Content uploaded by Peter Kok
Author content
All content in this area was uploaded by Peter Kok on May 09, 2024
Content may be subject to copyright.
Report
Predictions and errors are distinctly represented
across V1 layers
Highlights
dA 7T fMRI study presents expected and unexpected Gabor
orientations
dExpectation differentially modulates decoding across
primary visual cortex layers
dThis pattern supports predictive processing accounts of the
brain and mind
Authors
Emily R. Thomas, Joost Haarsma,
Jessica Nicholson, Daniel Yon,
Peter Kok, Clare Press
Correspondence
emilyrosethomas@outlook.com (E.R.T.),
c.press@ucl.ac.uk (C.P.)
In brief
Thomas et al. use 7T fMRI to find that
expected events are represented
similarly across deep, middle, and
superficial layers of the primary visual
cortex, while unexpected events are only
robustly represented in superficial layers.
These findings support accounts of
sensory processing requiring distinct
representations of predictions and errors.
Thomas et al., 2024, Current Biology 34, 1–7
May 20, 2024 ª2024 The Author(s). Published by Elsevier Inc.
https://doi.org/10.1016/j.cub.2024.04.036 ll
Report
Predictions and errors are distinctly
represented across V1 layers
Emily R. Thomas,
1,2,
*Joost Haarsma,
3
Jessica Nicholson,
2
Daniel Yon,
2
Peter Kok,
3
and Clare Press
2,3,4,5,6,
*
1
Neuroscience Institute, New York University Medical Center, 435 East 30
th
Street, New York 10016, USA
2
Department of Psychological Sciences, Birkbeck, University of London, Malet Street, London WC1E 7HX, UK
3
Wellcome Centre for Human Neuroimaging, Institute of Neurology, University College London, 12 Queen Square, London WC1N 3AR, UK
4
Department of Experimental Psychology, University College London, 26 Bedford Way, London WC1H 0AP, UK
5
X (formerly Twitter): @ClarePress
6
Lead contact
*Correspondence: emilyrosethomas@outlook.com (E.R.T.), c.press@ucl.ac.uk (C.P.)
https://doi.org/10.1016/j.cub.2024.04.036
SUMMARY
Popular accounts of mind and brain propose that the brain continuously forms predictions about future sen-
sory inputs and combines predictions with inputs to determine what we perceive.
1–6
Under ‘‘predictive pro-
cessing’’ schemes, such integration is supported by the hierarchical organization of the cortex, whereby
feedback connections communicate predictions from higher-level deep layers to agranular (superficial
and deep) lower-level layers.
7–10
Predictions are compared with input to compute the ‘‘prediction error,’’
which is transmitted up the hierarchy from superficial layers of lower cortical regions to the middle layers
of higher areas, to update higher-level predictions until errors are reconciled.
11–15
In the primary visual cortex
(V1), predictions have thereby been proposed to influence representations in deep layers while error signals
may be computed in superficial layers. Despite the framework’s popularity, there is little evidence for these
functional distinctions because, to our knowledge, unexpected sensory events have not previously been pre-
sented in human laminar paradigms to contrast against expected events. To this end, this 7T fMRI study con-
trasted V1 responses to expected (75% likely) and unexpected (25%) Gabor orientations. Multivariate decod-
ing analyses revealed an interaction between expectation and layer, such that expected events could be
decoded with comparable accuracy across layers, while unexpected events could only be decoded in super-
ficial laminae. Although these results are in line with these accounts that have been popular for decades, such
distinctions have not previously been demonstrated in humans. We discuss how both prediction and error
processes may operate together to shape our unitary perceptual experiences.
RESULTS
Due to previous work, partly from our group,
16,17
demonstrating
fast and robust learning of action-cue-outcome relationships in
human participants, we used an action-outcome paradigm to
establish predictions. Twenty-two participants (17 female, mean
age = 26.09 years, and SD = 3.41) were trained with perfect
relationships between fingeractions and visual Gabor orientations
(e.g., index finger abduction = clockwise-oriented Gabor [CW];
little finger abduction = counter-clockwise Gabor [CCW]). They
were presented at test (scanning phase, the following day)
with degraded contingencies to measure neural responses to
‘‘expected’’ (in line with perfect contingency training phase;
75% of trials in the scanner) and ‘‘unexpected’’ (25%) events
(see Figure 1). On half the trials they were asked to give a yes/no
response to whether the stimulus was oriented CW and on the
other half they were asked whether it was oriented CCW. This
design orthogonalized the Gabor orientation presentation from
the response. Linear support vector machines (SVMs) were
trained to discriminate Gabor orientations from V1 activation dur-
ing a localizer and were tested on the main task,
18
separately for
expected and unexpected events (Figure 2). Under the predictive
processingaccount outlined above, an interaction is hypothesized
between expectation and layer in decoding accuracy.
Influence of expectations on behavior
Reaction time (RT) data were collected for responses to ex-
pected and unexpected stimuli in the test session and median
RTs were calculated for correct trials, separately for each condi-
tion and participant. Similarly, the proportion of correct re-
sponses was analyzed for expected and unexpected conditions.
RT analyses revealed no difference between expected (M=
586.78 ms, SD = 73.42) and unexpected (M= 589.98 ms, SD =
75.60) trials (t(19) = 0.57, p= 0.58, and d = 0.13). Participants
were, however, more accurate on expected (M= 0.97, SD = 0.03)
than unexpected (M= 0.95, SD = 0.04) trials (t(19) = 2.67, p=
0.015, d = 0.60; see Figure 3A).
Distinct cortical representations of predictions and
errors
Using ultra-high-field 7T fMRI (spatial resolution: 0.8 mm
isotropic), we examined the brain activity patterns across
Current Biology 34, 1–7, May 20, 2024 ª2024 The Author(s). Published by Elsevier Inc. 1
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
ll
OPEN ACCESS
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
cortical layers of the V1 for expected and unexpected Gabor ori-
entations. To determine layer-specific activity patterns, cortical
V1 voxels were divided into three equivolume gray matter layer
bins—superficial, middle, and deep. The proportion of each vox-
el’s volume across the layers was used to create three cortical
layer V1 masks for each participant that were used as layer re-
gions of interest (ROIs) for the following decoding analyses.
To investigate how expectations are represented across the
cortical column in the V1, linear SVMs were trained to discrimi-
nate Gabor orientations (CW or CCW) from a short localizer
task that presented blocks of high-contrast flickering Gabors
and were tested on the main task (beta images from a first-level
generalized linear model [GLM]). Given that expected events
were presented with 75% likelihood, while unexpected events
were presented with 25% likelihood, modeling all expected trials
would render regressors that contained three times the data of
unexpected regressors. We therefore modeled three expected
regressors, all with an identical number of trials to unexpected
regressors in the GLM, to reduce decoding biases across com-
parisons. The accuracy scores for each of the three expected
decoding conditions were averaged for each participant,
providing one accuracy score for expected trials and one for un-
expected trials in each of the layer masks. These decoding accu-
racies were compared using a repeated measures ANOVA.
This analysis revealed a main effect of layer (F(1.45, 29.04) =
5.98, p= 0.012, np
2
= 0.23, Greenhouse-Geisser corrected,
ε= 0.73), no main effect of expectation (F(1, 20) = 0.39, p=
0.54, and np
2
= 0.019), and, crucially, an interaction between
expectation and layer (F(2, 40) = 4.45, p= 0.018, and np
2
=
0.18; see Figure 3B). Two control analyses were run at the voxel
selection stage to balance the number of voxels in each layer
mask, considering that there were more voxels in the superficial
Figure 1. Experiment design
(A) Schematic representation of proposed extrinsic feedforward (red) and feedback (blue) connections across layers in early visual areas.
(B) Experimental paradigm. A centrally presented visual cue instructed participants to abduct either their index or little finger. The imperative cue could be either a
triangle or square presented around the fixation cross. Each finger abduction predicted an oriented Gabor and participants were required to respond (yes/no) to
whether the stimulus was clockwise (CW) or counter-clockwise (CCW) oriented relative to the vertical. In the training phase, actions perfectly predicted the
stimulus orientation (100% contingency). This example demonstrates the relationship for a participant trained in index finger abduction to clockwise-oriented
Gabor mappings.
(C) In the scanning session 24 h after the training phase, participants completed the same task, but the action-outcome relationship was degraded to 75% to
produce unexpected (25%) as well as expected (75%) events.
ll
OPEN ACCESS
2Current Biology 34, 1–7, May 20, 2024
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
layers (due to known draining vein biases in gradient-echo echo-
planar imaging [EPI]
21–25
; see STAR Methods). The effects re-
mained in both a voxel-balanced control t-map analysis (this
analysis selected an equal number of the most active voxels
across the three layer bins; see STAR Methods; expectation 3
layer: F(2, 40) = 3.64, p= 0.035, np
2
= 0.15; layer: F(1.60,
31.98) = 4.83, p= 0.021, and np
2
= 0.20, Huynh-Feldt corrected,
ε= 0.80; expectation: F(1, 20) = 1.77, p= 0.20, and np
2
= 0.08),
and a random-sample analysis (selecting an equal number of
voxels across layers sampled randomly from each layer;
expectation 3layer: F(2, 40) = 4.35, p= 0.019, and np
2
= 0.18;
layer: F(2, 40) = 4.70, p= 0.015, and np
2
= 0.19; expectation:
F(1, 20) = 2.07, p= 0.17, and np
2
= 0.09).
This interaction was generated via relatively consistent decod-
ing across layers for expected events (F(2, 40) = 0.82, p= 0.45,
and np
2
= 0.04), while decoding performance differed for unex-
pected events (F(2, 40) = 6.90, p= 0.003, np
2
= 0.26)—increasing
from deep to superficial layers (Figure 3B). These effects are
complemented by one-sample t tests demonstrating that de-
coding of expected events was significantly different from
chance across all layers (deep: t(20) = 2.62, p= 0.016; middle:
t(20) = 4.13, p= 0.001; superficial: t(20) = 2.94, p= 0.008), while
unexpected events could only be decoded in superficial layers
(deep: t(20) = 0.73, p= 0.47; middle: t(20) = 1.03, p= 0.31;
superficial: t(20) = 3.97, p= 0.001). Additional post hoc tests re-
vealed no significant differences between expected and unex-
pected decoding within each layer (deep: t(20) = 2.01, p=
0.058, d = 0.44; middle: t(20) = 0.94, p= 0.36, d = 0.21; superfi-
cial: t(20) = 1.41, p= 0.18, and d = 0.31), though the numerical
differences reveal superior representation of expected events in
deep (expected: M= 3.27, SD = 5.73; unexpected: M=1.76,
SD = 11.21) and middle layers (expected: M= 5.36, SD = 5.95;
unexpected: M= 2.98, SD = 13.20), flipping to superior re-
presentation of the unexpected in superficial layers (expected:
M= 4.76, SD = 7.43; unexpected: M= 8.63, SD = 9.98).
Taken together, these tests demonstrate that representation
of unexpected events increases toward the superficial layers of
the V1, only becoming significantly decodable in these layers,
while expected events are represented similarly across the
cortical column.
DISCUSSION
This study examined how unexpected visual events are repre-
sented across cortical layers, in comparison with expected
events, in a high-resolution fMRI study. It found, in line with previ-
ous work,
19
that expected events were represented (decoded)
equivalently across deep, middle, and superficial bins but, more
Figure 2. Data analysis
(A) Visualization of the selected anatomical V1 ROI (light gray) on a mean functional image of an example participant . Overlaid red and yellow lines represent co-
registered anatomical WM (yellow) and pial surface (red) boundaries to the mean functional image, showing voxels that were significantly active against baseline
to the presented stimuli in the functional localizer task (green).
(B) A mean functional image overlaid with distributions of voxels in superficial (green), middle (blue), and deep (red) layers of the cortex.
(C) A schematic representing the level-set approach used to determine the volume distribution of a selected voxel (e.g., red square) over the superficial, middle,
and deep cortical layers.
19,20
(D) A schematic of the decoding approach adopted here. Voxel proportions across the three layer bins in (C) were used to separate voxels according to the
majority layer and formed layer masks for V1. Linear classifiers (SVMs) were trained on CW and CCW stimuli from the localizer task and tested on Gabors from the
main task. The procedure was repeated separately for expected and unexpected time courses and in each V1 layer mask.
ll
OPEN ACCESS
Current Biology 34, 1–7, May 20, 2024 3
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
novelly, that unexpected events were represented with varying fi-
delity—suchthat they were poorly representedin deep and middle
layers and could only be decoded above chance in superficial
layers.
These findings are in line with predictive processing accounts
in which predictions and errors are represented distinctly across
cortical laminae, such that predictions conveyed via feedback
projections inject input into hypothesis units in deep layers, while
feedforward connections transmit the error from the superficial
laminae.
12
This finding is also in line with data from mice and mon-
keys indicating superficial layer discrepancy signals with respect
to other types of feedback
26–28
(see also Gillon et al.
29
and Fiser
et al.
30
). Although this account of cortical processing has been
popular for a couple of decades, such distinctions have not pre-
viously been demonstrated in human cortical processing.
It is worth noting that alternatives have been proposed to this
account to explain precisely how prediction and input signals
are combined in the brain. Particularly, a number of fMRI studies
have observed improved sensory decoding of events expected
on the basis of preceding cues relative to unexpected
events,
16,17,31,32
which may suggest that channels tuned to ex-
pected inputs are more responsive than channelstuned to the un-
expected. Such a ‘‘global sharpening’’ account
33
proposes that
the precision weights are adjusted to increase the gain of ex-
pected channels, allowing them to respond more sensitively to
expected input and subsequently improving representation of
the expected across the cortical column. This account would pre-
dict facilitated processing of the expected
1–5
across layers, if we
relatively inhibit processing via other channels across layers, and
therefore is inconsistent with the patterns observed here.
Figure 3. Results
(A) Mean RTs and accuracy (± SEM) for expected and unexpected events alongside probability density estimates and individual participant data points. There
was no difference between conditions in RT, but participants were more accurate in expected than unexpected judgments (*p< 0.05).
(B) The results of the decoding analysis across cortical layer bins, where mean (± SEM) decoding accuracy percentage above chance (50%) is plotted for ex-
pected and unexpected trials. On the left panel, circles around mean data points indicate that the decoding accuracy was significantly above chance (p< 0.05),
which was the case across all layers for expected events but only in superficial layers for unexpec ted events. On the right panel, decoding accuracies are plotted
alongside probability density estimates and individual participant data points. The linear trend across layers was significant for unexpected (**p< 0.001) but not
expected events.
ll
OPEN ACCESS
4Current Biology 34, 1–7, May 20, 2024
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
Predictive processing accounts were initially developed to
explain inference about the present—removing redundancy in
the system—while the difference between our expected and un-
expected conditions pertains to cues that allow one to predict
statistical likelihoods about future events. The fact that we see
differences between conditions like these demonstrates that
future-based (e.g., cue-based) predictions inject hypotheses
into units in deep layers
19
in the same way as present-state pre-
dictions. These findings are in line with evidence from human
ultra-high resolution 7T MRI studies demonstrating activity in
deep cortical layers of the V1 for visual events that are expected
but never presented
19,34
(see also Muckli et al.
35
).
Forging functional conclusions about the operation of
mechanisms across cortical layers has become possible with
high-resolution MRI but is, of course, also plagued by interpreta-
tional issues due to venous draining of blood toward the pial
surface.
21–25
Specifically, gradient-echo blood-oxygen-level-
dependent (BOLD) signal is known to exhibit strong contributions
from large veins situated perpendicular to the cortical
surface as venous blood is drained from lower to upper
cortical layers.
15,22,24
Such venous issues render it likely, for
instance, that neural effects at deeper cortical layers contribute
to responses in superficial layers.
36
Other labs favor cerebral-
blood-volume-based vascular-space-occupancy (VASO)
37
fMRI
due to superior laminar separation, although this is accompanied
by reduced content-based sensitivity relative to gradient-echo
methods.
38
Here, we use gradient-echo BOLD but mitigate such
contributions by comparing responses between stimuli that are
identical—other than their expectedness due to a preceding
cue—because venous draining influences should be equivalent
for both expected and unexpected events. Another methodolog-
ical debate in the field surrounds separation of the signal into
three cortical layers,
19,20,36,39–41
such as here, versus more layers.
Given the voxel size of 0.8 mm and a cortical thickness of
2.5–3 mm, three layers could be conceived to be the most realistic
resolution to be achieved with this voxel size.
42
Nevertheless,
importantly for our conclusions, we know of no literature that
would suggest such an expectation 3layer interaction effect, as
observed here, would be generated by our methodological
choices and not reflective of true mechanistic differences, but
future work would, of course, be wise to investigate replicability
with different approaches.
Such distinct representation of prediction and error may be an
adaptive solution allowing predictions to shape perception to
serve a number of functions. Some of us have recently discussed
how predictions often need to exhibit quite distinct behavioral
shaping of perception to serve the organism.
4,43
To overcome
noise in sensory processing and generate broadly accurate ex-
periences rapidly, we may bias perception toward what we
expect.
44,45
However, larger error signals (that cannot have re-
sulted from noise) may require high perceptual resources to
enable accurate perception and resultant model updating. If
we represent the error signal separate from the prediction,
even in early sensory processing, this may be one way to enable
these large error signals to communicate deviation rapidly to
systems mediating model updating—such as the locus coeru-
leus.
46
Future work must establish how these error signals relate
to perception and model updating to truly test these accounts
and examine whether error signals in superficial layers are
calculated in the first feedforward sweep
47
or subsequent stim-
ulus-processing iterations.
It has been suggested in various theoretical accounts that
symptoms of psychosis, like hallucinations and delusions, can
be explained in terms of aberrant signaling of prediction error
as well as overweighting of expectations.
48–50
As demonstrated
in this study, laminar fMRI is capable of distinguishing the repre-
sentation of these signals across different cortical layers. There-
fore, laminar fMRI would be well suited to test the theoretical pre-
dictions from predictive coding models of psychosis, as well as
comparing these mechanisms in other clinical and neurological
populations characterized by aberrant perceptual inference,
like Parkinson’s disease.
51
In conclusion, this study provides evidence that expected and
unexpected visual events are distinctly represented across the
cortical column in the V1 via a novel 7T fMRI design that pre-
sented unexpected visual events alongside expected counter-
parts. Expected events were represented similarly across layers
but unexpected events were only represented well in superficial
layers. These findings contribute to our understanding of how
predictions can interact with sensory inputs to shape what we
perceive and how we interact with the world.
STAR+METHODS
Detailed methods are provided in the online version of this paper and include
the following:
dKEY RESOURCES TABLE
dRESOURCE AVAILABILITY
BLead contact
BMaterials availability
BData and code availability
dEXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS
BParticipants
dMETHOD DETAILS
BStimuli
BProcedure
BImage acquisition
dQUANTIFICATION AND STATISTICAL ANALYSIS
BBehavioural analyses
BfMRI data preprocessing
BCortical layer definition
BLayer-specific ROI definition
BDecoding analysis
ACKNOWLEDGMENTS
This work was supported by a Leverhulme Trust project grant (RPG-2016-105)
and European Research Council (ERC) consolidator grant (101001592) under
the European Union’s Horizon 2020 research and innovation programme, both
awarded to C.P. P.K. was supported by a Wellcome/Royal Society Sir Henry
Dale Fellowship (218535/Z/19/Z) and an ERC starting grant (948548). E.R.T.
and DY were supported by the Leverhulme Trust grant awarded to C.P. and
J.H. by the ERC grant awarded to P.K. The Wellcome Centre for Human Neu-
roimaging is supported by core funding from the Wellcome Trust (203147/Z/
16/Z). We are grateful to Martina Callaghan for useful discussions.
AUTHOR CONTRIBUTIONS
Conceptualization, E.R.T., D.Y., P.K., and C.P.; formal analysis, investigation,
and project administration, E.R.T., J.H., and J.N.; writing – original draft,
E.R.T.; writing – review and editing, all authors.
ll
OPEN ACCESS
Current Biology 34, 1–7, May 20, 2024 5
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
DECLARATION OF INTERESTS
The authors declare no competing interests.
Received: January 22, 2024
Revised: April 9, 2024
Accepted: April 13, 2024
Published: May 1, 2024
REFERENCES
1. Bar, M. (2004). Visual objects in context. Nat. Rev. Neurosci. 5, 617–629.
https://doi.org/10.1038/nrn1476.
2. Yuille, A., and Kersten, D. (2006). Vision as Bayesian inference: analysis by
synthesis? Trends Cogn. Sci. 10, 301–308. https://doi.org/10.1016/j.tics.
2006.05.002.
3. de Lange, F.P., Heilbron, M., and Kok, P. (2018). How Do Expectations
Shape Perception? Trends Cogn. Sci. 22, 764–779. https://doi.org/10.
1016/j.tics.2018.06.002.
4. Press, C., Kok, P., and Yon, D. (2020). The Perceptual Prediction Paradox.
Trends Cogn. Sci. 24, 13–24. https://doi.org/10.1016/J.TICS.2019.
11.003.
5. Den Ouden, H.E., Kok, P., and De Lange, F.P. (2012). How Prediction
Errors Shape Perception, Attention, and Motivation. Front. Psychol. 3,
548. https://doi.org/10.3389/fpsyg.2012.00548.
6. Kaiser, D., Quek, G.L., Cichy, R.M., and Peelen, M.V. (2019). Object Vision
in a Structured World. Trends Cogn. Sci. 23, 672–685. https://doi.org/10.
1016/j.tics.2019.04.013.
7. Felleman, D.J., and Van Essen, D.C. (1991). Distributed Hierarchical
Processing in the Primate Cerebral Cortex. Cereb. Cortex 1, 1–47.
https://doi.org/10.1093/cercor/1.1.1-a.
8. Rockland, K.S., and Virga, A. (1989). Terminal arbors of individual
‘‘Feedback’’ axons projecting from area V2 to V1 in the macaque monkey:
A study using immunohistochemistry of anterogradely transported
Phaseolus vulgaris-leucoagglutinin. J. Comp. Neurol. 285, 54–72. https://
doi.org/10.1002/cne.902850106.
9. van Kerkoerle, T., Self,M.W., and Roelfsema, P.R. (2017). Layer-specificity
in the effects of attention and working memory on activity in primary visual
cortex. Nat. Commun. 8, 13804. https://doi.org/10.1038/ncomms13804.
10. Yu, Y., Huber, L., Yang, J., Jangraw, D.C., Handwerker , D.A., Molfese,
P.J., Chen, G., Ejima, Y., Wu, J., and Bandettini, P.A. (2019). Layer-specific
activation of sensory input and predictive feedback in the human primary
somatosensory cortex. Sci. Adv. 5, eaav9053. https://doi.org/10.1126/
sciadv.aav9053.
11. Friston, K. (2005). A theory of cortical responses. Philos. Trans. R. Soc.
Lond. B Biol. Sci. 360, 815–836. https://doi.org/10.1098/rstb.2005.1622.
12. Bastos, A.M., Usrey, W.M., Adams, R.A., Mangun, G.R., Fries, P., and
Friston, K.J. (2012). Canonical Microcircuits for Predictive Coding.
Neuron 76, 695–711. https://doi.org/10.1016/j.neuron.2012.10.038.
13. Kanai, R., Komura, Y., Shipp, S., and Friston, K. (2015). Cerebral hierar-
chies: Predictive processing, precision and the pulvinar. Philos. Trans.
R. Soc. Lond. B Biol. Sci. 370.https://doi.org/10.1098/rstb.2014.0169.
14. Rao, R.P.N., and Ballard, D.H. (1999). Predictive coding in the visual cor-
tex: a functional interpretation of some extra-classical receptive-field ef-
fects. Nat. Neurosci. 2, 79–87. https://doi.org/10.1038/4580.
15. Stephan, K.E., Petzschner, F.H., Kasper, L., Bayer, J., Wellstein, K.V.,
Stefanics, G., Pruessmann, K.P., and Heinzle, J. (2019). Laminar fMRI
and computational theories of brain function. NeuroImage 197, 699–706.
https://doi.org/10.1016/j.neuroimage.2017.11.001.
16. Yon, D., Thomas, E.R., Gilbert, S.J., de Lange, F.P., Kok, P., and Press, C.
(2023). Stubborn Predictions in Primary Visual Cortex. J. Cogn. Neurosci.
35, 1133–1143. https://doi.org/10.1162/jocn_a_01997.
17. Yon, D., Gilbert, S.J., De Lange, F.P., and Press, C. (2018). Action sharpens
sensory representations of expected outcomes. Nat. Commun. 9, 4288.
https://doi.org/10.1038/s41467-018-06752-7.
18. Peelen, M., and Downing, P. (2023). Testing cognitive theories using multi-
variate pattern analysis of neuroimaging data. Nat Hum Behav. 7, 1430–
1441. https://doi.org/10.1038/s41562-023-01680-z.
19. Aitken, F., Menelaou, G., Warrington, O., Koolschijn, R.S., Corbin, N.,
Callaghan, M.F., and Kok, P. (2020). Prior expectations evoke stimulus-
specific activity in the deep layers of the primary visual cortex. PLoS
Biol. 18, e3001023. https://doi.org/10.1371/journal.pbio.3001023.
20. van Mourik, T., van der Eerden, J.P.J.M., Bazin, P.L., and Norris, D.G.
(2019). Laminar signal extraction over extended cortical areas by means
of a spatial GLM. PLOS ONE 14, e0212493. https://doi.org/10.1371/jour-
nal.pone.0212493.
21. Lawrence, S.J.D., Formisano, E., Muckli, L., and De Lange, F.P. (2019).
Laminar fMRI: Applications for cognitive neuroscience. NeuroImage 197,
785–791. https://doi.org/10.1016/j.neuroimage.2017.07.004.
22. Duvernoy, H.M., Delon, S., and Vannson, J.L. (1981). Cortical blood ves-
sels of the human brain. Brain Res. Bull. 7, 519–579. https://doi.org/10.
1016/0361-9230(81)90007-1.
23. Koopmans, P.J., Barth, M., and Norris, D.G. (2010). Layer-specific BOLD
activation in human V1. Hum. Brain Mapp. 31, 1297–1304. https://doi.org/
10.1002/hbm.20936.
24. Markuerkiaga, I., Barth, M., and Norris, D.G. (2016). A cortical vascular
model for examining the specificity of the laminar BOLD signal.
NeuroImage 132, 491–498. https://doi.org/10.1016/j.neuroimage.2016.
02.073.
25. Heinzle, J., Koopmans, P.J., den Ouden, H.E.M., Raman, S., and Stephan,
K.E. (2016). A hemodynamic model for layered BOLD signals. NeuroImag e
125, 556–570. https://doi.org/10.1016/j.neuroimage.2015.10.025.
26. Audette, N.J., Zhou, W., La Chioma, A., and Schneider, D.M. (2022).
Precise movement-based predictions in the mouse auditory cortex.
Curr. Biol. 32, 4925–4940.e6. https://doi.org/10.1016/j.cub.2022.09.064.
27. Jordan, R., and Keller, G.B. (2020). Opposing Influence of Top-down and
Bottom-up Input on Excitatory Layer 2/3 Neurons in Mouse Primary Visual
Cortex. Neuron 108, 1194–1206.e5. https://doi.org/10.1016/j.neuron.
2020.09.024.
28. Bastos, A.M., Lundqvist, M., Waite, A.S., Kopell, N., andMiller, E.K. (2 020).
Layer and rhythm specificity for predictive routing. Proc. Natl. Acad. Sci.
USA 117, 31459–31469. https://doi.org/10.1073/pnas.2014868117.
29. Gillon, C.J., Pina, J.E., Lecoq, J.A., Ahmed, R., Billeh, Y.N., Caldejon, S.,
Groblewski, P., Henley, T.M., Kato, I., Lee, E., et al. (2023). Learning
from unexpected events in the neocortical microcircuit. Preprint at
bioRxiv. https://doi.org/10.1101/2021.01.15.426915.
30. Fiser, A., Mahringer, D., Oyibo, H.K., Petersen, A.V., Leinweber, M., and
Keller, G.B. (2016). Experience-dependent spatial expectations in mouse
visual cortex. Nat. Neurosci. 19, 1658–1664. https://doi.org/10.1038/
nn.4385.
31. Kok, P., Jehee, J.F.M., and de Lange, F.P. (2012). Less Is More:
Expectation Sharpens Representations in the Primary Visual Cortex.
Neuron 75, 265–270. https://doi.org/10.1016/j.neuron.2012.04.034.
32. Brandman, T., and Peelen, M.V. (2023). Objects sharpe n visual scene rep-
resentations: evidence from MEG decoding. Cereb. Cortex 33, 9524–
9531. https://doi.org/10.1093/cercor/bhad222.
33. Friston, K. (2018). Does predictive coding have a future?Nat. Neurosci.21,
1019–1021. https://doi.org/10.1038/s41593-018-0200-7.
34. Haarsma, J., Deveci, N., Corbin, N., Callaghan, M.F., and Kok, P. (2023).
Perceptual expectations and false percepts generate stimulus-specific
activity in distinct layers of the early visual cortex. J Neurosci. 43, 7946–
7957. https://doi.org/10.1523/JNEUROSCI.0998-23.2023.
35. Muckli, L., De Martino, F., Vizioli, L., Petro, L.S., Smith, F.W., Ugurbil, K.,
Goebel, R., and Yacoub, E. (2015). Contextual Feedback to Superficial
Layers of V1. Curr. Biol. 25, 2690–2695. https://doi.org/10.1016/j.cub.
2015.08.057.
ll
OPEN ACCESS
6Current Biology 34, 1–7, May 20, 2024
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
36. Kok, P., Bains, L.J., van Mourik, T., Norris, D.G., and de Lange, F.P. (2016).
Selective Activation of the Deep Layers of the Human Primary Visual
Cortex by Top-Down Feedback. Curr. Biol. 26, 371–376. https://doi.org/
10.1016/j.cub.2015.12.038.
37. Lu, H., Golay, X., Pekar, J.J., and Van Zijl, P.C.M. (2003). Functional mag-
netic resonance imaging based on changes in vascular space occupancy.
Magn. Reson. Med. 50, 263–274. https://doi.org/10.1002/mrm.10519.
38. Huber, L., Goense, J., Kennerley, A.J., Trampel, R., Guidi, M., Reimer, E.,
Ivanov, D., Neef, N., Gauthier, C.J., Turner, R., and Mo
¨ller, H.E. (2015).
Cortical lamina-dependent blood volume changes in human brain at 7 T.
NeuroImage 107, 23–33. https://doi.org/10.1016/j.neuroimage.2014.
11.046.
39. Lawrence, S.J.D., van Mourik, T., Kok, P., Koopmans, P.J., Norris, D.G.,
and de Lange, F.P. (2018). Laminar Organization of Working Memory
Signals in Human Visual Cortex. Curr. Biol. 28, 3435–3440.e4. https://
doi.org/10.1016/j.cub.2018.08.043.
40. Lawrence, S.J., Norris, D.G., and de Lange, F.P. (2019). Dissociable
laminar profiles of concurrent bottom-up and top-down modulation in
the human visual cortex. eLife 8, e44422. https://doi.org/10.7554/eLife.
44422.
41. Haarsma, J., Deveci, N., Corbin, N., Callaghan, M.F., and Kok, P. (2022).
Perceptual expectations and false percepts generate stimulus-specific
activity in distinct layers of the early visual cortex. J Neurosci. 43, 7946–
7957. https://doi.org/10.1523/JNEUROSCI.0998-23.2023.
42. de Sousa, A.A., Sherwood, C.C., Schleicher, A., Amunts, K., MacLeod,
C.E., Hof, P.R., and Zilles, K. (2010). Comparative Cytoarchitectural
Analyses of Striate and Extrastriate Areas in Hominoids. Cereb. Cortex
20, 966–981. https://doi.org/10.1093/cercor/bhp158.
43. Press, C., Kok, P., and Yon, D. (2020). Learning to Perc eive and Perceiving
to Learn. Trends Cogn. Sci. 24, 260–261. https://doi.org/10.1016/j.tics.
2020.01.002.
44. Thomas, E.R., Yon, D., De Lange, F.P., and Press, C. (2022). Action
Enhances Predicted Touch. Psychol. Sci. 33, 48–59. https://doi.org/10.
1177/09567976211017505.
45. Press, C., and Yon, D. (2019). Perceptual Prediction: Rapidly Making
Sense of a Noisy World. Curr. Biol. 29, R751–R753. https://doi.org/10.
1016/j.cub.2019.06.054.
46. Yu, A.J., and Dayan, P. (2005). Uncertainty, neuromodulation, and atten-
tion. Neuron 46, 681–692. https://doi.org/10.1016/j.neuron.2005.04.026.
47. Alilovi
c, J., Timmermans, B., Reteig, L.C., van Gaal, S., and Slagter, H.A.
(2019). No Evidence that Predictions and Attention Modulate the First
Feedforward Sweep of Cortical Information Processing. Cereb. Cortex
29, 2261–2278. https://doi.org/10.1093/cercor/bhz038.
48. Randeniya, R., Oestreich, L.K.L., and Garrido, M.I. (2018). Sensory predic-
tion errors in the continuum of psychosis. Schizophr. Res. 191, 109–122.
https://doi.org/10.1016/j.schres.2017.04.019.
49. Sterzer, P., Adams, R.A., Fletcher, P., Frith, C., Lawrie, S.M., Muckli, L.,
Petrovic, P., Uhlhaas, P., Voss, M., and Corlett, P.R. (2018). The
Predictive Coding Account of Psychosis. Biol. Psychiatry 84, 634–643.
https://doi.org/10.1016/j.biopsych.2018.05.015.
50. Kafadar, E., Fisher, V.L., Quagan, B., Hammer, A., Jaeger, H., Mourgues,
C., Thomas, R., Chen, L., Imtiaz, A., Sibarium, E., et al. (2022). Conditioned
Hallucinations and Prior Overweighting Are State-Sensitive Markers of
Hallucination Susceptibility. Biol. Psychiatry 92, 772–780. https://doi.
org/10.1016/j.biopsych.2022.05.007.
51. Haarsma, J., Kok,P., and Browning, M. (2022). The promiseof layer-specific
neuroimaging for testing predictive codingtheories of psychosis. Schizophr.
Res. 245,68–76.https://doi.org/10.1016/j.schres.2020.10.009.
52. Greve, D.N., and Fischl, B. (2009). Accurate and robust brain image align-
ment using boundary-based registration. NeuroImage 48, 63–72. https://
doi.org/10.1016/j.neuroimage.2009.06.060.
53. van Mourik, T., Koopmans, P.J., and Norris, D.G. (2019). Improved cortical
boundary registration for locally distorted fMRI scans. PLOS ONE 14,
e0223440. https://doi.org/10.1371/journal.pone.0223440.
54. Hebart, M.N., Go
¨rgen, K., and Haynes, J.D. (2014). The Decoding Toolbox
(TDT): a versatile software package for multivariate analyses of functional
imaging data. Front. Neuroinform. 8, 88. https://doi.org/10.3389/fninf.
2014.00088.
55. He, H., and Garcia, E.A. (2009). Learning from Imbalanced Data. IEEE
Trans. Knowl. Data Eng. 21, 1263–1284. https://doi.org/10.1109/TKDE.
2008.239.
ll
OPEN ACCESS
Current Biology 34, 1–7, May 20, 2024 7
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
STAR+METHODS
KEY RESOURCES TABLE
RESOURCE AVAILABILITY
Lead contact
Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Clare Press (c.press@
ucl.ac.uk).
Materials availability
This study did not generate any new materials.
Data and code availability
dThe GLM-generated beta data will be deposited at OSF and will be publicly available as of the date of publication. DOIs are
listed in the key resources table.
dAll original code will be deposited at OSF and will be publicly available as of the date of publication. DOIs are listed in the key
resources table.
dAny additional information required to re-analyse the data reported in this paper is available from the lead contact upon request.
EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS
Participants
Twenty-two participants (17 female, mean age = 26.09 years, SD = 3.41) were recruited from UCL and Birkbeck, University of Lon-
don, and paid a small honorarium for participation. All participants reported normal or corrected to normal vision and had no history of
psychiatric or neurological illness. We did not analyse the effects according to demographic differences, because we only collected
gender and age information, and there was no plan to conduct such analyses at any point. In principle this limits the generalisability of
our findings, but there is no evidence from previous work to suggest such low level and fundamental visual functions would differ
according to these characteristics. One participant’s data were excluded due to a technical error during acquisition, which meant
that event onsets in one run could not be modelled. This resulted in a final sample of 21 participants. The experiment was approved
by the UCL ethics committee.
METHOD DETAILS
Stimuli
Sinusoidal grating (Gabor) stimuli were created using MATLAB and presented against a grey background using Cogent Graphics.
During pre-scanner training, stimuli were presented on a 14’’ LCD screen (resolution: 1280x1024; refresh rate: 60 Hz) at a viewing
distance of 45 cm, and during scanning on an LCD monitor (resolution: 1280x1024; refresh rate: 60 Hz) through a mirror at a viewing
distance of 91 cm. In both sessions, stimuli were viewed at 15 degrees of visual angle. A Gaussian filter enveloped the grating stimuli
to create Gabor patches of 80% Michelson contrast, at 1.5 cycles per degree, and with random spatial phase. The Gabor stimuli were
presented in an annulus around a fixation cross in the middle of the screen (see Figure 1B). Two stimulus orientations were generated
to appear in CW (45) and CCW (135) orientations (relative to the hypothetical vertical mid-point e.g., 90).
Procedure
Main task
Participants completed two sessions. First, they completed a training session in which finger abductions perfectly predicted visually
presented Gabor orientations. The following day, they completed the same task in the MRI scanner but the action-outcome relation-
ship was degraded to 75% validity to allow for presentation of unexpected (25%) as well as expected events.
Participants completed the training session on Gorilla (www.gorilla.sc) for online experiments, taking part on either a laptop or
desktop computer no more than 24 hrs before the scanning session. Instruction at the beginning of the experiment requested par-
ticipants to set screen brightness to the maximum level to reduce variability in viewing conditions. Each trial started with a white
Deposited information Source Identifier
Analysis code and data This paper DOI: http://osf.io/s2z8c
ll
OPEN ACCESS
e1 Current Biology 34, 1–7.e1–e4, May 20, 2024
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
fixation cross. Participants were instructed to depress the ‘c’ and ‘m’ computer keys with their right index and little fingers, respec-
tively, until an imperative cue (e.g., square or triangle overlaid around the fixation cross) indicated which finger to abduct. A right-hand
index finger abduction would involve the finger moving left of hand midline, while a right hand little finger abduction moves right of the
midline. After the appropriate action was executed, the imperative cue was replaced with an oriented Gabor for 500 ms, resulting in
apparent synchrony of stimulus onset with action execution. A variable 300 – 500 ms delay followed stimulus offset and preceded a
response screen which asked about Gabor orientation. On half the trials they were asked to give a yes/no response to whether the
stimulus was oriented CW and on the other half they were asked whether it was oriented CCW. This design orthogonalised the Gabor
from the response. Participants were required to respond to the question screen within 1500 ms and the next trial started after a var-
iable ITI of 2000-3000 ms. Responses were made using the left thumb on the ‘a’ and ‘z’ keys for ‘yes’ or ‘no’, respectively. The
response question alternated every block. Participants completed the training task in ten runs of 36 trials each.
The following day, participants completed the test session at the Wellcome Centre for Human Neuroimaging, UCL. The test ses-
sion task was largely similar to the training session except that participants’ abductions now predicted the stimulus orientation with
75% validity and they performed actions using MR-compatible button boxes instead of the keyboard. The right-hand button box was
positioned orthogonally to the screen in the scanner, in line with the body midline. A short refresher of the training session was pre-
sented immediately before the scanning session, using the MR compatible button boxes outside of the scanner. Responses were
now required within 1000 ms of the question screen (to reduce scanning time), and the response question was randomly selected
on each trial. The next trial started after a variable ITI of 2000-6000 ms. Participants completed the test session in four scanning
runs that contained 96 trials each, and a 30 s break was presented mid-way through each run.
There were 384 main experimental trials in the scanning session, 360 online training trials and 192 refresher training trials. This num-
ber of trials was determined based on a preceding 3T fMRI study using a comparable task design.
16
Participants completed 32 prac-
tice trials before proceeding to the main trials in the initial training session. Imperative cue order and trial order were randomised
within blocks and the specific action-Gabor (predictive) relationship was counterbalanced across participants. The imperative
cue-action mapping was also counterbalanced and reversed halfway through each session (e.g., at the beginning of the sixth block
in training, and beginning of the third block in scanning) to deconfound potential influences of cue-outcome learning and remove any
correlation between the imperative action cues and actual or expected Gabor orientations across the experiment.
Localiser task
At the end of the main experiment, participants completed a functional localiser task in an additional scanning run. This task pre-
sented flickering Gabor stimuli at approximately 1.8 Hz along with a fixation cross. These Gabors were identical to those presented
in the main experiment except that they were presented at 100% contrast, and in blocks of 14 s. Each block containing flickering
Gabors was followed by a blank screen containing only the fixation cross for the same duration. In each stimulus block, Gabor orien-
tation was either CW or CCW and the presentation order was pseudorandomised. The task required participants to respond by
pressing any button when the central fixation cross changed colour from white to grey, ensuring that their fixation remained central.
In total, 32 blocks of flickering Gabors were presented, 16 of each orientation.
Image acquisition
Images were acquired using a 7T Magnetom MRI scanner (Siemens Healthcare GmbH, Erlangen, Germany) using a 32-channel head
coil at the Wellcome Centre for Human Neuroimaging, UCL. Functional images were acquired using T2*-weighted 3D gradient-echo
EPI sequence (3,552 ms volume acquisition time, TR = 74 ms, TE = 26.95 ms, 48 slices, 15flip angle, voxel size: 0.8 x 0.8 x 0.8 mm,
field of view: 192 x 192 x 39 mm). Structural images were acquired using a Magnetization Prepared Two Rapid Acquisition Gradient
Echo (MP2RAGE) sequence (TR = 5,000 ms, TE = 2.60 ms, TI = 900 ms, 240 slices, voxel size 0.7 30.7 30.7 mm, 5flip angle, field of
view 208 3208 3156 mm).
QUANTIFICATION AND STATISTICAL ANALYSIS
Behavioural analyses
RT data were collected for responses to expected and unexpected stimuli in the test session and median RTs were calculated for
correct trials separately for each condition, for each participant. Similarly, the proportion of correct responses was analysed for ex-
pected and unexpected conditions for each participant. One participant was removed from the behavioural analysis due to missing
almost half of the responses (44% of trials) and performing similarly to chance on the remainder (62% accuracy; note that this partic-
ipant was maintained for the imaging analysis, but the significance patterns were identical if they were removed).
fMRI data preprocessing
Preprocessing of the images was conducted in SPM12 and Freesurfer (http://surfer.nmr.mgh.harvard.edu/). Functional images were
cropped to select only the occipital lobe, to account for distortions in the frontal lobes. These cropped functional images were
spatially realigned to the mean image within runs, but also across runs. The temporal signal-to-noise ratio (tSNR, defined as
mean signal/SD over time) was calculated before and after spatial realignment and was found to be significantly higher after
(M=14.31, SD = 1.23) than before (M= 10.21, SD = 1.05) realignment (t(20) = -22.10, p<.001).
The realigned functional images were co-registered to the cortical surfaces estimated in participants’ MP2RAGE scans in several
steps. First, boundaries between grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF) were detected using Freesurfer
ll
OPEN ACCESS
Current Biology 34, 1–7.e1–e4, May 20, 2024 e2
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
on re-constructed structural scans (skull removed) and were manually corrected to remove any dura that was inaccurately classified
as part of the GM surface. A rigid body boundary-based registration (BBR)
52
was used to register GM boundaries to the mean func-
tional image, and a further recursive boundary-based registration (RBR)
53
applied the BBR recursively to portions of cortical mesh in
6 iterations.
Cortical layer definition
The level set method was used to divide GM into three equivolume layers for cortical layer definition (for details see
20
). This method
was used to separate five cortical bins (3 GM, WM and CSF) and determine three GM layers (deep, middle, and superficial) by calcu-
lating two intermediate surfaces between the WM and pial boundaries. In human V1, these three layer bins have been suggested to
correspond to histological layers 1 to 3, layer 4, and layers 5 and 6, respectively.
19
Layer-specific ROI definition
Freesurfer was used to define V1 based on anatomical landmarks in the MP2RAGE scans. ROIs were restricted to voxels from the
preprocessed functional localiser data that were most active during blocked presentation of the stimuli. This was achieved by model-
ling regressors for blocks of CW and CCW stimuli against baseline in a temporal GLM to identify voxels that expressed a significant
response to these stimuli (t> 2.3, p< 0.05; M= 5420.57, SD = 2228.12 number of voxels). A V1 mask of active voxels was created this
way for each participant.
Next, these active voxel masks were used to design a matrix of distributed voxels across each layer bin using the level-set defi-
nition described earlier.
20
These participant-specific design matrices specified the proportion of each active voxel across the 5 layer
bins specified above (3GM, WM, CSF), where each voxel was binned into one of the three GM layer bins according to its majority
proportion (Figure 2D; see online materials for a complementary univariate analysis approach). For example, a voxel that was spatially
located 7% in superficial, 76% in middle and 17% in deep layers would be labelled as a middle layer voxel and selected to contribute
to the voxels in the middle layer mask. An arbitrary threshold was set such that the majority proportion for a voxel to be included in a
layer mask was >0.4 (40%). This meant that any voxels with roughly equal proportion in each layer bin would not be selected. Impor-
tantly, the results did not change when this threshold was removed, since the majority of voxels’ ‘winning’ proportion was greater
than 0.4. Using this approach, three layer masks were created from the active V1 voxels for each participant.
This method of defining layer specific ROIs yields V1 layer masks that differ in the number of voxels that contribute to each layer in
each participant. Notably, there is a consistently greater number of voxels in superficial (M= 1613.95, SD = 678.17) than middle
(M= 1172.48, SD = 530.94) and deep (M= 907.62, SD = 442.68) layer masks (one-way ANOVA: F(2,40) = 120.50, p<.001,
np
2
=.86). We therefore performed another analysis to control for these differences, considering that greater information contributing
to the decoding signals in superficial layers relative to the other layers may confound our interpretations. Here, the steps are identical
to above, except that an additional step was performed to equalise the number of voxels present in each layer ROI mask. Specifically,
we defined the number of voxels to select in each mask as the maximum number common to all layers. For example, if the deep mask
had the fewest and contained 831 voxels, 831 voxels would be selected across all layers. Next, we loaded in an orientation prefer-
ence t-map from the GLM specified above, that contrasted CW and CCW regressors against each other, to select the (e.g., 831) most
orientation-tuned voxels from each layer. These voxels were those that contributed to each layer mask, such that each layer mask
contained an equivalent number of voxels in each layer. Another control version selected the (e.g. 831) voxels randomly from all the
active voxels in each layer. Importantly, the results did not change across these selection methods, suggesting that differences in
voxel numbers across layers should not alter interpretation (see Results).
Decoding analysis
Multivariate decoding analyses were implemented using the TDT toolbox
54
in MATLAB. We used a cross-classification approach
whereby a linear SVM was trained to discriminate Gabor orientations (CW or CCW) presented during the localizer task. This indepen-
dent dataset ensured that the trained classifier was not biased with any information about the predictability of stimuli. For this step,
we reran the GLM that we used for ROI definition above, but instead specified the onsets for each block in the localizer task as sepa-
rate regressors. Movement parameters were also modelled as nuisance regressors. This GLM resulted in 16 beta images for each
orientation that were fed into the SVM for training.
Next, we specified the test data in our cross-classification decoding approach from our main experimental task data. Specifically,
we reran and modified the GLM previously run on the main task data that included separate regressors for each condition type
(expected, unexpected) and stimulus type (CW, CCW) in each experiment run, in two ways. First, considering that we only had 4
scanning runs, yet it is well established that decoding data is more reliable with increased number of samples, we modelled each
condition according to the first and second halves of each run (since there was a 30s break in between continuous scanning;
note also that all trial types were balanced within each run half). This resulted in 8 condition regressors for each scanning run
(2x ExpCW1, ExpCCW1, UnexpCW1, UnexpCCW1). Second, we ensured that each modelled regressor would have equal weight
in terms of the number of trials contributing to each image, considering known biases in decoding performance with unequal numbers
of trials.
55
We therefore modelled expected conditions with the same number of trials as those that contribute to unexpected regres-
sors, by randomly sampling from expected trials to form three different expected regressors (see Results). Again, movement param-
eters were modelled as nuisance regressors.
ll
OPEN ACCESS
e3 Current Biology 34, 1–7.e1–e4, May 20, 2024
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
In total, this GLM resulted in 16 beta images (3x ExpCW, ExpCCW, 1x UnexpCW, UnexpCCW, twice in each scanning run). The
beta images from this GLM were grouped according to our main experimental conditions such that we repeated the decoding pro-
cedure separately four times (Expected [x3], Unexpected) to determine whether stimulus orientation classification differed across
each of these conditions. We tested each of these four decoding iterations separately, restricting the voxels to each of our three
V1 layer masks. This procedure resulted in 12 testing iterations (4 conditions in each of 3 masks). Accuracy of the SVM was calculated
as the proportion of correctly classified images across all decoding steps and was conducted separately for each participant. The
accuracy scores for each of the three expected conditions were averaged for each participant, providing one accuracy score for ‘ex-
pected’ trials, and one for ‘unexpected’ trials, in each of the layer masks. These scores were then compared between expected and
unexpected conditions, and across layers, to determine whether information about presented stimuli varied as a function of learned
expectation across the cortical layer bins.
The results were then analysed with a 2x3 repeated measures ANOVA with the factors experimental condition (expected, unex-
pected) and cortical layer (deep, middle, superficial). Follow up tests examined differences across layers, separately for expected
and unexpected events.
ll
OPEN ACCESS
Current Biology 34, 1–7.e1–e4, May 20, 2024 e4
Please cite this article in press as: Thomas et al., Predictions and errors are distinctly represented across V1 layers, Current Biology (2024), https://
doi.org/10.1016/j.cub.2024.04.036
Report
