Supplementary Materials for
Domain-specific representation of social inference by neurons in the human
amygdala and hippocampus
Runnan Cao
et al.
Corresponding author: Runnan Cao, r.cao@wustl.edu; Ueli Rutishauser, ueli.rutishauser@cshs.org
Sci. Adv.
10
, eado6166 (2024)
DOI: 10.1126/sciadv.ado6166
This PDF file includes:
Supplementary Text
Tables S1 to S3
Figs. S1 to S5
Supplementary
Materials
Control
analysis
for
the
selection
of
inference
and
category
neurons.
The
response
time
for
the
why
question
was
significantly
longer
than
that
for
the
how
questions
(
Fig.
1D
).
To
rule
out
the
possibility
that
the
inference
-
type
neurons
were
driven
by
different
difficulty
as
indicated
by
differential
response
time
(RT;
see
Fig.
1D
),
we
used
a
linear
mixed
-
effect
(LME)
model
to
examine
whether
the
selected
inference
-
type
neurons
could
still
maintain
significant
discrimination
between
Why
and
How
questions
(
see
Materials
and
Methods
).
In
the
LME
model,
RT
was
considered
as
a
random
factor.
As
a
result,
83
out
of
105
(79.05%)
of
the
selected
inference
-
type
neurons
maintained
selective
in
the
LME
model,
suggesting
that
RT
was
not
critical
for
their
selection.
Participants
had
a
higher
accuracy
in
answering
“How”
questions
than
“Why”
questions.
To
control
another
potential
confounding
factor
of
difficulty,
we
also
selected
inference
-
type
neurons
using
correct
trials
only
using
the
ANOVA
procedure.
As
a
result,
80
out
of
105
(76.19%)
of
the
selected
inference
-
type
neurons
maintained
selective
when
using
correct
trials
only.
Taken
together,
the
difference
in
difficulty
could
not
explain
the
discrimination
between
Why
and
How
questions.
In
other
words,
the
inference
-
type
neurons
we
identified
were
tuned
to
inference
type
rather
than
difficulty
level
of
the
questions.
Table
S1.
Patients.
Patient
ID
Age
Sex
Epilepsy
Diagnosis
#
Session
#
MTL
Neurons
#
MFC
Neurons
AMY
HIPP
dACC
pre
-
SMA
P42CS
25
F
Not
Localized
1
11
3
23
18
P43CS
42
F
Left
Medial
Temporal
2
1
9
0
0
P44CS
53
F
Right
Medial
Temporal
3
1
7
8
21
P47CS
32
M
Right
Medial
Temporal
4
31
0
0
5
P48CS
32
F
Left
Medial
Temporal
5
30
40
36
31
P49CS
24
F
Left
Medial
Temporal
6
3
0
3
5
7
1
8
0
3
P51CS
17
M
Not
Localized
8
7
8
10
5
9
0
9
10
1
P53CS
60
M
Bilateral
Independent
Temporal
10
11
1
3
16
11
6
4
18
20
P54CS
59
F
Right
Medial
Temporal
12
37
39
4
21
P55CS
43
F
Right
Medial
Temporal
13
8
3
0
4
14
28
6
2
16
P56CS
48
M
Bilateral
Independent
Temporal
15
16
2
2
5
16
16
1
6
7
P57CS
46
M
Left
Other
17
8
1
7
8
P58CS
32
F
Right
Lateral
Frontal
18
9
13
9
5
P65CS
55
F
Not
Localized
19
12
4
0
0
Sum
14
patients
(5
Males)
236
158
141
191
Table
S2.
List
of
questions
in
the
Why/How
task.
Category
Social
inference
(“Why”)
Perceptual
judgment
(“How”)
Face
Is
the
person
admiring
someone?
Is
the
person
smiling?
Is
the
person
proud
of
themselves?
Is
the
person
looking
at
the
camera?
Is
the
person
in
an
argument?
Is
the
person
opening
their
mouth?
Is
the
person
expressing
self
-
doubt?
Is
the
person
looking
to
their
side?
Hand
Is
the
person
concerned
with
their
health?
Is
the
person
lifting
something?
Is
the
person
helping
someone?
Is
the
person
using
both
hands?
Is
the
person
competing
against
others?
Is
the
person
pressing
a
button?
Is
the
person
protecting
themselves?
Is
the
person
reaching
for
something?
Table
S3.
List
of
questions
in
the
updated
version
of
Why/How
task.
Category
Social
inference
(“Why”)
Perceptual
judgment
(“How”)
Face
Is
the
person
being
affectionate?
Is
the
person
gazing
up?
Is
the
person
celebrating
something?
Is
the
person
looking
at
the
camera?
Is
the
person
expressing
gratitude?
Is
the
person
looking
to
their
side?
Is
the
person
expressing
self
-
doubt?
Is
the
person
opening
their
mouth?
Is
the
person
angry
with
someone?
Is
the
person
showing
their
teeth?
Is
the
person
proud
of
themselves?
Is
the
person
smiling?
Hand
Is
the
person
competing
against
others?
Is
the
person
carrying
something?
Is
the
person
doing
their
job?
Is
the
person
lifting
something?
Is
the
person
expressing
themselves?
Is
the
person
putting
something
on?
Is
the
person
helping
someone?
Is
the
person
reaching
for
something?
Is
the
person
protecting
themselves?
Is
the
person
using
a
writing
utensil?
Is
the
person
sharing
knowledge?
Is
the
person
using
both
hands?
Scene
Is
it
a
result
of
a
drought?
Is
the
photo
showing
clouds?
Is
it
a
result
of
a
forest
fire?
Is
the
photo
showing
colorful
flowers?
Is
it
a
result
of
a
hurricane?
Is
the
photo
showing
dry
ground?
Is
it
a
result
of
a
rainstorm?
Is
the
photo
showing
moving
water?
Is
it
a
result
of
Spring
season?
Is
the
photo
showing
palm
trees?
Is
it
going
to
result
in
a
rainstorm?
Is
the
photo
showing
smoke?
Figure
S1.
Additional
results
for
inference
-
type
neurons.
(
A
-
F
)
Average
normalized
firing
rate
for
inference
-
type
neurons.
(
A
-
B
)
Responses
were
aligned
to
the
stimul
us
onset
of
each
trial.
(
C
-
F
)
Responses
were
aligned
to
the
stimul
us
onset
of
each
block.
(
A,
C,
E
)
High
-
level
inference
preferring
neurons
(i.e.,
neurons
had
greater
activity
to
“Why”
conditions).
(
B,
D,
F
)
Low
-
level
inference
preferring
neurons
(i.e.,
neurons
with
greater
activity
to
“How”
conditions
).
Legend
conventions
as
in
Figure
2
.
(
G
)
Distribution
of
inference
neurons
(
separat
ed
into
Why
-
preferring
neurons
and
How
-
preferring
neurons)
among
the
four
recording
sites.
Overall,
we
found
a
trend
of
higher
proportion
of
Why
-
preferring
neurons
than
How
-
preferring
neurons
across
different
brain
regions
(AMY:
7.11%
vs.
5.33%
for
Why
-
preferring
and
How
-
preferring
respectively
,
χ
2
-
test
of
proportion:
P
=
0.29
;
HIPP:
7.59
%
vs.
6.90%
,
P
=
0.
76
;
dACC:
6.02%
vs.
6.02%
,
P
=
1.0
;
and
pre
-
SMA:
13.89%
vs.
8.33%
,
P
=
0.025,
not
significant
after
multiple
-
comparison
correction).
(
H
)
Proportion
of
inference
neurons
in
each
area
separating
for
left
and
right
hemisphere.
A
b
inomial
test
was
performed
to
determine
the
significance
of
inference
neurons
selection
.
*:
P
<
0.05;
**:
P
<
0.01;
***:
P
<
0.001;
****:
P
<
0.0001.
(
I
-
J
)
Single
-
trial
analysis
using
the
response
selectivity
index
for
MTL
(
I
)
and
MFC
(
J
)
respectively.
(
K
)
Population
decoding
of
inference
type
in
each
subregion
of
the
MTL
and
MFC.
(
L
)
Temporal
decoding
of
inference
type
on
the
population
of
MTL
and
MFC
respectively.
Legend
conventions
as
in
Figure
2
.
Figure
S2.
Domain
-
specific
inference
type
coding
in
each
subregion.
(
A
-
B
)
Single
-
trial
response
selectivity
index
of
inference
type
(
“Why”
vs.
“How”
),
calculated
separately
for
face
(dark
colors)
and
hand
(light
colors)
trials.
(
A
)
Inference
-
type
neurons
in
the
MTL
(left,
n
=
45)
and
MFC
(right,
n
=
45)
selected
using
face
stimuli.
(
B
)
Inference
-
type
neurons
in
the
MTL
(left,
n
=
31)
and
MFC
(right,
n
=
48)
selected
using
hand
stimuli.
Legend
conventions
as
in
Fig.
3A
.
(
C
)
Proportion
of
inference
-
type
neuron
selected
with
face
or
hand
stimuli
only.
(
D
)
Decoding
performance
of
inference
type
within
-
(filled
bars)
and
cross
-
category
(opened
bars).
Legend
conventions
as
in
Fig.
3H
.
(
E
)
Quantification
of
the
generalizability
of
inference
decoding
across
categories.
Figures
S3.
Representation
of
scene
inferences.
(
A
-
B
)
Example
inference
-
type
neurons
selected
in
the
MTL
(
A
)
and
MFC
(
B
)
with
non
-
social
scene
stimuli.
(
C
)
Proportion
of
inference
-
type
neurons
selected
with
scene
stimuli
in
the
MTL
and
MFC
respectively
.
(
D
)
Proportion
of
inference
-
type
neurons
selected
with
scene
stimuli
in
each
subregion
,
respectively
.
(
E
)
Single
-
trial
analysis
using
the
response
selectivity
index
(RSI;
see
Materials
and
Methods
).
Shown
is
the
cumulative
distribution
of
RSI.
(
F
-
G
)
RSI
analysis
of
scene
inference
(
“Why”
vs.
“How”
)
for
face
-
selected
inference
-
type
neurons
(
F
)
and
hand
-
selected
inference
-
type
neurons
(
G
)
suggested
the
encoding
of
inference
-
typle
does
not
generalize
between
social
(face
and
hand)
vs.
non
-
social
stimuli
(scene).
Figure
S4.
Interaction
of
inference
representation
and
category
representation
in
the
MTL
and
MFC.
Upper
row:
MTL;
Lower
row:
MFC.
(
A,
D
)
Distribution
of
inference
neurons
and
category
neurons.
Legend
conventions
as
in
Fig.
5I
.
(
B,
E
)
Scatter
plot
of
the
importance
index
(
see
Materials
and
Methods
)
assigned
by
a
decoder
to
each
cell
for
inference
-
decoding
(x
-
axis)
and
category
-
decoding
(y
-
axis).
(
C,
F
)
Distribution
of
the
weight
vector
angle
(
see
Materials
and
Methods
)
for
the
top
25%
cells
in
MTL
and
MFC
in
either
decoder.
Figure
S5.
Independent
representation
of
inference
type
and
category
in
subregions
of
the
MTL
and
MFC.
(
A
)
Proportion
of
category
-
selective
neurons.
(
B
)
Distribution
of
CS
neurons
selected
with
“Why”
and
“How”
trials
.
(
C
)
Population
decoding
of
category
using
a
within
-
inference
decoder
(i.e.,
train
and
test
with
“Why”
or
“How”
trials
only)
and
a
cross
-
inference
decoder
(i.e.,
train
with
one
inference
type
and
test
on
the
other).
Category
decoding
was
significantly
above
chance
when
decoded
across
different
inference
types
in
the
AMY
(train
with
“Why”
:
78.40
±
5.12
%
[mean
±
SD],
P
<0.0001;
train
with
“
How
”
:
81.86
±
5.
00
%
[mean
±
SD],
P
<0.0001)
and
pre
-
SMA
(train
with
“Why”
:
74.83
±
5.3
9%
[mean
±
SD],
P
<0.0001;
train
with
“
How
”
:
79.42
±
4.89
%
[mean
±
SD],
P
<0.0001),
but
was
only
significant
when
trained
on
“Why”
trials
in
the
HIPP
(
“Why”
:
60.26
±
4.86%
[mean
±
SD],
P
=
0.01;
train
with
“
How
”
:
50.22
±
5.10%
[mean
±
SD],
P
=
0.40)
and
dACC
(train
with
“Why”
:
64.70
±
5.44
%
[mean
±
SD],
P
=
0.003
;
train
with
“
How
”
:
56.60
±
5.
01
%
[mean
±
SD],
P
=
0.07
)
.
(
D
)
Distribution
of
inference
-
t
ype
neurons
and
category
neurons
among
each
subregion.
(
E)
Scatter
plot
of
the
importance
index
assigned
by
a
decoder
to
each
cell
for
inference
-
decoding
(x
-
axis)
and
category
-
decoding
(y
-
axis).
Legend
conventions
as
in
Fig.
5.