ACKNOWLEDGMENTS. We thank all the members of the Cambridge and
In study 2, the 40 healthy observers who participated in the study were Peterborough Assessing, Managing, and Enhancing Outcomes (CAMEO)
recruited from the local population through local mailing lists. For further services for their help and support in conducting this study. This work was
details, see SI Materials and Methods. carried out within the Behavioural and Clinical Neuroscience Institute
(BCNI), funded by the Wellcome Trust and the Medical Research Council,
and the Cambridgeshire and Peterborough National Health Service Foun-
Stimuli. Ideal two-tone images have the following two properties: (i) they dation Trust (CPFT). This study was supported by grants from the Wellcome
should be impossible or very difficult to disambiguate before being given prior Trust and the Bernard Wolfe Health Neuroscience fund (to P.C.F.). I.M.G. is
knowledge about image content; and (ii) once participants have seen the supported by the Wellcome Trust.

1. Kersten D, Mamassian P, Yuille A (2004) Object perception as Bayesian inference. 31. Huang Y, Rao RPN (2011) Predictive coding. Wiley Interdiscip Rev Cogn Sci 2(5):580–593.
Annu Rev Psychol 55(1):271–304. 32. Spratling MW (2010) Predictive coding as a model of response properties in cortical
2. Neri P (2014) Semantic control of feature extraction from natural scenes. J Neurosci area V1. J Neurosci 30(9):3531–3543.
34(6):2374–2388. 33. Van de Cruys S, et al. (2014) Precise minds in uncertain worlds: Predictive coding in
3. Cavanagh P (2011) Visual cognition. Vision Res 51(13):1538–1551. autism. Psychol Rev 121(4):649–675.
4. Dakin SC (2009) Vision: Thinking globally, acting locally. Curr Biol 19(18):R851–R854. 34. Lawson RP, Rees G, Friston KJ (2014) An aberrant precision account of autism. Front
5. Friston K (2010) The free-energy principle: A unified brain theory? Nat Rev Neurosci Hum Neurosci 8:302.
11(2):127–138. 35. Adams RA, Stephan KE, Brown HR, Frith CD, Friston KJ (2013) The computational
6. Schmack K, et al. (2013) Delusions and the role of beliefs in perceptual inference. anatomy of psychosis. Front Integr Neurosci 4(47):1–26.
J Neurosci 33(34):13701–13712. 36. Kapur S (2003) Psychosis as a state of aberrant salience: A framework linking biology,
7. Teufel C, Subramaniam N, Fletcher PC (2013) The role of priors in Bayesian models of phenomenology, and pharmacology in schizophrenia. Am J Psychiatry 160(1):13–23.
perception. Front Comput Neurosci 7(25):25. 37. Dakin S, Carlin P, Hemsley D (2005) Weak suppression of visual context in chronic
8. Fletcher PC, Frith CD (2009) Perceiving is believing: A Bayesian approach to explaining schizophrenia. Curr Biol 15(20):R822–R824.
the positive symptoms of schizophrenia. Nat Rev Neurosci 10(1):48–58. 38. Tibber MS, et al. (2013) Visual surround suppression in schizophrenia. Front Integr
9. Grossberg S (2013) Adaptive Resonance Theory: How a brain learns to consciously Neurosci 4(88):1–13.
attend, learn, and recognize a changing world. Neural Netw 37:1–47. 39. Kantrowitz JT, Butler PD, Schecter I, Silipo G, Javitt DC (2009) Seeing the world dimly:
10. Jardri R, Denève S (2013) Computational models of hallucinations. The Neuroscience of The impact of early visual deficits on visual experience in schizophrenia. Schizophr
Hallucinations, eds Jardri R, Cachia A, Thomas P, Pins D (Springer, Berlin), pp 289–313. Bull 35(6):1085–1094.
11. Corlett PR, Frith CD, Fletcher PC (2009) From drugs to deprivation: A Bayesian framework 40. Schneider U, et al. (2002) Reduced binocular depth inversion in schizophrenic pa-
for understanding models of psychosis. Psychopharmacology (Berl) 206(4):515–530. tients. Schizophr Res 53(1-2):101–108.
12. Clark A (2013) Whatever next? Predictive brains, situated agents, and the future of 41. Keane BP, Silverstein SM, Wang Y, Papathomas TV (2013) Reduced depth inversion
cognitive science. Behav Brain Sci 36(3):181–204. illusions in schizophrenia are state-specific and occur for multiple object types and
13. Waters F, et al. (2014) Visual hallucinations in the psychosis spectrum and comparative
viewing conditions. J Abnorm Psychol 122(2):506–512.
information from neurodegenerative disorders and eye disease. Schizophr Bull 40
42. Barlow HB (1997) The knowledge used in vision and where it comes from. Philos Trans
(Suppl 4):S233–S245.
R Soc Lond B Biol Sci 352(1358):1141–1147.
14. Klosterkötter J, Hellmich M, Steinmeyer EM, Schultze-Lutter F (2001) Diagnosing
43. Hill H, Johnston A (2007) The hollow-face illusion: Object-specific knowledge, general
schizophrenia in the initial prodromal phase. Arch Gen Psychiatry 58(2):158–164.
assumptions or properties of the stimulus? Perception 36(2):199–223.
15. Hui C, et al. (2013) Psychiatric morbidity, functioning and quality of life in young
44. Pawełczyk A, et al. (2015) Figural fluency and immediate visual memory in patients with
people at clinical high risk for psychosis. Schizophr Res 148(1-3):175–180.
at-risk mental state for psychosis: Empirical study. Early Interv Psychiatry 9(4):324–330.
16. Bell V, Halligan PW, Ellis HD (2006) The Cardiff Anomalous Perceptions Scale (CAPS): A new
45. Schmidt-Hansen M, Honey RC (2009) Working memory and multidimensional schizotypy:
validated measure of anomalous perceptual experience. Schizophr Bull 32(2):366–377.
Dissociable influences of the different dimensions. Cogn Neuropsychol 26(7):655–670.
17. Peters E, Joseph S, Day S, Garety P (2004) Measuring delusional ideation: The 21-item
46. Grützner C, et al. (2013) Deficits in high- (>60 Hz) gamma-band oscillations during
Peters et al. Delusions Inventory (PDI). Schizophr Bull 30(4):1005–1022.
visual processing in schizophrenia. Front Hum Neurosci 7(88):88.
18. Kwapil TR, Gross GM, Silvia PJ, Barrantes-Vidal N (2013) Prediction of psychopathol-
47. Sun L, et al. (2013) Evidence for dysregulated high-frequency oscillations during sensory
ogy and functional impairment by positive and negative schizotypy in the Chapmans’
processing in medication-naïve, first episode schizophrenia. Schizophr Res 150(2-3):519–525.
ten-year longitudinal study. J Abnorm Psychol 122(3):807–815.
48. Javitt DC (2009) When doors of perception close: Bottom-up models of disrupted
19. Johns LC, van Os J (2001) The continuity of psychotic experiences in the general
population. Clin Psychol Rev 21(8):1125–1141. cognition in schizophrenia. Annu Rev Clin Psychol 5(1):249–275.
20. van Os J, Linscott RJ, Myin-Germeys I, Delespaul P, Krabbendam L (2009) A systematic 49. Butler PD, Silverstein SM, Dakin SC (2008) Visual perception and its impairment in
review and meta-analysis of the psychosis continuum: Evidence for a psychosis prone- schizophrenia. Biol Psychiatry 64(1):40–47.
ness-persistence-impairment model of psychotic disorder. Psychol Med 39(2):179–195. 50. Doniger GM, Silipo G, Rabinowicz EF, Snodgrass JG, Javitt DC (2001) Impaired sensory
21. Stochl J, et al. (2015) Mood, anxiety and psychotic phenomena measure a common processing as a basis for object-recognition deficits in schizophrenia. Am J Psychiatry
psychopathological factor. Psychol Med 45(7):1483–1493. 158(11):1818–1826.
22. Cuthbert BN, Insel TR (2013) Toward the future of psychiatric diagnosis: The seven 51. Robol V, et al. (2013) Reduced crowding and poor contour detection in schizophrenia
pillars of RDoC. BMC Med 11(1):126. are consistent with weak surround inhibition. PLoS ONE 8(4):e60951.
23. Moore C, Cavanagh P (1998) Recovery of 3D volume from 2-tone images of novel 52. Silverstein SM, Keane BP (2011) Perceptual organization impairment in schizophrenia
objects. Cognition 67(1-2):45–71. and associated brain mechanisms: Review of research from 2005 to 2010. Schizophr
24. Hegdé J, Kersten D (2010) A link between visual disambiguation and visual memory. Bull 37(4):690–699.
J Neurosci 30(45):15124–15133. 53. Uhlhaas PJ, Silverstein SM (2005) Perceptual organization in schizophrenia spectrum
25. Dolan RJ, et al. (1997) How the brain learns to see objects and faces in an impov- disorders: Empirical research and theoretical implications. Psychol Bull 131(4):618–632.
erished context. Nature 389(6651):596–599. 54. Cattell RB, Cattell AK (1973) Measuring Intelligence with the Culture Fair Tests (In-
26. Hsieh PJ, Vul E, Kanwisher N (2010) Recognition alters the spatial pattern of FMRI stitute for Personality and Ability Testing, Champaign, IL).
activation in early retinotopic cortex. J Neurophysiol 103(3):1501–1507. 55. Kleiner M, Brainard D, Pelli D (2007) What’s new in psychtoolbox-3? Perception 36(14):1.
27. Mason O, Claridge G, Jackson M (1995) New scales for the assessment of schizotypy. 56. Brainard DH (1997) The psychophysics toolbox. Spat Vis 10(4):433–436.
Pers Individ Dif 18(1):7–13. 57. Macmillan NA, Creelman CD (2005) Detection Theory: A User’s Guide (Lawrence Erbaum
28. Raine A (1991) The SPQ: A scale for the assessment of schizotypal personality based Associates, Mahwah, NJ), 2nd Ed.
on DSM-III-R criteria. Schizophr Bull 17(4):555–564. 58. Beck AT, Ward CH, Mendelson M, Mock J, Erbaugh J (1961) An inventory for mea-
29. Chapman J (1966) The early symptoms of schizophrenia. Br J Psychiatry 112(484):225–251. suring depression. Arch Gen Psychiatry 4(6):561–571.
30. Leitman DI, et al. (2010) Sensory deficits and distributed hierarchical dysfunction in 59. Spielberger CD, Gorsuch RL, Lushene R, Vagg PR, Jacobs GA (1983) Manual for the
schizophrenia. Am J Psychiatry 167(7):818–827. State-Trait Anxiety Inventory (Consulting Psychologists Press, Palo Alto, CA).

6 of 6 | www.pnas.org/cgi/doi/10.1073/pnas.1503916112 Teufel et al.