of 23
NeuroImage
272
(2023)
120002
Contents
lists
available
at
ScienceDirect
NeuroImage
journal
homepage:
www.elsevier.com/locate/neuroimage
Registered
Report
Characterizing
habit
learning
in
the
human
brain
at
the
individual
and
group
levels:
A
multi-modal
MRI
study
Rani
Gera
a
,
b
,
c
,
Maya
Bar
Or
a
,
c
,
Ido
Tavor
a
,
d
,
Dana
Roll
c
,
Jeffrey
Cockburn
e
,
Segev
Barak
a
,
b
,
Elizabeth
Tricomi
f
,
John
P.
O’Doherty
e
,
g
,
Tom
Schonberg
a
,
c
,
a
Sagol
School
of
Neuroscience,
Tel
Aviv
University,
Tel
Aviv,
Israel
b
School
of
Psychological
Sciences,
Tel
Aviv
University,
Tel
Aviv,
Israel
c
School
of
Neurobiology,
Biochemistry
&
Biophysics,
Faculty
of
Life
Sciences,
Tel
Aviv
University,
Tel
Aviv,
Israe
d
Sackler
School
of
Medicine,
Tel
Aviv
University,
Tel
Aviv,
Israel
e
Division
of
Humanities
and
Social
Sciences,
California
Institute
of
Technology,
Pasadena,
CA,
United
States
of
America
f
Department
of
Psychology,
Rutgers
University,
Newark,
NJ,
United
States
of
America
g
Computation
and
Neural
Systems
Program,
California
Institute
of
Technology,
Pasadena,
CA,
United
States
of
America
a
r
t
i
c
l
e
i
n
f
o
Keywords:
Habit
Goal-directed
behavior
Functional
plasticity
Microstructural
plasticity
Striatum
Visual
cortex
a
b
s
t
r
a
c
t
The
dual-process
theory
of
action
control
postulates
that
there
are
two
competitive
and
complementary
mecha-
nisms
that
control
our
behavior:
a
goal-directed
system
that
executes
deliberate
actions,
explicitly
aimed
toward
a
particular
outcome,
and
a
habitual
system
that
autonomously
execute
well-learned
actions,
typically
following
an
encounter
with
a
previously
associated
cue.
In
line
with
the
dual-process
theory,
animal
studies
have
provided
convincing
evidence
for
dissociable
neural
mechanisms,
mainly
manifested
in
cortico-striatal
regions,
involved
in
goal-directed
and
habitual
action
control.
While
substantial
progress
has
been
made
in
characterizing
the
neural
mechanism
underlying
habit
learning
in
animals,
we
still
lack
knowledge
on
how
habits
are
formed
and
main-
tained
in
the
human
brain.
Thus
far
only
one
study,
conducted
more
than
a
decade
ago
by
Tricomi
et
al.
(2009),
has
been
able
to
induce
habitual
behavior
in
humans
via
extensive
training.
This
study
also
implicated
the
pos-
terior
putamen
in
the
process,
using
functional
magnetic
resonance
imaging
(fMRI).
However,
recent
attempts
to
replicate
the
behavioral
results
of
this
study
were
not
successful.
This
leaves
the
research
of
human
habits,
and
particularly
the
research
of
their
formation
through
extensive
repetition,
as
well
as
their
neural
basis,
limited
and
far
behind
the
animal
research
in
the
field.
This
motivated
us
to
(1)
attempt
to
replicate
the
behavioral
and
imaging
main
findings
of
Tricomi
et
al.,
(2)
identify
further
functional
and
microstructural
neural
modifications
associated
with
habit
formation
and
manifestation,
and
(3)
investigate
the
relationships
between
functional
and
structural
plasticity
and
individual
differences
in
habit
expression.
To
this
end,
in
this
registered
report
we
used
Tricomi
et
al.’s
free-operant
task
along
with
multi-modal
MRI
methods
in
a
well-powered
sample
(n
=
123).
In
this
task
participants’
sensitivity
to
outcome
devaluation
(an
index
of
goal-directed/habitual
action
control)
is
tested
following
either
short
or
extensive
training.
In
contrast
to
our
hypothesis,
we
were
not
able
to
demonstrate
habit
formation
as
a
function
of
training
duration
nor
were
we
able
to
relate
any
functional
or
microstructural
plasticity
in
the
putamen
with
individual
habit
expression.
We
found
that
a
pattern
of
increased
activations
in
the
left
head
of
caudate
that
reoccurred
across
each
day’s
training
was
associated
with
goal
directed
behavior
and
that
increased
processing
of
devalued
cues
in
low-level
visual
regions
was
indicative
of
goal-directed
behavior.
In
a
follow-up
exploratory
analysis
comparing
habitual
and
goal-directed
subgroups
within
each
experimental
group,
we
found
that
elevated
activations
in
frontoparietal
regions
during
early
stages
of
training,
as
well
as
increased
reactivity
towards
still-valued
cues
in
somatosensory
and
superior
parietal
regions,
were
found
in
in-
dividuals
that
were
more
inclined
to
perform
goal-directed
behavior
(compared
with
more
habitual
individuals).
Taken
together,
regions
commonly
implicated
in
goal-directed
behavior
were
most
predictive
of
individual
habit
expression.
Finally,
we
also
found
that
differential
patterns
of
training-related
microstructural
plasticity,
as
mea-
Corresponding
author.
E-mail
address:
schonberg@post.tau.ac.il
(T.
Schonberg)
.
https://doi.org/10.1016/j.neuroimage.2023.120002
.
Received
2
November
2022;
Received
in
revised
form
28
February
2023;
Accepted
1
March
2023
Available
online
11
March
2023.
1053-8119
R.
Gera,
M.
Bar
Or,
I.
Tavor
et
al.
NeuroImage
272
(2023)
120002
sured
with
diffusion
MRI,
in
midbrain
dopaminergic
regions
were
associated
with
habit
expression.
This
work
provides
new
insights
into
the
neural
dynamics
involved
in
individual
habit
formation/expression
and
encourages
the
development
and
testing
of
new,
more
sensitive,
procedures
for
experimental
habit
induction
in
humans.
1.
Introduction
Action
control,
acquired
through
instrumental
learning,
is
hypothe-
sized
to
be
determined
by
an
interplay
between
two
distinct
systems
(
De
Wit
and
Dickinson,
2009
,
Dickinson,
1985
):
one
responsible
for
goal-directed
behaviors
and
another
that
forms
and
executes
habitual
behaviors.
Goal-directed
behavior
relies
on
a
relatively
careful
consid-
eration
of
the
available
information
and
is
particularly
dependent
on
the
association
between
a
response
and
its
outcome
(R-O)
(
Dickinson,
1985
,
Balleine
and
O’Doherty,
2010
).
Therefore,
it
allows
executing
carefully
planned
actions
and
the
flexibility
to
adjust
them
in
order
to
maxi-
mize
desired
outcomes
when
circumstances
change
(e.g.,
(
Valentin
et
al.,
2007
)).
However,
this
behavior
is
cognitively
taxing.
In
contrast,
habitual
behavior
is
considered
to
be
automatic
and
is
cognitively
un-
demanding
(
Dickinson,
1985
,
Graybiel,
2008
).
It
relies
on
the
associa-
tion
between
a
stimulus
and
a
response
(S-R)
and
is
thus
relatively
fixed
and
insensitive
to
changes
in
outcome
value.
Newly
acquired
instru-
mental
actions
are
at
first
goal-directed
and
are
sensitive
to
changes
in
outcome
value.
However,
with
extensive
training,
namely
through
be-
havioral
repetition,
a
qualitative
shift
emerges
as
responding
becomes
autonomous
and
is
automatically
triggered
by
an
associated
stimulus,
regardless
of
changes
in
outcome
value
or
in
R-O
contingency
(
Adams,
1982
,
Dickinson
et
al.,
1995
)
(for
review
see
(
Balleine
and
O’Doherty,
2010
)).
Notably,
the
vast
majority
of
the
evidence
supporting
this
dual-
system
account
of
action
control
and
its
dynamics
has
been
accumulated
from
animal
research
(
Corbit,
2018
)
(mostly
rats),
whereas
it
is
yet
to
be
well-established
and
characterized
in
humans.
Habitual
action
control
is
typically
adaptive,
liberating
mental
re-
sources
while
automating
usually-beneficial
actions.
However,
an
im-
balance
with
the
goal-directed
system
in
the
“struggle
”for
action
con-
trol
and
more
specifically,
overreliance
on
the
habitual
system,
consti-
tutes
a
key
feature
in
several
psychopathologies.
Such
malfunction
has
been
implicated
in
addiction
(
McKim
et
al.,
2016
,
Hogarth
et
al.,
2012
,
Hogarth
and
Chase,
2011
,
Sjoerds
et
al.,
2013
,
Ersche
et
al.,
2016
),
obsessive-compulsive
disorder
(OCD)
(
Gillan
et
al.,
2011
,
Gillan
et
al.,
2015
,
Snorrason
et
al.,
2016
,
Gillan
et
al.,
2014
),
schizophrenia
(
Morris
et
al.,
2015
),
autism
(
Alvares
et
al.,
2016
),
social
anxiety
(
Alvares
et
al.,
2016
,
Alvares
et
al.,
2014
),
Tourette
syndrome
(
Berman
et
al.,
2016
)
and
obesity
(
Horstmann
et
al.,
2015
).
Thus,
characterizing
the
interplay
be-
tween
these
two
systems
and
specifically
the
shift
from
goal-directed
to
habitual
control,
as
well
as
understanding
the
underlying
neural
mech-
anisms,
are
of
great
importance
for
psychotherapeutical
and
clinical
in-
terventions.
A
substantial
landmark
in
the
habit
learning
field
has
been
the
devel-
opment
of
the
reinforcer
devaluation
paradigm
in
rodents
(
Dickinson,
1985
).
This
paradigm
successfully
dissociates
goal-directed
from
habit-
ual
behavior,
based
on
whether
a
learned
action
had
become
insen-
sitive
to
outcome
devaluation
(usually
through
food
satiation
or
con-
ditioned
taste
aversion).
It
became
a
pivotal
measurement
method
of
habits,
yielding
fruitful
insights
in
animal
research.
A
major
contribu-
tion
of
this
paradigm
has
been
its
capacity
to
demonstrate
the
transition
from
goal-directed
behavior
(sensitive
to
outcome
devaluation)
to
habit-
ual
behavior
(insensitive
to
outcome
devaluation),
as
a
result
of
exten-
sive
instrumental
training
(
Adams,
1982
).
Different
human
paradigms,
aimed
to
utilize
the
devaluation
sensitivity
criterion,
have
been
devel-
oped
since
(e.g.
(
Valentin
et
al.,
2007
,
Hogarth
et
al.,
2012
,
Gillan
et
al.,
2014
,
Alvares
et
al.,
2016
,
Tricomi
et
al.,
2009
,
Schwabe
and
Wolf,
2009
,
Liljeholm
et
al.,
2015
,
Reber
et
al.,
2017
)).
The
majority
of
these
tasks
aimed
to
distinguish
S-R
from
R-O
action
control
to
point
at
inter-
individual
differences
or
pathologic
group
tendencies
(e.g.
(
Hogarth
et
al.,
2012
,
Gillan
et
al.,
2014
,
Alvares
et
al.,
2016
)),
identify
relevant
modulators
(e.g.
(
Schwabe
and
Wolf,
2009
)),
or
discern
their
neural
correlates
(e.g.
(
Valentin
et
al.,
2007
,
Liljeholm
et
al.,
2015
,
Reber
et
al.,
2017
)).
However,
apart
from
the
one
study
conducted
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
),
habit
learning,
as
acquired
through
behavioral
repetition,
has
not
been
clearly
demonstrated.
Neuroscientific
research
has
identified
distinct
brain
regions
corre-
sponding
with
the
formation
and
execution
of
goal-directed
behavior
and
habits.
In
rodents,
the
dorsomedial
striatum
(DMS),
the
prelimbic
cortex
(PL)
and
the
nucleus
accumbens
are
the
main
regions
implicated
in
goal-directed
control
(
Yin
et
al.,
2005
,
Yin
et
al.,
2005
,
Corbit
and
Balleine,
2003
,
Corbit
et
al.,
2012
,
Corbit
et
al.,
2001
).
The
execution
of
habitual
responding
is
particularly
dependent
on
the
dorsolateral
stria-
tum
(DLS)
(
Yin
et
al.,
2004
,
Yin
et
al.,
2006
),
gradually
gaining
response-
control
across
the
course
of
learning
(
Balleine
and
Ostlund,
2007
).
In
humans,
homologous
to
the
rodent
DMS,
the
anterior
caudate
nucleus,
and
homologous
to
the
PL,
the
ventromedial
prefrontal
cortex
(vmPFC)
(
Balleine
and
O’Doherty,
2010
),
were
implicated
in
goal
directed
be-
havior
(
Valentin
et
al.,
2007
,
Tricomi
et
al.,
2009
,
de
Wit
et
al.,
2009
,
de
Wit
et
al.,
2012
,
McNamee
et
al.,
2015
).
In
contrast,
the
research
of
the
neural
mechanisms
underlying
habitual
control
in
humans
is
lim-
ited.
There
is
evidence
that
the
human
putamen
is
homologous
to
the
rodent
DLS
in
habitual
action
control
(
Tricomi
et
al.,
2009
,
de
Wit
et
al.,
2012
,
McNamee
et
al.,
2015
).
McNamee
et
al.
implicated
the
poste-
rior
putamen
in
stimulus-triggered
actions,
suggesting
it
has
a
specific
role
in
habit-associated
S-R
encoding
(
McNamee
et
al.,
2015
).
De
Wit
et
al.
found
that
elevated
gray
matter
density
in
the
posterior
putamen
and
white
matter
tract
strength
between
this
region
and
the
premotor
cortex
are
linked
to
individual
tendency
to
habit-like
performance
(
de
Wit
et
al.,
2012
).
However,
thus
far
only
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
implicated
the
posterior
putamen
in
habit
learning,
using
func-
tional
magnetic
resonance
imaging
(fMRI).
Based
on
accumulated
knowledge,
demonstrating
habit
induction
in
healthy
humans
and
as
follows,
characterizing
its
underlying
neural
mechanisms,
is
still
a
considerable
challenge.
Although
it
was
exten-
sively
demonstrated
in
animals
for
decades,
only
the
above-mentioned
procedure
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
was
able
thus
far
to
demonstrate
the
shift
from
goal-directed
to
habitual
control
through
ex-
tensive
training
in
humans
and
point
at
relevant
neural
mechanisms.
However,
these
prominent
findings
have
yet
to
be
successfully
repli-
cated.
Moreover,
a
recent
study
reported
five
failures
in
experimental
habit
induction
(
de
Wit
et
al.,
2018
),
two
of
which
were
attempts
to
replicate
the
behavioral
findings
of
Tricomi
et
al.
(
Tricomi
et
al.,
2009
).
Thus,
the
reliability
of
the
habit
induction
procedure
utilized
by
Tri-
comi
et
al.
(
Tricomi
et
al.,
2009
)
is
currently
unclear.
Furthermore,
to
date,
there
is
no
other
paradigm
that
has
been
able
to
reliably
induce
habits
through
extensive
training
in
humans.
Consequently,
the
neural
mechanisms
underlying
the
formation
(through
behavioral
repetition)
and
manifestation
of
habits
have
yet
to
be
characterized.
The
discrepancy
in
results
across
human
research,
the
gap
between
animal
and
human
literature
on
habits
and
the
fact
that
habits
are
such
a
fundamental
feature
of
human
behavior,
motivated
us
to
establish
the
experimental
induction
of
habit
learning
and
characterize
its
underlying
neural
mechanisms
in
humans.
In
the
current
work
we
aimed
to
replicate
and
expand
the
findings
of
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
with
a
well-powered
sample
us-
ing
multi-modal
MRI
methods.
We
conducted
a
power
analysis
using
the
fMRIPower
tool
(
Mumford
and
Nichols,
2008
)
of
the
effect
demon-
strated
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
in
the
right
putamen,
2
R.
Gera,
M.
Bar
Or,
I.
Tavor
et
al.
NeuroImage
272
(2023)
120002
as
defined
by
the
automated
anatomical
labeling
(AAL)
atlas
(
Tzourio-
Mazoyer
et
al.,
2002
).
This
analysis
yielded
an
n
=
61
for
the
extensive
training
group
where
the
effect
in
the
posterior
putamen
was
observed.
Therefore,
we
aimed
to
obtain
data
from
122
valid
participants:
61
in
each
of
two
groups,
differ
in
their
training
duration:
short
vs.
extensive.
Apart
from
the
replication
attempt
of
the
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
findings,
this
well-powered
sample
also
allowed
us
to
test
for
fur-
ther
functional
correlates
with
habit
formation
and
manifestation
(see
Materials
and
methods).
Additionally,
we
used
diffusion
tensor
imaging
(DTI)
scans
to
probe
micro-structural
brain
plasticity
related
to
habit
formation.
We
chose
to
use
this
method
as
in
recent
years
it
has
been
repeatedly
shown
that
DTI
indices
can
point
at
learning-induced
neu-
roplasticity
in
gray
matter
regions
(
Sagi
et
al.,
2012
,
Hofstetter
et
al.,
2013
,
Tavor
et
al.,
2013
,
Brodt
et
al.,
2018
).
Finally,
we
constructed
a
parametric
index
based
on
task
performance,
that
measures
the
level
of
individual
habit
expression
and
correlated
this
measure
with
structural
and
functional
measurements
to
identify
neural
determinants
of
habit
formation.
Our
target
sample
size
was
four
times
larger
than
the
one
used
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
and
thus
had
the
poten-
tial
to
reliably
identify
relevant
effects
at
both
the
individual
and
group
levels
that
were
not
possible
in
the
original
study.
1.1.
Hypothesis
We
hypothesized
that
extensive
training
will
render
responding
ha-
bitual.
The
study
conducted
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
was
carried
out
in
a
considerably
different
environment
than
the
unsuccess-
ful
replication
attempts
of
its
behavioral
findings
(
de
Wit
et
al.,
2018
).
The
original
study
was
performed
inside
an
MRI
scanner,
whereas
the
replication
attempts
were
performed
in
common
behavioral
settings.
Critically,
such
discrepancy
with
regard
to
the
induction
of
habits,
which
heavily
rely
on
the
associations
between
responses
and
cues
and
con-
texts,
may
have
led
to
different
behavioral
effects.
Factors
that
may
have
potentially
biased
the
behavioral
effect
as
a
function
of
the
differ-
ent
environment
include:
(1)
participants
inside
the
MRI
scanner
may
experience
more
stress
which
promotes
habit
formation
(
Schwabe
and
Wolf,
2009
);
(2)
on
the
other
hand,
volunteers
for
MRI
experiments
may
be
self-selected
to
have
low
rates
of
stress
and
anxiety,
allowing
the
manifestation
of
more
goal-directed
behavior
following
short
training,
thereby
sharpening
the
differential
effect
of
short
and
extensive
train-
ing
on
action
control;
(3)
The
unusual
and
salient
context
of
the
MRI
environment
may
impose
S-R
associations
more
robustly;
(4)
The
loud
noise
in
the
MRI
scanner
may
exploit
some
cognitive
resources
which
may
reduce
the
reliance
on
the
goal-directed
system
(As
evidenced
by
the
effect
of
cognitive
load
on
characteristics
and
strategies
related
to
goal-directed
action
control
(
Foerde
et
al.,
2006
,
Otto
et
al.,
2013
));
(5)
reduced
arousal
inside
the
MRI
scanner
may
negatively
affect
goal
di-
rected
action
control.
Taken
together,
we
presumed
that
at
least
some
of
these
factors
enhance
the
formation
and/or
manifestation
of
habitual
responding
as
a
function
of
training
duration.
Thus,
we
expected
to
suc-
cessfully
replicate
the
behavioral
findings
of
Tricomi
et
al.
(
Tricomi
et
al.,
2009
).
Nonetheless,
we
considered
the
consequences
of
the
possibility
that
the
behavioral
data
would
not
support
our
hypothesis.
This
should
not
affect
the
individual
level
analyses
we
planned
to
perform;
however,
it
is
crucial
for
the
interpretation
of
the
group
level
analyses
of
the
imag-
ing
data.
Therefore,
we
stated
that
in
case
we
would
not
observe
the
hypothesized
behavioral
effect,
we
would
conduct
an
exploratory
anal-
ysis,
in
which
we
would
define
two
well-distinct
subgroups
for
each
experimental
group.
One
subgroup
would
include
participants
who
had
expressed
habitual
responding
and
the
other
would
include
those
who
had
not.
The
clustering
would
be
based
on
a
habit
index
calculated
from
the
behavioral
data
(see
in
Individual
differences
in
functional
MRI
for
details
on
the
generation
of
this
index).
Then,
we
would
compare
these
subgroups
within
training
conditions
to
identify
functional
and
micro-
structural
differences.
We
hypothesized
that
regions
within
the
corticostriatal
network
will
be
implicated
in
habit
formation
and
expression.
For
an
elaborated
de-
piction
of
the
hypotheses
and
their
corresponding
confirmatory
anal-
yses
of
the
neuroimaging
data
see
Table
1
.
We
expected
that
changes
in
activity
and
micro-structural
plasticity
in
the
putamen
will
be
in-
volved
in
habit
learning
while
similar
indices
in
the
anterior
caudate
and
vmPFC
will
be
implicated
in
goal-directed
action
control.
Neverthe-
less,
the
(goal-directed
action
control
associated)
R-O
contingencies
in
our
task
are
very
easy
to
learn.
Thus,
it
is
highly
likely
that
the
anterior
caudate
and
the
vmPFC
would
not
be
employed
differentially
enough
throughout
the
task
to
yield
an
effect
in
most
of
our
planned
analyses
(see
Table
1
and
Data
analysis).
Therefore,
the
analysis
of
these
regions
is
considered
exploratory
unless
noted
differently
in
Table
1
.
2.
Materials
and
methods
2.1.
Data
Sharing
Registered
report
protocol
pre-registration
is
available
at
the
Open
Science
Framework:
https://osf.io/385dx
.
This
protocol
received
in-principal
acceptance
on
21
June
2019.
Behavioral
data,
analy-
sis
codes
and
task
codes
are
available
through
the
Github
repository:
https://github.com/ranigera/MultiModalMRI
_
Habits
.
The
imaging
data
in
Brain
Imaging
Data
Structure
(BIDS)
format
is
available
at
Open-
Neuro
(
https://openneuro.org/datasets/ds004299/versions/1.0.0
)
and
unthresholded
statistical
maps
are
available
at
NeuroVault
(
https://neurovault.org/collections/13090
).
2.2.
Participants
We
aimed
to
collect
a
sample
size
of
122
valid
participants,
ran-
domly
assigned
to
two
groups
receiving
short
(1-day)
or
extensive
(3-
day)
training,
each
with
61
participants.
This
number
was
based
on
a
power
analysis
calculated
using
the
fMRIPower
tool
(
Mumford
and
Nichols,
2008
)
of
the
effect
demonstrated
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
in
the
right
putamen
(
Fig.
1
),
as
defined
by
the
automated
anatomical
labeling
(AAL)
atlas
(
Tzourio-Mazoyer
et
al.,
2002
),
yielding
n
=
61
for
the
extensive
training
group
(on
which
the
effect
was
originally
demonstrated).
To
minimize
a
possible
within-participant
variability
derived
from
diurnal
variation,
participants
from
the
3-day
group
were
scheduled
to
participate
at
as
similar
time
as
possible
on
each
day
of
the
experiment.
We
confined
participants
to
either
morning
or
afternoon
sessions.
In
ad-
dition,
the
experimenter
collecting
the
data
was
the
same
person
across
all
days
of
the
experiment
for
each
participant.
The
study
was
approved
by
the
institutional
review
board
at
the
Sheba
Tel
Hashomer
Medical
Center
and
the
ethics
committee
at
Tel
Aviv
University.
All
procedures
were
performed
in
compliance
with
the
relevant
laws
and
institutional
guidelines.
We
obtained
informed
con-
sent
from
all
participants
prior
to
their
participation
in
the
experiment.
Recruitment:
As
food
rewards
were
used
in
the
experimental
proce-
dure
(see
Experimental
procedure),
participants
were
prescreened
prior
to
their
recruitment
to
ensure
that
they
generally
like
eating
snacks,
do
not
restrict
or
limit
their
food
consumption
to
avoid
high
calorie
foods,
are
not
vegan,
do
not
suffer
from
food
allergies
in
relation
to
snacks
used
in
the
experiment
and
are
willing
not
to
consume
any
food
for
6
hours
before
arriving
to
each
day
of
the
experiment.
Failing
to
comply
with
any
of
these
criteria
prevented
participation
in
the
experi-
ment.
They
also
rated
their
liking
on
a
Likert
pleasantness
scale
(ranging
from
-5,
very
unpleasant,
to
5,
very
pleasant)
toward
three
sweet
and
three
savory
snacks
which
were
later
used
to
choose
from
at
the
be-
ginning
of
the
experiment.
To
ensure
participants’
desire
to
earn
and
eat
snacks,
participants
who
did
not
rate
the
highest
sweet
and
high-
est
savory
snacks
with
at
least
+
2
did
not
participate
in
the
study.
Fi-
nally,
participants
were
asked
to
fill
out
the
eating
attitudes
test
(EAT-26
3
R.
Gera,
M.
Bar
Or,
I.
Tavor
et
al.
NeuroImage
272
(2023)
120002
Table
1
A
mapping
between
hypotheses
and
confirmatory
analyses
of
the
neuroimaging
data.
#
Hypothesis
Confirmatory
analysis
/
examined
contrast
Neuroimaging
modality
subtitle
in
the
text
Within-group
analysis
(3-day
group)
1
The
putamen
increases
its
cue
sensitivity
after
extensive
training
when
habitual
action
control
is
expressed
(replicating
the
finding
of
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
)
.
[task
onset
-
rest
onset]
of
the
last
two
vs.
first
two
sessions
of
training:
ROI
analysis
of
the
average
signal
in
the
putamen.
fMRI
Replicating
the
effect
found
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
in
the
posterior
(right)
putamen
2
The
putamen
increases
its
cue
sensitivity
after
extensive
training
when
habitual
action
control
is
expressed.
[task
onset
-
rest
onset]
of
the
last
two
vs.
first
two
sessions
of
training.
fMRI
Replicating
the
effect
found
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
in
the
posterior
(right)
putamen
3
The
putamen
gradually
increases
its
cue
sensitivity
as
training
progresses.
Linear
trend
analysis
on
the
contrast
of
[task
onset
-
rest
onset]
by
assigning
increasing
linear
trend
weights
to
the
12
training
sessions
according
to
their
chronological
order.
fMRI
Training
duration
induced-effects
4
The
putamen
increases
its
cue
sensitivity
after
extensive
training
within
each
day.
[task
onset
-
rest
onset]
of
the
last
vs.
first
sessions
of
training,
averaged
across
days.
fMRI
Within-day
training
duration
induced-effects
5
The
putamen
gradually
increases
its
cue-sensitivity
as
training
progresses
within
each
day.
Linear
trend
analysis
on
the
contrast
of
[task
onset
-
rest
onset]
by
assigning
increasing
linear
trend
weights
across
the
four
within-day
training
sessions
averaged
across
days.
fMRI
Within-day
training
duration
induced-effects
Between-group
analysis
6
Devaluation
exerts
larger
changes
in
activations
in
the
anterior
caudate
and
vmPFC
following
short
training
compared
to
extensive
training.
Compare
groups
(3-day
group
vs.
1-day
group)
for
[post-devaluation
difference
-
pre-devaluation
difference]
formed
by
first
contrasting
[valued
snack
onset
-
devalued
snack
onset]
for
pre-and
post-devaluation
separately.
fMRI
Effect
of
devaluation
as
a
function
of
training
duration
7
Extensive
training
induces
micro-structural
plasticity
in
the
putamen
compared
to
short
training,
expressed
as
a
reduction
in
MD.
A
voxel-wise
mixed-design
ANOVA
with
factors
of
time
(before
/
after
training)
and
group
(1-day
or
3-day)
with
participant
as
a
random
factor
for
extracted
MD
maps.
DTI
DTI
analysis
Individual
differences
analysis
(Behavioral
and
Neuroimaging)
8
The
putamen
activations,
as
hypothesized
in
points
2-5,
and
anterior
caudate
and
vmPFC
activations,
as
hypothesized
in
point
6,
positively
and
negatively
correlate
with
individual
expression
of
habits,
respectively.
Each
hypothesis
corresponding
to
points
2-6
was
tested
separately.
For
point
6
the
analysis
was
done
after
collapsing
the
data
across
both
groups.
We
used
the
individual
maps
generated
in
points
2-6
and
correlated
them
with
the
individual
behavioral
habit
index.
fMRI
Individual
differences
in
functional
MRI
9
Micro-structural
changes
in
the
putamen
correlate
with
individual
expression
of
habits,
i.e.,
the
larger
the
change
(reduction
in
MD),
the
larger
the
habitual
action
control.
Correlation
between
the
individual
difference
between
extracted
MD
maps
(before
vs.
after)
and
the
individual
behavioral
habit
index.
We
ran
this
analysis
separately
for
each
group.
DTI
Individual
differences
in
micro-structural
MRI
In
point
1
the
ROI
analysis
is
based
on
the
automated
anatomical
labeling
atlas,
which
was
used
for
the
power
analysis
of
the
effect
demonstrated
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
in
the
right
putamen.
For
all
other
analyses
we
used
small
volume
correction
analysis
using
a
mask
based
on
the
Harvard-Oxford
atlas.
All
the
within-group
and
between-group
analyses
require
that
the
behavioral
results
demonstrate
habitual
responding
as
a
function
of
training
duration
(when
comparing
sensitivity
to
outcome
devaluation
between
groups)
in
order
to
relate
their
neuroimaging
results
to
habits.
Abbreviations:
ROI,
region
of
interest;
vmPFC,
ventromedial
prefrontal
cortex;
fMRI,
functional
magnetic
resonance
imaging;
DTI,
diffusion
tensor
imaging;
MD,
mean
diffusivity.
(
Garner
et
al.,
1982
))
to
account
for
eating
disorders.
Participants
with
a
score
of
20
or
above
were
excluded
from
participation
in
the
study.
2.2.1.
Experimental
procedure
Participants
were
scanned
before,
during
and
following
a
free-
operant
task
(
Fig.
2
),
identical
to
the
one
used
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
),
aimed
to
render
responding
habitual
as
a
function
of
exten-
sive
training.
Upon
their
arrival
and
before
scanning,
participants
were
asked
to
taste
three
sweet
and
three
savory
snacks
and
choose
their
fa-
vorite
one
of
each
type.
We
used
snacks
comprised
of
small
pieces:
the
sweet
set
included
M&M,
Skittles
and
a
Maltesers-like
Israeli
snack
of
small
chocolate
balls;
the
savory
set
included
potato
chips,
Doritos
and
4
R.
Gera,
M.
Bar
Or,
I.
Tavor
et
al.
NeuroImage
272
(2023)
120002
Fig.
1.
Results
of
the
power
analysis
of
the
effect
found
by
Tricomi
et
al.
(
Tricomi
et
al.,
2009
)
in
the
right
putamen
(as
defined
by
the
automated
anatomical
labeling
atlas
(
Tzourio-Mazoyer
et
al.,
2002
)),
using
the
fMRIPower
tool
(
Mumford
and
Nichols,
2008
).
The
horizontal
dashed
line
rep-
resents
an
estimated
power
of
80%.
The
vertical
dashed
line
represents
the
number
of
participants
required
to
cross
the
80%
power
criterion
(n
=
61).
X
represents
where
these
lines
meet.
Fig.
2.
Procedure
general
outline.
The
behavioral
tasks
are
presented
along
the
arrow
and
below
the
line
are
the
imaging
scans.
For
the
3-day
group
the
parts
marked
in
blue
were
performed
only
on
the
third
day.
The
other
parts,
including
the
magnetic
resonance
imaging
(MRI)
scans,
were
conducted
on
each
of
the
three
days,
except
for
the
resting-state
scans,
which
were
conducted
only
at
the
beginning
of
the
first
day
and
after
completing
the
free-operant
task
on
the
last
day.
The
1-day
group
performed
all
stages
in
one
day.
Post
experiment
tasks
included
the
administration
of
questionnaires,
working
memory
assessment
and
a
variant
of
the
two-step
sequential
decision-making
task
(
Daw
et
al.,
2011
).
Abbreviations:
DWI,
diffusion
weighted
imaging;
fMRI,
functional
magnetic
resonance
imaging.
cashews.
The
one
sweet
and
one
savory
snack
chosen
by
each
partici-
pant
were
then
used
throughout
the
entire
experiment.
Afterwards,
par-
ticipants
entered
the
MRI
scanner
and
first
underwent
DTI
and
resting-
state
scans.
Then,
they
performed
the
free-operant
task,
consisting
of
three
main
stages:
(1)
training,
(2)
outcome
devaluation
and
(3)
extinc-
tion
test.
The
training
was
comprised
of
8-minute
sessions.
The
amount
of
training
sessions
was
varied
between
two
experimental
groups:
ei-
ther
two
sessions
on
a
single
day
(1-day
group)
or
12
sessions,
four
on
each
of
three
consecutive
days
(3-day
group).
During
all
training
sessions
and
the
extinction
test,
participants
were
scanned
using
fMRI,
whereas
the
devaluation
procedure
was
conducted
outside
the
scanner.
Before
starting
and
after
completing
the
task
phase
on
each
day,
par-
ticipants’
DTI
data
was
obtained.
Anatomical
scans
were
completed
at
the
end
of
each
day.
An
additional
resting-state
scan
was
performed
after
completing
the
task
phase
on
the
last
day.
Following
the
comple-
tion
of
the
free-operant
task
and
all
scanning
procedures,
participants
underwent
a
working
memory
capacity
assessment
and
were
asked
to
fill
out
a
battery
of
questionnaires
aimed
at
estimating
the
relationships
between
individual
factors
and
tendencies
to
manifest
habits
as
well
as
obtain
self-report
indices
of
habits.
Participants
also
performed
a
vari-
ant
of
the
two-step
sequential
decision-
making
task
(
Daw
et
al.,
2011
),
which
has
been
shown
to
dissociate
the
use
of
model-free
and
model-
based
decision-making
strategies.
These
strategies
are
hypothesized
to
reflect
goal-directed
and
habitual
action
control.
The
experiment
was
programmed
and
run
in
Matlab
(The
MathWorks,
Natick,
Massachusetts,
USA)
using
the
Psychophysics
toolbox
(
Brainard,
1997
).
For
the
working
memory
capacity
assessment,
we
used
a
computerized
task
(see
Working
memory
assessment)
programmed
and
run
using
Inquisit
Lab
(Millisec-
ond
Software,
Seattle,
WA,
USA).
2.3.
Free-operant
task
Training:
Each
training
session
was
comprised
of
12
task
blocks
of
20
or
40
seconds
and
eight
rest
blocks
of
20
seconds.
During
the
task
blocks,
participants
were
trained
on
two
Stimulus-Response-Outcome
(S-R-O)
associative
structures
to
induce
instrumental
learning
of
two
5