ArticlePDF Available

Learning to optimize perceptual decisions through suppressive interactions in the human brain

Authors:

Abstract and Figures

Translating noisy sensory signals to perceptual decisions is critical for successful interactions in complex environments. Learning is known to improve perceptual judgments by filtering external noise and task-irrelevant information. Yet, little is known about the brain mechanisms that mediate learning-dependent suppression. Here, we employ ultra-high field magnetic resonance spectroscopy of GABA to test whether suppressive processing in decision-related and visual areas facilitates perceptual judgments during training. We demonstrate that parietal GABA relates to suppression of task-irrelevant information, while learning-dependent changes in visual GABA relate to enhanced performance in target detection and feature discrimination tasks. Combining GABA measurements with functional brain connectivity demonstrates that training on a target detection task involves local connectivity and disinhibition of visual cortex, while training on a feature discrimination task involves inter-cortical interactions that relate to suppressive visual processing. Our findings provide evidence that learning optimizes perceptual decisions through suppressive interactions in decision-related networks.
Task design, and MR spectroscopy (MRS) voxels. a Design: Each participant took part in a single session during which we acquired resting-state functional MRI (rs-fMRI) and MRS data. We collected MRS GABA during one block before (Baseline) and three blocks during training (T1, T2, T3). Each block comprised two MRS acquisitions: one from occipito-temporal (OCT voxel-black squares) and one from posterior parietal (PPC voxel-gray squares) cortex. The order of the voxels within each block was counterbalanced across participants. During each block, participants were presented with stimuli for 400 trials (200 trials per MRS voxel acquisition). b MRS voxel placement: We positioned the MRS voxels using anatomical landmarks (superior temporal gyrus and middle occipital gyrus for OCT and intraparietal sulcus for PPC) on the acquired T1 scan to ensure that voxel placement was consistent across participants. The mean GM tissue fraction was 44 ± 8% for OCT and 46 ± 7% for PPC and GM tissue content did not differ significantly between the two MRS Voxels (t(81) = 1.17, p = 0.24). The average distance of individual MRS voxels from the mean Montreal Neurological Institute (MNI) coordinates (OCT: x = −38.6 ± 4.4 mm, y = −67.8 ± 4.8 mm, z = 1.8 ± 4.4 mm; PPC: x = −31.6 ± 4.0 mm, y = −50.5 ± 6.5 mm, z = 40.7 ± 4.7 mm) across participants was small (i.e., lower than the MRS spatial resolution; 6.9 ± 3.6 mm for OCT and 7.9 ± 4.0 mm for PPC and did not differ between the two tasks (t(34) = 0.02, p = 0.99 for OCT; t(38) = 0.65, p = 0.52 for PPC). We computed the overlap across participant MRS voxels for OCT (yellow) and PPC (green) separately. We illustrate a group MRS mask (sagittal, coronal, axial view) that covers a cortical area that is common in at least 50% of the participants' MRS voxels. c MRS spectra: Example spectra from the OCT and PPC voxel for one participant (see Supplementary Figure 1 for an average MRS spectrum across participants and Supplementary Figure 2 for individual participant MRS spectra). We show the GABA fit using LC model (GABA peaks at 1.89 ppm, 2.29 ppm and 3.01 ppm), the residuals, the spectrum comprising all metabolites, the LC Model fit, and the baseline
… 
MR spectroscopy (MRS) measurements of γ-aminobutyric acid (GABA) and glutamate during training. a MRS-measured GABA over time is shown from two voxels (occipito-temporal, posterior parietal cortex) per task (signal-in-noise, feature differences). For each MRS voxel, we calculated % GABA change: we normalized GABA/tCr per training block (T1, T2, T3) to GABA/tCr recorded during the baseline block; that is, we computed GABA/tCr change subtracting GABA/tCr measurements in each of the three training blocks from the baseline block and then divided by GABA/tCr in the baseline block. We observed maximum (across blocks) 20% GABA change (mean across participants). This is consistent with previous studies measuring GABA at 3T that have reported changes in GABA between 10 and 15% within a single session (duration of 20–30 min) of stimulation⁷⁵ or training¹³. Ultra-high field imaging (i.e., 7T) has been shown to have increased sensitivity for MR spectroscopy measurements19,20 and may result in enhanced and more accurate detection of learning-dependent changes in GABA. b MRS-measured glutamate over time is shown from two voxels (occipito-temporal, posterior parietal cortex) per task (signal-in-noise, feature differences). For each MRS voxel, we calculated % glutamate change: we normalized glutamate/tCr per training block (T1, T2, T3) to glutamate/tCr recorded during the baseline block; that is, we computed glutamate/tCr change subtracting glutamate/tCr measurements in each of the three training blocks from the baseline block and then divided by glutamate/tCr in the baseline block. Error bars indicate standard error of the mean across participants. Error bars are not visible for small changes in metabolite concentrations
… 
Correlating γ-aminobutyric acid (GABA) change with behavioral improvement. a Skipped Pearson’s correlations showing a significant negative correlation of GABA/tCr change in occipito-temporal cortex with behavioral improvement for the signal-in-noise task (n = 16, r = -0.43, CI = [–0.75, –0.002]), whereas a significant positive correlation for the feature differences task (n = 19, r = 0.55, CI = [0.10, 0.78]). We computed GABA/tCr change for each participant as the difference between GABA/tCr in the training block with maximum performance and GABA/tCr in the baseline block. Note that the temporal resolution of GABA measurements does not allow us to separate the different processes associated with different events in a trial (e.g., stimulus vs. response). Thus, the correlations we report here relate learning-dependent changes in GABA levels to learning-dependent changes in overall task performance. The plots indicate that for a small number of participants the data deviated from the overall pattern of the correlation; e.g., for some participants in the SD task, GABA/tCr values were higher rather than lower compared with baseline. Our treatment of the data (i.e., GABA data are expressed as percent over baseline; behavioral improvement is expressed as percent over early performance) accounts for potential differences across participants in performance early in training or baseline GABA before training. It is possible that this individual variability was due to the single training session employed in our study during which participant performance did not saturate (i.e., average 72% maximum performance across participants). b Correlations of posterior parietal GABA/tCr change from baseline with behavioral improvement were not significant for the signal-in-noise (n = 17, r = –0.23, CI = [–0.61, 0.19]) or the feature differences task (n = 21, r = 0.05, CI = [–0.37, 0.43]). Significant correlations are indicated by closed symbols; nonsignificant correlations by open symbols
… 
This content is subject to copyright. Terms and conditions apply.
ARTICLE
Learning to optimize perceptual decisions through
suppressive interactions in the human brain
Polytimi Frangou 1, Uzay E. Emir 2,3, Vasilis M. Karlaftis 1, Caroline Nettekoven 4, Emily L. Hinson 3,4,
Stephanie Larcombe 3, Holly Bridge 3, Charlotte J. Stagg 3,4 & Zoe Kourtzi 1
Translating noisy sensory signals to perceptual decisions is critical for successful interactions
in complex environments. Learning is known to improve perceptual judgments by ltering
external noise and task-irrelevant information. Yet, little is known about the brain mechan-
isms that mediate learning-dependent suppression. Here, we employ ultra-high eld mag-
netic resonance spectroscopy of GABA to test whether suppressive processing in decision-
related and visual areas facilitates perceptual judgments during training. We demonstrate
that parietal GABA relates to suppression of task-irrelevant information, while learning-
dependent changes in visual GABA relate to enhanced performance in target detection and
feature discrimination tasks. Combining GABA measurements with functional brain con-
nectivity demonstrates that training on a target detection task involves local connectivity and
disinhibition of visual cortex, while training on a feature discrimination task involves inter-
cortical interactions that relate to suppressive visual processing. Our ndings provide evi-
dence that learning optimizes perceptual decisions through suppressive interactions in
decision-related networks.
https://doi.org/10.1038/s41467-019-08313-y OPEN
1Department of Psychology, University of Cambridge, Cambridge, CB2 3EB, UK. 2Purdue University School of Health Sciences, 550 Stadium Mall Drive,
West Lafayette, IN 47907, USA. 3Wellcome Centre for Integrative Neuroimaging, FMRIB, Nufeld Department of Clinical Neurosciences, University of
Oxford, Oxford OX3 9DU, UK. 4Oxford Centre for Human Brain Activity, Wellcome Centre for Integrative Neuroimaging, Department of Psychiatry,
University of Oxford, Oxford OX3 7JX, UK. Correspondence and requests for materials should be addressed to Z.K. (email: zk240@cam.ac.uk)
NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications 1
1234567890():,;
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Making successful decisions entails extracting meaningful
information from multiple sources in the environment
that are inherently noisy and ambiguous. Experience
and training have been shown to play a key role in optimizing
perceptual decisions13by ltering external noise (e.g., when
detecting targets in cluttered scenes) and retuning task-relevant
feature templates (e.g., when discriminating highly similar
objects)46. Previous functional magnetic resonance imaging
(fMRI) studies have demonstrated learning-dependent changes in
functional brain activity due to training on perceptual tasks that
involve detecting targets in clutter or discriminating ne feature
differences (for reviews7,8). However, fMRI does not allow us to
distinguish excitatory from inhibitory mechanisms of experience-
dependent plasticity, as BOLD reects aggregate activity from
both excitatory and inhibitory signals across large neural popu-
lations9. Thus, the inhibitory brain plasticity mechanisms that
support our ability to improve our perceptual decisions by
learning to suppress noisy and task-irrelevant information
through training remain largely unknown.
To investigate inhibitory mechanisms of learning-dependent
plasticity, we employed magnetic resonance spectroscopy (MRS)
that has only recently made it possible to measure γ-aminobutyric
acid (GABA), the primary inhibitory neurotransmitter in the
brain. Previous animal studies have linked decreased GABAergic
activity to learning and synaptic plasticity in primary motor
cortex10,11. In accordance with these ndings, human MRS stu-
dies have shown that GABA levels in the primary motor cortex
decrease following interventions that facilitate cortical reorgani-
sation12 and motor training13,14. In the visual cortex, human
MRS studies have shown that GABA levels relate positively to
performance in perceptual tasks1517. Further, decrease in visual
GABA has been shown to relate to homeostatic plasticity18. Here,
we took advantage of the high spectral resolution afforded by
ultra-high eld (7T) MRS to reliably resolve GABA19,20 and take
fast and reliable repeated measurements of functional GABA
during training. This allowed us to test changes in GABA during
training (i.e., while the participants were trained on a task),
extending beyond standard correlational approaches that relate
single measurements of GABA at baseline (i.e., when participants
are at rest) to behavior. Further, we tested whether changes in
GABAergic inhibition during task-specic training relate to
improvement in perceptual decisions.
To probe the brain mechanisms that support learning by
suppressing noisy and irrelevant signals, we employed two
learning tasks that have been shown to rely either on noise l-
tering or feature template retuning: (1) a signal-in-noise task that
involves extracting a target masked by noise, (2) a feature dif-
ferences task that involves judging ne differences21. Recent
computational investigations22,23 and animal studies propose
dissociable roles for inhibition in learning to interpret noisy
sensory signals vs. tuning ne feature processing. Based on this
work, we hypothesized that distinct GABAergic inhibition
mechanisms are involved in task-dependent learning and plasti-
city. Specically, we reasoned that decreased GABAergic inhibi-
tion during training would relate to improved ability to detect
targets in clutter, as changes in GABAergic inhibition have been
linked to neural gain, (i.e., changes in information transmission
between neurons24 or the slope of the neural input-output rela-
tionship25). In contrast, we reasoned that increased GABAergic
inhibition would relate to improved performance in ne feature
discrimination, as increased GABAergic inhibition has been
linked to enhanced orientation selectivity in visual cortex16,2628.
Further, previous neuroimaging and neurophysiology studies
have implicated distinct functional roles for the visual and pos-
terior parietal cortex (PPC) in sensory processing vs. perceptual
decision making, respectively29,30. To test the role of inhibitory
processing in learning for both visual and parietal cortex, we
implemented an imaging protocol that measured GABA in two
voxels (one in occipito-temporal (OCT), one in PPC) in alter-
nating order and allowed us to track changes in GABA in both
areas during training. Interestingly, previous studies have pro-
posed that perceptual learning is implemented by top-down
inuences from decision-related areas that re-weight processing
in sensory areas30,31. To test whether learning involves local
processing within visual cortex or suppressive interactions
between decision-related and sensory areas, we combined GABA
measurements in occipito-temporal and posterior parietal cortex
with functional brain connectivity, as measured by resting-state
fMRI. In particular, we tested the hypothesis that learning is
implemented by local inhibitory processing in visual cortex that is
gated by functional interactions between sensory and decision-
related areas. Specically, we tested whether learning-dependent
changes in visual cortex GABA relate to functional connectivity
between visual and posterior parietal cortex.
Our results reveal distinct GABAergic inhibition mechanisms
in a cortical network that is known to be involved in perceptual
decisions. In particular, increased parietal GABA with training
suggests suppression of task-irrelevant information. In contrast,
changes in occipito-temporal GABA with training relate to
enhanced target detection and discriminability, suggesting
learning-dependent changes in the processing of task-relevant
features. Further, analysis of functional brain connectivity at rest
reveals interactions within this network that relate to GABA
changes and behavioral improvement during training. Learning
to detect targets from clutter is implemented by local connectivity
and disinhibition of the visual cortex. In contrast, learning feature
differences is implemented by interactions between parietal and
visual areas that relate to increased GABAergic inhibition in
visual cortex. Our results provide evidence that learning improves
perceptual decisions through suppressive interactions within
decision-related circuits in the human brain.
Results
Training improves behavioral performance. We tested two
groups of participants on (a) a signal-in-noise (SN) task that
involves extracting shapes (radial vs. concentric Glass patterns)
masked by noise or (b) a feature differences (FD) task that
involves judging ne differences induced by morphing between
two stimulus classes (Fig. 1a). Participants were asked to judge the
identity of the stimulus presented per trial (i.e., radial vs. con-
centric). Our results showed that participants improved in their
judgments within a single training session during scanning
(Fig. 1b), consistent with previous reports showing fast behavioral
improvement early in the training (for a review, see ref. 32). A
linear mixed effects (LME) model with Task and Training Block
(6 blocks, 200 trials per block) as xed effects showed signicantly
improved performance during training across tasks (main effect
of Block: F(1, 249) =9.35, p=0.002). No signicant interaction
between Task × Block (F(1, 249) =0.10, p=0.75) was observed,
suggesting similar improvement across tasks (Fig. 1b).
To quantify behavioral improvement due to training in
individual participants, we compared performance at the
beginning of training (i.e., rst training block) to the maximum
performance achieved (i.e., when participants achieved highest
accuracy) by each participant during training (Fig. 1b) (see Sup-
plementary Methods). We chose this measure to capture
individual variability across participants that may be more
pronounced in our data, as participants were trained only for a
single training session in contrast to our previous studies that
have shown that performance saturates after multiple training
sessions on similar perceptual tasks33,34. A repeated measures
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y
2NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved
two-way analysis of variance (Task × Block) showed signicantly
improved performance during training across tasks (main effect
of block: F(1, 45) =59.88, p< 0.0001) but no signicant interac-
tion between Task × Block (F(1, 45) =1.07, p=0.31).
Learning-dependent changes in GABA. To test whether
GABAergic inhibition in the visual or parietal cortex changes
with training, we measured GABA (Fig. 2a) before (baseline
block) and during training. During the baseline block, partici-
pants were presented with random dot stimuli, while during the
training blocks participants performed either the SN or the FD
task on Glass pattern stimuli. We tested two MRS voxelsone
centered on the occipito-temporal cortex (OCT voxel) and the
other on the PPC (PPC voxel) (Fig. 2b)following previous
studies showing that these areas are involved in learning using the
same tasks and stimuli as in our study33,34. Each block comprised
one MRS acquisition per voxel and the order of the voxels within
each block was counterbalanced across participants.
Figure 3shows that OCT GABA/tCr changed in the two tasks
during training in opposite directions. In contrast, PPC GABA/
tCr changed in the same direction (i.e., increased during training)
for both tasks. These effects were supported by an LME analysis
(Task, Voxel, and MRS Blocks as xed effects) that modeled
GABA before (baseline block) and during (three training blocks)
training in both OCT and PPC. This analysis showed that
learning-dependent changes in GABA levels differed between
tasks and regions (Task × Voxel × Block: F(1, 264) =6.24, p=
0.01). In particular, we found that changes in OCT GABA (OCT
GABA) levels during training differed between tasks (LME model
for OCT GABA with Task and MRS Block as xed effects; Task ×
Block: F(1, 119) =10.77, p=0.001) (Fig. 3a). In contrast, GABA
changes in the PPC (PPC GABA) during training did not differ
signicantly between tasks (LME model for PPC GABA with
Task and MRS Block as xed effects; Task × Block: F(1, 145) =
0.18, p=0.68) (Fig. 3a). That is, PPC GABA increased during
training independent of Task (LME model for PPC GABA with
MRS Block as xed effect; main effect of Block: F(1, 147) =6.27,
t=2.50, p=0.01).
Specically, for the FD task, GABA changes during training did
not differ signicantly between OCT and PPC (LME model for
FD task GABA with Voxel and MRS Block as xed effects; Voxel
× Block: F(1, 143) =0.10, p=0.74). That is, GABA increased
signicantly during training independent of Voxel (LME model
for FD GABA with MRS Block as xed effect; main effect of MRS
Block: F(1, 145) =6.13, t=2.48, p=0.01). In contrast, for the SN
task GABA changes during training differed in the two regions
(LME model for SN task with Voxel and MRS Block as xed
effects; Voxel × Block: F(1, 121) =13.06, p=0.0004). That is, we
found a signicant decrease for OCT GABA (LME model for
SN task with MRS Block as xed effect; main effect of Block:
Radial
Concentric
Signal-in-noise Feature differences
a
Stimulus
prototypes
0.50 0.60 0.70 0.80 0.90
0.50
0.60
0.70
0.80
0.90
0.50 0.60 0.70 0.80 0.90
Maximum performance
Early performance Early performance
Signal-in-noise (SN)
Feature differences (FD)
b
400 600 800 1000 1200
−4
−2
0
2
4
6
8
10
12
14
Trials
Behavioral improvement (%)
200
Behavioral performance
Stimuli
Fig. 1 Tasks, stimuli, and behavioral results. aStimuli: Example stimuli comprising radial and concentric Glass patterns (stimuli are presented with inverted
contrast for illustration purposes). Stimuli are shown for the signal-in-noise task (SN task, 25% signal, spiral angle 0° for radial and 90° for concentric) and
the feature differences task (FD task, 100% signal, spiral angle 38° for radial and 52° for concentric). Prototype stimuli (100% signal, spiral angle 0° for
radial and 90° for concentric) are shown for illustration purposes only. bBehavioral performance across participants: We calculated improvement in task
performance during training as the difference in mean performance (i.e., mean accuracy) during each training block (200 trials) from the rst training block
(200 trials), divided by performance in the rst training block. Further, we compared individual participant accuracy early in training (rst 200 trials) to
the maximum accuracy achieved per participant during training (200 trials). Participants did not differ signicantly in their performance early in training
(t(45) =0.56, p=0.57) between the two tasks. Note that most participants (85%) achieved maximum performance during the last two MRS
measurements. Error bars indicate standard error of the mean across participants
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y ARTICLE
NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications 3
Content courtesy of Springer Nature, terms of use apply. Rights reserved
F(1,58) =16.65, t=4.08, p=0.0001), but a nonsignicant
increasing trend for PPC GABA (LME model for SN task with
MRS Block as xed effect; main effect of Block: F(1,63) =3.26, t
=1.80, p=0.08).
We conducted the following control analyses that corroborated
our results (see Supplementary Methods for more details). First,
we demonstrated that the learning-dependent changes we
observed in GABA levels: (a) were not simply driven by GABA
measurements at baseline, (b) could not be simply due to the
order with which the MRS voxels were acquired during training.
Second, we tested whether the learning-dependent changes we
observed in GABA, were simply due to differences in data quality
across multiple MRS measurements. In particular, we showed
that there were no signicant differences in the signal-to-noise
Design
MRS
rs-fMRI
Baseline T1 T2 T3
a
Blocks
b
Occipito-temporal (OCT)
Posterior parietal (PPC)
c
Occipito-temporal (OCT) Posterior parietal (PPC)
GABA
Raw data
Baseline
1234
Residual
1234
Fit
ppm ppm
MRS spectra
MRS voxel placement
Fig. 2 Task design, and MR spectroscopy (MRS) voxels. aDesign: Each participant took part in a single session during which we acquired resting-state
functional MRI (rs-fMRI) and MRS data. We collected MRS GABA during one block before (Baseline) and three blocks during training (T1, T2, T3). Each
block comprised two MRS acquisitions: one from occipito-temporal (OCT voxelblack squares) and one from posterior parietal (PPC voxelgray squares)
cortex. The order of the voxels within each block was counterbalanced across participants. During each block, participants were presented with stimuli for
400 trials (200 trials per MRS voxel acquisition). bMRS voxel placement: We positioned the MRS voxels using anatomical landmarks (superior temporal
gyrus and middle occipital gyrus for OCT and intraparietal sulcus for PPC) on the acquired T1 scan to ensure that voxel placement was consistent across
participants. The mean GM tissue fraction was 44 ± 8% for OCT and 46 ± 7% for PPC and GM tissue content did not differ signicantly between the two
MRS Voxels (t(81) =1.17, p=0.24). The average distance of individual MRS voxels from the mean Montreal Neurological Institute (MNI) coordinates
(OCT: x=38.6 ± 4.4 mm, y=67.8 ± 4.8 mm, z=1.8 ± 4.4 mm; PPC: x=31.6 ± 4.0 mm, y=50.5 ± 6.5 mm, z=40.7 ± 4.7 mm) across participants
was small (i.e., lower than the MRS spatial resolution; 6.9 ± 3.6 mm for OCT and 7.9 ±4.0 mm for PPC and did not differ between the two tasks (t(34) =
0.02, p=0.99 for OCT; t(38) =0.65, p=0.52 for PPC). We computed the overlap across participant MRS voxels for OCT (yellow) and PPC (green)
separately. We illustrate a group MRS mask (sagittal, coronal, axial view) that covers a cortical area that is common in at least 50% of the participants
MRS voxels. cMRS spectra: Example spectra from the OCT and PPC voxel for one participant (see Supplementary Figure 1 for an average MRS spectrum
across participants and Supplementary Figure 2 for individual participant MRS spectra). We show the GABA t using LC model (GABA peaks at 1.89 ppm,
2.29 ppm and 3.01 ppm), the residuals, the spectrum comprising all metabolites, the LC Model t, and the baseline
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y
4NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved
ratio (SNR) of MRS measurements over time, nor BOLD effects
on MRS spectra as represented by narrowing of the linewidth. In
addition, we conducted a no-training control experiment on an
independent group of participants (n=8). We measured OCT
GABA before and after exposing the participants to Glass pattern
stimuli for a duration comparable to the training in our main
study. During this time participants performed a xation task
rather than the SN or FD tasks. We did not observe any
signicant differences (t(7) =1.56, p=0.16) between two mea-
surements of OCT GABA, suggesting that the GABA changes we
observed in our main study were specic to task-training. Third,
we showed that the learning-dependent changes we observed in
GABA/tCr (a) could not be explained simply by changes in tCr
concentration over time (Supplementary Figure 5), (b) remained
signicant when we referenced GABA to water rather than tCr
concentration (Supplementary Figure 4), (c) were specicto
GABA and did not generalize to Glutamate (Fig. 3b).
Learning-dependent changes in GABA relate to behavior.We
next tested whether the learning-dependent changes we observed
in occipito-temporal and parietal GABA related to behavioral
improvement. Correlating change in OCT GABA
(see Supplementary Methods) to behavioral improvement showed
differences between tasks. Specically, we observed a signicant
negative correlation of OCT GABA change with behavioral
improvement (r=0.43, CI =[0.75, 0.02]) for the SN task,
whereas a signicant positive correlation (r=0.55, CI =[0.10,
0.78]) for the FD task (Fig. 4a). A linear regression analysis
conrmed this dissociation (OCT GABA change × Task inter-
action: F(1, 29) =9.03, p=0.005), suggesting that lower vs.
higher occipito-temporal GABA relates to improved performance
when learning to detect targets vs. discriminate feature differ-
ences, respectively. This dissociable result between tasks cannot
be simply explained by differences in task difculty, as partici-
pants showed similar behavioral improvement across tasks
(Fig. 1b).
In contrast to these correlations of OCT GABA change with
behavioral improvement, we did not observe any signicant
correlations between PPC GABA change and behavioral
improvement for either task (SN: r=0.23, CI =[0.61, 0.19];
FD: r=0.05, CI =[0.37, 0.43]) nor a signicant PPC GABA
change × Task interaction (F(1, 34) =0.57, p=0.45) (Fig. 4b).
Following previous work on the role of the PPC early rather than
later in training35, we next tested the link between PPC GABA
−40
−30
−20
−10
0
10
20
30
40
Baseline T1 T2 T3
Occipito-temporal cortex
GABA/tCr
GABA/tCr change (%)
−40
−30
−20
−10
0
10
20
30
40
Baseline T1 T2 T3
Signal-in-noise
Feature differences
Posterior parietal cortex
a
Glutamate/tCr
Glutamate/tCr change (%)
−40
−30
−20
−10
0
10
20
30
40
Baseline T1 T2 T3 −40
−30
−20
−10
0
10
20
30
40
Baseline T1 T2 T3
b
Fig. 3 MR spectroscopy (MRS) measurements of γ-aminobutyric acid (GABA) and glutamate during training. aMRS-measured GABA over time is shown
from two voxels (occipito-temporal, posterior parietal cortex) per task (signal-in-noise, feature differences). For each MRS voxel, we calculated % GABA
change: we normalized GABA/tCr per training block (T1, T2, T3) to GABA/tCr recorded during the baseline block; that is, we computed GABA/tCr change
subtracting GABA/tCr measurements in each of the three training blocks from the baseline block and then divided by GABA/tCr in the baseline block. We
observed maximum (across blocks) 20% GABA change (mean across participants). This is consistent with previous studies measuring GABA at 3T that
have reported changes in GABA between 10 and 15% within a single session (duration of 2030 min) of stimulation75 or training13. Ultra-high eld imaging
(i.e., 7T) has been shown to have increased sensitivity for MR spectroscopy measurements19,20 and may result in enhanced and more accurate detection
of learning-dependent changes in GABA. bMRS-measured glutamate over time is shown from two voxels (occipito-temporal, posterior parietal cortex) per
task (signal-in-noise, feature differences). For each MRS voxel, we calculated % glutamate change: we normalized glutamate/tCr per training block (T1, T2,
T3) to glutamate/tCr recorded during the baseline block; that is, we computed glutamate/tCr change subtracting glutamate/tCr measurements in each of
the three training blocks from the baseline block and then divided by glutamate/tCr in the baseline block. Error bars indicate standard error of the mean
across participants. Error bars are not visible for small changes in metabolite concentrations
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y ARTICLE
NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications 5
Content courtesy of Springer Nature, terms of use apply. Rights reserved
and performance early in training for the two tasks. A linear
regression model with baseline PPC GABA and Task as
predictors of behavioral performance (rst training block) showed
asignicant effect of PPC GABA (F(1, 37) =4.69, p=0.04), but no
interaction between PPC GABA and Task (F(1, 37) =0.03, p=
0.85). This result was conrmed by a signicant positive
correlation between baseline PPC GABA and performance in
the rst training block (r=0.34, CI =[0.06, 0.59]), suggesting
GABAergic inhibition in the parietal cortex before training relates
to suppression of task-irrelevant information early in the training
across tasks. We did not nd a signicant effect of OCT GABA
(F(1, 33) =2.14, p=0.15) nor an interaction between OCT
GABA and Task (F(1, 33) =2.33, p=0.14), suggesting that this
result linking GABA to performance early in training was specic
to parietal cortex.
Functional connectivity at rest relates to behavior and GABA.
Previous studies have shown that functional connectivity in
motor14 and visual36 networks relates to behavioral improvement
and learning-dependent plasticity. Further, GABAergic inhibition
has been suggested to shape network connectivity37. In particular,
extra-synaptic GABA has been shown to relate to local oscillatory
activity in the high gamma frequency range38 and to inter-
regional functional connectivity39. In the human brain, MRS-
−1 −0.5 0 0.5 1
−10
0
10
20
30
40
OCT GABA/tCr change
Signal-in-noise
Behavioral improvement (%)
Correlations of OCT GABA/tCr change with behavioral improvement
−1 −0.5 0 0.5 1
Feature differences
OCT GABA/tCr change
−10
0
10
20
30
40
a
Correlations of PPC GABA/tCr change with behavioral improvement
PPC GABA/tCr change
Behavioral improvement (%)
Signal-in-noise
−10
0
10
20
30
40
−1 −0.5 0 0.5 1
Feature differences
PPC GABA/tCr change
−1 −0.5 0 0.5 1
−10
0
10
20
30
40
b
Fig. 4 Correlating γ-aminobutyric acid (GABA) change with behavioral improvement. aSkipped Pearsons correlations showing a signicant negative
correlation of GABA/tCr change in occipito-temporal cortex with behavioral improvement for the signal-in-noise task (n=16, r=-0.43, CI =[0.75,
0.002]), whereas a signicant positive correlation for the feature differences task (n=19, r=0.55, CI =[0.10, 0.78]). We computed GABA/tCr change
for each participant as the difference between GABA/tCr in the training block with maximum performance and GABA/tCr in the baseline block. Note that
the temporal resolution of GABA measurements does not allow us to separate the different processes associated with different events in a trial (e.g.,
stimulus vs. response). Thus, the correlations we report here relate learning-dependent changes in GABA levels to learning-dependent changes in overall
task performance. The plots indicate that for a small number of participants the data deviated from the overall pattern of the correlation; e.g., for some
participants in the SD task, GABA/tCr values were higher rather than lower compared with baseline. Our treatment of the data (i.e., GABA data are
expressed as percent over baseline; behavioral improvement is expressed as percent over early performance) accounts for potential differences across
participants in performance early in training or baseline GABA before training. It is possible that this individual variability was due to the single training
session employed in our study during which participant performance did not saturate (i.e., average 72% maximum performance across participants).
bCorrelations of posterior parietal GABA/tCr change from baseline with behavioral improvement were not signicant for the signal-in-noise (n=17,
r=0.23, CI =[0.61, 0.19]) or the feature differences task (n=21, r=0.05, CI =[0.37, 0.43]). Signicant correlations are indicated by closed symbols;
nonsignicant correlations by open symbols
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y
6NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved
assessed GABA has been linked to functional connectivity as
measured by resting-state fMRI37,40,41, suggesting that GABAer-
gic inhibition may relate to local neural dynamics42. Here we test
whether functional connectivity between the OCT and PPC
relates to behavioral improvement and GABA changes during
training to detect targets in clutter vs. discriminate ne features.
First, we tested whether functional connectivity between these
regions relates to behavioral improvement in each of the two
learning tasks. We extracted the resting-state fMRI (rs-fMRI)
time course only from the gray matter voxels within the OCT
MRS voxel and the PPC MRS voxel (see Supplementary
Methods). We measured functional connectivity by correlating
the rs-fMRI time courses at rest between these two regions (OCT-
PPC connectivity) (see Supplementary Methods). We observed a
signicant positive correlation between OCT-PPC connectivity
and behavioral improvement for the FD task (r=0.37, CI =
[0.03, 0.69]), whereas a signicant negative correlation for the SN
task (r=0.72, CI =[0.90, 0.29]) (Fig. 5). This dissociation
in the relationship of OCT-PPC connectivity and behavioral
improvement between tasks was conrmed by a linear regression
showing a signicant interaction between OCT-PPC connectivity
and Task (F(1, 39) =10.72, p=0.002), suggesting that higher
connectivity between parietal and visual cortex facilitates learning
of ne feature differences. We interpret these results according to
previous studies that show both positive and negative values for
resting-state connectivity (e.g.,36). Positive connectivity values
(i.e., correlated signals) are typically interpreted to indicate
integration or cooperation within or between brain regions. In
contrast, negative connectivity values (i.e., anti-correlated signals)
are typically interpreted to indicate segregation of processing or
competition between networks. Thus, the correlations we
observed between resting-state connectivity and behavioral
improvement suggest higher cooperation between OCT and
PPC for learning ne features differences (FD task) than learning
to detect targets from noise (SN task). Further, to test whether
this link between functional connectivity and behavioral
improvement is specic to interactions between parietal and
visual areas, we extracted rs-fMRI for two additional control
regions: early visual cortex and motor cortex. We did not observe
any signicant results for correlations of behavioral improvement
and rs-fMRI connectivity between OCT and (a) early visual
cortex (SN: r=0.37, CI =[0.17, 0.70]; FD: r=0.22, CI =
[0.56, 0.15]) or (b) motor cortex (SN: r=0.26, CI =[0.61,
0.12]; FD: r=0.06, CI =[0.39, 0.42]) (see Supplementary
Methods).
Second, we tested whether OCT-PPC functional connectivity
relates to changes in visual cortex GABA during training. Our
results showed a signicant positive correlation between func-
tional connectivity and OCT GABA change for the FD task (r=
0.55, CI =[0.13, 0.87]), but not the SN task (r=0.27, CI =
[0.08, 0.60]) (Fig. 6a). These correlations were specic to GABA
changes in OCT. That is, there were no signicant correlations
between OCT-PPC connectivity and PPC GABA change for the
FD task (r=0.29, CI =[0.37, 0.69]) nor the SN task (r=0.20,
CI =[0.31, 0.60]). A multivariate linear regression showed that
OCT-PPC connectivity had a signicant effect on GABA change
in OCT (F(1, 14) =20.74, p=0.0005), but not PPC (F(1, 14) =
0.46, p=0.51) for the FD task. The correlation between
functional connectivity and OCT GABA change for the FD task
remained signicant (r=0.49, CI =[0.10, 0.82]) when we tested
for percentage GABA change (GABA change/baseline GABA),
suggesting that our results could not be due to differences in
baseline GABA. These results show that parietal-visual cortex
connectivity relates to GABAergic inhibition in visual cortex,
suggesting top-down inuences to suppressive processing in
visual cortex for learning ne feature differences.
Interestingly, for the SN task we observed a signicantly
negative correlation between connectivity within the OCT (i.e.,
OCT connectivity: connectivity measured as rs-fMRI correlations
across voxels within the OCT MRS voxel, see Supplementary
Methods) and OCT GABA change (r=0.55, CI =[0.86,
0.01]) (Fig. 6b). That is, higher OCT connectivity related to
decreased OCT GABA with training, suggesting that learning to
detect targets in clutter is supported by local connectivity and
decreased GABAergic inhibition within visual cortex. Correla-
tions of functional connectivity and GABA change in the OCT
were not signicant for the FD task (r=0.12, CI =[0.29, 0.51])
and were signicantly different from correlations for the SN task
(Z=2.33, p=0.02). The correlation between OCT connectivity
and GABA change for the SN task remained signicant
(r=0.55, CI =[0.88, 0.003]), when we used percentage
GABA change (GABA change/baseline GABA), suggesting that
our results could not be due to differences in baseline GABA.
Correlating OCT connectivity with PPC GABA change did not
show any signicant results (r=0.18, CI =[0.39, 0.60]),
suggesting that local connectivity relates specically to GABA
changes in visual cortex.
Taken together, our results suggest that learning ne
discriminations involves interactions between decision-related
(posterior parietal) and sensory (visual) areas that may facilitate
retuning of task-relevant features in visual cortex. In contrast,
learning to detect targets from clutter involves local processing
and decreased inhibition in visual cortex. To further test this
proposal, we performed moderation analyses43 that allowed us to
test whether the inuence that an independent variable (i.e.,
GABA change) has on the outcome (i.e., behavioral improve-
ment) is moderated by a moderator variable (i.e., connectivity).
Our results showed that OCT-PPC connectivity moderates the
relationship between OCT GABA and behavior for the FD task (F
(1, 15) =6.19, p=0.03) but not the SN task (F(1, 12) =0.74, p=
0.41). In contrast, local connectivity within OCT moderates the
relationship between GABA change and behavior for the SN task
(F(1, 12) =7.65, p=0.02) but not the FD task (F(1, 15) =2.80,
p=0.11). These moderation analyses suggest that the relation-
ship between learning-dependent changes in GABA and behavior
is moderated by functional connectivity; that is, interactions
between parietal and visual areas for the FD task, whereas local
connectivity within visual cortex for the SN task.
Discussion
Here, we provide evidence that learning improves perceptual
decisions (i.e., learning to detect targets in clutter or discriminate
highly similar features) by suppressive interactions in the human
brain. First, we demonstrate dissociable GABAergic inhibition
mechanisms for learning in a posterior cortical network (i.e.,
occipito-temporal and posterior-parietal cortex) known to be
involved in perceptual decisions. In particular, increased
GABAergic inhibition in the PPC with training suggests sup-
pressive processing of task-irrelevant information. In contrast,
changes in occipito-temporal GABA with training relate to
enhanced target detection and discriminability, suggesting
learning-dependent changes in the processing of behaviorally
relevant features. Second, we provide evidence that interactions
within this network, as measured by functional brain connectivity
at rest, gate suppressive processing of sensory signals. Learning to
detect targets from clutter involves local processing and disin-
hibition in visual cortex, while learning feature differences
involves suppressive interactions between decision-related (par-
ietal) and visual areas.
Previous studies have used MRS to measure GABA in the
context of visual15, sensory-motor4446, and learning tasks47,48.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y ARTICLE
NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications 7
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Signal-in-noise
Correlations of OCT-PPC connectivity with behavioral improvement
−0.8 −0.4 0 0.4 0.8
−10
0
10
20
30
40
Behavioral improvement (%)
Feature differences
−0.8 −0.4 0 0.4 0.8
−10
0
10
20
30
40
OCT−PPC connectivity
OCT−PPC connectivity
Fig. 5 Correlating occipito-temporal cortex (OCT)-posterior parietal cortex (PPC) functional connectivity with behavioral improvement. Skipped Pearsons
correlations showing a signicant negative correlation of OCT-PPC connectivity with behavioral improvement for the signal-in-noise task (n=21,
r=0.72, CI =[0.90, 0.29]), whereas a signicant positive correlation for the feature differences task (n=25, r=0.37, CI =[0.03, 0.69])
−0.6 −0.4 −0.2 0 0.2 0.4 0.6
−1
−0.5
0
0.5
1
aCorrelations of OCT-PPC connectivity with OCT GABA/tCr change
OCT GABA/tCr change
OCT−PPC connectivity
Signal-in-noise
−1
−0.5
0
0.5
1
−0.6 −0.4 −0.2 0 0.2 0.4 0.6
OCT−PPC connectivity
Feature differences
Correlations of OCT connectivity with OCT GABA/tCr change
Signal-in-noise
−1
−0.5
0
0.5
1
0.4 0.6 0.8 1 1.2
OCT connectivity
Feature differences
0.4 0.6 0.8 1 1.2
−1
−0.5
0
0.5
1
OCT connectivity
OCT GABA/tCr change
b
Fig. 6 Correlating functional connectivity with occipito-temporal cortex (OCT) γ-aminobutyric acid (GABA)/tCr change. aSkipped Pearsons correlations
showing a signicant positive correlation of OCT-posterior parietal cortex (PPC) connectivity with OCT GABA/tCr change for the feature differences task
(n=19, r=0.55, CI =[0.13, 0.87]), but not the signal-in-noise task (n=16, r=0.27, CI =[0.08, 0.60]). We computed GABA/tCr change for each
participant as the difference between GABA/tCr in the training block with maximum performance and GABA/tCr in the baseline block. bSkipped Pearsons
correlations showing a signicant negative correlation of functional connectivity within the occipito-temporal cortex with OCT GABA/tCr change for the
signal-in-noise task (n=16, r=0.55, CI =[0.86, 0.01]), but not the feature differences task (n=19, r=0.12, CI =[0.29, 0.51]). Signicant
correlations are indicated by closed symbols; nonsignicant correlations by open symbols
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y
8NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Here, we exploited the high spectral resolution afforded by ultra-
high eld (7T) MRS to reliably resolve GABA from other meta-
bolites19,20 and measure GABA changes over time at faster time
scales than typically recorded at lower eld strength. However, as
the temporal resolution of MRS-assessed GABA is in the order of
minutes, it is not possible to separate GABAergic inhibition
related to different processes within a trial (i.e., stimulus pro-
cessing, perceptual judgment, response, feedback); thus, it is likely
that the observed GABA changes relate to aggregate signals across
all processes involved in the tasks performed by the participants.
Despite this, our methodology provides two main advantages
over previous work. First, moving beyond the standard correla-
tional approach that relates single measurements of GABA at
baseline to behavior, we implemented an interventional approach
to test whether GABAergic inhibition changes due to task-specic
training. In particular, we captured learning-dependent changes
in functional GABA during training. Two previous studies
reporting measurements of GABA during training have focused
on motor cortex GABAergic inhibition during training on a
motor learning task13,49. Second, we implemented an MRS pro-
tocol that allows us to track longitudinal changes in GABA during
training from both OCT and PPC that are thought to play distinct
functional roles in perceptual learning29,30. Using this protocol, in
combination with functional connectivity as measured by rs-
fMRI, we tested for the role of suppressive interactions between
sensory and decision-related areas in perceptual learning.
Our ndings provide evidence for distinct GABAergic inhibi-
tion mechanisms in PPC vs. OCT for learning. In particular, we
demonstrate a common mechanism across tasks in the PPC. That
is, GABA increases during training may facilitate suppression of
task-irrelevant information (i.e., background clutter when learn-
ing to detect targets, task-irrelevant features when learning ne
differences). It is unlikely that changes in GABA levels with
training reect reduced attention to the task, as the task remained
sufciently demanding (i.e., mean maximum performance was
72%) during the single training session employed in our study.
Our ndings are consistent with the known role of parietal cortex
in perceptual decision-making and attentional selection. In par-
ticular, the PPC has been implicated in detecting low-saliency
targets by suppressing distractors50, providing a salient repre-
sentation of the environment and top-down attentional feed-
back51, accumulating sensory information52 and directing
attention to task-relevant features29,51. Interestingly, we show that
parietal cortex GABA before training relates to performance early
in training across tasks. This is in contrast with visual cortex
changes in GABA that relate to improvement in behavioral per-
formance during training. This nding suggests that suppression
of task-irrelevant information in the PPC may precede suppres-
sion in the visual cortex related to the processing of task-relevant
features. This is consistent with previous studies showing that
transcranial magnetic stimulation (TMS) in the parietal cortex
disrupts performance in visual discrimination tasks early in the
training compared with TMS in the visual cortex that disrupts
performance after training35.
In contrast to this common GABAergic inhibition mechanism
across tasks in the PPC, our results in the visual cortex suggest
distinct GABAergic mechanisms between tasks that relate to
improved behavioral performance. In particular, decreased
occipito-temporal GABA related to improved performance when
training to detect targets in clutter, while increased occipito-
temporal GABA related to improved performance when training
to discriminate ne feature differences. These results suggest that
learning to detect targets in clutter is implemented by decreased
local GABAergic inhibition that may facilitate noise ltering and
feature detection. Disinhibition of the visual cortex may facilitate
cortical recruitment that enhances probability summation and
SNR for target detection53. This mechanism is consistent with
previous animal studies linking GABAergic inhibition to neural
gain25 and interventional studies showing that blocking
GABAergic inhibition increases neural gain24. In contrast, dis-
criminating highly similar targets is implemented by increased
GABAergic inhibition in the visual cortex that may facilitate
learning through feature template retuning. This mechanism is
consistent with neurophysiological studies linking GABAergic
inhibition to cortical tuning26 and pharmacological interventions
showing that GABA agonists enhance orientation selectivity in
visual cortex27,28, whereas blocking GABAergic inhibition results
in broader neural tuning27. Interestingly, recent work has
implicated different populations of inhibitory interneurons in
neural gain vs. tuning54 that may contribute differentially to
learning by noise ltering vs. feature template retuning. In par-
ticular somatostatin (SOM)-positive interneurons have been
implicated in spatial summation55 and have been shown to gate
plasticity by providing contextual information56. In contrast,
parvalbumin (PV)-positive interneurons have been implicated in
selective inhibition16 that sharpens feature representations after
training57. It is therefore possible that SOM interneurons may
support learning to detect targets from clutter (SN task) through
noise ltering, whereas PV interneurons may support learning
ne differences (FD task) through feature retuning. Thus, our
ndings propose testable hypotheses for further animal studies on
the micro-circuits that mediate adaptive behavior and underlie
the macroscopic learning-dependent plasticity as measured by
human brain imaging.
We interpret these links between MRS-assessed GABA, beha-
vior and possible neural mechanisms with caution, as the precise
mechanisms that underlie changes in GABA as measured by MRS
remain under investigation. In particular, it is unclear whether
MRS-assessed GABA represents the entire pool of GABA avail-
able in a voxel. It is possible that the individual GABA pools are
not equally visible using MRS13,58. Another possibility is that
MRS-assessed GABA reects the exchange between the intra-
cellular and synaptic pools of GABA59. A number of studies using
paired-pulse TMS (e.g., 60,61) have shown that MRS-assessed
GABA does not relate to GABA
A
or GABA
B
activity and there-
fore it is unlikely to reect synaptic transmission of GABA. It is
more likely that MRS-assessed GABA reects ambient extra-
cellular GABA that contributes to tonic GABAergic activity (for
reviews e.g., 62,63). This is consistent with animal studies showing
that GABA synthesis64 is associated with Glutamic Acid Dec-
arboxylase (GAD)
67
that is predominantly found throughout the
cell, rather than GAD
65
that is found in axon terminals65. Finally,
we did not observe changes in glutamate simultaneously with the
change in GABA observed during training. Thus, it is unlikely
that the changes in GABA reect overall changes in the meta-
bolite cycling. GABA undergoes rapid turnover in the mamma-
lian cortex, and GAD activity has been shown to be rapidly
modulated in a variety of physiological processes66 in both
human67 and animal68 studies. Future studies combining invasive
investigations in animals (e.g., two photon imaging of inter-
neurons) with non-invasive MRS are necessary to shed more light
on the basis MRS-assessed GABA.
Finally, investigating functional connectivity within this pos-
terior cortical network revealed suppressive interactions that
relate to our ability to improve perceptual judgments through
training. We demonstrate that higher connectivity between par-
ietal and visual cortex at rest relates to increased GABA levels in
visual cortex during training and behavioral improvement in ne
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y ARTICLE
NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications 9
Content courtesy of Springer Nature, terms of use apply. Rights reserved
feature discrimination. This nding suggests top-down inuences
from the parietal cortex on suppressive processing in the visual
cortex for retuning feature templates. Previous theoretical inves-
tigations have proposed that sensory selectivity is enhanced by
suppressive feedback mechanisms that change recurrent proces-
sing in visual cortex23 and relate to attention-guided selection of
behaviorally relevant information69. Our ndings suggest that
suppression of task-irrelevant information in the parietal cortex
enhances tuning of task-relevant features in visual cortex through
top-down feedback, consistent with previous proposals that per-
ceptual learning re-weights sensory processing30,31. In contrast,
we show that higher connectivity within the visual cortex (rather
than connectivity between visual and parietal cortex) relates to
decreased GABA in visual cortex and behavioral improvement in
detecting targets from clutter. Lateral interactions within the early
visual cortex are shown to support contextual processing and
shape integration53,70. Further, recurrent processing has been
implicated in robust representations of ambiguous stimuli in
higher visual cortex71, suggesting that local connectivity in occi-
pito-temporal cortex facilitates visual processing under
uncertainty.
In sum, our ndings provide evidence for distinct GABAergic
inhibition mechanisms that support our ability to optimize per-
ceptual decisions through training. We propose that local sup-
pressive processing within visual cortex enhances target detection,
whereas top-down suppression from decision-related areas (i.e.,
posterior parietal cortex) enhances retuning of task-relevant
features for ne discrimination in the visual cortex. Decision-
making models have proposed suppressive mechanisms that
resolve competition between neuronal ensembles that represent
behavioral choices72. Our ndings provide novel insights in
understanding how these suppressive mechanisms are imple-
mented in decision-related and sensory areas in the human brain
and optimize our ability for perceptual decisions through
training.
Methods
Participants. Fifty participants (32 females; mean age 25.6 ± 3.3 years) took part in
this study. Data from three participants were not useable due to technical pro-
blems. All participants were right handed, had normal or corrected-to-normal
vision, and gave written informed consent. The study was approved the University
of Oxford Central University Research Ethics committee (MSD-IDREC-C1-2014-
081).
Stimuli. We presented participants with Glass patterns generated using previously
described methods33,34. In particular, stimuli were dened by white dot pairs
(dipoles) displayed within a square aperture on a black background. Experiment
stimuli (size =7.9o× 7.9o) were presented in the center of the screen. The dot
density was 3%, and the Glass shift (i.e., the distance between two dots in a dipole)
was 16.2 arc min. The size of each dot was 2.3 × 2.3 arc min2. For each dot dipole,
the spiral angle was dened as the angle between the dot dipole orientation and the
radius from the center of the dipole to the center of the stimulus aperture. Each
stimulus comprised dot dipoles that were aligned according to the specied spiral
angle (signal dipoles) for a given stimulus and noise dipoles for which the spiral
angle was randomly selected. The proportion of signal dipoles dened the stimulus
signal level.
We generated radial (0ospiral angle) and concentric (90ospiral angle) Glass
patterns by placing dipoles orthogonally (radial stimuli) or tangentially (concentric
stimuli) to the circumference of a circle centered on the xation dot. Further, we
generated intermediate patterns between these two Glass pattern types by
parametrically varying the spiral angle of the pattern from 0° (radial pattern) to 90°
(concentric pattern). We randomized the presentation of clockwise (0° to 90° spiral
angle) and counterclockwise patterns (0° to 90° spiral angle) across participants. A
new pattern was generated for each stimulus presented in a trial, resulting in
stimuli that were locally jittered in their position.
For the SN task, radial and concentric stimuli (spiral angle: 0° ± 90°) were
presented at 24% ± 1% signal level; that is, 76% of the dipoles were presented at
random position and orientation. For the FD task, stimuli were presented at 100%
signal and spiral angle of ± 38o(radial) or ± 52o(concentric) (Fig. 1a). To control
for stimulus-specic training effects, we presented each participant with a newly
generated set of stimuli. To control for potential local adaptation due to stimulus
repetition, we jittered ( ± 13°) the spiral angle across stimuli. These procedures
ensured that learning related to global shape rather than local stimulus features.
Experiment design. All participants took part in a single brain imaging session
(Fig. 2a) during which they were randomly assigned and trained on either the SN
or the FD task. We recorded whole-brain rs-fMRI data before training while
participants xated on a cross at the center for the screen.
Following the rs-fMRI scan, we recorded MRS GABA before and during
training (Fig. 2a). We collected MRS GABA from one baseline block before
training and three blocks during task training. Each block comprised two MRS
acquisitions: one from OCT and one from PPC. The order of the voxels within
each block was counterbalanced across participants. During each training block,
participants were presented with stimuli for 400 trials (200 trials per MRS voxel
acquisition). During the baseline MRS block (400 trials) participants engaged in a
task with similar stimuli as those presented during the training; that is, participants
viewed random dot patterns (0% signal dipoles) and were asked to respond (button
press) as soon as a pattern appeared. This ensured that differences in GABA
between blocks could not be simply attributed to differences in overall alertness.
During the MRS training blocks (3 MRS blocks, 400 trials each), participants were
presented with Glass patterns and were asked to judge and indicate by button press
whether the presented stimulus in each trial was radial or concentric. Two stimulus
conditions (radial vs. concentric Glass patterns; 200 trials per condition), were
presented for each training block. For each trial, a stimulus was presented for
300 ms and was followed by xation (i.e., blank screen with a central xation dot)
while waiting for the participants response (self-paced training paradigm). Trial-
by-trial feedback was provided by means of a visual cue (green tick for correct, red
xfor incorrect) followed by a xation dot for 500 ms before the onset of the next
trial. Average time to complete a trial ranged between 1.4 s and 1.9s across
participants (1.66 ± 0.16 s for SN; 1.69 ± 0.16 s for FD). Mean trial duration across
participants decreased across MRS blocks for both tasks (LME model for
Trial duration with Task and MRS Block as xed effects; main effect of Block:
F(1, 113) =4.68, p=0.03) but did not differ between tasks (main effect of Task:
F(1, 113) =0.12, p=0.73; Task × Block: F(1, 113) =0.002, p=0.97), suggesting
that participant became faster in their judgments with training for both tasks. Each
MRS acquisition lasted for 5 min and 56 s, and each MRS training block (400 trials)
for 11 min and 11 s ± 63 s. Thus, in most cases training was completed within the
duration of the training block (i.e., 2 MRS acquisitions × 5 min 56 s). In the event
that the training took longer than the MRS block, the next MRS acquisition was
delayed until completion of the previous training block.
Data acquisition. Experiments were conducted at the Wellcome Centre for Inte-
grative Neuroimaging, using a Siemens 7T Magnetom (Siemens, Erlangen) with a
32-channel head coil.
We acquired structural data (MPRAGE; TR 2200 ms; TE 2.82 ms; slice
thickness 1.0 mm; in-plane resolution 1.0 × 1.0 mm2; GRAPPA factor =4) and
echo planar imaging data (gradient echo-pulse sequences) from 40 slices (TR
2250 ms; TE 28 ms; slice thickness 3.0 mm; in-plane resolution 3.0 × 3.0 mm2;
GRAPPA factor =2, 140 volumes).
We acquired MRS data using a semi-localization by adiabatic selective
refocusing (semi-LASER) sequence73 (64 averages, TR 5010 ms, TE 36 ms). We
measured two MRS voxels (2 × 2 × 2 cm3isotropic), one in the left OCT (OCT
voxel) and one in the left PPC (PPC voxel), avoiding contact with the dura to
minimize macromolecule contamination. We focused on the left hemisphere as
previous fMRI50 and TMS35 studies have shown that the left PPC is involved in
suppressing distracting signals. To cover both areas with the same dielectric pad,
we placed both MRS voxels on the left hemisphere.
Statistical analyses. To compare changes in behavioral performance and neu-
rotransmitter concentrations across blocks, tasks, and MRS voxels, we used a LME
approach. LME models allow modeling of longitudinal data (i.e., multiple mea-
surements over time) and can account for missing values across participants. In
each of the models, we tested both for random (Participants) and xed (MRS
Block, Voxel, and Task) effects. To relate behavioral improvement to GABA
changes and rs-fMRI connectivity, we computed Pearson skipped correlations
using the Robust Correlation Toolbox74. This method accounts for potential
bivariate outliers and determines statistical signicance using bootstrapped con-
dence intervals (CI) for 1000 permutations. Note that bivariate outliers are not
shown in the data gures. To directly compare the relationship of GABA change
and rs-fMRI connectivity with behavioral improvement between the two tasks, we
used linear regression models with interaction terms. Data distribution assump-
tions of normality and heteroscedasticity of variance were veried using
ShapiroWilk and Levenes tests, respectively. All statistical tests are two sided.
Data availability
Data les have been uploaded on the Cambridge Data Repository: https://doi.org/
10.17863/CAM.33382.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y
10 NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Received: 16 May 2018 Accepted: 16 December 2018
References
1. Fine, I. & Jacobs, R. A. Comparing perceptual learning tasks: a review. J. Vis. 2,
190203 (2002).
2. Gilbert, C. D., Sigman, M. & Crist, R. R. The neural basis of perceptual
learning. Neuron 31, 681697 (2001).
3. Goldstone, R. L. Perceptual learning. Annu. Rev. Psychol. 49, 585612 (1998).
4. Dosher, B. A., Jeter, P., Liu, J. & Lu, Z.-L. An integrated reweighting theory of
perceptual learning. Proc. Natl Acad. Sci. USA 110, 1367813683 (2013).
5. Dosher, B. A. & Lu, Z.-L. Perceptual learning reects external noise ltering
and internal noise reduction through channel reweighting. Proc. Natl Acad.
Sci. USA 95, 1398813993 (1998).
6. Li, R. W., Levi, D. M. & Klein, S. A. Perceptual learning improves efciency by
re-tuning the decision templatefor position discrimination. Nat. Neurosci. 7,
178183 (2004).
7. Kourtzi, Z. Visual learning for perceptual and categorical decisions in the
human brain. Vision. Res. 50, 433440 (2010).
8. Welchman, A. E. & Kourtzi, Z. Linking brain imaging signals to visual
perception. Vis. Neurosci.30, 229241 (2013).
9. Logothetis, N. K. What we can do and what we cannot do with fMRI. Nature
453, 869878 (2008).
10. Castro-Alamancos, Ma, Donoghue, J. P. & Connors, B. W. Different forms of
synaptic plasticity in somatosensory and motor areas of the neocortex. J.
Neurosci. 15, 53245333 (1995).
11. Trepel, C. & Racine, R. J. GABAergic modulation of neocortical long-term
potentiation in the freely moving rat. Synapse 35, 120128 (2000).
12. Levy, L. M., Ziemann, U., Chen, R. & Cohen, L. G. Rapid modulation of
GABA in sensorimotor cortex induced by acute deafferentation. Ann. Neurol.
52, 755761 (2002).
13. Floyer-Lea, A. et al. Rapid modulation of GABA concentration in human
sensorimotor cortex during motor learning. J. Neurophysiol. 95, 16391644
(2006).
14. Sampaio-Baptista, C. et al. Changes in functional connectivity and GABA
levels with long-term motor learning. Neuroimage 106,1520 (2015).
15. Edden, R. A. E., Muthukumaraswamy, S. D., Freeman, T. C. A. & Singh, K. D.
Orientation discrimination performance is predicted by GABA concentration
and gamma oscillation frequency in human primary visual cortex. J. Neurosci.
29, 1572115726 (2009).
16. Rokem, A. et al. Broader visual orientation tuning in patients with
schizophrenia. Front. Hum. Neurosci. 5,19 (2011).
17. Song, C., Sandberg, K., Andersen, L. M., Blicher, J. U. & Rees, G. Human
occipital and parietal GABA selectively inuence visual perception of
orientation and size. J. Neurosci. 37, 89298937 (2017).
18. Lunghi, C., Emir, U. E., Morrone, M. C. & Bridge, H. Short-term monocular
deprivation alters GABA in the adult human visual cortex. Curr. Biol. 25,
14961501 (2015).
19. Puts, N. A. J. J. & Edden, R. A. E. E. In vivo magnetic resonance spectroscopy
of GABA: a methodological review. Prog. Nucl. Magn. Reson. Spectrosc. 60,
2941 (2012).
20. Tkác, I., Oz, G., Adriany, G., Uğurbil, K. & Gruetter, R. In vivo 1H NMR
spectroscopy of the human brain at high magnetic elds: metabolite
quantication at 4T vs. 7T. Magn. Reson. Med. 62, 868879 (2009).
21. Chang, D. H. F., Kourtzi, Z. & Welchman, A. E. Mechanisms for extracting a
signal from noise as revealed through the specicity and generality of task
training. J. Neurosci. 33, 1096210971 (2013).
22. Chalk, M., Masset, P., Gutkin, B. & Deneve, S. Sensory noise predicts divisive
reshaping of receptive elds. PLoS Comput. Biol. 13,126 (2017).
23. Moldakarimov, S., Bazhenov, M. & Sejnowski, T. J. Top-down inputs enhance
orientation selectivity in neurons of the primary visual cortex during
perceptual learning. PLoS Comput. Biol.https://doi.org/10.1371/journal.
pcbi.1003770 (2014).
24. Hamann, M., Rossi, D. J. & Attwell, D. Tonic and spillover inhibition of
granule cells control information ow through cerebellar cortex. Neuron 33,
625633 (2002).
25. Mitchell, S. J. & Silver, R. A. Shunting inhibition modulates neuronal gain
during synaptic excitation. Neuron 38, 433445 (2003).
26. Wehr, M. & Zador, A. M. Balanced inhibition underlies tuning and sharpens
spike timing in auditory cortex. Nature 426, 442446 (2003).
27. Leventhal, A. G., Wang, Y., Pu, M., Zhou, Y. & Ma, Y. GABA and its agonists
improved visual cortical function in senescent monkeys. Science 300, 812815
(2003).
28. Li, G. et al. GABA-mediated inhibition correlates with orientation selectivity
in primary visual cortex of cat. Neuroscience 155, 914922 (2008).
29. Freedman, D. J. & Ibos, G. An integrative framework for sensory, motor, and
cognitive functions of the posterior parietal cortex. Neuron 97, 12191234
(2018).
30. Law, C.-T. & Gold, J. I. Shared mechanisms of perceptual learning and
decision making. Top. Cogn. Sci. 2, 226238 (2010).
31. Ahissar, M. & Hochstein, S. The reverse hierarchy theory of visual perceptual
learning. Trends Cogn. Sci. 8, 457464 (2004).
32. Sagi, D. Perceptual learning in vision research. Vision Res. 51, 15521566
(2011). https://doi.org/10.1016/j.visres.2010.10.01
33. Li, S. et al. Learning shapes the representation of behavioral choice in the
human brain. Neuron 62, 441452 (2009).
34. Mayhew, S. D., Li, S. & Kourtzi, Z. Learning acts on distinct processes for
visual form perception in the human brain. J. Neurosci. 32, 775786 (2012).
35. Chang, D. H. F., Mevorach, C., Kourtzi, Z. & Welchman, A. E. Training
transfers the limits on perception from parietal to ventral cortex. Curr. Biol.
24, 24452450 (2014).
36. Baldassarre, A. et al. Individual variability in functional connectivity predicts
performance of a perceptual task. Proc. Natl Acad. Sci. USA 113, E6723E6723
(2012).
37. Stagg, C. et al. Local GABA concentration is related to network-level resting
functional connectivity. eLife 2014,19 (2014).
38. Towers, S. K. et al. α5 subunit-containing GABAA receptors affect the
dynamic range of mouse hippocampal kainate-induced gamma frequency
oscillations in vitro. J. Physiol. 559, 721728 (2004).
39. Shmuel, A. & Leopold, D. A. Neuronal correlates of spontaneous uctuations
in fMRI signals in monkey visual cortex: implications for functional
connectivity at rest. Hum. Brain. Mapp. 29, 751761 (2008).
40. Bachtiar, V., Near, J., Johansen-Berg, H. & Stagg, C. Modulation of GABA and
resting state functional connectivity by transcranial direct current stimulation.
eLife 4,19 (2015).
41. Kapogiannis, D., Reiter, D. A., Willette, A. A. & Mattson, M. P. Posteromedial
cortex glutamate and GABA predict intrinsic functional connectivity of the
default mode network. Neuroimage 64, 112119 (2013).
42. Cabral, J., Hugues, E., Sporns, O. & Deco, G. Role of local network oscillations
in resting-state functional connectivity. Neuroimage 57, 130139 (2011).
43. Hayes, A. F. PROCESS: A Versatile Computational Tool for Observed Variable
Mediation, Moderation, and Conditional Process Modeling (Guildford Press,
2012). https://doi.org/978-1-60918-230-4
44. Kolasinski, J. et al. A mechanistic link from GABA to cortical architecture and
perception. Curr. Biol.27, 16851691 (2017).
45. Heba, S. et al. Local GABA concentration predicts perceptual improvements
after repetitive sensory stimulation in humans. Cereb. Cortex 26, 12951301
(2016).
46. Stagg, C., Bachtiar, V. & Johansen-Berg, H. The role of GABA in human
motor learning. Curr. Biol. 21, 480484 (2011).
47. Scholl, J. et al. Excitation and inhibition in anterior cingulate predict use of
past experiences. eLife 6, e20365 (2017).
48. Shibata, K. et al. Overlearning hyperstabilizes a skill by rapidly making
neurochemical processing inhibitory-dominant. Nat. Neurosci.20, 470475
(2017).
49. Kolasinski, J. et al. The dynamics of cortical GABA in human motor learning.
J. Physiol.597, 271282 (2019).
50. Mevorach, C., Humphreys, G. W. & Shalev, L. Reexive and preparatory
selection and suppression of salient information in the right and left posterior
parietal cortex. J. Cogn. Neurosci. 21, 12041214 (2009).
51. Gottlieb, J. From thought to action: the parietal cortex as a bridge between
perception, action, and cognition. Neuron 53,916 (2007).
52. Mazurek, M. E., Roitman, J. D., Ditterich, J. & Shadlen, M. N. A role for neural
integrators in perceptual decision making. Cereb. Cortex 13, 12571269
(2003).
53. Gilbert, C. D. & Li, W. Adult visual cortical plasticity. Neuron 75, 250264
(2012).
54. Phillips, E. A. & Hasenstaub, A. R. Asymmetric effects of activating and
inactivating cortical interneurons. eLife 5, e18383 (2016).
55. Adesnik, H., Bruns, W., Taniguchi, H., Huang, Z. J. & Scanziani, M. A neural
circuit for spatial summation in visual cortex. Nature 490, 226231 (2012).
56. van Versendaal, D. & Levelt, C. N. Inhibitory interneurons in visual cortical
plasticity. Cell. Mol. Life Sci.73, 36773691 (2016).
57. Khan, A. G. et al. Distinct learning-induced changes in stimulus selectivity and
interactions of GABAergic interneuron classes in visual cortex. Nat. Neurosci.
21, 851859 (2018).
58. Stagg, C. et al. Neurochemical effects of theta burst stimulation as assessed by
magnetic resonance spectroscopy. J. Neurophysiol. 101, 28722877 (2009).
59. Ashton, A. C. & Ushkaryov, Y. A. Properties of synaptic vesicle pools in
mature central nerve terminals. J. Biol. Chem. 280, 3727837288 (2005).
60. Stagg, C. et al. Relationship between physiological measures of excitability and
levels of glutamate and GABA in the human motor cortex. J. Physiol. 589,
58455855 (2011).
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y ARTICLE
NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved
61. Tremblay, S. et al. Relationship between transcranial magnetic stimulation
measures of intracortical inhibition and spectroscopy measures of GABA and
glutamate+glutamine. J. Neurophysiol.109, 13431349 (2013).
62. Johnstone, A., Levenstein, J. M., Hinson, E. L. & Stagg, C. Neurochemical
changes underpinning the development of adjunct therapies in recovery after
stroke: a role for GABA? J. Cereb. Blood Flow Metab. 38, 15641583 (2017).
63. Stagg, C. Magnetic resonance spectroscopy as a tool to study the role of GABA
in motor-cortical plasticity. Neuroimage 86,1927 (2014).
64. Mason, G. F. et al. Decrease in GABA synthesis rate in rat cortex following
GABA-transaminase inhibition correlates with the decrease in GAD67
protein. Brain Res. 914,8191 (2001).
65. Kaufman, D. L., Houser, C. R. & Tobin, A. J. Two forms of the γaminobutyric
acid synthetic enzyme glutamate decarboxylase have distinct intraneuronal
distributions and cofactor interactions. J. Neurochem. 56, 720723 (1991).
66. Garraghty, P. E., Lachica, E. A. & Kaas, J. H. Injury-induced reorganization of
somatosensory cortex is accompanied by reductions in GABA staining.
Somatosens. Mot. Res. 8, 347354 (1991).
67. Shen, J. et al. Determination of the rate of the glutamate/glutamine cycle in the
human brain by in vivo 13C NMR. Proc. Natl Acad. Sci. USA 96, 82358240
(1999).
68. Manor, D. et al. The rate of turnover of cortical GABA from [1-13C]glucose is
reduced in rats treated with the GABA -transaminase inhibitor vigatrin (g-
vinyl GABA). Neurochem. Res. 21, 10311041 (1996).
69. Roelfsema, P. R. & Ooyen, A. Attention-gated reinforcement learning of
internal representations for classication. Neural Comput. 17, 21762214
(2005).
70. Stettler, D. D., Das, A., Bennett, J. & Gilbert, C. D. Lateral connectivity and
contextual interactions in macaque primary visual cortex. Neuron 36, 739750
(2002).
71. OReilly, R. C., Wyatte, D., Herd, S., Mingus, B. & Jilk, D. J. Recurrent
processing during object recognition. Front. Psychol. 4,114 (2013).
72. Bogacz, R., Brown, E., Moehlis, J., Holmes, P. & Cohen, J. D. The physics of
optimal decision making: a formal analysis of models of performance in two-
alternative forced-choice tasks. Psychol. Rev. 113, 700765 (2006).
73. Scheenen, T., Klomp, D., van de Moortele, P., Adriany, G. & Heerschap, A.
Semi-LASER 1H-MR spectroscopic imaging of the human brain at 7T. Proc.
16th Sci. Meet. Int. Soc. Magn. Reson. Med.59,16 (2008).
74. Pernet, C. R., Wilcox, R., Rousselet, G. A., Finch, H. & Budden, J. S. Robust
correlation analyses: false positive and power validation using a new open
source Matlab toolbox. Front Psychol. https://doi.org/10.3389/
fpsyg.2012.00606 (2013).
75. Stagg, C. et al. Polarity-sensitive modulation of cortical neurotransmitters by
transcranial stimulation. J. Neurosci. 29, 52025206 (2009).
Acknowledgements
We would like to thank Adam Berrington and Clark Lemke for help with the data
collection, Joseph Giorgio and Ke Jia for help on preliminary data analysis, Rogier Kievit,
Delia Fuhrmann, and Mark Haggard for advice on statistical analyses, Adrian Garcia for
preliminary work, Jasper Poort for helpful comments. This work was supported by
funding to Z.K. from the Alan Turing Institute, the Biotechnology and Biological Sci-
ences Research Council (grants: H012508, P021255), the European Communitys
Seventh Framework Programme (grant FP7/20072013 under agreement PITN-GA
2011290011), and the Wellcome Trust (grant 205067). C.J.S. holds a Sir Henry Dale
Fellowship, funded by the Wellcome Trust and the Royal Society (102584/Z/13/Z). E.L.
H. is supported by the NIHR Oxford Health Biomedical Research Centre. The Wellcome
Centre for Integrative Neuroimaging is supported by core funding from the Wellcome
Trust (203139/Z/16/Z).
Author contributions
P.F. conceived the study, collected the data, performed the analysis and wrote the
manuscript. U.E.E. provided acquisition and analysis tools for MR Spectroscopy. V.K.
performed the analysis of the rs-fMRI data. C.N., E.L.H. and S.L. collected the data. H.B.
and C.J.S guided the study design and analysis. Z.K. conceived the study, guided the
analysis, and wrote the manuscript.
Additional information
Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467-
019-08313-y.
Competing interests: The authors declare no competing interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
Journal peer review information:Nature Communications thanks the anonymous
reviewers for their contribution to the peer review of this work.
Publishers note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional afliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the articles Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
articles Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit http://creativecommons.org/
licenses/by/4.0/.
© The Author(s) 2019
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-08313-y
12 NATURE COMMUNICATIONS | (2019) 10:474 | https://doi.org/10.1038/s41467-019-08313-y | www.nature.com/naturecommunications
Content courtesy of Springer Nature, terms of use apply. Rights reserved
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com

Supplementary resource (1)

... Provencher (2021) 70 . According to this criteria, the following data were excluded: within the SMA voxel, GABA in 1 healthy and 1 OCD subjects, and Gln in 2 OCD patients, within the OCC voxel, 2 Glu and Gln in 2 healthy subjects, and 1 GABA in a healthy subject, and 3 GABA in patients. ...
... Next, values larger than 2SD from the mean of each group were excluded for both measures. The latter were according toFrangou et al. (2019) ...
Preprint
Full-text available
There has been relatively little analysis of possible neurochemical correlates of compulsive behavior to illuminate its underlying neural mechanisms. We utilised 7-Tesla proton magnetic resonance spectroscopy ( ¹ H-MRS) to assess the balance of excitatory and inhibitory neurotransmission by measuring glutamate and GABA levels in anterior cingulate cortex and supplementary motor area (SMA) of healthy volunteers and patients with Obsessive-Compulsive Disorder (OCD). Within the SMA, trait and clinical measures of compulsive behavior were related to glutamate levels, whereas a behavioral index of habitual control correlated with the glutamate:GABA ratio. OCD patients additionally exhibited elevated glutamate levels and glutamate:GABA ratios in anterior cingulate cortex which also correlated with habitual control. This study highlights important underlying relationships between SMA mechanisms of habitual control relevant to compulsive behavior, common to the healthy sub-clinical and OCD populations. The results also demonstrate additional involvement of anterior cingulate in the balance between goal-directed and habitual responding in OCD.
... This means, both of these criteria needed to be respected for the data to be included. This led to exclusion of quite a few metabolite per voxel, however, this is comparable to other similar studies using the same exclusion criteria (Frangou et al., 2019) and helps avoiding introduction of noise in the data and loss of informative but unusual data. For ACC, in OCD patients, 2 GABA, 1 Gln, 2 Gln, and 1 NAA, and in healthy group, 2 GABA, 1 Glu, and 1 NAA values were excluded. ...
Thesis
Full-text available
The introductory Chapter 1 reviewed several possible explanations of compulsive behaviour as manifested especially in Obsessive Compulsive Disorder (OCD), and schizophrenia, following treatment with the second generation antipsychotic medication clozapine. Particular focus is placed on compulsive checking behaviours and their relationship to current theories of compulsivity based on the hypothesis of imbalance between the goal-directed and habit systems and aberrant prediction-error learning. Chapter 2 describes experimental attempts in this thesis to measure human checking behaviour in the laboratory. Initially, a previously published test of checking was employed which however failed to show significant increases in OCD. I then designed a new testing procedure to measure checking, based on perceptual decision-making under a time constraint. This was administered together with other cognitive tests to patients with OCD, clozapine treated schizophrenia patients without and with obsessive compulsive symptoms, and a healthy volunteer group. In general, there were no major differences compared to controls, although patients with schizophrenia performed worse. In a second study, contingency degradation learning and checking were measured using a second variant of the task in which there was no time constraint. However, significant increases in checking behaviour were shown in another group of OCD patients compared with healthy volunteers. In Chapter 3, after a review of previous findings in OCD using the Magnetic Resonance Spectroscopy (MRS) technique at 7T, the same participants employed in the last checking study, were subjected to MRS scans to measure GABA, Glutamate (Glu), Glutamine (Gln), and NAA in three areas of brain including the Anterior Cingulate Cortex (ACC), Supplementary Motor Area (SMA) and the Visual Cortex. The most important findings were in the ACC, where significantly higher levels of Glu and Gln and lower levels of GABA and GABA:Glu ratio were found in OCD patients compared to the healthy group. In Chapter 4, the relationship between the behavioural results from chapter 2 and the neurometabolites measured with MRS in our OCD and healthy participants in chapter 3 was examined. The major findings were: 1) Higher ACC GABA/Glu ratio was related to superior accuracy of decision-making as well as increased checking on the checking task in OCD patients. 2) Checking was negatively correlated with SMA Glu in the healthy group but not in OCD. Moreover, in a test of goal directed behaviour and habit learning based on contingency degradation, a positive relationship was evident between performance and the ACC GABA/Glu ratios in patients for full degradation of the task contingencies. A similar positive relationship was observed for healthy volunteers for GABA/Glu ratios in SMA for partial degradation of the contingencies. Chapter 5 discusses neuropsychological interpretations of our findings in relation to the symptomatology of OCD and schizophrenia, together with their implications for understanding the role of the anterior cingulate cortex in decision-making and compulsive behaviour.
... This is in agreement with previous studies demonstrating a negative relationship between regional neural activation and GABA, and a deactivation of GABAergic mechanisms when excitation is required (Duncan et al., 2014;Kiemes et al., 2021). It has been suggested that task-related GABA changes are more robustly observed in stimulus paradigms with a change in behavioural performance (Ip and Bridge, 2021), such as learning (Frangou et al., 2018(Frangou et al., , 2019, motor or sensory performance (Stagg et al., 2011;Heba et al., 2016;Kolasinski et al., 2019), and stress (Houtepen et al., 2017;Lynn et al., 2018a); this is reflected in our meta-analysis results. fMRS studies in pain appeared to be most consistent, but most domains show huge variation in their responses. ...
Preprint
Full-text available
Functional magnetic resonance spectroscopy (fMRS) can be used to investigate neurometabolic responses to external stimuli in-vivo, but findings to date are inconsistent. We performed a systematic review and meta-analysis on 49 human fMRS studies that measured levels of the primary neurotransmitters Glutamate (Glu), Glx (Glutamate + Glutamine), and GABA. Data were extracted, grouped by metabolite, stimulus domain, and brain region, and then analysed by determining standardised effect sizes. The quality of individual studies (such as the choice of acquisition parameters and reporting quality) was rated with a standardised check list. When results were analysed by metabolite type across all studies, small to moderate effect sizes of 0.29-0.47 (p < 0.05) were observed for changes in Glu and Glx regardless of stimulus domain and brain region, but no significant effects were observed for GABA. Further analysis suggests that Glu, Glx and GABA responses differ by stimulus domain or task and vary depending on the time course of stimulation and data acquisition technique. In summary, we establish effect sizes and directionality of GABA, Glu and Glx response in all currently available fMRS studies. This work provides normative ground to improve consensus and reproducibility of fMRS research and highlights the importance of standardised reporting and minimal best practice.
... Given that increased sensorimotor GABA has been associated with both modulating motor variability and performance (Stagg et al., 2011;Kolasinski et al., 2019), and increasing perceptual acuity (Puts et al., 2011;Kolasinski et al., 2017), we hypothesized that GABA may influence the relationship between EV and subjective assessments of effort. Importantly, a recent study of visual perception found a similar relationship by which GABA influenced the interaction between visual processing and judgments of visual stimuli (Frangou et al., 2019). This result in the visual system further suggests a general role for GABA in mediating the relationship between sensory processing and perception. ...
Article
Experiences of physical exertion guide our assessments of effort. While these assessments critically influence our decisions to engage in daily activities, little is known about how they are generated. We had female and male human participants exert grip force and assess how effortful these exertions felt; and used magnetic resonance spectroscopy (MRS) to measure their brain GABA concentration. We found variability in exertion (i.e., the coefficient of variation in their force exertion profile) was associated with increases in assessments of effort, making participants judge efforts as more costly. GABA levels in the sensorimotor cortex (SM1) moderated the influence of exertion variability on over-assessments of effort. In individuals with higher sensorimotor GABA, exertion variability had a diminished influence on over-assessments of effort. Essentially, sensorimotor GABA had a protective effect on the influence of exertion variability on inflations of effort assessment. Our findings provide a neurobiological account of how the brain's GABAergic system integrates features of physical exertion into judgments of effort, and how basic sensorimotor properties may influence higher-order judgments of effort.SIGNIFICANCE STATEMENTFeelings of effort critically shape our decisions to partake in activities of daily living. It remains unclear how the brain translates physical activity into judgments about effort (i.e., "How effortful did that activity feel?"). Using modeling of behavior and neuroimaging, we show how the nervous system uses information about physical exertion to generate assessments of effort. We found that higher variability in exertion was associated with increases in assessments of effort, making participants judge efforts as more costly. GABA, the brain's main inhibitory neurotransmitter, moderated the influence of exertion variability on over-assessments of effort. These findings illustrate how low-level features of motor performance and sensorimotor neurochemistry influence higher-order cognitive processes related to feelings of effort.
... From the perspective of MRS, GABA levels have been associated with various types of behavior, for example, perceptual acuity (Edden et al., 2009), motor response performance (Kolasinski et al., 2017) and executive function (Marsman et al., 2017). Furthermore, expanding abilities and enhancing skills through short-term learning have been reported to depend on the modulation of GABA levels in task-related brain areas (Bezalel et al., 2019;Frangou et al., 2019;Kolasinski et al., 2019). Across larger time scales (up to lifetime development), age-related changes/differences in brain GABA levels have been implicated in the decline of various behaviors, such as inhibitory control and other higher cognitive functions (Hermans et al., 2018;Marenco et al., 2018;Porges et al., 2017;Simmonite et al., 2019). ...
Article
Full-text available
Understanding the neurophysiological mechanisms that drive human behavior has been a long-standing focus of cognitive neuroscience. One well-known neuro-metabolite involved in the creation of optimal behavioral repertoires is GABA, the main inhibitory neurochemical in the human brain. Converging evidence from both animal and human studies indicates that individual variations in GABAergic function are associated with behavioral performance. In humans, one increasingly used in vivo approach to measuring GABA levels is through Magnetic Resonance Spectroscopy (MRS). However, the implications of MRS measures of GABA for behavior remain poorly understood. In this respect, it is yet to be determined how GABA levels within distinct task-related brain regions of interest account for differences in behavioral performance. This review summarizes findings from cross-sectional studies that determined baseline MRS-assessed GABA levels and examined their associations with performance on various behaviors representing the perceptual, motor and cognitive domains, with a particular focus on healthy participants across the lifespan. Overall, the results indicate that MRS-assessed GABA levels play a pivotal role in various domains of behavior. Even though some converging patterns emerge, it is challenging to draw comprehensive conclusions due to differences in behavioral task paradigms, targeted brain regions of interest, implemented MRS techniques and reference compounds used. Across all studies, the effects of GABA levels on behavioral performance point to generic and partially independent functions that refer to distinctiveness, interference suppression and cognitive flexibility. On one hand, higher baseline GABA levels may support the distinctiveness of neural representations during task performance and better coping with interference and suppression of preferred response tendencies. On the other hand, lower baseline GABA levels may support a reduction of inhibition, leading to higher cognitive flexibility. These effects are task-dependent and appear to be mediated by age. Nonetheless, additional studies using emerging advanced methods are required to further clarify the role of MRS-assessed GABA in behavioral performance.
... Because our study was focused on quantifying both GABA and glutamate, we utilized a sequence capable of detecting both of these neurotransmitters. For example, previous studies have shown that the method we employed can detect GABA in a valid and reliable manner (Barron et al. 2016;Kolasinski et al. 2017;Joers et al. 2018;Frangou et al. 2019;Hong et al. 2019;Ip et al. 2019). A study demonstrated that short-TE MRS can be employed for the reproducible detection of GABA at 3T (Near et al. 2013). ...
Article
Full-text available
The increased prevalence of test anxiety in our competitive society makes it a health issue of public concern. However, its neurobiological basis, especially during the years of formal education, is currently scant. Previous research has highlighted the association between neural excitation/inhibition balance and psychopathology and disease. We examined whether the glutamate/GABA profile tracks test anxiety levels in development, using a cross-sectional and longitudinal design in a cohort spanning from early childhood to early adulthood (N = 289), reassessed approximately 21 months later (N = 194). We used magnetic resonance spectroscopy to noninvasively quantify glutamate and gamma-Aminobutyric acid (GABA) levels in the intraparietal sulcus (IPS) and the middle frontal gyrus. We show that the glutamate/GABA balance within the IPS relates to current individual variation in test anxiety levels and predict future test anxiety approximately 21 months later. Critically, this relationship was observed during early childhood but not during the later developmental stages. Our results extend the use of the excitation/inhibition balance framework to characterize the psychopathology mechanisms of test anxiety, an underexplored yet widespread and debilitating condition that can impact early child development. Our findings provide a better understanding of the neurotransmitter basis underlying the emergence of anxiety disorders during development.
Article
It is generally thought that children learn more efficiently than adults. One way to accomplish this is to have learning rapidly stabilized such that it is not interfered with by subsequent learning. Although γ-aminobutyric acid (GABA) plays an important role in stabilization, it has been reported that GABAergic inhibitory processing is not fully matured yet in children compared with adults. Does this finding indicate that more efficient learning in children is not due to more rapid stabilization? Here, we measured the concentration of GABA in early visual cortical areas in a time-resolved fashion before, during, and after visual perceptual learning (VPL) within subjects using functional MRS (fMRS) and then compared the concentrations between children (8 to 11 years old) and adults (18 to 35 years old). We found that children exhibited a rapid boost of GABA during visual training that persisted after training ended, whereas the concentration of GABA in adults remained unchanged. Moreover, behavioral experiments showed that children exhibited rapid development of resilience to retrograde interference, which indicates that children stabilize VPL much faster than adults. These results together suggest that inhibitory processing in children’s brains is more dynamic and adapts more quickly to stabilize learning than in adults, making learning more efficient in children.
Preprint
Full-text available
Previous research has highlighted the role of glutamate and gamma-aminobutyric acid (GABA) in perceptual, cognitive, and motor tasks. However, the exact involvement of these neurochemical mechanisms in the chain of information processing, and across human development, are unclear. In a cross-sectional longitudinal design, we used a computational approach to dissociate cognitive, decision, and visuomotor processing in 293 individuals spanning early childhood to adulthood. We found that glutamate and GABA within the intraparietal sulcus (IPS) explained unique variance in visuomotor processing, with higher glutamate predicting poorer visuomotor processing in younger participants but better visuomotor processing in mature participants, while GABA showed the opposite pattern. These findings were functionally, neuroanatomically, and neurochemically specific, and were replicated ~1.5 years later and were generalized in two further different behavioral tasks. Using resting fMRI, we revealed that the relationship between IPS neurochemicals and visuomotor processing is mediated by functional connectivity in the visuomotor network. We then extended our findings to high-level cognitive behavior by predicting fluid intelligence performance. We showed that fluid intelligence performance is explained by IPS GABA and glutamate and is mediated by visuomotor processing, and moderated by the developmental stage. These results provide an integrative biological and psychological mechanistic explanation that links cognitive processes and neurotransmitters across human development and establishes their involvement in intelligent behavior.
Article
Functional magnetic resonance spectroscopy (fMRS) can be used to investigate neurometabolic responses to external stimuli in-vivo, but findings are inconsistent. We performed a systematic review and meta-analysis on fMRS studies of the primary neurotransmitters Glutamate (Glu), Glx (Glutamate + Glutamine), and GABA. Data were extracted, grouped by metabolite, stimulus domain, and brain region, and analysed by determining standardized effect sizes. The quality of individual studies was rated. When results were analysed by metabolite type small to moderate effect sizes of 0.29-0.47 (p < 0.05) were observed for changes in Glu and Glx regardless of stimulus domain and brain region, but no significant effects were observed for GABA. Further analysis suggests that Glu, Glx and GABA responses differ by stimulus domain or task and vary depending on the time course of stimulation and data acquisition. Here, we establish effect sizes and directionality of GABA, Glu and Glx response in fMRS. This work highlights the importance of standardised reporting and minimal best practice for fMRS research.
Article
Learning and experience are known to improve our ability to make perceptual decisions. Yet, our understanding of the brain mechanisms that support improved perceptual decisions through training remains limited. Here, we test the neurochemical and functional interactions that support learning for perceptual decisions in the context of an orientation identification task. Using magnetic resonance spectroscopy (MRS), we measure neurotransmitters that are known to be involved in visual processing and learning (i.e. glutamate, GABA) in sensory (early visual cortex: EV) and decision-related (dorsolateral prefrontal cortex: DLPFC) brain regions. Using resting-state functional magnetic resonance imaging (rs-fMRI), we test for functional interactions between these regions that relate to decision processes. We demonstrate that training improves perceptual judgments (i.e. orientation identification) as indicated by faster rates of evidence accumulation after training. These learning-dependent changes in decision processes relate to lower EV glutamate levels and EV-DLPFC connectivity, suggesting that glutamatergic excitation and functional interactions between visual and dorsolateral prefrontal cortex facilitate perceptual decisions. Further, anodal transcranial direct current stimulation (tDCS) in early visual cortex impairs learning, suggesting a direct link between visual cortex excitation and perceptual decisions. Our findings advance our understanding of the role of learning in perceptual decision making, suggesting that glutamatergic excitation for efficient sensory processing and functional interactions between sensory and decision-related regions support improved perceptual decisions.
Article
Full-text available
Key points: The ability to learn new motor skills is supported by plasticity in the structural and functional organisation of the primary motor cortex in the human brain. Changes inhibitory to signalling by GABA are thought to be crucial in inducing motor cortex plasticity. This study used magnetic resonance spectroscopy (MRS) to quantify the concentration of GABA in human motor cortex during a period of motor learning, as well as during a period of movement and a period at rest. We report evidence for a reduction in the MRS-measured concentration of GABA specific to learning. Further, the GABA concentration early in the learning task was strongly correlated with the magnitude of subsequent learning: higher GABA concentrations were associated with poorer learning. The results provide initial insight into the neurochemical correlates of cortical plasticity associated with motor learning, specifically relevant in therapeutic efforts to induce cortical plasticity during recovery from stroke. Abstract: The ability to learn novel motor skills is a central part of our daily lives and can provide a model for rehabilitation after a stroke. However, there are still fundamental gaps in our understanding of the physiological mechanisms that underpin human motor plasticity. The acquisition of new motor skills is dependent on changes in local circuitry within the primary motor cortex (M1). This reorganisation has been hypothesised to be facilitated by a decrease in local inhibition via modulation of the neurotransmitter GABA, but this link has not been conclusively demonstrated in humans. Here, we used 7 T magnetic resonance spectroscopy to investigate the dynamics of GABA concentrations in human M1 during the learning of an explicit, serial reaction time task. We observed a significant reduction in GABA concentration during motor learning that was not seen in an equivalent motor task lacking a learnable sequence, nor during a passive resting task of the same duration. No change in glutamate was observed in any group. Furthermore, M1 GABA measured early in task performance was strongly correlated with the degree of subsequent learning, such that greater inhibition was associated with poorer subsequent learning. This result suggests that higher levels of cortical inhibition may present a barrier that must be surmounted in order to achieve an increase in M1 excitability, and hence encoding of a new motor skill. These results provide strong support for the mechanistic role of GABAergic inhibition in motor plasticity, raising questions regarding the link between population variability in motor learning and GABA metabolism in the brain.
Article
Full-text available
How learning enhances neural representations for behaviorally relevant stimuli via activity changes of cortical cell types remains unclear. We simultaneously imaged responses of pyramidal cells (PYR) along with parvalbumin (PV), somatostatin (SOM), and vasoactive intestinal peptide (VIP) inhibitory interneurons in primary visual cortex while mice learned to discriminate visual patterns. Learning increased selectivity for task-relevant stimuli of PYR, PV and SOM subsets but not VIP cells. Strikingly, PV neurons became as selective as PYR cells, and their functional interactions reorganized, leading to the emergence of stimulus-selective PYR-PV ensembles. Conversely, SOM activity became strongly decorrelated from the network, and PYR-SOM coupling before learning predicted selectivity increases in individual PYR cells. Thus, learning differentially shapes the activity and interactions of multiple cell classes: while SOM inhibition may gate selectivity changes, PV interneurons become recruited into stimulus-specific ensembles and provide more selective inhibition as the network becomes better at discriminating behaviorally relevant stimuli.
Article
Full-text available
Throughout the history of modern neuroscience, the parietal cortex has been associated with a wide array of sensory, motor, and cognitive functions. The use of non-human primates as a model organism has been instrumental in our current understanding of how areas in the posterior parietal cortex (PPC) modulate our perception and influence our behavior. In this Perspective, we highlight a series of influential studies over the last five decades examining the role of the PPC in visual perception and motor planning. We also integrate long-standing views of PPC functions with more recent evidence to propose a more general model framework to explain integrative sensory, motor, and cognitive functions of the PPC.
Article
Full-text available
Dorsal anterior cingulate cortex (dACC) mediates updating and maintenance of cognitive models of the world used to drive adaptive reward-guided behavior. We investigated the neurochemical underpinnings of this process. We used magnetic resonance spectroscopy in humans, to measure levels of glutamate and GABA in dACC. We examined their relationship to neural signals in dACC, measured with fMRI, and cognitive task performance. Both inhibitory and excitatory neurotransmitters in dACC were predictive of the strength of neural signals in dACC and behavioral adaptation. Glutamate levels were correlated, first, with stronger neural activity representing information to be learnt about the tasks' costs and benefits and, second, greater use of this information in the guidance of behavior. GABA levels were negatively correlated with the same neural signals and the same indices of behavioral influence. Our results suggest that glutamate and GABA in dACC affect the encoding and use of past experiences to guide behavior.
Article
Full-text available
Anatomically plausible networks of functionally inter-connected regions have been reliably demonstrated at rest, although the neurochemical basis of these 'resting state networks' is not well understood. In this study, we combined magnetic resonance spectroscopy (MRS) and resting state fMRI and demonstrated an inverse relationship between levels of the inhibitory neurotransmitter GABA within the primary motor cortex (M1) and the strength of functional connectivity across the resting motor network. This relationship was both neurochemically and anatomically specific. We then went on to show that anodal transcranial direct current stimulation (tDCS), an intervention previously shown to decrease GABA levels within M1, increased resting motor network connectivity. We therefore suggest that network-level functional connectivity within the motor system is related to the degree of inhibition in M1, a major node within the motor network, a finding in line with converging evidence from both simulation and empirical studies. DOI: http://dx.doi.org/10.7554/eLife.01465.001.
Article
Full-text available
GABA is the primary inhibitory neurotransmitter in human brain. The level of GABA varies substantially across individuals, and this variability is associated with interindividual differences in visual perception. However, it remains unclear whether the association between GABA level and visual perception reflects a general influence of visual inhibition or whether the GABA levels of different cortical regions selectively influence perception of different visual features. To address this, we studied how the GABA levels of parietal and occipital cortices related to interindividual differences in size, orientation, and brightness perception. We used visual contextual illusion as a perceptual assay since the illusion dissociates perceptual content from stimulus content and the magnitude of the illusion reflects the effect of visual inhibition. Across individuals, we observed selective correlations between the level of GABA and the magnitude of contextual illusion. Specifically, parietal GABA level correlated with size illusion magnitude but not with orientation or brightness illusion magnitude; in contrast, occipital GABA level correlated with orientation illusion magnitude but not with size or brightness illusion magnitude. Our findings reveal a regionand feature-dependent influence of GABA level on human visual perception. Parietal and occipital cortices contain, respectively, topographic maps of size and orientation preference in which neural responses to stimulus sizes and stimulus orientations are modulated by intraregional lateral connections. We propose that these lateral connections may underlie the selective influence of GABA on visual perception.
Article
Full-text available
In order to respond reliably to specific features of their environment, sensory neurons need to integrate multiple incoming noisy signals. Crucially, they also need to compete for the interpretation of those signals with other neurons representing similar features. The form that this competition should take depends critically on the noise corrupting these signals. In this study we show that for the type of noise commonly observed in sensory systems, whose variance scales with the mean signal, sensory neurons should selectively divide their input signals by their predictions, suppressing ambiguous cues while amplifying others. Any change in the stimulus context alters which inputs are suppressed, leading to a deep dynamic reshaping of neural receptive fields going far beyond simple surround suppression. Paradoxically, these highly variable receptive fields go alongside and are in fact required for an invariant representation of external sensory features. In addition to offering a normative account of context-dependent changes in sensory responses, perceptual inference in the presence of signal-dependent noise accounts for ubiquitous features of sensory neurons such as divisive normalization, gain control and contrast dependent temporal dynamics.
Article
Full-text available
Understanding both the organization of the human cortex and its relation to the performance of distinct functions is fundamental in neuroscience. The primary sensory cortices display topographic organization, whereby receptive fields follow a characteristic pattern, from tonotopy to retinotopy to somatotopy [1]. GABAergic signaling is vital to the maintenance of cortical receptive fields [2]; however, it is unclear how this fine-grain inhibition relates to measurable patterns of perception [3, 4]. Based on perceptual changes following perturbation of the GABAergic system, it is conceivable that the resting level of cortical GABAergic tone directly relates to the spatial specificity of activation in response to a given input [5–7]. The specificity of cortical activation can be considered in terms of cortical tuning: greater cortical tuning yields more localized recruitment of cortical territory in response to a given input. We applied a combination of fMRI, MR spectroscopy, and psychophysics to substantiate the link between the cortical neurochemical milieu, the tuning of cortical activity, and variability in perceptual acuity, using human somatosensory cortex as a model. We provide data that explain human perceptual acuity in terms of both the underlying cellular and metabolic processes. Specifically, higher concentrations of sensorimotor GABA are associated with more selective cortical tuning, which in turn is associated with enhanced perception. These results show anatomical and neurochemical specificity and are replicated in an independent cohort. The mechanistic link from neurochemistry to perception provides a vital step in understanding population variability in sensory behavior, informing metabolic therapeutic interventions to restore perceptual abilities clinically.
Article
We have studied vertical synaptic pathways in two cytoarchitectonically distinct areas of rat neocortex--the granular primary somatosensory (SI) area and the agranular primary motor (MI) area--and tested their propensity to generate long-term potentiation (LTP), long-term depression (LTD), and related forms of synaptic plasticity. Extracellular and intracellular responses were recorded in layer II/III of slices in vitro while stimulating in middle cortical layers (in or around layer IV). Under control conditions, 5 Hz theta-burst stimulation produced LTP in the granular area, but not in the agranular area. Agranular cortex did generate short-term potentiation that decayed within 20 min. Varying the inter-burst frequency from 2 Hz to 10 Hz reliably yielded LTP of 21-34% above control levels in granular cortex, but no lasting changes were induced in agranular cortex. However, the agranular cortex was capable of generating LTP if a GABAA receptor antagonist was applied locally at the recording site during the induction phase. In contrast to LTP, an identical form of homosynaptic LTD could be induced in both granular and agranular areas by applying low frequency stimulation (1 Hz for 15 min) to the middle layers. Under control conditions, both LTP and LTD were synapse-specific; theta-burst or low-frequency stimulation in the vertical pathway did not induce changes in responses to stimulation of a layer II/III horizontal pathway. Application of the NMDA receptor antagonist D-2-amino-5-phosphonovaleric acid (AP5) blocked the induction of both LTP and LTD in granular and agranular cortex. In the presence of AP5, low-frequency conditioning stimuli yielded a short-term depression in both areas that decayed within 10-15 min. Nifedipine, which blocks L-type, voltage-sensitive calcium channels, slightly depressed the magnitudes of LTP and LTD but did not abolish them. Synaptic responses evoked during theta-burst stimulation were strikingly different in granular and agranular areas. Responses in granular cortex were progressively facilitated during each sequence of 10 theta-bursts, and from sequence-to-sequence; in contrast, responses in agranular cortex were stable during an entire theta-burst tetanus. The results suggest that vertical pathways in primary somatosensory cortex and primary motor cortex express several forms of synaptic plasticity. They were equally capable of generating LTD, but the pathways in somatosensory cortex much more reliably generated LTP, unless inhibition was reduced. LTP may be more easily produced in sensory cortex because of the pronounced synaptic facilitation that occurs there during repetitive stimulation of the induction phase.(ABSTRACT TRUNCATED AT 400 WORDS)
Article
Stroke is a leading cause of long-term disability, with around three-quarters of stroke survivors experiencing motor problems. Intensive physiotherapy is currently the most effective treatment for post-stroke motor deficits, but much recent research has been targeted at increasing the effects of the intervention by pairing it with a wide variety of adjunct therapies, all of which aim to increase cortical plasticity, and thereby hope to maximize functional outcome. Here, we review the literature describing neurochemical changes underlying plasticity induction following stroke. We discuss methods of assessing neurochemicals in humans, and how these measurements change post-stroke. Motor learning in healthy individuals has been suggested as a model for stroke plasticity, and we discuss the support for this model, and what evidence it provides for neurochemical changes. One converging hypothesis from animal, healthy and stroke studies is the importance of the regulation of the inhibitory neurotransmitter GABA for the induction of cortical plasticity. We discuss the evidence supporting this hypothesis, before finally summarizing the literature surrounding the use of adjunct therapies such as non-invasive brain stimulation and SSRIs in post-stroke motor recovery, both of which have been show to influence the GABAergic system.