BRAIN CONNECTIVITY

Volume 4, Number 1, 2014

Mary Ann Liebert, Inc.
DOI: 10.1089/brain.2013.0172

Brain Networks Shaping Religious Belief

Dimitrios Kapogiannis,1 Gopikrishna Deshpande,2 Frank Krueger,3
Matthew P. Thornburg,4 and Jordan Henry Grafman5

Abstract

We previously demonstrated with functional magnetic resonance imaging (fMRI) that religious belief depends upon
three cognitive dimensions, which can be mapped to specific brain regions. In the present study, we considered these
co-activated regions as nodes of three networks each one corresponding to a particular dimension, corresponding to
each dimension and examined the causal flow within and between these networks to address two important hypotheses that remained untested in our previous work. First, we hypothesized that regions involved in theory of mind
(ToM) are located upstream the causal flow and drive non-ToM regions, in line with theories attributing religion
to the evolution of ToM. Second, we hypothesized that differences in directional connectivity are associated
with differences in religiosity. To test these hypotheses, we performed a multivariate Granger causality-based directional connectivity analysis of fMRI data to demonstrate the causal flow within religious belief-related networks.
Our results supported both hypotheses. Religious subjects preferentially activated a pathway from inferolateral to
dorsomedial frontal cortex to monitor the intent and involvement of supernatural agents (SAs; intent-related
ToM). Perception of SAs engaged pathways involved in fear regulation and affective ToM. Religious beliefs are
founded both on propositional statements for doctrine, but also on episodic memory and imagery. Beliefs based
on doctrine engaged a pathway from Brocas to Wernickes language areas. Beliefs related to everyday life experiences engaged pathways involved in imagery. Beliefs implying less involved SAs and evoking imagery activated a
pathway from right lateral temporal to occipital regions. This pathway was more active in non-religious compared to
religious subjects, suggesting greater difficulty and procedural demands for imagining and processing the intent of
SAs. Insights gained by Granger connectivity analysis inform us about the causal binding of individual regions activated during religious belief processing.
Key words: belief

formation; emotion; imagery; semantic memory; social cognition; theory of mind

of perceived SAs (D2); and D3 refers to the mixed foundation of religious beliefs on abstract semantic processing
(i.e., processing of propositional statements for knowledge
of religious doctrines) and episodic memory and imagery.
Using a functional magnetic resonance imaging (fMRI) paradigm, in which subjects had to indicate whether they agreed
or not to a range of religious beliefs, we demonstrated sets of
brain regions that became active in association with these dimensions (Kapogiannis et al., 2009b).
The main regions activated by D1 were bilateral inferior
frontal gyrus (IFG, BA 45), right (R) middle temporal
gyrus (MTG, BA 21), R inferior temporal gyrus (ITG, BA
20), R precuneus (BA 7), and R superior medial frontal
gyrus (SMFG), BA 8 (part of dorsomedial PFC) and 10
(frontopolar PFC). These areas play key roles in action

Introduction

eligious behavior is a uniquely human trait, the cornerstone of which is religious belief (Boyer and Bergstrom, 2008). Religious beliefs refer to supernatural agents
(SAs, exemplified by God) and to cosmological concepts
and domains (such as Heaven and Hell). Belief representations involve multiple elemental cognitive and affective
processes recruited in parallel. In a previous study on religious belief, we identified the three most readily demonstrable (and, arguably, most important) among these processes,
which constitute Dimensions of religious belief (Baylor
Institute for Studies of Religion, 2006; Kapogiannis et al.,
2009b): Dimension 1 (D1) monitors the level of involvement
and intent of perceived SAs; D2 monitors the love and anger
1

Laboratory of Clinical Investigation, National Institute on Aging (NIA/NIH), Baltimore, Maryland.

Department of Electrical and Computer Engineering, MRI Research Center, Auburn University, Auburn, Alabama.
3
National Institute of Neurological Disorders and Stroke, Bethesda, Maryland.
4
George Mason University, Arlington, Virginia.
5
Cognitive Neuroscience Laboratory, Rehabilitation Institute of Chicago, Chicago, Illinois.
2

70

BRAIN NETWORKS SHAPING RELIGIOUS BELIEF

understanding and intent-related theory of mind (ToM) (German et al., 2004; Han et al., 2008; Molnar-Szakacs et al.,
2005). The main regions activated by D2 were R middle
frontal gyrus (MFG, BA 11, part of ventrolateral PFC)
with perception of SAs love, and left (L) MTG (BA 21)
with perception of SAs anger; these areas play roles in affective ToM and emotional regulation. Regarding D3, bilateral
calcarine (CaG) and L fusiform (FG) gyri (BA 17, 18, and
19); L precuneus (BA 7); and L IFG (BA 44, Brocas area)
were activated with beliefs mainly founded on episodic
memory and imagery (Desai et al., 2010; Szpunar et al.,
2007). Beliefs mainly processed as propositional statements
for doctrine activated lateral temporal areas, including the L
superior temporal gyrus (STG, BA 22, Wernickes area).
In the present study, we considered the coactivated regions
as nodes of three networks, each one corresponding to a particular dimension and hypothesized that religious beliefs
emerge from causal flow within and between these networks.
According to modern theories that attribute the development
of religion to the evolution of ToM (Boyer, 2003; Boyer and
Bergstrom, 2008), we hypothesized that regions involved in
intent-related (such as the IFG) and affective ToM are located
upstream the causal flow, i.e., they are drivers of activity in
non-ToM regions. In addition, the previous fMRI analysis
generated the unexpected finding of common brain activation
patterns in religious and non-religious subjects (Kapogiannis
et al., 2009b). To address this paradox, we hypothesized that
differences in the causal flow within and between networks is
associated with differences in religiosity.
To address these hypotheses, the present study utilized a
multivariate Granger causality (GC)-based directional connectivity analysis of our fMRI data. Techniques based on
the principle of GC (Granger, 1969) have been successfully
employed to demonstrate the effective connectivity of brain
networks involved in sensory (Deshpande et al., 2008; Roebroeck et al., 2005; Stilla et al., 2007, 2008), motor (Abler
et al., 2006; Deshpande et al., 2009), and cognitive processing (Hampstead et al., 2010; Krueger et al., 2011; Sridharan
et al., 2008; Strenziok et al., 2011), but they have not been
applied to the study of religious cognition.
Materials and Methods

Detailed information on subjects and fMRI acquisition

methods can be found in our original article (Kapogiannis
et al., 2009b). All subjects provided written informed consent in compliance with the Institutional Review Board of
the National Institute of Neurological Disorders and Stroke
(Bethesda, MD). Briefly, first, we used a data-reduction approach (Multidimensional Scaling) to identify dimensions
underlying religious belief in 13 religious and 13 nonreligious healthy volunteers; this allowed us to create a
three-dimensional cognitive space where a set of statements
describing religious beliefs was represented. Then, fMRI was
performed in a different cohort of 20 religious and 20 nonreligious healthy volunteers (matched for age, sex, and education), using the same set of statements as stimuli; subjects
had to read and indicate whether they agreed or not with
each statement. We employed a General Linear Model analysis, in which the dimension coordinates of the statements (D1,
D2, and D3) were treated as parametric modulators of the hemodynamic response function. Eight areas showed a positive

71

linear association with D1; one area had a positive and another
a negative linear association with D2; and six areas had a negative and five a positive linear association with D3.
To pursue GC analysis, we considered these 21 areas as regions of interest (ROIs) and extracted their representative
time series (first eigenvariate), which were input to a single
dynamic multivariate autoregressive model (dMVAR).
Dynamic correlation-purged Granger causality

GC analysis, in this context, attempts to determine

whether there is a causal relationship between activity in
different nodes of a neural network. Suppose xm, m = 1..k
correspond to the k selected ROI time series and X(t) =
(x1 (t), x2 (t) . . . xk (t))T , then the dMVAR using X(t) was defined such that its coefficients are a function of time
p

X(t) = V(t) + A(n, t)X(t  n) E(t)

(1)

n=0

where V is the intercept vector, E(t) is the vector corresponding to the residuals, and t represents discrete time. The model
order p was determined to be one, using the Bayesian information criterion (Deshpande et al., 2009). Being a multivariate model, the dMVAR is less sensitive to indirect causal
relationships due to two regions being influenced from a
third variable (Kus et al., 2004). In accordance with previous
studies (Sato et al., 2006), the elements of A(n,t), that is,
aij(n,t), can be expanded using a wavelet basis as follows
aij (n, t) = cn 1, 0 /(t)

X 2x  1

+ + cnx, y wx, y (t)

(2)

x=0 y=0

where cnx, y (x =  1, 0, 1 . . . T  1, y = 0, 1, 2 . . . 2x  1, and

n = 1 . . . p) are the wavelet coefficients, u(t) is the scaling function, and wx,y(t) are orthonormal basis functions derived from a
mother wavelet. We chose the Daubechies wavelet as the
mother wavelet owing to its regularity and compact support
(Daubechies, 1988). The choice of the specific Daubechies
wavelet (D2D20) is dictated by the expected order of polynomial behavior in the data, given the fact that the number of vanishing moments of DN is N/2. For example, D4 is most suited
for modeling a constant and linear component in the data
(polynomial with two coefficients) because it has four wavelet
filter coefficients and two vanishing moments. In particular, we
chose the D8 Daubechies wavelet as the mother wavelet in this
study because previous studies have indicated that fMRI activation data may be appropriately modeled by polynomials of
an order of 3 to 5 (Clark, 2002; Gibbons et al., 2004). Both parameter T and maximum resolution parameter X must be a
power of two. An iterative generalized least squares estimation
procedure (Sato et al., 2006) was adopted to solve for the
wavelet coefficients to obtain A(n,t) and V(t). As shown before
(Sato et al., 2006), the number of temporal observations we
have (i.e., 280 volumes per scan) is enough to reliably estimate
the unknown parameters via this procedure. Dynamic correlation-purged Granger causality (CPGC) (Deshpande et al.,
2010a, 2010b; Lacey et al., 2010b) was then obtained as
shown in Equation (3). A custom implementation of the
dMVAR model was performed using MATLAB.
p

CPGCij (t) = + [aij(n, t)]2

n=1

(3)

72

KAPOGIANNIS ET AL.

Stimulus entrained dynamic CPGC analysis

Dynamic CPGC between all 21 ROIs was obtained for

every run and every subject using a first order dMVAR
model using wavelets as described above. Using the GLM
design matrix used in the previous study (Kapogiannis
et al., 2009b), the beta values indicating the strength of covariance between CPGC pathways (as opposed to fMRI
time series in activation analysis) and the experimental paradigm, specifically D1, D2, and D3, were determined for
each individual subject. Linear contrasts were computed to
assess the effect of each dimensional regressor compared
to the baseline, as in (Kapogiannis et al., 2009b). This

yielded pathways significant for contrasts D1 > baseline,

D2 > baseline, and D3 > baseline. Such paths had connectivity that covaried significantly more with the effect of interest
than with the baseline. For such paths, a t-test was carried out
on the betas to find out connections that were significantly
different between the religious and non-religious groups.
Significance was set to p < 0.001 (FDR corrected) in all
cases. A schematic of this procedure is shown in Figure 1.
The motivation for using the stimulus entrained dynamic
connectivity analysis in place of other simpler techniques,
which do not model the dynamics, is as follows. First, our approach formulates connectivity investigation within the
methodological framework of activity detection, which

Judgment Task

Stimuli dimensional co-ordinates

D1 regressor

D2 regressor

D3 regressor

Activation GLM
8 activated ROIs

FIG. 1. A schematic illustrating the

analysis methods adopted in this article.
Specifically, the activation analysis performed (Kapogiannis et al., 2009b) is
summarized first. The time series from the
activated regions of interest (ROIs) are
input to the dynamic connectivity model
following which the covariance of the
obtained dynamic connectivities with dimensional regressions is modeled using a
GLM. Individual subject-level bs
obtained from the GLM are subject to
dimension > baseline contrasts. The significant paths resulting from this are again
tested for significant differences between
religious and non-religious groups.

2 activated ROIs

11 activated ROIs

Dynamic CPGC Analysis

Dynamic CPGCij

D2 regressor

D1 regressor

D3 regressor

Connectivity GLM
Individual subject-level bs for covariance of ji with D1/D2/D3
Linear Contrasts
D1>baseline paths

D2>baseline paths

D3>baseline paths

Significant? If yes
T-test on bs using subject sample

Religious > Non-religious paths

Non-religious > Religious paths

Kapogiannis et al., 2009b

Parametric modulation of HRF

BRAIN NETWORKS SHAPING RELIGIOUS BELIEF

makes it easier to interpret the relationship between activity

and connectivity (Lacey et al., 2010a). Second, intrinsic causality that is not entrained to the external stimulus, though interesting, is not relevant to the specific brain mechanism
being investigated when using a task to evoke brain activity.
Our method allows the characterization of stimulus-evoked
connectivity changes. Third, hemodynamic variability arising due to non-neuronal sources is structural, rather than
functional, in nature and hence does not change with time.
Therefore, as shown previously (Deshpande and Hu, 2012),
the results obtained from this model are not influenced by
the variability of the hemodynamic response across regions
and subjects.

73
Results

The effective connectivity analysis demonstrates the

causal flow between the 21 ROIs. The results are graphically
depicted as causal pathways linking network nodes in Figure
2 (Fig. 2a for D1 network nodes and Fig. 2b for pathways
linking D1 with D2 and D3 nodes), Figure 3 (for pathways
linking D2 with D1 and D3 nodes, given that we did not
find a significant pathway between the two D2 nodes) and
Figure 4 (Fig. 4a for D3 network nodes and Fig. 4b for pathways linking it with D1 and D2 nodes). Our analysis identified pathways for which the directional connectivity
significantly covaried with the corresponding dimension

FIG. 2. (a) Pathways between dimension 1 (D1) network nodes; (b) Pathways
between D1 and other networks. Red-Yellow: pathway
strength positively covaries
with D1. All covariances
were significant at p < 0.001,
FDR corrected. The color bar
represents the p-values
obtained in the D1 > baseline
contrast. The pathways labeled R > NR and NR > R
were stronger in religious
compared to non-religious
subjects and vice versa, respectively.

74

KAPOGIANNIS ET AL.

FIG. 3. Pathways between

D2 and other networks. RedYellow: pathway strength
positively covaries with D2;
Blue-Aquamarine: pathway
strength negatively covaries
with D2. All covariances
were significant at p < 0.001,
FDR corrected. The color bar
represents the p-values
obtained in the D2 > baseline
contrast. The pathways labeled R > NR and NR > R
were stronger in religious
compared to non-religious
subjects and vice versa, respectively.

(Table 1). Moreover, for some causal pathways this covariance was significantly greater for religious or non-religious
subjects. Note that splitting the pathways whose connectivity
covaried with a given dimension into inter-network paths and
intra-network paths is only for convenience of interpretation
and visual display; all the paths were computed using a single model with all 21 ROI time series.

from there to R SMG, BA 40 (IPL), was greater with greater

perception of SAs love ( + D2) for all subjects. In addition,
in non-religious compared to religious subjects, the pathway
from L MTG, BA 21, to R ITG, BA 21, had a greater
negative association with perception of SAs anger (D2)
(Fig. 3).
D3 network, content of religious belief

D1 network, perceived SAs level of involvement

D1 network nodes showed higher activation in association

with perception of SAs lack of involvement ( + D1). The
pathways from R IFG, BA 45, to R precuneus, BA 7, and
R SMFG, BA 10, had a positive association with D1 for all
subjects. In other words, the connectivity from R IFG to R
precuneus and R SMFG were stronger with processing statements describing relatively less involved SAs. Moreover, the
pathway from R ITG, BA 20 to R IFG was stronger in nonreligious compared to religious subjects. Religious subjects
compared to non-religious subjects showed greater connectivity from L IFG, BA 45 to R SMFG, BA 8 (Fig. 2a). Examining pathways between D1, D2, and D3 network nodes
revealed that, in religious compared to non-religious subjects, the pathway from R SMFG, BA 10, to L IFG, BA 44
(a D3 node that activates with religious beliefs based on episodic memory/imagery) had a greater positive association
with perception of SAs lack of involvement. Conversely,
in non-religious compared to religious subjects, the pathway
from R MTG, BA 21, to R CaG, BA 18 (a D3 node that also
activates with religious beliefs based on episodic memory/
imagery) had a greater positive association with perception
of SAs lack of involvement (Fig. 2b).
D2 network, perceived SAs love and anger

The strength of the pathways from R SMFG, BA 8 (dorsomedial PFC), to R MFG, BA 11 (ventrolateral PFC), and

Beliefs founded on doctrine ( + D3) had a negative association with activation of the pathway from L STG, BA 22, to
L FG, BA 19, while beliefs founded on episodic memory/imagery (D3) had a positive association with activation of the
pathway from L precuneus, BA 7, to L IFG, BA 45, and from
there to L STG, BA 22, in all subjects (Fig. 4a). Further,
pathways linking D3 and other networks, and covarying
with D3, were observed (Fig. 4b). The pathway from R
MTG, BA 21, to L FG, BA 18, had a positive association with
+ D3, in all subjects. Religious compared to non-religious
subjects showed a greater positive association of the pathway
from R SMFG, BA 10, to L IFG, BA 44/47, with + D3. Nonreligious compared to religious subjects showed greater
positive association of the pathway from R MTG, BA 21,
to R CaG, BA 17, with D3.
Discussion

Our findings that D1 regions involved in action understanding and intent-related ToM (German et al., 2004; Han
et al., 2008; Molnar-Szakacs et al., 2005) are located upstream the causal flow and drive non-ToM regions support
theories attributing religion to evolution of ToM for SAs
(Boyer, 2003; Boyer and Bergstrom, 2008). Connectivity
within the D1 network originated in the R IFG, a key area
of the mirror-neuron system, consistently activated by intent-related, and affective, ToM (Mason and Just, 2011;
Mier et al., 2010). The right sided predominance of the D1

BRAIN NETWORKS SHAPING RELIGIOUS BELIEF

75

FIG. 4. (a) Pathways between D3 network nodes; (b)

Pathways between D3 and
other networks. Red-Yellow:
pathway strength positively
covaries with D3; BlueAquamarine: pathway
strength negatively covaries
with D3. All covariances
were significant at p < 0.001,
FDR corrected. The color bar
represents the p-values
obtained in the D3 > baseline
contrast. The pathways labeled R > NR and NR > R
were stronger in religious
compared to non-religious
subjects and vice versa, respectively.

network suggests that it is monitoring intent over simple action understanding (Ortigue et al., 2010). Pathways from IFG
modulated the precuneus and dmPFC (BAs 8 and 10), areas
heavily interconnected with each other, which play key roles
in processing of self versus other (Cavanna and Trimble,
2006; Margulies et al., 2009; Nahab et al., 2011). The pathway to dmPFC was more active with perception of less involved SAs, perhaps, because of increased uncertainty
about their intent ( Jenkins and Mitchell, 2010). Strikingly
similar networks monitor the protagonists intent during discourse processing (Mason and Just, 2011) and self-agency
during movement (Nahab et al., 2011).
Interestingly, all subjects shared the R-sided IFG to
dmPFC pathway, but religious subjects also possessed a sim-

ilar L-sided network, therefore, providing bilateral IFG input

to this self-referential area. This double input may support a
more anthropomorphic representation of SAs, since the more
human-like an observed agent is, the higher medial frontal
activation it evokes (Steinbeis and Koelsch, 2009).
Moreover, in non-religious subjects, a pathway from R
ITG to R IFG positively covaried with + D1. White matter
connections between these regions [especially the arcuate
fasciculus, which connects the IFG, pars triangularis, with
the lateral temporal lobe (Kaplan et al., 2010)] have been
strengthened in recent phylogenesis, presumably, in association with the evolution of language (to the L) and other symbolic representation systems (to the R) (Rilling et al., 2008).
Preferential recruitment of this pathway by non-religious

76

+ D3 (founded on
doctrine/semantics)
D3 (founded on episodic
memory/imagery)

D2 (more angry SA)

D1 (more involved SA)

+ D2 (more loving SA)

+ D1 (less involved SA)

L STG/L
FG =  0.17
L PREC/L
IFG = 0.29
L IFG/L
STG = 0.08

R IFG/R
PREC = 0.38
R IFG/R
SMFG (BA
10) = 0.34

All

R ITG/R
IFG = 0.02

Non-religious >
religious

L IFG/R
SMFG (BA
8) = 0.09

Religious >
non-religious

Within-network pathways

R MTG/L
FG = 0.21

R SMFG
(BA 8)/R
MFG = 0.29
R MFG/R
IP/smG = 0.16

All

R SMFG/L
IFG = 0.37

R SMFG (10)/L
IFG = 0.25

Religious >
non-religious

Between-network pathways

R MTG/R
CalcG = 0.34

L MTG/R
ITG =  0.30

R MTG/R
CalcG = 0.21

Non-religious >
religious

SA, supernatural agent; IFG, inferior frontal gyrus; MTG, middle temporal gyrus; ITG, inferior temporal gyrus; SMFG, superior medial frontal gyrus; MFG, middle frontal gyrus; STG, superior
temporal gyrus.

D3 (foundation
of religious belief)

D2 (SAs perceived
love/anger)

D1 (SAs perceived
involvement)

Dimension

Covariance of the pathway

with the dimension
(positive or negative)

Table 1. Beta Values for the Positive or Negative Covariance of Each Pathway with the Dimensions ( p < 0.001)

BRAIN NETWORKS SHAPING RELIGIOUS BELIEF

subjects suggests that processing of symbolic representations

may have driven their understanding of SAs intent. By contrast, with higher + D1 (and + D3), religious subjects
recruited a pathway that assesses plausibility and resolves
conceptual ambiguities (Ye and Zhou, 2009): this pathway
originates at R frontopolar PFC (BA 10) and terminates at
Brocas area, a key area for action understanding and verbal
representation of intended behaviors (Mason and Just, 2011;
Ortigue et al., 2010).
With increased detection of perceived SAs love over
anger ( + D2), both religious and non-religious subjects
recruited a pathway running from R dmPFC (BA 8) to
vlPFC (BA 11) and from there to the R IPL (BA 40). We
speculate that the role of this pathway during religious belief
consideration is reappraisal and suppression of fear induced
by SAs. The dmPFC participates in emotional regulation by
guiding attention to resolve emotional conflicts and orchestrate emotional reappraisal (Mitchell, 2011). This reappraisal, in turn, may be carried out by vlPFC (Blair et al.,
2007; Levesque et al., 2003). In line with the above interpretation of this pathway, its third node, the IPL, is involved in
suppression of fear, such as fear induced by faces (Amting
et al., 2010; Bayle and Taylor, 2010). Failure in emotional
regulation through this pathway may result in susceptibility
to fear, including fear of SAs; as we have already shown,
subjects with decreased dlPFC volume are more prone to experience fear of SAs (Kapogiannis et al., 2009a). In addition,
vlPFC is a key region for detection of punishment cues that
require a change in behavior (Kringelbach and Rolls, 2003;
Mitchell, 2011), therefore this pathway may also play a
role in linking emotion-inducing aspects of religious belief
with behavioral guidance.
Previously, we saw that perceived SAs anger over love
(D2) correlated with activity at the L MTG, BA 21 (Kapogiannis et al., 2009b). Although the lateral temporal lobes do
not generate fear responses, they play a role in fear modulation (Goldin et al., 2008; Meletti et al., 2006). The present
pathway analysis demonstrated that, with D2, religious
subjects preferentially activate a pathway from the L MTG
to the contralateral lateral temporal lobe (R ITG, BA 20),
an area important for accurate characterization of emotions
(Rosen et al., 2006).
Dimension 3, of doctrine/semantics versus episodic memory/imagery recruited a pathway running from L precuneus
to L IFG (Brocas area) and from there to L STG (Wernickes area), nodes of the semantic processing network
(Binder et al., 2009), in both religious and non-religious subjects. In particular, the activation of the pathway from Brocas to Wernickes language areas may signify decoding of
abstract content (Chen et al., 2008; Pobric et al., 2008). In addition, the strength of a pathway running from Wernickes
area to L FG involved in script-driven imagery (Esterman
and Yantis, 2010) decreased with higher doctrinal semantic
content ( + D3), in both religious and non-religious subjects.
These findings suggest a broader systems-level function for
the D3 network, in establishing a balance between episodic
memory and semantic processing systems (Battaglia and
Pennartz, 2011). Activation of the R MTG to R CaG pathway
differentiated non-religious from religious subjects. This
pathway was preferentially activated, in non-religious compared to religious subjects, with D3 (i.e., beliefs evoking
imagery) and, + D1 (i.e., less involved SAs), suggesting

77

greater difficulty and procedural demands for imagining

and processing the intent of SAs among non-religious
subjects.
It is worth noting a few limitations of this study. The methodology employed in this study was only applied to predefined functional ROIs. Therefore, it is possible that other
regions modulate the activity and causal flow in the networks
under consideration. It is also possible that the networks and
pathways engaged during religious belief processing may
change depending on the subjects state, situational context,
and the task at hand [such as activation of analytical thinking
(Gervais and Norenzayan, 2012)].
Conclusions

This study demonstrated how insights gained by Granger

connectivity analysis inform us about the causal binding of
individual regions activated during religious belief processing. More broadly, this study enriches our understanding of
the cognitive processes and networks involved in religious
belief. It demonstrated that intent-related ToM for SAs is
causally upstream in religious belief processing, perception
of SAs engages pathways involved in fear regulation and affective ToM, and both semantic processing and imagery are
foundations of religious beliefs. These processes are dynamic and constantly inform each other at multiple levels;
this cross feeding of information varies among individuals
in association with their religiosity.
Acknowledgments

This research was supported in part by the Intramural

Research Program of the National Institute of Neurological
Disorders and Stroke (NINDS/NIH) and in part by the Intramural Research Program of the National Institute on Aging
(NIA/NIH).
Author Disclosure Statement

No competing financial interests exist for any of the authors.

References

Abler B, Roebroeck A, Goebel R, Hose A, Schonfeldt-Lecuona

C, Hole G, Walter H. 2006. Investigating directed influences
between activated brain areas in a motor-response task using
fMRI. Magn Reson Imaging 24:181185.
Amting JM, Greening SG, Mitchell DG. 2010. Multiple mechanisms of consciousness: the neural correlates of emotional
awareness. J Neurosci 30:1003910047.
Battaglia FP, Pennartz CM. 2011. The construction of semantic
memory: grammar-based representations learned from relational episodic information. Front Comput Neurosci 5:36.
Bayle DJ, Taylor MJ. 2010. Attention inhibition of early cortical
activation to fearful faces. Brain Res 1313:113123.
Baylor Institute for Studies of Religion. 2006. American Piety in
the 21st Century; New Insights to the Depth and Complexity
of Religion in the US. Houston, TX: Baylor Institute for Studies of Religion.
Binder JR, Desai RH, Graves WW, Conant LL. 2009. Where is the
semantic system? A critical review and meta-analysis of 120
functional neuroimaging studies. Cereb Cortex 19:27672796.
Blair KS, Smith BW, Mitchell DG, Morton J, Vythilingam M,
Pessoa L, Fridberg D, Zametkin A, Sturman D, Nelson EE,

78

Drevets WC, Pine DS, Martin A, Blair RJ. 2007. Modulation

of emotion by cognition and cognition by emotion. Neuroimage 35:430440.
Boyer P. 2003. Religious thought and behaviour as by-products
of brain function. Trends Cogn Sci 7:119124.
Boyer P, Bergstrom B. 2008. Evolutionary perspectives on religion. Annu RevAnthropol 37:111130.
Cavanna AE, Trimble MR. 2006. The precuneus: a review of its
functional anatomy and behavioural correlates. Brain 129:
564583.
Chen E, Widick P, Chatterjee A. 2008. Functional-anatomical
organization of predicate metaphor processing. Brain Lang
107:194202.
Clark VP. 2002. Orthogonal polynomial regression for the detection of response variability in event-related MRI. Neuroimage 17:344363.
Daubechies I. 1988. Orthonormal bases of compactly supported
wavelets. Commun Pure Appl Math 41:909996.
Desai RH, Binder JR, Conant LL, Seidenberg MS. 2010. Activation of sensory-motor areas in sentence comprehension.
Cereb Cortex 20:468478.
Deshpande G, Hu X, 2012. Investigating effective brain connectivity from FMRI data: past findings and current issues with reference to granger causality analysis. Brain Connect 2:235245.
Deshpande G, Hu XP, Lacey S, Stilla R, Sathian K. 2010a.
Object familiarity modulates effective connectivity during
haptic shape perception. Neuroimage 49:19912000.
Deshpande G, Hu XP, Stilla R, Sathian K. 2008. Effective connectivity during haptic perception: a study using Granger
causality analysis of functional magnetic resonance imaging
data. Neuroimage 40:18071814.
Deshpande G, LaConte S, James GA, Peltier S, Hu XP. 2009.
Multivariate Granger causality analysis of fMRI data. Hum
Brain Mapp 30:13611373.
Deshpande G, Sathian K, Hu XP. 2010b. Assessing and compensating for Zero-Lag correlation effects in time-lagged
Granger causality analysis of fMRI. IEEE Trans Biomed
Eng 57:14461456.
Esterman M, Yantis S. 2010. Perceptual expectation evokes category-selective cortical activity. Cereb Cortex 20:12451253.
German TP, Niehaus JL, Roarty MP, Giesbrecht B, Miller MB.
2004. Neural correlates of detecting pretense: automatic engagement of the intentional stance under covert conditions.
J Cogn Neurosci 16:18051817.
Gervais WM, Norenzayan A. 2012. Analytic thinking promotes
religious disbelief. Science 336:493496.
Gibbons RD, Lazar NA, Bhaumik DK, Sclove SL, Chen HY,
Thulborn KR, Sweeney JA, Hur K, Patterson D. 2004. Estimation and classification of fMRI hemodynamic response
patterns. Neuroimage 22:804814.
Goldin PR, McRae K, Ramel W, Gross JJ. 2008. The neural
bases of emotion regulation: reappraisal and suppression of
negative emotion. Biol Psychiatry 63:577586.
Granger CWJ. 1969. Investigating causal relations by econometric models and cross-spectral methods. Econometrica 37:
424438.
Hampstead BM, Stringer AY, Stilla RF, Deshpande G, Hu X,
Moore AB, Sathian K. 2010. Activation and effective connectivity changes following explicit-memory training for
face-name pairs in patients with mild cognitive impairment:
a pilot study. Neurorehabil Neural Repair 25:210222.
Han S, Mao L, Gu X, Zhu Y, Ge J, Ma Y. 2008. Neural consequences of religious belief on self-referential processing. Soc
Neurosci 3:115.

KAPOGIANNIS ET AL.

Jenkins AC, Mitchell JP. 2010. Mentalizing under uncertainty:

dissociated neural responses to ambiguous and unambiguous
mental state inferences. Cereb Cortex 20:404410.
Kaplan E, Naeser MA, Martin PI, Ho M, Wang Y, Baker E,
Pascual-Leone A. 2010. Horizontal portion of arcuate fasciculus fibers track to pars opercularis, not pars triangularis, in
right and left hemispheres: a DTI study. Neuroimage 52:
436444.
Kapogiannis D, Barbey AK, Su M, Krueger F, Grafman J.
2009a. Neuroanatomical variability of religiosity. PLoS
One 4:e7180.
Kapogiannis D, Barbey AK, Su M, Zamboni G, Krueger F, Grafman J. 2009b. Cognitive and neural foundations of religious
belief. Proc Natl Acad Sci U S A 106:48764881.
Kringelbach ML, Rolls ET. 2003. Neural correlates of rapid reversal learning in a simple model of human social interaction.
Neuroimage 20:13711383.
Krueger F, Landgraf S, van der Meer E, Deshpande G, Hu X.
2011. Effective connectivity of the multiplication network:
a functional MRI and multivariate Granger Causality Mapping study. Hum Brain Mapp 32:14191431.
Kus R, Kaminski M, Blinowska KJ. 2004. Determination of
EEG activity propagation: pair-wise versus multichannel estimate. IEEE Trans Biomed Eng 51:15011510.
Lacey S, Flueckiger P, Stilla R, Lava M, Sathian K. 2010a.
Object familiarity modulates the relationship between visual
object imagery and haptic shape perception. Neuroimage
49:19771990.
Lacey S, Hagtvedt H, Patrick VM, Anderson A, Stilla R, Deshpande G, Hu X, Sato JR, Reddy S, Sathian K. 2010b. Art for
rewards sake: visual art recruits the ventral striatum. Neuroimage 55:420433.
Levesque J, Eugene F, Joanette Y, Paquette V, Mensour B,
Beaudoin G, Leroux JM, Bourgouin P, Beauregard M.
2003. Neural circuitry underlying voluntary suppression of
sadness. Biol Psychiatry 53:502510.
Margulies DS, Vincent JL, Kelly C, Lohmann G, Uddin LQ,
Biswal BB, Villringer A, Castellanos FX, Milham MP,
Petrides M. 2009. Precuneus shares intrinsic functional architecture in humans and monkeys. Proc Natl Acad Sci U S A
106:2006920074.
Mason RA, Just MA. 2011. Differentiable cortical networks for
inferences concerning peoples intentions versus physical
causality. Hum Brain Mapp 32:313329.
Meletti S, Tassi L, Mai R, Fini N, Tassinari CA, Russo GL.
2006. Emotions induced by intracerebral electrical stimulation of the temporal lobe. Epilepsia 47 Suppl 5:4751.
Mier D, Lis S, Neuthe K, Sauer C, Esslinger C, Gallhofer B,
Kirsch P. 2010. The involvement of emotion recognition in
affective theory of mind. Psychophysiology 47:10281039.
Mitchell DG. 2011. The nexus between decision making and
emotion regulation: a review of convergent neurocognitive
substrates. Behav Brain Res 217:215231.
Molnar-Szakacs I, Iacoboni M, Koski L, Mazziotta JC. 2005.
Functional segregation within pars opercularis of the inferior
frontal gyrus: evidence from fMRI studies of imitation and
action observation. Cereb Cortex 15:986994.
Nahab FB, Kundu P, Gallea C, Kakareka J, Pursley R, Pohida T,
Miletta N, Friedman J, Hallett M. 2011. The neural processes
underlying self-agency. Cereb Cortex 21:4855.
Ortigue S, Sinigaglia C, Rizzolatti G, Grafton ST. 2010. Understanding actions of others: the electrodynamics of the left and
right hemispheres. A high-density EEG neuroimaging study.
PLoS One 5:e12160.

BRAIN NETWORKS SHAPING RELIGIOUS BELIEF

Pobric G, Mashal N, Faust M, Lavidor M. 2008. The role of the

right cerebral hemisphere in processing novel metaphoric expressions: a transcranial magnetic stimulation study. J Cogn
Neurosci 20:170181.
Rilling JK, Glasser MF, Preuss TM, Ma X, Zhao T, Hu X, Behrens TE. 2008. The evolution of the arcuate fasciculus
revealed with comparative DTI. Nat Neurosci 11:426428.
Roebroeck A, Formisano E, Goebel R. 2005. Mapping directed
influence over the brain using Granger causality and fMRI.
Neuroimage 25:230242.
Rosen HJ, Wilson MR, Schauer GF, Allison S, Gorno-Tempini
ML, Pace-Savitsky C, Kramer JH, Levenson RW, Weiner M,
Miller BL. 2006. Neuroanatomical correlates of impaired recognition of emotion in dementia. Neuropsychologia 44:365373.
Sato JR, Amaro E, Takahashi DY, Felix MD, Brammer MJ,
Morettin PA. 2006. A method to produce evolving functional
connectivity maps during the course of an fMRI experiment
using wavelet-based time-varying Granger causality. Neuroimage 31:187196.
Sridharan D, Levitin DJ, Menon V. 2008. A critical role for the right
fronto-insular cortex in switching between central-executive
and default-mode networks. Proc Natl Acad Sci U S A 105:
1256912574.
Steinbeis N, Koelsch S. 2009. Understanding the intentions behind man-made products elicits neural activity in areas dedicated to mental state attribution. Cereb Cortex 19:619623.

79

Stilla R, Deshpande G, LaConte S, Hu XP, Sathian K. 2007.

Posteromedial parietal cortical activity and inputs predict tactile spatial acuity. J Neurosci 27:1109111102.
Stilla R, Hanna R, Hu XP, Mariola E, Deshpande G, Sathian K.
2008. Neural processing underlying tactile microspatial discrimination in the blind: A functional magnetic resonance
imaging study. J Vis 8:119.
Strenziok M, Krueger F, Deshpande G, Lenroot RK, van der
Meer E, Grafman J. 2011. Fronto-parietal regulation of
media violence exposure in adolescents: a multi-method
study. Soc Cogn Affect Neurosci 6:537547.
Szpunar KK, Watson JM, McDermott KB. 2007. Neural substrates of envisioning the future. Proc Natl Acad Sci U S A
104:642647.
Ye Z, Zhou X. 2009. Conflict control during sentence comprehension: fMRI evidence. Neuroimage 48:280290.

Address correspondence to:

Dimitrios Kapogiannis
Laboratory of Clinical Investigation
National Institute on Aging (NIA/NIH)
3001 South Hanover Street, NM 531
Baltimore, MD 21230
E-mail: [email protected]