Validating the Why/How contrast for functional MRI studies of Theory

advertisement
- SUPPLEMENTARY MATERIALS -
Validating the Why/How Contrast for Functional MRI Studies of Theory of Mind
Robert P. Spunt & Ralph Adolphs
California Institute of Technology
S1. Supplemental Methods
1.1. Ruling Out Performance-Related Variability in the Why/How Contrast
Compared to How questions, Why questions reliably elicit lower response accuracy and
longer response times (RT). The analyses reported in the main text were conducted using singlesubject models that include covariates of no interest corresponding to variability in accuracy and RT.
In order to verify that this method of control was achieving its intended effect, and to further confirm
that performance-related variability cannot explain the neural responses typically observed in the
Why/How contrast, we estimated two additional models for each participant. Except for the
differences described below, these models including the same nuisance covariates as the models
described in the main text, and were estimated using the same procedures.
In the first, we took each condition and evenly divided blocks into those eliciting the highest
and the lowest accuracy rates. This yielded four regressors corresponding to High and Low Accuracy
Why questions, and High and Low Accuracy How questions. Given that blocks were divided based
on accuracy, a nuisance covariate corresponding to accuracy rates was not included in these models.
Following estimation, we computed a contrast amongst high accuracy Why questions and low
accuracy How questions.
In the second, we took each condition and evenly divided blocks into those eliciting the fastest
and the slowest RTs. This yielded four regressors corresponding to Fast and Slow RT Why questions,
and Fast and Slow RT How questions. Given that blocks were divided based on RT, a nuisance
covariate corresponding to RT variability was not included in these models. Following estimation, we
computed a contrast amongst Fast Why questions and Slow How questions.
S2. Supplemental Results and Discussion
2.1. Analyses of Performance Variability in the Why/How Contrast
A paired t-test confirmed that participants were more accurate for High Accuracy Why
questions (M = 99.185%, SD = 2.000) than for Low Accuracy How questions M = 92.342%, SD =
5.823), t(49) = 7.817, p = 0.000, 95% CI [5.083, 8.601]. Paralleling this effect, a paired t-test
confirmed that participants had faster RTs during Fast Why blocks (M = 852 ms, SD = 150 ms) than
for Slow How blocks (M = 758 ms, SD = 151 ms), t(49) = 8.011, p = 0.000, 95% CI [0.071, 0.118].
These effects effectively reverse those observed in the main text.
Table S2 shows the results of whole-brain comparisons of High Accuracy Why to Low
Accuracy How blocks, and of Fast Why to Slow åHow blocks. Both comparisons replicate the results
that were observed in both Study 1 and Study 3. This conclusively demonstrates that performance
Page 2
variability does not provide a sufficient alternative explanation of the regional brain activity
modulated in the Why/How contrast. Moreover, they verify the efficacy of the methods that our
primary analyses employed to control for performance-related variability.
To be clear, it remains possible that some variance in the Why/How contrast is explained by
performance-related effects. This analysis simply shows that these performance-related effects do not
provide a sufficient explanation of the effects observed in the Why/How contrast. In future research, it
will be worthwhile to systematically examine the source of these behavorial effects, for instance by
testing targeted hypotheses about brain-behavior relationships, as well as parametrically manipulating
task variables tied to task difficulty.
2.2. Analyses of Functional Lateralization
In an effort to increase the precision of our anatomical definition of the Why/How contrast, we
tested the contrast for hemispheric lateralization. Table S3 lists regions showing significant functional
lateralization in the Why > How contrast based on the pooled data from Study 1 and Study 3 (N = 50).
For each participant, we used their Why/How contrast image to calculate a laterality index at each
voxel in the entire brain. Positive values indicate stronger responses in the left compared to right
hemisphere; a value of zero indicates equivalent responses in the two hemispheres; and negative
values indicate stronger responses in the right compared to left hemispheres. Then, we submitted
these images to a one-sample t-test to determine those regions demonstrating a statistically
meaningful level of functional lateralization. As can be seen from Table S3, the functional anatomy of
the Why > How contrast is highly lateralized to the left hemisphere. In fact, of all the cortical regions
associated with the Why > How contrast in the three studies, only the posterior cingulate cortex failed
to show left hemisphere selectivity. The single region to show evidence of right hemisphere
selectivity was in the posterior lobe of the cerebellum.
2.3. Whole-Brain Multivariate Similarity Analysis
To evaluate differences in the pairwise similaries across the two Why/How contrasts and the
Belief/Photo contrasts for the 10 subjects included in Study 2, we used the Fisher's z-transformed
Pearson correlation of the multivariate response pattern across the whole-brain (grey-matter masked).
Whereas we observed no evidence for a correlation of the Why/How Localizer with the Belief/Photo
Localizer (rmean = 0.17, rsd = 0.18), such a correlation was apparent with Why/How at the first
timepoint (rmean = 0.49, rsd = 0.16), and the difference between these two sets of correlations was
significant, t(9) = 5.100, p = 0.001, 95% CI [0.176, 0.457]. Figure S3 illustrates this difference using a
Page 3
representational dissimilarity matrix (RDM). The dissimilarity metric used is 1 minus the Pearson
correlation (r) and ranges from 0 (identical) to 2 (complete anti-correlation). Figure S3 also shows a
representation of each participants' response pattern in two-dimensions using Multidimensional
Scaling.
Page 4
References
Grinband, J., Savitskaya, J., Wager, T. D., Teichert, T., Ferrera, V. P., & Hirsch, J. (2011). The dorsal
medial frontal cortex is sensitive to time on task, not response conflict or error likelihood.
NeuroImage, 57(2), 303-311. doi:10.1016/j.neuroimage.2010.12.027
Grinband, J., Wager, T. D., Lindquist, M., Ferrera, V. P., & Hirsch, J. (2008). Detection of timevarying
signals
in
event-related
fmri
doi:10.1016/j.neuroimage.2008.07.065
Page 5
designs.
NeuroImage,
43(3),
509-520.
Table S1
The question used to manipulate attention to why vs. how for actions vs. expressions in Study 2. All
questions began with the string “Is the person “.
Why
Hand Actions
How
Facial Expressions
Hand Actions
Facial Expressions
competing against others?
being affectionate?
carrying something?
gazing up?
doing their job?
celebrating something?
lifting something up?
looking at the camera?
expressing themselves?
expressing gratitude?
putting something on?
looking to the side?
helping someone?
expressing self-doubt?
reaching for something?
opening their mouth?
protecting themselves?
in an argument?
using a writing utensil?
showing their teeth?
sharing knowledge?
proud of themselves?
using both hands?
smiling?
Page 6
Table S2
Analyses that test the hypothesis that the network observed in the Why > How contrast is explained
by differences in mean response time (RT) or accuracy in the two conditions. Each contrast was
estimated using a separate model. The first compares high accuracy Why blocks and low accuracy
How blocks, while the second compares the fastest Why blocks to the slowest How blocks. Results
are based on pooled data from Study 1 and Study 3 (N = 50). All peaks survive a whole-brain search
thresholded at a voxel-wise family-wise error rate of .05 and a cluster extent (k) of at least 10 voxels.
PFC = Prefrontal Cortex; OFC = Orbitofrontal Cortex; STS = Superior Temporal Sulcus; x, y, and z =
Montreal Neurological Institute (MNI) coordinates in the left-right, anterior-posterior, and inferiorsuperior dimensions, respectively.
Contrast Name
Region Name
L/R
k
t-value
Why (High Accuracy) > How (Low Accuracy)
Dorsomedial PFC
L 2926 12.714
L
8.404
R
6.288
Ventromedial PFC
L/R
9.877
Lateral OFC
L
256 7.607
Temporoparietal Junction
L
270 7.658
Posterior Cingulate Cortex
L
516 10.644
Posterior STS
L
86
7.079
Anterior STS
L
311 8.794
R
138 7.634
Temporal Pole
R
57
6.934
Why (Fastest) > How (Slowest)
Dorsomedial PFC
Ventromedial PFC
Lateral OFC
Temporoparietal Junction
Posterior Cingulate Cortex
Anterior STS
Temporal Pole
L 1586 12.366
L
6.042
R
21
5.809
L/R 330 7.992
L
56
7.009
L
83
7.307
L
385 10.318
L
226 8.133
R
13
5.869
R
19
6.690
Page 7
MNI Coordinates
x
y
z
-4
-8
16
0
-46
-42
-2
-58
-52
60
46
58
28
42
56
26
-62
-56
-30
-4
-4
14
28
60
42
-14
-6
30
32
-4
-20
-16
-32
-6
-8
8
0
-42
-44
-4
-62
52
48
60
26
50
46
22
-62
-54
-10
-2
8
26
62
42
-18
-14
28
32
-18
-22
-30
Table S3
Regions showing significant functional lateralization in the Why > How contrast based on the pooled
data from Study 1 and Study 3 (N = 50). Clusters were defined using a voxel-wise thresholded of p <
.001, and all reported clusters survive cluster-level correction at a family-wise error rate of .05. PFC =
Prefrontal Cortex; OFC = Orbitofrontal Cortex; STS = Superior Temporal Sulcus; x, y, and z =
Montreal Neurological Institute (MNI) coordinates in the left-right, anterior-posterior, and inferiorsuperior dimensions, respectively.
MNI Coordinates
Region Name
Dorsomedial PFC
L/R
k
t-value
x
y
z
L
1605
10.954
-6
60
38
-
6.550
-10
38
52
-
4.037
-20
62
12
Ventromedial PFC
L
239
5.459
-6
40
-14
Lateral OFC
L
940
6.804
-46
20
-12
Temporoparietal Junction
L
710
6.556
-48
-72
36
Temporal Pole
L
84
5.799
-44
6
-36
Anterior STS
L
383
6.992
-64
-14
-8
Posterior STS
L
94
4.755
-58
-46
2
Dorsolateral PFC
L
224
5.641
-34
22
50
Cerebellum (Posterior Lobe)
R
324
6.448
34
-78
-34
Page 8
Figure S1
Design of the False-Belief Localizer. Each row shows actual stimuli and timing for the Story and
Judgment elements of the Belief and Photo conditions.
Page 9
Figure S2
Dissimilarity analysis of the whole-brain (grey-matter masked) multivariate response patterns
produced by the Why/How and Belief/Photo contrasts. This supplements panels (b) and (c) of Figure
3, which shows the same analysis within a meta-analytically defined mask of the Theory-of-Mind
Network. (a) A representational dissimilarity matrix (RDM) showing the degree of pairwise
dissimilarity among the three contrasts estimated for each of the ten participants: the Why/How
contrast from Study 1 (rows/columns 1-10; Why/HowS1); the Why/How contrast from Study 2
(rows/columns 11-20; Why/HowS2); and the Belief/Photo contrast (rows/columns 21-30). The
dissimilarity metric is 1 minus the Pearson correlation (r), where a value of 0 indicates perfect
correlation; 1 indicates non-correlation; and 2 indicates perfect anti-correlation. Because the order of
participants is constant across the three blocks of contrasts, the diagonals within each block represent
within-subject
pattern
dissimilarities,
while
the
off-diagonals
represent
between-subject
dissimilarities. (b) A two-dimensional representation of the similarity structure based on
multidimensional scaling applied to the RDM. Each colored circle represents a single contrast image,
and constrast images for the same participant are connected by dashed colored lines. The length of
these lines is the Euclidean distance between them, with longer lines representing more dissimilar
multivariate patterns.
Page 10
Download