Accepted Manuscript

Effective connectivity in depression

Edmund T. Rolls, Wei Cheng, Matthieu Gilson, Jiang Qiu, Zicheng Hu, Hongtao
Ruan, Yu Li, Chu-Chung Huang, Albert C. Yang, Shih-Jen Tsai, Xiaodong Zhang,
Kaixiang Zhuang, Ching-Po Lin, Gustavo Deco, Peng Xie, Jianfeng Feng

PII: S2451-9022(17)30195-7
DOI: 10.1016/j.bpsc.2017.10.004
Reference: BPSC 204

To appear in: Biological Psychiatry: Cognitive Neuroscience and

Neuroimaging

Received Date: 5 May 2017

Revised Date: 10 October 2017
Accepted Date: 11 October 2017

Please cite this article as: Rolls E.T., Cheng W., Gilson M., Qiu J., Hu Z., Ruan H., Li Y., Huang C.-C.,
Yang A.C., Tsai S.-J., Zhang X., Zhuang K., Lin C.-P., Deco G., Xie P. & Feng J., Effective connectivity
in depression, Biological Psychiatry: Cognitive Neuroscience and Neuroimaging (2017), doi: 10.1016/
j.bpsc.2017.10.004.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT
Effective connectivity in depression

Edmund T. Rolls1,2, #; Wei Cheng1,3 #; Matthieu Gilson13, #; Jiang Qiu4,5, #; Zicheng Hu7,8,9,#;
Hongtao Ruan3,6; Yu Li5; Chu-Chung Huang7; Albert C. Yang11; Shih-Jen Tsai11; Xiaodong
Zhang7,8,9; Kaixiang Zhuang5; Ching-Po Lin3,7, 15,*; Gustavo Deco13,14; Peng Xie8,9,10,
*
;Jianfeng Feng1, 3, 6,12, *

PT
1. Department of Computer Science, University of Warwick, Coventry CV4 7AL, UK
2. Oxford Centre for Computational Neuroscience, Oxford, UK

RI
3. Institute of Science and Technology for Brain-inspired Intelligence, Fudan University,
Shanghai, 200433, PR China
4. Key Laboratory of Cognition and Personality (SWU), Ministry of Education, Chongqing,

SC
China
5. Department of Psychology, Southwest University, Chongqing, China
6. School of Mathematical Sciences, Fudan University, Shanghai, 200433, PR China
7. Institute of Neuroscience, National Yang-Ming University, Taipei, Taiwan

U
8. Institute of Neuroscience, Chongqing Medical University, Chongqing, China
9. Chongqing Key Laboratory of Neurobiology, Chongqing, China
AN
10. Department of Neurology, The First Affiliated Hospital of Chongqing Medical University,
Chongqing, China
11. Department of Psychiatry, Taipei Veterans General Hospital, Taipei, Taiwan
M

12. School of Life Science and the Collaborative Innovation Center for Brain Science, Fudan
University, Shanghai, 200433, PR China
13. Center for Brain and Cognition, Computational Neuroscience Group, Department of
D

Information and Communication Technologies, Universitat Pompeu Fabra, Roc Boronat 138,
Barcelona, 08018, Spain Brain and Cognition, Pompeu Fabra University, Barcelona, Spain.
TE

14. Instituci Catalana de la Recerca i Estudis Avanats (ICREA), Universitat Pompeu Fabra,
Passeig Llus Companys 23, Barcelona, 08010, Spain.
15. Brain Research Center, National Yang-Ming University, Taipei, Taiwan
EP

#
These authors contributed equally to this work.
C

Short title: effective connectivity and depression

AC

Keywords: depression; effective connectivity; orbitofrontal cortex; functional connectivity;

resting state functional neuroimaging; medial temporal lobe; precuneus

ORCID ID of Edmund T Rolls: 0000-0003-3025-1292

Corresponding author: Professor Edmund T Rolls, Oxford Centre for Computational

Neuroscience, Oxford, UK. Edmund.Rolls@oxcns.org www.oxcns.org

1
 ACCEPTED MANUSCRIPT
* Correspondence may also be addressed after publication to:
Professor Jianfeng Feng,
Centre for Computational Systems Biology,
School of Mathematical Sciences, Fudan University, Shanghai 200433, China, and
Department of Computer Science, University of Warwick, Coventry CV4 7AL, UK
E-mail: jianfeng64@gmail.com

or
Professor Peng Xie, Department of Neurology, The First Affiliated Hospital of Chongqing

PT
Medical University,
Chongqing, 400016, PR China
E-mail: xiepeng@cqmu.edu.cn

RI
or
Professor Ching-Po Lin, Institute of Neuroscience,

SC
National Yang-Ming University, Taipei, Taiwan
E-mail: cplin@ym.edu.tw

Words in abstract 250

U
AN
Words in text 3921
Number of Tables 2
Number of Figures 3
M

Supplementary Material 1.
D
TE
C EP
AC

2
 ACCEPTED MANUSCRIPT
Abstract
Background.
We go beyond resting state functional connectivity which reflects correlations in the
activity between brain areas, to effective connectivity between different brain areas to
measure directed influences of human brain regions on each other, and compare the results in
depression and controls.

Methods.

PT
We utilize a new approach to the measurement of effective connectivity in which each
brain area has a simple dynamical model, and known anatomical connectivity is used to
provide constraints. This helps the approach to measure the effective connectivity between

RI
the 94 AAL2 brain areas using resting state functional magnetic resonance imaging.
Moreover, we show how the approach can be used to measure the differences in effective
connectivity between different groups of individuals, using as an example effective

SC
connectivity in the healthy brain and in individuals with depression. The first brain-wide
resting state effective-connectivity neuroimaging analysis of depression, with 350 healthy
individuals, and 336 patients with major depressive disorder, is described.

U
AN
Results.
Key findings are that the medial orbitofrontal cortex, implicated in reward and
subjective pleasure, has reduced effective connectivity from temporal lobe input areas in
depression; that the lateral orbitofrontal cortex, implicated in non-reward, has increased
M

activity (variance) in depression, with decreased effective connectivity to and from cortical
areas contralateral to language-related areas; and that the hippocampus, implicated in
D

memory, has increased activity (variance) in depression and increased effective connectivity
from the temporal pole.
TE

Conclusions.
Measurements with the new method for effective connectivity provide a new
EP

approach to causal mechanisms in the brain in depression.

C
AC

3
 ACCEPTED MANUSCRIPT
Introduction
Resting state functional connectivity measured with functional magnetic resonance
imaging and which reflects correlations in the activity between brain areas is widely used to
help understand human brain function in health and disease (1, 2). Here we go beyond
functional connectivity to effective connectivity between different brain areas to measure
directed influences of human brain regions on each other. Effective connectivity is
conceptually very different, for it measures the effect of one brain region on another in a
particular direction, and can in principle therefore provide information more closely related
to the causal processes that operate in brain function, that is, how one brain region influences

PT
another. In the context of disorders of brain function, the effective connectivity may provide
evidence on which brain regions may have altered function, and then influence other brain
regions, by comparing effective connectivity in patients and controls participants.

RI
In this paper we utilize a new approach to the measurement of effective connectivity in
which each brain area has a simple dynamical model, and known anatomical connectivity is
used to provide constraints (3). This helps the approach to measure the effective connectivity

SC
between the 94 automated anatomical atlas (AAL2) (4) brain areas using resting state
functional magnetic resonance imaging. Moreover, we show how the approach can be used to
measure the differences in effective connectivity between different groups of individuals,

U
using as an example effective connectivity in the healthy brain and in individuals with
depression. This results in the first brain-wide resting state effective-connectivity
AN
neuroimaging analysis of depression, with 350 healthy individuals, and 336 patients with
major depressive disorder.
Major depressive disorder is ranked by the World Health Organization as the leading
M

cause of years-of-life lived with disability (5-8). Major depressive episodes, found in both
major depressive disorder and bipolar disorder are pathological mood states characterized by
persistently sad or depressed mood. Major depressive disorders are generally accompanied
D

by: (a) altered incentive and reward processing, evidenced by amotivation, apathy, and
anhedonia; (b) impaired modulation of anxiety and worry, manifested by generalized, social
TE

and panic anxiety, and oversensitivity to negative feedback; (c) inflexibility of thought and
behavior in association with changing reinforcement contingencies, apparent as ruminative
thoughts of self-reproach, pessimism, and guilt, and inertia toward initiating goal-directed
EP

behavior; (d) altered integration of sensory and social information, as evidenced by mood-
congruent processing biases; (e) impaired attention and memory, shown as performance
deficits on tests of attention set-shifting and maintenance, and autobiographical and short-
term memory; and (f) visceral disturbances, including altered weight, appetite, sleep, and
C

endocrine and autonomic function (5, 7).

AC

Patients with depression show impairments in the coordinated activity of several brain
regions considered to be important for several domains of mental functioning such as
emotional processing (amygdala, subgenual anterior cingulate and pallidum) (9, 10), self-
referential processes (medial prefrontal cortex (MPFC), precuneus and posterior cingulate
cortex) (10-12), cognitive functions such as memory (hippocampus, parahippocampal cortex)
(13), visual processing (fusiform gyrus, lingual gyrus and lateral temporal cortex) (14), and
attention processing (dorsolateral prefrontal cortex, anterior cingulate cortex, thalamus and
insula) (15).
Research into the pathophysiology of depression has included the analysis of possible
differences in the functional connectivity of different brain areas to elucidate some of the
brain changes that may relate to depression. Resting-state fMRI provides a task-free approach
that removes some performance-related confounds, and provides a reliable measure of
4
 ACCEPTED MANUSCRIPT
baseline brain activity and connectivity (16). A meta-analysis of previous investigations of
resting state functional connectivity in depression was based on seed based studies each with
tens of participants, and described hypoconnectivity within a frontoparietal network, and
hyperconnectivity within the default mode network, a network believed to support internally
oriented and self-referential thought (17). For comparison, a recent study included almost as
many participants as this meta-analysis, was not forced because of small numbers of
participants to rely on a priori, seed-based analyses, and was able given the voxel-based
approach to provide detailed information about the exact brain regions involved (2), rather
than brain systems identified for example as the default mode network or fronto-parietal

PT
control systems. In that first investigation using a voxel-based unbiased brain-wide
association study (BWAS) approach on resting-state functional magnetic resonance imaging
(fMRI) data in 421 patients with major depressive disorder compared to 488 controls (2), we

RI
found decreased functional connectivity between the medial orbitofrontal cortex (which has
functions related to reward) with medial temporal lobe memory-related areas including the
perirhinal cortex BA 36 and entorhinal cortex BA 28. We also found that the lateral

SC
orbitofrontal cortex BA 47/12 (which has functions related to non-reward and punishment)
had increased functional connectivity with the precuneus, the angular gyrus, and the temporal
visual cortex BA 21 (2).

U
In the present research, we go beyond the functional connectivity (FC) approach, which
reflects the inter-region correlations of the observed activity, to effective connectivity (EC),
AN
which estimates causal/directed interactions between brain regions. In essence, our model-
based approach infers two sets of parameters from FC: the local fluctuating activity for each
ROI (e.g., excitability, described by the diagonal parameters of the matrix ) and the matrix
M

of EC weights between the ROIs (the existence of connections is determined from DTI data
beforehand). Our dynamic model combines these parameters to generate FC, taking network
effects into account at the whole-brain level. This approach may be useful in understanding
D

the changes underlying depression, as it is not clear whether causes are circumscribed to the
activity of a few nodes, or connectivity within a subnetwork. The new method that we use (3,
TE

18) has advantages and limitations compared to dynamic causal modelling (DCM) (19-21),
and is described in the Supplementary Material.
To help understand some of the implications of the new findings described here, we note
EP

that there is considerable evidence that the medial and middle orbitofrontal cortex is involved
in reward, and that the lateral orbitofrontal cortex is involved in non-reward and punishment
(2, 8, 22-24). The hypotheses being investigated were that effective connectivity might be
different between patients with depression, and that the identification of which effective
C

connectivities were different may be useful in understanding the neural bases of depression.
AC

Methods
The methods are described in the Supplementary Material.

Results

The fMRI resting state effective connectivity analyses were performed with 336 patients
with a diagnosis of major depression, and 350 controls, and this large population was
sufficient to allow FDR corrected statistics as described in detail elsewhere (2, 25, 26) with
the 94 areas in the AAL2 brain atlas (4) (see Table S2 for the abbreviations for each area).

5
 ACCEPTED MANUSCRIPT
Differences in Effective Connectivity between patients with Major Depressive Disorder
and controls
The results of the comparison of Effective Connectivity between patients with Major
Depressive Disorder (MDD) and controls are shown in Fig. 1 and Table 1. The results shown
are those with significantly different effective connectivity links after FDR p<0.05 correction,
and with a threshold for the EC >=0.01 in either or both healthy controls and patients with
depression. (This thresholding on effect size precludes reporting trivial effects.) In Table 1,
the forward direction is the direction with the higher effective connectivity (see
Supplementary Material for further explanation). In Table 1, the results are grouped usefully

PT
according to the target region of the altered effective connectivity. We use this grouping by
target brain regions to help describe the results for the main groups of differences of effective
connectivity between patients and controls. The matrices of Effective Connectivity are shown

RI
in Fig. S1 for reference. We focus below on Figures of the brain and tables showing the
differences in effective connectivity, but Fig. S1 shows for example that temporal lobe areas
85-94 in AAL2 (Table S2) tend to have high effective connectivity directed to the

SC
orbitofrontal areas (25-32) in healthy controls. Another interesting effect is that the functional
connectivity from frontal areas including the inferior frontal gyrus (3-12) and lateral
orbitofrontal cortex (31, 32) are strong in the direction to supramarginal and angular gyri (65-

U
70). Thus effective connectivity provides useful information, emphasizing that medial
orbitofrontal cortex areas receive from the temporal cortex, and that the lateral orbitofrontal
AN
cortex / inferior frontal gyrus has strong forward connections to language areas. We
emphasize that for most links that are different in depressed patients, the differences are in
both the forward and backward effective connectivities (see Table 1 and Fig. S2, FDR
M

corrected p<0.05). What is especially new about the findings presented here is the direction
of the forward vs the backward connectivity of these links that are different in depression,
and this is emphasized in Fig. 1 by larger arrow heads in the direction of the forward
D

connectivity, defined as the direction with the greater effective connectivity.

We summarize some of the main points evident by inspection of Fig. 1, and then provide
TE

a more detailed analysis referring also to Table 1 below. One feature apparent in Fig. 1 is that
in depression a number of areas including the parahippocampal gyrus, inferior temporal
gyrus, and temporal pole have decreased effective connectivity directed to the medial and
EP

middle orbitofrontal cortex areas. Another feature is that the fusiform gyrus (FFG) has
decreased effective connectivity directed to earlier visual cortical areas (occipital).

Medial and Middle Orbitofrontal Cortex

C

The AAL2 regions included in this group and shown in Table 1 are the OFC_med,
AC

OFC_ant, OFC_post, Rectus, and OLF (the olfactory tubercle region at the posterior end of
the orbitofrontal cortex). This medial and middle orbitofrontal cortex region has decreased
effective connectivity (shown by a negative value for z in Table 1 and a blue arrow in Fig. 1
into the target) from brain regions including the parahippocampal gyrus, temporal pole,
inferior temporal gyrus, and amygdala. This implies less strong positive driving influences of
these regions on the medial and middle orbitofrontal cortex (see columns 7 and 8 of Table 1).
Many of these effective connectivities were much greater in the forward direction into the
medial orbitofrontal cortex than in the backward direction (Table 1). Both the forward and
the backward effective connectivities were in general lower in the depressed group than in
the controls (Table 1).
There is also reduced effective connectivity between some of these different AAL2
regions in the medial and middle orbitofrontal cortex (see Table 1).
6
 ACCEPTED MANUSCRIPT

Lateral orbitofrontal cortex

The AAL2 regions included in this group and shown in Table 1 are the OFC_lat and
Frontal_Inf_Orb. The OFC_post (one of the middle OFC areas) has increased effective
connectivity directed to the OFC_lat. Given that the medial orbitofrontal cortex (which
includes OFC_post) tends to be activated by rewards and the lateral orbitofrontal cortex by
non-rewards and punishers (23, 27) and even that they are reciprocally activated by reward
and loss (28), we sought to elucidate the interpretation of this increase in effective
connectivity from medial to lateral orbitofrontal cortex in depression. The effective

PT
connectivity measure does not specify whether this should be interpreted as increased
excitatory input from the medial to the lateral orbitofrontal cortex; or an increased
connectivity which might reflect that any change in medial OFC produces a larger change,

RI
but in the opposite (reciprocal) direction. We reasoned that the functional connectivity
between the medial and lateral orbitofrontal cortex might provide relevant evidence. What we
found in summary is that all the medial orbitofrontal cortex areas (OFCmed, OFCant,

SC
OFCpost, Rectus, and OLF) have a high functional connectivity with each other that is on
average 0.58 (std 0.13) (in the control group). Similarly, the two lateral orbitofrontal cortex
areas (OFClat and IFG_Orb) have high functional connectivity with each other that is on

U
average 0.68 (std 0.08). However, the mean FC between the medial orbitofrontal cortex areas
and lateral orbitofrontal cortex areas was much lower, 0.36 (std 0.16), and the difference was
AN
significant (t test, p<10-12). Further, this relates to an average functional connectivity value
across all pairs in the brain of 0.35. This evidence provides an indication that the medial and
lateral orbitofrontal cortex areas are not positively coupled to each other, but can operate in
M

opposite directions, and even could operate reciprocally. We thus interpret the increased
effective connectivity from medial to lateral orbitofrontal cortex as consistent with the
hypothesis that underactivity in the medial orbitofrontal cortex in depression (2, 8) may be
D

one of the causes of lateral orbitofrontal cortex activity being high in depression for which
evidence is described below and elsewhere (2, 8).
TE

In addition, the Inferior frontal gyrus opercular part back-connection to the lateral
orbitofrontal cortex is reduced in depression. The Supramarginal gyrus has decreased
effective connectivity with the Frontal_Inf_Orb_2 in depression. The Supramarginal gyrus_R
EP

also has decreased effective connectivity in both directions with OFClat_R.

Temporal lobe
The temporal lobe areas with different effective connectivity in depression include the
C

temporal pole, inferior, and middle temporal gyrus. Most of these areas have reduced forward
AC

effective connectivity directed to medial and middle orbitofrontal cortex areas including
OFCmed and OFCant. (Although Table 1 shows significant increases in the backprojection to
the temporal areas from the precuneus, we discount these because these backprojection ECs
are so very low.)

Hippocampus and parahippocampal gyrus

The effective connectivity directed from the temporal pole to the hippocampus is
increased in depression. As noted above, the effective connectivity from the
parahippocampal gyrus to the medial orbitofrontal cortex areas (and to the superior parietal
lobule) is decreased in depression.

Precuneus
7
 ACCEPTED MANUSCRIPT
Four forward links from the left inferior/mid temporal gyrus to the precuneus have
increased effective connectivity in depression (Table 1 and Fig. 1). It is notable that these
links have a very much greater strength in the forward than in the backward direction, with a
mean ratio of > 20 (Table 1). (It is noted that separately each of these forward links did not
quite reach the threshold required for FDR correction, although the strengths in the backward
direction did.)

Sensori-motor cortical areas

The precentral gyrus (motor cortex) has increased EC directed to some other motor areas

PT
including the Supplementary Motor Area.

Differences in both forward and backward effective connectivity in depression

RI
The results of the comparison of EC (forward - backwards) between MDD and HC are
shown in Fig. S2. The main implication of this Figure is that links change similarly in both

SC
directions in depression. That is, if an effective connectivity link is stronger in one direction
in depression, it is likely to be stronger in the other direction too; and if a link is weaker in
one direction in depression, it is likely to be weaker in the opposite direction too (r=0.44,

U
p<0.0001). AN
Differences in , the spontaneous activity parameter, between patients with Major
Depressive Disorder and controls
M

The results of the comparison of between MDD and HC are shown in Fig. 2 and Table
2. values for AAL2 regions significantly different (FDR corrected p<0.05) between
depressed patients and controls are shown.
D

One point of particular interest is that for the right and left hippocampus is
significantly increased in patients with major depressive disorder. This is in the context that
TE

the effective connectivity directed from the temporal pole to the hippocampus is increased in
depression.
A second point of particular interest is that for the lateral orbitofrontal cortex
EP

(OFClat_L) is significantly increased in patients with major depressive disorder. This effect
spread as far medially as at least a part of OFCant_L. For comparison, the value of for
OFClat_R was also increased in depression (p<0.05 uncorrected).
C

This is consistent with the hypothesis that in depression there is increased activity in
the lateral orbitofrontal cortex (a region involved in non-reward and punishment), and the
AC

hippocampus (a region involved in memory) (2, 8).

Correlations between the effective connectivity links and the depression severity

Correlations between the effective connectivity links and the depression symptom
severity scores, in particular, the illness duration, are shown in Table S3. These results
provide an indication that the differences in effective connectivity that were found are related
to the severity of the depression. Further evidence consistent with this is that some of the
effective connectivity links were correlated with the scores on the Beck Depression Inventory
(BDI), the Hamilton Depression rating scale (HAM-D) and the Hamilton Anxiety rating scale
8
 ACCEPTED MANUSCRIPT
(HAM-A), as shown in as shown in Table S3. (This information is provided to help interpret
the findings, though we do not rely on these correlations because they are not corrected for
multiple comparisons.)

Summary diagram

A summary of the networks that show different effective connectivity in patients with
depression is shown in Fig. 3. A decrease in effective connectivity is shown in blue, and an
increase in red.

PT
Discussion

RI
The main findings include the following in this investigation of effective

SC
connectivity with 336 patients with major depressive disorder and 350 controls. The key
findings are that the medial orbitofrontal cortex, implicated in reward and subjective pleasure,
has reduced effective connectivity from temporal lobe areas in depression; that the lateral
orbitofrontal cortex, implicated in non-reward, has increased activity in depression, with

U
decreased effective connectivity to and from areas contralateral to language-related areas
AN
(including supramarginal gyrus); and that the hippocampus, implicated in memory, has
increased activity in depression, and increased effective connectivity from the temporal pole.
In more detail, it was found that effective connectivity directed to the medial
M

orbitofrontal cortex from areas including the parahippocampal gyrus, temporal pole, inferior
temporal gyrus, and amygdala were decreased in depression. This is the forward direction for
most of these links. This implies less strong positive driving influences of these input regions
D

on the medial and middle orbitofrontal cortex, regions implicated in reward, and thus helps to
elucidate part of the decreased feelings of happy states in depression (8). The forward links
TE

from temporal cortical areas to the precuneus are increased in depression (and are close to
significant after FDR correction), and this may relate to representations of the sense of self
(29), which become more negative in depression (2, 8). The lateral orbitofrontal cortex areas
EP

have reduced effective connectivity with the (mainly right) inferior frontal gyrus opercular
part and directed to the supramarginal gyrus. In addition, the lateral orbitofrontal cortex, an
area implicated in non-reward and punishment, had an increased level of activity as reflected
C

in in the depressed group. A notable finding was that was also increased in the right and
left hippocampus of patients with depression, reflecting it is suggested some type of
AC

heightened memory-related processing. This is in the context that the effective connectivity
directed from the temporal pole to the hippocampus is increased in depression. Together
these differences are consistent with the hypothesis that some aspects of hippocampal
processing, perhaps those related to unpleasant memories, are increased in depression (2, 8),
and that the influence of temporal lobe memory systems on specifically the medial
orbitofrontal cortex is reduced in depression. The value of effective connectivity in
understanding the operation of these systems in depression is that although the functional
connectivity (which reflects correlations) between these areas has been shown to be reduced
in depression (2), it is only by using effective connectivity that we understand better the
direction of the major influence between these brain regions (from the temporal lobe to the

9
 ACCEPTED MANUSCRIPT
medial orbitofrontal cortex), and for example that this directed connectivity is reduced in
depression (Fig. 3 and Table 1).
The findings for different brain systems are now considered, putting together the results
not only of the effective connectivity analysis described here, but also of the large analysis of
functional connectivity in patients with depression (2).
A very interesting finding of the investigation is that the medial (which include the
middle) orbitofrontal cortex-related areas receive forward projections from the temporal
cortex areas as shown by the effective connectivity measure. This is consistent with macaque

PT
neuroanatomy (30, 31), and with the fact that these medial orbitofrontal cortex areas have
responses to visual, taste, olfactory, somatosensory and auditory inputs, which must originate
from temporal, insular, olfactory etc areas. The medial orbitofrontal cortex areas have

RI
neuronal responses in macaques and fMRI activations in humans which show that they
represent the reward value of these stimuli (22, 23, 27, 32). The implication is that the
reduced forward inputs into the medial orbitofrontal cortex in depression relate to the

SC
decrease in positively affective states that are present in depression, and that this is one of the
key brain changes related to depression (2, 8, 33-35). This hypothesis is supported by the
finding that the decrease in the effective connectivity to the anterior orbitofrontal cortex from

U
temporal lobe areas is correlated with the severity of the depression as assessed by the
AN
duration of the illness (Table S3).
With respect to the lateral orbitofrontal cortex, we previously reported that there is
increased functional connectivity between the lateral orbitofrontal cortex and the precuneus,
angular gyrus, and inferior temporal cortex (2). In the context of the functions of the lateral
M

orbitofrontal cortex in non-reward and punishment (8, 24), this increased functional
connectivity was related to increased negative value of the self (low self-esteem) (precuneus),
D

to increased language-based negative thoughts (rumination) (angular gyrus), and to increased

aversive or non-rewarding effects of some visual stimuli (inferior temporal cortex) (2). The
TE

new findings presented here provide supporting complementary evidence. For example, the
activity as reflected by was increased in the lateral orbitofrontal cortex of patients with
major depressive disorder (Table 2), consistent with increased non-reward / aversive
EP

processing in depression being implemented by the lateral orbitofrontal cortex (8, 24). The
right Inferior frontal gyrus opercular part (area 44) connection from the lateral orbitofrontal
cortex is reduced in depression. The Frontal_Inf_Orb_2_R (a lateral part of the lateral
C

orbitofrontal cortex) has reduced effective connectivity from the Supramarginal gyrus_R.
Thus the lateral orbitofrontal cortex has a number of reduced effective connectivities with
AC

areas mainly contralateral to language-related areas. The most interesting finding was the
increase in activity (assessed by ) in the lateral orbitofrontal cortex in depression, which
taken with the increased functional connectivity with the precuneus and language areas in
depression (2), support the hypothesis of low self-esteem and high rumination being related
to the connections to these two areas in depression.
The link from the inferior temporal gyrus and temporal pole to the right posterior
cingulate cortex is increased in depression. These complementary findings serve to draw
attention to the altered functioning of the precuneus (and connected posterior cingulate
cortex), which is involved in representations of the self (29), in depression. The relevant

10
 ACCEPTED MANUSCRIPT
circuit may include the lateral orbitofrontal cortex, precuneus, posterior cingulate, and
temporal lobe cortical areas.
Although it was not a primary aim of this investigation, and following a suggestion, the
effects of medication were assessed by comparing the functional connectivity in 125 patients
not receiving medication, and 157 patients receiving medication. The overall pattern of
functional connectivity differences between patients and controls is similar for the
unmedicated and the medicated subgroups of patients (Fig. S4), providing evidence that the
main differences between patients and controls shown in Fig. 1 were found in depressed

PT
patients whether or not they were receiving medication. Further details are provided in the
Supplementary Material.
Finally, in this large-scale test of the effective connectivity algorithm (3), we show that it

RI
has potential to elucidate processing in the brain that goes beyond correlations between brain
areas (functional connectivity) to directed connectivity between brain areas (effective
connectivity). The approach thus provides evidence on how one brain area may influence

SC
another. Part of the power of the approach compared to other approaches is that evidence on
the anatomical connectivity of the brain is taken into account. The research described here
thus makes a contribution to understanding brain structure and function, and indeed how

U
structure and function are related in both normal and disordered brain function.
AN
M
D
TE
C EP
AC

11
 ACCEPTED MANUSCRIPT
Contributors
Edmund T. Rolls, Matthieu Gilson, Wei Cheng and Jianfeng Feng contributed to the design
of the study. Jiang Qiu, Zicheng Hu, Hongtao Ruan, Yu Li, Chu-Chung Huang, Albert C.
Yang, Shih-Jen Tsai, Xiaodong Zhang, Kaixiang Zhuang, Ching-Po Lin and Peng Xie
contributed to the collection of the data. Wei Cheng, Edmund T. Rolls, and Matthieu Gilson
contributed to the analysis of the data and the preparation of the manuscript. Edmund T.
Rolls, Wei Cheng, and Matthieu Gilson participated in writing the paper, with Gustavo Deco
involved in the interpretation of the findings. All collaborators had an opportunity to
contribute to the interpretation of the results and to the drafting of the manuscript.

PT
Financial Disclosures
All authors report no biomedical financial interests or potential conflicts of interest.

RI
Acknowledgements
J.Feng is a Royal Society Wolfson Research Merit Award holder. J.Feng is also partially

SC
supported by the National High Technology Research and Development Program of China
No. 2015AA020507 and the key project of Shanghai Science & Technology Innovation
Plan (No. 15JC1400101). The research was partially supported by the National Centre for

U
Mathematics and Interdisciplinary Sciences (NCMIS) of the Chinese Academy of Sciences,
Key Program of National Natural Science Foundation of China (No. 91230201), and the
AN
Shanghai Soft Science Research Program (No. 15692106604). Wei Cheng is supported by
grants from the National Natural Sciences Foundation of China (No.81701773, 11771010,
11471081, 11101429 and 71661167002), Sponsored by Shanghai Sailing Program (No.
M

17YF1426200) and the Research Fund for the Doctoral Program of Higher Education of
China (No. 2017M610226). CP.Lin was supported in part by funding from Ministry of
Science and Technology, Taiwan (NSC100-2911-I-010-010, NSC101-2911-I-010-009,
D

NSC100-2628-E-010-002-MY3, NSC102-2321-B-010-023, and NSC103-2911-I-010-501),

National Health Research Institutes (NHRI-EX103-10310EI), Ministry of Health and
TE

Welfare of Taiwan (DOH102-TD-PB-111-NSC006), and Academia Sinica, Taipei, Taiwan.

J.Qiu was supported by the National Natural Science Foundation of China (31271087;
31470981; 31571137; 31500885), National Outstanding young people plan, the Program for
EP

the Top Young Talents by Chongqing, the Fundamental Research Funds for the Central
Universities (SWU1509383), Natural Science Foundation of Chongqing
(cstc2015jcyjA10106), General Financial Grant from the China Postdoctoral Science
Foundation (2015M572423). P.Xie is supported by National Science Foundation of China
C

(NSFC 31271189). The effective connectivity algorithm work was supported by the Human
AC

Brain Project (grant FP7-FET-ICT-604102 and H2020-720270 HBP SGA1 to GD) and the
Marie Sklodowska-Curie Action (grant H2020-MSCA-656547 to MG).

12
 ACCEPTED MANUSCRIPT
Table 1. Effective connectivity links between depressed patients and controls.
Forward refers to the direction in which the link is strongest, in the direction from
AAL2 Region 1 to Region 2. Links are shown if their EC value in either direction
exceeds the threshold of 0.01, and if there is a significant different in at least one
direction using FDR correction for multiple comparisons, for which the significance
level must be p<1.6E-02. Significant differences are shown in red font. A negative
value for z indicates a weaker effective connectivity link in patients with depression.
EC of EC of EC of EC of
z value for p value for z value for p value for EC ratio in HC
Region 1 Region 2 forward in forward in backward in backward in
forward forward backward backward (forward/backward)

PT
HC MDD HC MDD
OFCpost_L Amygdala_L -2.917 3.53E-03 -3.914 9.09E-05 0.012 0.009 0.009 0.007 1.345
Temporal_Pole_Mid_R Cingulate_Post_R 2.225 2.61E-02 3.611 3.05E-04 0.021 0.022 0.003 0.003 7.433
Temporal_Inf_L Cingulate_Post_R 1.903 5.70E-02 3.557 3.75E-04 0.011 0.012 0.001 0.001 18.872
Frontal_Mid_2_L Frontal_Sup_Medial_R -3.7 2.15E-04 -3.32 9.01E-04 0.011 0.008 0.004 0.003 2.635
Insula_L Insula_R 0.193 8.47E-01 3.449 5.63E-04 0.028 0.028 0.018 0.018 1.534
Fusiform_L Occipital_Mid_L -4.032 5.52E-05 0.31 7.56E-01 0.024 0.022 0.008 0.008 3.06
Fusiform_R Occipital_Mid_R -3.717 2.02E-04 -1.211 2.26E-01 0.023 0.02 0.011 0.01 2.076

RI
ParaHippocampal_R Occipital_Mid_R -3.328 8.74E-04 -0.24 8.11E-01 0.016 0.013 0.007 0.006 2.386
Fusiform_R Occipital_Sup_L -3.487 4.89E-04 -1.317 1.88E-01 0.023 0.02 0.005 0.005 4.167
Fusiform_L Occipital_Sup_L -3.338 8.43E-04 -0.285 7.76E-01 0.024 0.021 0.005 0.005 4.77
Temporal_Inf_L OFCant_L -2.309 2.10E-02 -3.334 8.55E-04 0.014 0.012 0.006 0.005 2.225
OFCpost_L OFClat_L 3.466 5.29E-04 -0.767 4.43E-01 0.018 0.02 0.008 0.008 2.238
Frontal_Inf_Oper_R OFClat_R -1.462 1.44E-01 -3.74 1.84E-04 0.014 0.011 0.007 0.005 1.988
ParaHippocampal_L OFCmed_L -3.247 1.17E-03 -1.209 2.27E-01 0.014 0.011 0.005 0.004 2.548

SC
OFCant_L OFCmed_L -3.168 1.54E-03 1.507 1.32E-01 0.019 0.016 0.015 0.015 1.224
ParaHippocampal_R OFCmed_R -4.749 2.05E-06 -2.854 4.32E-03 0.013 0.01 0.005 0.003 2.815
Temporal_Pole_Mid_R OFCmed_R -5.084 3.70E-07 -4.014 5.98E-05 0.011 0.007 0.005 0.003 2.481
Olfactory_R OFCmed_R -3.255 1.14E-03 -2.897 3.76E-03 0.019 0.016 0.015 0.013 1.221
Temporal_Pole_Mid_L OFCpost_L -3.443 5.76E-04 -3.051 2.28E-03 0.012 0.01 0.011 0.009 1.074
Temporal_Inf_L OFCpost_L -3.354 7.97E-04 -1.493 1.35E-01 0.013 0.01 0.006 0.005 2.081
Temporal_Inf_R OFCpost_R -2.765 5.70E-03 -3.383 7.18E-04 0.011 0.009 0.007 0.006 1.517
OFCmed_R Olfactory_L -2.716 6.61E-03 -3.535 4.07E-04 0.014 0.012 0.013 0.011 1.045
OFCmed_L Olfactory_L -2.697 7.00E-03 -3.41 6.49E-04 0.017 0.014 0.016 0.013 1.079

U
Temporal_Pole_Mid_L Olfactory_L -3.272 1.07E-03 -2.919 3.51E-03 0.01 0.008 0.006 0.004 1.814
OFCant_L Olfactory_L -3.208 1.34E-03 -1.499 1.34E-01 0.011 0.009 0.007 0.006 1.622
Temporal_Pole_Mid_R Olfactory_R -3.761 1.69E-04 -3.283 1.03E-03 0.011 0.008 0.006 0.004 1.84
SupraMarginal_R Parietal_Inf_R 0.706 4.80E-01 3.443 5.74E-04 0.023 0.022 0.013 0.014 1.741
AN
ParaHippocampal_L Parietal_Sup_L -3.983 6.81E-05 -1.636 1.02E-01 0.014 0.01 0.003 0.002 4.694
Precentral_L Precentral_R -2.222 2.63E-02 3.32 9.01E-04 0.021 0.018 0.02 0.021 1.032
Temporal_Inf_L Precuneus_L 1.735 8.28E-02 3.179 1.48E-03 0.02 0.02 0.001 0.001 23.087
Temporal_Inf_L Precuneus_R 2.879 3.99E-03 3.917 8.97E-05 0.014 0.015 0 0.001 31.401
Temporal_Mid_L Precuneus_R 2.915 3.56E-03 3.258 1.12E-03 0.011 0.012 0.001 0.001 12.235
Temporal_Inf_R Precuneus_R 1.524 1.28E-01 3.215 1.31E-03 0.015 0.015 0.001 0.001 16.205
Olfactory_L Rectus_L -1.421 1.55E-01 -3.443 5.74E-04 0.017 0.016 0.012 0.01 1.42
OFCmed_R Rectus_L -0.657 5.11E-01 -3.237 1.21E-03 0.015 0.013 0.01 0.007 1.503
Precentral_R Rolandic_Oper_R -0.551 5.82E-01 -3.552 3.82E-04 0.02 0.02 0.015 0.012 1.347
M

Precentral_R Supp_Motor_Area_L 3.786 1.53E-04 -0.046 9.63E-01 0.019 0.02 0.009 0.008 2.167
Precentral_R Supp_Motor_Area_R 3.205 1.35E-03 -0.924 3.55E-01 0.019 0.02 0.013 0.012 1.412
Frontal_Inf_Orb_2_R SupraMarginal_R -2.88 3.98E-03 -4.056 5.00E-05 0.015 0.012 0.01 0.008 1.507
Temporal_Pole_Mid_R Temporal_Pole_Mid_L 0.509 6.11E-01 3.3 9.67E-04 0.026 0.025 0.019 0.02 1.376
Hippocampus_L Temporal_Pole_Mid_L 1.611 1.07E-01 3.267 1.09E-03 0.012 0.013 0.007 0.008 1.625
Precentral_R Temporal_Sup_R -0.437 6.62E-01 -4.033 5.51E-05 0.017 0.016 0.009 0.007 1.787
D
TE
C EP
AC

13
 ACCEPTED MANUSCRIPT
Table 2. values for AAL2 regions significantly different (FDR corrected) between
depressed patients and controls. The of HC shown is the mean after normalization
within each participant.

Region z value of p value of of HC of MDD

Precentral_R -3.946 7.96E-05 -0.288 -0.481
Hippocampus_L 3.926 8.64E-05 -0.942 -0.891
Occipital_Mid_R -3.295 9.83E-04 -0.309 -0.378

PT
Putamen_L 3.212 1.32E-03 -0.896 -0.854
Postcentral_R -3.154 1.61E-03 -0.258 -0.400
OFClat_L 3.068 2.16E-03 1.086 1.253

RI
Paracentral_Lobule_R -3.058 2.23E-03 1.001 0.647
Hippocampus_R 2.980 2.88E-03 -0.989 -0.942

SC
Paracentral_Lobule_L -2.936 3.33E-03 0.776 0.461
OFCant_L 2.777 5.48E-03 -0.464 -0.356

U
AN
M
D
TE
C EP
AC

14
 ACCEPTED MANUSCRIPT
Figure legends

Figure 1. Differences in Effective Connectivity between patients with major depressive

disorder and controls. MDD and HC. The links shown are those with significantly different
effective connectivity after FDR p<0.05 correction. Red indicates that the effective
connectivity is increased in patients, and blue that it is decreased. The direction of the
stronger effective connectivity is indicated by an arrow head in only one direction. If a link is
decreased in strength in one direction in patients with depression, it is usually decreased in
strength in the other direction, as shown in Table 1; and vice versa. If the effective

PT
connectivities were similar (the ratio was less than 1.5), then arrow heads are shown in both
directions. The exact values and statistics for these links are provided in Table 1. Table 1
shows for example that although the forward connectivity from the visual areas classed as

RI
calcarine to the orbitofrontal cortex is increased in patients, the actual values for this effective
connectivity are small. Only AAL2 regions are shown that have significantly different EC
values in patients and controls on at least one side of the brain. The glass brains were

SC
generated using BrainNet Viewer (36).

Figure 2. The results of the comparison of between patients with major depressive disorder

U
and healthy controls. This figure shows the significant AAL2 areas after FDR 0.05 correction.
AN
Normalisation of was used, applied in the same way as for the effective connectivity. Red-
yellow indicates AAL2 regions with increased , and blue with decreased (see Table 2).

Figure 3. Summary of the networks that show different effective connectivity in patients
M

with depression, shown on a ventral view of the brain. A decrease in effective connectivity in
patients with major depressive disorder is shown in blue, and an increase in red. In most
D

cases there was a similar change in the effective connectivity in both directions in depression.
The direction of the arrows shows though the direction of the stronger (termed forward)
TE

effective connectivity. Regions with an increased value of , reflecting increased activity, are
indicated by a red circle; and regions with a decreased value of , are indicated by a red
circle. For further details of the differences in the effective connectivities, and the side of the
EP

brain on which they are present, are provided in Table 1 and Fig. 1. (ECFig3a.eps)
C
AC

15
 ACCEPTED MANUSCRIPT

References

1. Deco G, Kringelbach ML (2014): Great expectations: using whole-brain

computational connectomics for understanding neuropsychiatric disorders. Neuron.
84:892-905.

PT
2. Cheng W, Rolls ET, Qiu J, Liu W, Tang Y, Huang CC, et al. (2016): Medial
reward and lateral non-reward orbitofrontal cortex circuits change in opposite
directions in depression. Brain. 139:3296-3309.

RI
3. Gilson M, Moreno-Bote R, Ponce-Alvarez A, Ritter P, Deco G (2016):
Estimation of directed effective connectivity from fMRI functional connectivity hints
at asymmetries in the cortical connectome. PLoS Comput Biol. 12:e1004762.

SC
4. Rolls ET, Joliot M, Tzourio-Mazoyer N (2015): Implementation of a new
parcellation of the orbitofrontal cortex in the automated anatomical labeling atlas.
Neuroimage. 122:1-5.

U
5. Drevets WC (2007): Orbitofrontal cortex function and structure in depression.
AN
Ann N Y Acad Sci. 1121:499-527.
6. Hamilton JP, Chen MC, Gotlib IH (2013): Neural systems approaches to
understanding major depressive disorder: an intrinsic functional organization
perspective. Neurobiol Dis. 52:4-11.
M

7. Gotlib IH, Hammen CL (2009): Handbook of Depression. New York: Guilford

Press.
D

8. Rolls ET (2016): A non-reward attractor theory of depression. Neurosci

Biobehav Rev. 68:47-58.
TE

9. Disner SG, Beevers CG, Haigh EA, Beck AT (2011): Neural mechanisms of the
cognitive model of depression. Nat Rev Neurosci. 12:467-477.
10. Sheline YI, Price JL, Yan Z, Mintun MA (2010): Resting-state functional MRI in
EP

depression unmasks increased connectivity between networks via the dorsal nexus.
Proc Natl Acad Sci U S A. 107:11020-11025.
11. Kuhn S, Gallinat J (2013): Resting-state brain activity in schizophrenia and
C

major depression: a quantitative meta-analysis. Schizophr Bull. 39:358-365.

12. Price JL, Drevets WC (2010): Neurocircuitry of mood disorders.
AC

Neuropsychopharmacology. 35:192-216.
13. Lorenzetti V, Allen NB, Fornito A, Yucel M (2009): Structural brain
abnormalities in major depressive disorder: a selective review of recent MRI studies.
J Affect Disord. 117:1-17.
14. Veer IM, Beckmann CF, van Tol MJ, Ferrarini L, Milles J, Veltman DJ, et al.
(2010): Whole brain resting-state analysis reveals decreased functional connectivity
in major depression. Front Syst Neurosci. 4.
15. Hamilton JP, Etkin A, Furman DJ, Lemus MG, Johnson RF, Gotlib IH (2012):
Functional neuroimaging of major depressive disorder: a meta-analysis and new
integration of base line activation and neural response data. Am J Psychiatry.
16
 ACCEPTED MANUSCRIPT
169:693-703.
16. Gusnard DA, Raichle ME, Raichle ME (2001): Searching for a baseline:
functional imaging and the resting human brain. Nat Rev Neurosci. 2:685-694.
17. Kaiser RH, Andrews-Hanna JR, Wager TD, Pizzagalli DA (2015): Large-Scale
Network Dysfunction in Major Depressive Disorder: A Meta-analysis of Resting-
State Functional Connectivity. JAMA Psychiatry. 72:603-611.
18. Gilson M, Deco G, Friston K, Hagmann P, Mantini D, Betti V, et al. (2017):
Effective connectivity inferred from fMRI transition dynamics during movie viewing

PT
points to a balanced reconfiguration of cortical interactions. bioRxiv.110015.
19. Valdes-Sosa PA, Roebroeck A, Daunizeau J, Friston K (2011): Effective
connectivity: influence, causality and biophysical modeling. Neuroimage. 58:339-361.

RI
20. Bajaj S, Adhikari BM, Friston KJ, Dhamala M (2016): Bridging the Gap:
Dynamic Causal Modeling and Granger Causality Analysis of Resting State
Functional Magnetic Resonance Imaging. Brain Connect.

SC
21. Friston K (2009): Causal modelling and brain connectivity in functional
magnetic resonance imaging. PLoS Biol. 7:e33.
22. Rolls ET (2014): Emotion and Decision-Making Explained. Oxford: Oxford

U
University Press.
AN
23. Grabenhorst F, Rolls ET (2011): Value, pleasure, and choice in the ventral
prefrontal cortex. Trends Cogn Sci. 15:56-67.
24. Rolls ET (2017): The roles of the orbitofrontal cortex via the habenula in non-
reward and depression, and in the responses of serotonin and dopamine neurons.
M

Neurosci Biobehav Rev. 75:331-334.

25. Cheng W, Rolls ET, Gu H, Zhang J, Feng J (2015): Autism: reduced functional
D

connectivity between cortical areas involved in face expression, theory of mind, and
the sense of self. Brain. 138:1382-1393.
TE

26. Cheng W, Palaniyappan L, Li M, Kendrick KM, Zhang J, Luo Q, et al. (2015):

Voxel-based, brain-wide association study of aberrant functional connectivity in
schizophrenia implicates thalamocortical circuitry. NPJ Schizophr. 1:15016.
EP

27. Rolls ET, Grabenhorst F (2008): The orbitofrontal cortex and beyond: from
affect to decision-making. Prog Neurobiol. 86:216-244.
28. O'Doherty J, Kringelbach ML, Rolls ET, Hornak J, Andrews C (2001): Abstract
C

reward and punishment representations in the human orbitofrontal cortex. Nat

Neurosci. 4:95-102.
AC

29. Cavanna AE, Trimble MR (2006): The precuneus: a review of its functional
anatomy and behavioural correlates. Brain. 129:564-583.
30. Pandya DN, Seltzer B, Petrides M, Cipolloni PB (2015): Cerebral Cortex:
Architecture, Connections, and the Dual Origin Concept. New York: Oxford
University Press.
31. Saleem KS, Kondo H, Price JL (2008): Complementary circuits connecting the
orbital and medial prefrontal networks with the temporal, insular, and opercular
cortex in the macaque monkey. J Comp Neurol. 506:659-693.
32. Rolls ET (2017): The orbitofrontal cortex and emotion in health and disease,
including depression.
17
 ACCEPTED MANUSCRIPT
Neuropsychologia.doi:10.1016/j.neuropsychologia.2017.1009.1021.
33. Eshel N, Roiser JP (2010): Reward and punishment processing in depression.
Biol Psychiatry. 68:118-124.
34. Nusslock R, Young CB, Damme KS (2014): Elevated reward-related neural
activation as a unique biological marker of bipolar disorder: assessment and treatment
implications. Behav Res Ther. 62:74-87.
35. Price JL, Drevets WC (2012): Neural circuits underlying the pathophysiology of
mood disorders. Trends Cogn Sci. 16:61-71.

PT
36. Xia M, Wang J, He Y (2013): BrainNet Viewer: a network visualization tool for
human brain connectomics. PLoS One. 8:e68910.

RI
U SC
AN
M
D
TE
C EP
AC

18
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

Occipital

PT
RI
Supra
marginal

SC
gyrus

U
Hippocampus

AN
Para Postcentral
Temporal
gyrus and
M
hippocampal gyrus
D
gyrus
pole
TE
CEP
AC

medial
OFC lateral
OFC
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

Effective Connectivity in Depression

Supplementary Information

Supplementary Methods

PT
Participants
There were 336 patients with a diagnosis of major depression, and 350 controls. The patients

RI
were from Xinan (First Affiliated Hospital of Chongqing Medical School in Chongqing, China),

SC
and Taiwan (Veteran General Hospital, Taipei). All participants were diagnosed according to
the Diagnostic and Statistical Manual of Mental Disorder-IV criteria for major depressive
disorder. Depression severity and symptomatology were evaluated by the Hamilton Depression

U
Rating Scale (HAMD, 17 items) (1) and the Beck Depression Inventory (BDI) (2). Table S1
AN
provides a summary of the demographic information and the psychiatric diagnosis (showing
how they were diagnosed) of the participants. The data collection was approved by the local
M

ethical review committees, was in accordance with the Code of Ethics of the World Medical
Association (Declaration of Helsinki), and informed consent was obtained. This is a subset of
D

patients from our previous functional connectivity investigation (3), but the analysis used here
TE

is completely different and novel in its application to depression, for it assesses effective or
directed connectivity, and goes beyond correlations. With respect to age and sex, Table S1
EP

shows that there were no significant differences in the age and sex of the depressed groups and
the controls. Further, the effects of age and sex were regressed out in all analyses and the
C

comparison of results between male and female, younger and older participants are shown in
AC

Fig. S5 and Fig. S6. 125 of the patients were not receiving medication at the time of the
neuroimaging. Further details follow.

Xinan: Patients with MDD were recruited from the outpatient department of the First Affiliated
Hospital of Chongqing Medical School in Chongqing, China. All were diagnosed according to
the Structured Clinical Interview for DSM-IV, by independent assessments of two psychiatrists.
They were also assessed for disease severity using the Hamilton Depression Rating Scale

1
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

(HAMD) (1) and Beck Depression Inventory (BDI), illness duration and the medication status
of the patients. Before the investigation, we excluded individuals who were not suitable for
MRI scanning by interview and by the self-reported checklist. The MRI related exclusion
criteria include claustrophobia, metallic implants, Menieres Syndrome and a history of
fainting within the previous half year. Exclusion criteria for both groups were as follows:

PT
current psychiatric disorders (except for MDD) and neurological disorders; substance abuse;
and stroke or serious encephalopathy. Of note, all of the subjects in the control group did not

RI
meet DSM-IV criteria for any psychiatric disorders and did not use any drugs that could affect
brain function. This study was approved by the Research Ethics Committee of the Brain

SC
Imaging Center of Southwest University and First Affiliated Hospital of Chongqing Medical
School. Informed written consent was obtained from each subject. This study was conducted

U
in accordance with the Helsinki Declaration as revised in 1989.
AN
Taiwan: Patients were recruited from the Veteran General Hospital in Taipei, Taiwan. All
M

participants were diagnosed according to the Diagnostic and Statistical Manual of Mental
Disorder-IV criteria for depression, and each participant's history of medical disease,
D

psychiatric illness, and medication use was evaluated by interview and medical charts carefully.
TE

Experiments were conducted in accordance with the Declaration of Helsinki and approved by
the Institutional Review Board of Taipei Veterans General Hospital. Written informed consent
EP

was obtained from all participants after ensuring adequate understanding of the study. Any
participants with the following conditions were excluded: 1) a comorbid substance-related
disorder, 2) presence of neurobiological disorders, such as dementia, head injury, stroke, or
C

Parkinsons disease; 3) presence of hypertension, diabetes, hyperlipidemia or coronary heart

AC

disease; 4) severe medical illness, such as malignancy, heart failure, or renal failure; 4)
presence of ferromagnetic foreign bodies or implants that were anywhere in the body.
Depression severity was evaluated by the psychiatrist-assessed Hamilton Depression Rating
Scale (HAMD, 17 items) (1).

2
Rolls et al. Supplement
ACCEPTED MANUSCRIPT
Table S1. A summary of the demographic information and the psychiatric diagnosis in the
present study.
Sex Education Medication Duration
Sites Group Age (years) HAMD BDI Mean FD
(male/female) (years) (yes / no) of illness
Healthy 49.188.58 60 / 36 15.04 3.83 / / / / 0.133 0.054

Patient 52.6414.86 33 / 21 12.66 3.95 54 / 0 9.34 6.99 / 8.63 6.92 0.116 0.056
Taiwan
Statistic
-1.810 / 0.072 0.028 / 0.866 3.60 / 4.3e-4 / / / 1.833 / 0.0687
(t / p)

PT
Healthy 39.65 15.80 166 / 88 13.01 3.89 / / / / 0.133 0.063

Patient 38.74 13.65 183 / 99 11.91 3.58 157 / 125 20.8 5.87 20.42 9.33 4.16 5.51 0.125 0.054
Xinan
Statistic

RI
0.719 / 0.472 0.013 / 0.911 3.41 / 6.9e-4 / / / / 1.729 / 0.084
(t / p)

SC
Image Acquisition and Preprocessing

U
Data for resting state connectivity analysis were collected in 3T MRI scanners in an 8 min
AN
period in which the participants were awake in the scanner not performing a task using standard
protocols described here.
M

Data preprocessing was performed using DPARSF (4) (http:// restfmri.net) which is a
toolbox based on the SPM8 software package. The first 10 EPI scans were discarded to
D

suppress equilibration effects. The remaining scans of each subject underwent slice timing
TE

correction by sinc interpolating volume slices, motion correction for volume to volume
displacement, spatial normalization to standard Montreal Neurological Institute (MNI) space
using affine transformation and nonlinear deformation with a voxel size of 3 3 33,
EP

followed by spatial smoothing (8 mm Full Width Half Maximum). To remove the sources of
spurious correlations present in resting-state BOLD data, all fMRI time-series underwent band-
C

pass temporal filtering (0.01-0.1 Hz), nuisance signal removal from the ventricles, and deep
AC

white matter, and regressing out any effects of head motion using the Friston et al 24 head
motion parameters procedure (5). We note that with a TR of 2 s the effective connectivity
method may benefit from band-pass temporal filtering in the range 0.01-1.0 Hz, to increase the
use of the high frequency fluctuations in the times series, but with the filtering applied did
obtain sufficient information to obtain asymmetric weights in the links between connected
brain areas to be detected. This filtering used though did preserve sufficient of the high-
frequency fluctuations that can be successfully captured by the model.

3
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

Finally, we implemented additional careful volume censoring (scrubbing) movement

correction as reported by Power et al. (6) to ensure that head-motion artifacts are not driving
observed effects. The mean framewise displacement (FD) was computed with FD threshold for
displacement being 0.5. In addition to the frame corresponding to the displaced time point, 1
preceding and 2 succeeding time points were also deleted to reduce the spill-over effect of

PT
head movements. Subjects with >10% displaced frames flagged were completely excluded
from the analysis as it is likely that such high-level of movement would have had an influence

RI
on several volumes. Global signals were not regressed out (3).
Any effects of gender ratio, years of education, head motion, and age between the patient

SC
and control groups were regressed out in the analysis. In fact, there were no differences in the
gender ratios, though the number of years of education was lower in the patients than controls.

U
We note that the Taiwanese sample included patients with depression in remission while under
AN
antidepressant treatment, and thus their scores on the Hamilton Depression Rating Scale
(HAMD) assessment were in the low range.
M

Xinan: All images were acquired on a 3.0-T Siemens Trio MRI scanner using a 16-channel
D

whole-brain coil (Siemens Medical, Erlangen, Germany). High- resolution T1-weighted 3D

TE

images were acquired using a magnetization-prepared rapid gradient echo (MPRAGE)

sequence (echo time (TE) = 2.52 ms; repetition time (TR) = 1900 ms; inversion time (TI) =
EP

900 ms; flip angle = 9 degrees; slices = 176; thickness = 1.0 mm; resolution matrix = 256256;
voxel size = 111 mm3). For each participant, 242 functional images were acquired with a
gradient echo type Echo Planar Imaging (EPI) sequence (echo time (TE) = 30 ms; repetition
C

time (TR) = 2000 ms; flip angle = 90 degrees; slices = 32; slice thickness = 3.0 mm; slice gap
AC

= 1 mm; resolution matrix = 6464; voxel size 3.43.43mm3). During image acquisition,
participants were instructed to keep their eyes closed while keeping their head as still as
possible without falling asleep. All participants stayed awake during the MRI imaging as
confirmed by the participants after the session.

Taiwan: fMRI scanning was performed at National Yang-Ming University in Taiwan using a
3.0-T Siemens MRI Scanner (Siemens Magnetom Tim Trio, Erlangen, Germany) with a 12-

4
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

channel head coil. During the experiments, the participants were instructed to relax with their
eyes closed, without falling asleep. After the resting state experiment, participants were asked
whether they fell asleep during the resting state scan session, and participants were rescanned
if they had fallen asleep during the resting state scan. T2*-weighted images with BOLD
contrast were measured using a gradient echo- planar imaging (EPI) sequence (repetition time,

PT
TR: 2,500 ms, echo time, TE: 27 ms, field of view, FoV: 220 mm, flip angle: 77 degree, matrix
size: 64 x 64, and voxel size: 3.443.443.40 mm). For each run, 200 EPI volume images were

RI
acquired in the anterior and posterior commissure (ACPC) plane. High-resolution structural
T1 images were acquired with three-dimensional (3D) magnetization-prepared rapid gradient-

SC
echo sequence (3D-MPRAGE; TR: 2,530 ms, TE: 3.5 ms, TI: 1,100 ms, FoV: 256 mm, and
flip angle: 7 degree, 192 sagittal slices, voxel size = 1.0 mm x 1.0 mm 1.0 mm, no gap). For

U
each participant, the whole fMRI scanning lasted about 16 min (T1: 8min, Resting: 8min).
AN
Effective connectivity measurement
M

Introduction
A classical approach to measuring effective connectivity is dynamic causal modelling (DCM)
D

(7-9). DCM is often used with circuits consisting of a-priori selected brain regions to test
TE

hypotheses on the interactions between the considered regions. Here we instead use a network
model with simpler assumptions than those typically used in DCM to perform a whole-brain
EP

connectivity analysis (10). This allows for the very efficient calculation of maximum-
likelihood EC estimates for a large number (94) of nodes, individually for a large cohort
of >500 subjects. In this way we target significant EC differences for all existing connections
C

(as determined by DTI) that characterize MDD with FDR correction and without preliminary
AC

knowledge, expecting a distributed pattern of abnormal EC links across the brain. Our
estimation procedure (10) iteratively optimizes a network model such that it reproduces the
empirical cross-covariances between ROIs, which are canonically related to the cross spectral
density used in recent studies that apply DCM to resting state fMRI data (11, 12). The proposed
model uses an exponential approximation of BOLD autocovariance (locally over a few TRs)
and discards very slow-frequency fluctuations. Moreover, by focusing on BOLD fluctuations
in the frequency range (0.1-1 Hz), we were able to dispense with a model of haemodynamic

5
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

mapping neuronal activity to fMRI signals, as the corresponding time constants are faster (13).
Finally, we place positivity constraints on extrinsic or between node connections - in line with
known neuroanatomy and previous modeling studies (14). A last simplification compared to
DCM includes a fixed (but plausible) form of endogenous neuronal fluctuations ( in our model)
that were characterized by a single (variance) parameter in each region or node. In spite of

PT
these differences, we still borrow the term effective connectivity from the DCM literature as
our connectivity estimates relate to directional interactions between ROIs in the brain network.

RI
This model-based approach has been successfully applied to identify changes in the cortical
coordination between rest and movie viewing (15).

SC
Compared to DCM the new method used here (10) is more powerful because it limits the
degrees of freedom for each brain region by utilizing a simpler model of each brain region, and

U
because it uses some structural connectivity information from for example diffusion tensor
AN
imaging. Further, the new effective connectivity method focuses on transitions between fMRI
activity states across successive time points (16) and does not include details about
M

hemodynamics like the Balloon model (17). The estimated effective connectivity measures the
strengths of causal interactions from one brain area to another, via the proxy of BOLD
D

fluctuations: it provides a single number that lumps together the effects of the strength of the
TE

synapse, and neurotransmitter release, etc. The synaptic conductivity interpretation also relates
to our earlier neuron-level models in which the synaptic conductivity between modules is a key
EP

parameter that specifies how much one module influences another module (18). The new
method has the additional advantage that each brain region or module has its own parameter
which specifies the variance of the modules activity, which may be related to the intrinsic
C

excitability of the region. In relation to our integrate-and-fire models, the parameter w+ that
AC

defines the strength of the recurrent collateral synapses within the attractor network (18) may
relate to the parameter in the current effective connectivity approach (10), because the local
feedback influenced by w+ influences the fluctuations of the activity, for example how readily
an area will transition to a high firing rate state.
Within a cortical hierarchy of connectivity (for example from primary visual cortex V1 to
the inferior temporal cortex (19)), the forward connections between any pair of cortical areas
up through the hierarchy are thought to be stronger than the backprojections based on a wealth

6
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

of evidence (20), and there are useful asymmetries in the terminations of the forward and
backward projections that facilitate this (20, 21). This ensures that sensory input dominates the
processing, rather than imagination. In the present investigation, we follow this lead and refer
to the direction in which the effective connectivity is the stronger as the forward direction. The
concept of forward relates to function as much as to anatomy. Further, the findings in this paper

PT
strengthen the evidence for the use of the term forward connections, in that the measured
ECs described here are what would be expected from our understanding of processing in

RI
cortical hierarchies (20).

SC
Overview
The approach used to calculate effective connectivity (EC) follows that described by Gilson et

U
al (10). Effective connectivity measures the individual efficacy of each existing connection
AN
between two brain regions, that is, how much one brain region influences another. Our
approach provides a signature for each subject in the high-dimensional space of EC connections
M

(>3000), which is then used to investigate differences between MDD patients and healthy
controls. The estimated effective connectivity values reflect the combined effects of synaptic
D

efficacies between the regions, the types and concentrations of neurotransmitters in the target
TE

regions, etc.
The dynamics for each brain region are described by a multivariate Ornstein-Uhlenbeck
EP

process: each region receives fluctuating inputs (white noise) that propagates via the effective
connectivity to other nodes, which shapes the correlation pattern at the global level, that is, the
functional connectivity (FC). Here the focus is on transitions of fMRI measurements across
C

successive TRs, which have been shown to convey information about conditions such as
AC

waking versus sleep (16). The effective connectivity model captures this information via the
covariances with non-zero time shifts (spatiotemporal FC) and the resulting EC contains
information about directed connectivity. Both EC and the local input variance are optimized
such that the model best reproduces statistics of observed fMRI signals measured by the
empirical spatiotemporal FC, which are canonically related to the cross spectra used to tune a
resting-state DCM (12).

7
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

Details about the optimization are provided by Gilson et al (10) for resting-state fMRI data
and are summarized next. The skeleton for the EC is provided by structural data obtained using
diffusion tensor imaging (DTI), from which we infer the existence of connections. This usefully
reduces the number of parameters to estimate and enhances the estimation procedure at the
level of individual subjects: from all possible 942 = 8836 connections, we specify that many

PT
are not present anatomically as direct projections, so in the model need to optimize only 39%
of the possible connections. The DTI connectivity matrix was set to just 0 (no connection) or

RI
1 (for a connection) between the automated anatomical atlas AAL2 94 regions (22), based on
the DTI atlas used by Gilson et al (2016). Because DTI may miss inter-hemispheric connections

SC
between homotopic regions on the two hemispheres (23, 24), we set these as being present, and
allowed the algorithm to tune the strengths of these just as for the other effective connectivities.

U
The automated anatomical atlas version 2 (AAL2) (22) was used to parcellate the brain into 94
AN
regions, because this number of regions provides a suitable number of functional connectivity
links without too many degrees of freedom; because its parcellation of the orbitofrontal cortex
M

region which is of special interest in relation to depression (25) has been remade to relate to
useful divisions and descriptions; and because it has been found to be useful in related
D

investigations (3). Limiting the number of parameters to estimate in the whole-brain dynamic
TE

model is crucial to obtain robust individualized EC estimates. On the other hand, AAL2
corresponds to about 3000 EC parameters (for 39% density), which is a sufficiently rich space
EP

to extract complex patterns to differentiate between patients and controls. Our choice aimed to
solve this trade-off.
C

Empirical covariances:
AC

For the resting-state session of each individual, we denote the centered (zero-mean) time series
by for region 1 with time 1 ; the duration is T=180. The zero-lag and
1-lag covariances are calculated as follows:

^ij0 = 1 11 and
^ij1 = 1 11 +1 . (1)
2 2

For each subject, we evaluate the time constant associated with the exponential decay of
^ii averaged over all regions:
the autocovariance function

8
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

= ^ 0 ^ii1
. (2)
logii log

Dynamic cortical model:

The model comprised = 94 interconnected cortical regions. The activity of each region
is governed by an Ornstein-Uhlenbeck process and evolves depending on the activity of other

PT
dx
populations: = + ij + dB . Here, the time constant corresponds to an
dt

exponential decay and is calibrated from the empirical data (see Eq. 2); dB is white Gaussian

RI
noise with covariance matrix , where the input variances sit on the diagonal and are zero

SC
elsewhere. These input fluctuations propagate via the effective connectivity embodied by the
matrix (its skeleton is determined by DTI). All variables have zero mean and the

theoretical spatiotemporal covariances are defined by ij = + , where the angular

U
AN
brackets denote averaging over randomness of the inputs; we use two time shifts: = 0 and
= 1 TR.
The mathematical mappings between matrices , 0 and 1 are given by Lyapunov
M

equation JQ0 + 0 + = 0 and 1 = 0 expm( ), where the Jacobian of the dynamical

D

ij
system ij = + ij depends on the mean activity of the network (ij is the Kronecker

TE

delta); the superscript T denotes the matrix transpose; expm denotes the matrix exponential.
These two consistency equations allow for the quick estimation of the predicted FC matrices,
EP

without simulating the network.

C

Parameter estimation procedure:

^0
AC

We tune the model such that its covariance matrices 0 and 1 reproduce the empirical
^1 . We summarize the essential steps of the procedure described in Gilson et al. (2016)
and
that iteratively optimizes the network parameters and . At each step, the Jacobian is
calculated from the current value of . Then, the model FC matrices 0 and are
calculated from the consistency equations, using the Bartels-Stewart algorithm to solve the
Lyapunov equation. The desired Jacobian update is the matrix = ( 0 )1 [ 0 +
1 expm(-J )], which aims to reduce the FC error between the empirical and model FC, as

9
Rolls et al. Supplement
ACCEPTED MANUSCRIPT
^0 0 and 1 =
determined by the two difference matrices 0 = ^1 1 . Finally, the

connectivity update is ij = ij for existing connections. We impose non-negativity of

the EC values during the optimization. The input variances are tuned according to ii =
(ii0 + ii0 ). We use = 0.0001 and = 0.1.

PT
Normalisation of model estimates
Normalisation of the effective connectivity within each individual was performed by

RI
performing the z-score over the matrix elements for each EC matrix within each participant:
(ECij - mean(ECij)) / std(ECij) for all EC links (performed over existing links corresponding to

SC
1s in the structural connectivity matrix). The aim of this was to enable each participants data
to contribute similarly to the statistics calculated across participants. We note that small

U
effective connectivities will appear in the tables in this paper as negative, but this is only due
AN
to the removal of the mean value. All effective connectivity links computed by the algorithm
are in fact positive. Any difference between patients and controls that is describe as negative
M

in fact refers only to a decrease of effective connectivity. A similar normalization within each
individual was used for the values. An increase of a value can be interpreted as an increase
D

of the variance in an AAL2 region. These normalisations were used for the statistical
TE

calculations. Table 1 shows the mean of the EC values not normalized (because a negative EC
would have no meaning). Table 2 shows the mean of the values normalized within each
EP

participant because this better reflects the statistical values.

C

Statistical analysis on effective connectivity

AC

Two-tailed, two-sample t-tests were performed on the normalized effective connectivity to

identify significantly altered effective connectivity links in patients compared to controls
within each imaging centre that provided resting-state fMRI data. The effects of age, gender
ratios, head motion and education were regressed within each dataset in this step by general
linear models (26, 27). After obtaining the t-test results for each centre, the Liptak-Stouffer z
score method (28), which has been described in detail in our previous studies (3, 29, 30), was
then used to combine the results from the individual datasets. Further, given that a reasonable

10
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

effect size for effective connectivity is in the range .05 to 1 Hz for EC strengths not corrected
by the time constant (31), we consider here only EC differences greater than a threshold value
of 0.01 (calculated by taking 0.05 and dividing it by the approximate time constant of
approximately 4 s).

PT
Robustness of the effective connectivity analyses
The effective connectivity algorithm (10) used here produced robust results, in that with this

RI
size of dataset, many links were found with different effective connectivity values in depressed
patients vs controls that were significant fully corrected (FDR) across all AAL2 areas. The

SC
results were also robust in that when the data set was split into two half splits, the two splits
had a correlation for the values of the 3,490 effective connectivity links of r = 0.26 with each

U
other (p<0.0001) (Fig. S3). Of the 50 effective connectivity links that were significantly
AN
different between depressed patients and controls in the full dataset, 6 of the same links were
significant in one half split, and 13 in the other half split. The implication is that the results are
M

reasonably reliable when a comparison involving differences between 336 patients and 350
controls.
D
TE

Validation of the effective connectivity analyses

The effective connectivity approach used here was validated in the sense that many of the key
EP

effective connectivity links in the human brain were stronger in the forward direction than the
backward direction where this might be expected on the basis of anatomical findings in non-
human primates (20, 32, 33). For example, the orbitofrontal cortex receives important forward
C

visual inputs from the temporal lobe visual cortical areas (20, 34), and the effective
AC

connectivity was greater in this direction. The forward connectivity is indicated anatomically
by connections from superficial pyramidal cells that terminate primarily in layers and 2-3; and
backprojections are from deep pyramidal cells that terminate primarily in layer 1 on the apical
dendrites of the pyramidal cells, far from the cell bodies, where they have relatively weak
effects (20, 21). (Indeed, for the operation of coupled excitatory and inhibitory neuronal pools
computational simulations show that modulatory effects that implement attentional biases need
to be approximately 0.4 of the strength of the driving connections that convey sensory input

11
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

connections (20, 35-37).) The actual values for the ratio of the backprojection to the forward
connection effective connectivities shown in Table 1, consistent with this theory, have a mean
value of 0.467 for the orbitofrontal cortex backprojections to the temporal lobe cortical areas.
(The ratio of these forward to their backprojections in Table 1 is 2.14.) The fact that this
baseline value for the effective connectivity in the forward vs backward directions is evident

PT
in resting-state fMRI is of considerable interest.

RI
U SC
AN
M
D
TE
C EP
AC

12
Rolls et al. Supplement
ACCEPTED MANUSCRIPT
Table S2. The anatomical regions defined in each hemisphere and their label in the automated
anatomical labelling atlas AAL2 (22). Column 4 provides a set of possible abbreviations for
the anatomical descriptions.
POSSIBLE
NO. ANATOMICAL DESCRIPTION LABEL aal2.nii.gz)
ABBREVIATION
1, 2 Precentral gyrus Precentral PreCG
3, 4 Superior frontal gyrus, dorsolateral Frontal_Sup SFG
5, 6 Middle frontal gyrus Frontal_Mid MFG
7, 8 Inferior frontal gyrus, opercular part Frontal_Inf_Oper IFGoperc

PT
9, 10 Inferior frontal gyrus, triangular part Frontal_Inf_Tri IFGtriang
11, 12 IFG pars orbitalis, Frontal_Inf_Orb IFGorb
13, 14 Rolandic operculum Rolandic_Oper ROL
15, 16 Supplementary motor area Supp_Motor_Area SMA

RI
17, 18 Olfactory cortex Olfactory OLF
19, 20 Superior frontal gyrus, medial Frontal_Sup_Med SFGmedial
21, 22 Superior frontal gyrus, medial orbital Frontal_Med_Orb PFCventmed

SC
23, 24 Gyrus rectus Rectus REC
25, 26 Medial orbital gyrus OFCmed OFCmed
27, 28 Anterior orbital gyrus OFCant OFCant
29, 30 Posterior orbital gyrus OFCpost OFCpost

U
31, 32 Lateral orbital gyrus OFClat OFClat
33, 34 Insula Insula INS
AN
35, 36 Anterior cingulate & paracingulate gyri Cingulate_Ant ACC
37, 38 Middle cingulate & paracingulate gyri Cingulate_Mid MCC
39, 40 Posterior cingulate gyrus Cingulate_Post PCC
41, 42 Hippocampus Hippocampus HIP
M

43, 44 Parahippocampal gyrus ParaHippocampal PHG

45, 46 Amygdala Amygdala AMYG
47, 48 Calcarine fissure and surrounding cortex Calcarine CAL
49, 50 Cuneus Cuneus CUN
D

51, 52 Lingual gyrus Lingual LING

53, 54 Superior occipital gyrus Occipital_Sup SOG
TE

55, 56 Middle occipital gyrus Occipital_Mid MOG

57, 58 Inferior occipital gyrus Occipital_Inf IOG
59, 60 Fusiform gyrus Fusiform FFG
61, 62 Postcentral gyrus Postcentral PoCG
EP

63, 64 Superior parietal gyrus Parietal_Sup SPG

Inferior parietal gyrus, excluding
65, 66 Parietal_Inf IPG
supramarginal and angular gyri
67, 68 SupraMarginal gyrus SupraMarginal SMG
C

69, 70 Angular gyrus Angular ANG

71, 72 Precuneus Precuneus PCUN
AC

73, 74 Paracentral lobule Paracentral_Lobule PCL

75, 76 Caudate nucleus Caudate CAU
77, 78 Lenticular nucleus, Putamen Putamen PUT
79, 80 Lenticular nucleus, Pallidum Pallidum PAL
81, 82 Thalamus Thalamus THA
83, 84 Heschls gyrus Heschl HES
85, 86 Superior temporal gyrus Temporal_Sup STG
87, 88 Temporal pole: superior temporal gyrus Temporal_Pole_Sup TPOsup
89, 90 Middle temporal gyrus Temporal_Mid MTG
91, 92 Temporal pole: middle temporal gyrus Temporal_Pole_Mid TPOmid
93, 94 Inferior temporal gyrus Temporal_Inf ITG

13
Rolls et al. Supplement
ACCEPTED MANUSCRIPT
Table S3. Correlations between the effective connectivity links and the depression severity
scores, assessed by the duration of the illness. The r value shows the correlation between the
effective connectivity and the illness duration calculated just in patients. The z value show the
difference in effective connectivity for each link for the comparison depression group control
group, and its p value follows in column 7. Thus for example for the OFCant to temporal pole
effective connectivity link, this shows a greater decrease with an increase of illness duration.
In addition, correlations of some of the functional connectivities with the Beck Depression
Inventory (BDI), the Hamilton Depression rating scale (HAM-D), and the Hamilton Anxiety

PT
rating scale (HAM-A) are shown.
Region 1 Region 2 clinical r value p value of r z value p value of z

RI
variables

Precentral_R Precentral_L Illness duration 0.108 0.0491 3.320 9.01E-04

SC
OFCant_L Temporal_Inf_L Illness duration -0.135 0.0123 -3.334 8.55E-04

Temporal_Pole_Mid_L Temporal_Pole_Mid_R Illness duration 0.121 0.0266 3.300 9.67E-04

U
Frontal_Mid_2_L Frontal_Sup_Medial_R Illness duration -0.134 0.0144 -3.700 2.15E-04

Olfactory_L OFCmed_L HAMA -0.139 0.0111 -3.410 6.49E-04

AN
Olfactory_R OFCmed_R HAMA -0.112 0.0412 -3.255 1.14E-03

OFCpost_L OFClat_L HAMD 0.120 0.0287 3.466 5.29E-04

M

Temporal_Pole_Mid_L Temporal_Pole_Mid_R HAMD 0.111 0.0425 3.300 9.67E-04

OFCmed_R Temporal_Pole_Mid_R HAMD -0.125 0.0226 -4.014 5.98E-05

D

Amygdala_L OFCpost_L BDI -0.138 0.0353 -3.914 9.09E-05

TE

Cingulate_Post_R Temporal_Pole_Mid_R BDI 0.130 0.0473 3.611 3.05E-04

ParaHippocampal_R Occipital_Mid_R BDI -0.158 0.0153 -3.328 8.74E-04

C EP
AC

14
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M

Figure S1. The matrices of effective connectivity of healthy controls. The axes are the
D

AAL2 areas, shown numbered and with their names in Table S1. The effective connectivity
matrix has the index j for the columns and the index i for the rows. The matrix is thus non-
TE

symmetric, and the effective connectivity is always from j to i. The effective connectivity
between any pair of links is shown in one direction in the upper right of the matrix, and in the
opposite direction in the lower left. The table shows for example that AAL2 areas 3-12 which
EP

include the inferior frontal gyrus reading along the horizontal axis tend to have strong forward
effective connectivity to angular and supramarginal gyri areas 67-70 reading on the vertical
axis.
C
AC

15
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M

Figure S2. Comparison of the difference between patients with major depressive disorder
and controls in the strength of the effective connectivity in the forwards vs the backwards
D

direction. The scattergram shows for every link the z value for the difference between patients
and controls in the forward direction on the abscissa and the backward direction on the ordinate.
TE

For this diagram, forwards indicates only i to j in the effective connectivity matrix shown in
Fig. S1, and backwards indicates j to i. The linear regression line and correlation value show
that if a link is weaker in one direction in patients, it is likely to be weaker in the opposite
EP

direction; and that if a link is stronger in patients in one direction, it is likely to be stronger in
the other direction too.
C
AC

16
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D

Fig. S3. Comparison of effective connectivity differences between patients and controls
when the dataset was split into two halves. The two splits had a correlation for the values
TE

of the 3,490 effective connectivity links of r = 0.70 with each other (p = 1.1 x 10-251).
EP

Effects of medication
Although it was not a primary aim of this investigation, and following a suggestion, the effects
C

of medication were assessed by comparing the functional connectivity in 125 patients not
AC

receiving medication, and 157 patients receiving medication. The medication consisted in most
cases of selective serotonin reuptake inhibitors (SSRIs) including fluoxetine, paroxetine,
sertraline, citalopram and escitalopram; or serotonin-norepinephrine reuptake inhibitors
(SNRIs) such as venflaxine, or a tetracyclic antidepressant such as mirtazepine. The overall
pattern of functional connectivity differences between patients and controls is similar for the
unmedicated and the medicated subgroups of patients (Fig. S4), providing evidence that the
main differences between patients and controls shown in Fig. 1 were found in depressed

17
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

patients whether or not they were receiving medication. This was shown by a correlation of
0.68 (p<10-10) between the functional connectivities across the whole brain in medicated and
unmedicated patients. Further, the changes shown in Table 1 for the differences between all
the patients and controls were similar, with similar signs, in both the medicated and the
unmedicated patients vs the controls.

PT
RI
U SC
AN
M
D
TE
C EP

Fig. S4. There is a correlation between the strengths of the functional connectivity links
AC

in the non-medicated patients and of the links in the medicated population, showing that
a similar pattern of differences from controls is found in medicated and un-medicated
patients. The bold points are for links that were significantly different between all patients and
controls. The faint points show the links between all other AAL2 areas. The r value is for all
links. The red points indicate links included in Table 1.

18
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D

Fig. S5. Comparison of effective connectivity differences between patients and controls
when the dataset was split into two groups, male and female. The red points indicate links
TE

included in Table 1.
C EP
AC

19
Rolls et al. Supplement
ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D

Fig. S6. Comparison of effective connectivity differences between patients and controls
when the dataset was split into two groups, younger and older. The red points indicate links
TE

included in Table 1.
C EP
AC

20
Rolls et al. Supplement
ACCEPTED MANUSCRIPT
Supplemental References

1. Hamilton M (1960): A rating scale for depression. J Neurol Neurosurg Psychiatry. 23:56-
62.
2. Beck AT, Beamesderfer A (1974): Assessment of depression: the depression inventory.
Mod Probl Pharmacopsychiatry. 7:151-169.
3. Cheng W, Rolls ET, Qiu J, Liu W, Tang Y, Huang CC, et al. (2016): Medial reward and
lateral non-reward orbitofrontal cortex circuits change in opposite directions in depression.

PT
Brain. 139:3296-3309.
4. Chao-Gan Y, Yu-Feng Z (2010): DPARSF: a MATLAB toolbox for pipeline data
analysis of resting-state fMRI. Front Syst Neurosci. 4:13.

RI
5. Friston KJ, Williams S, Howard R, Frackowiak RS, Turner R (1996): Movement-related
effects in fMRI time-series. Magn Reson Med. 35:346-355.
6. Power JD, Mitra A, Laumann TO, Snyder AZ, Schlaggar BL, Petersen SE (2014): Methods

SC
to detect, characterize, and remove motion artifact in resting state fMRI. Neuroimage. 84:320-
341.
7. Valdes-Sosa PA, Roebroeck A, Daunizeau J, Friston K (2011): Effective connectivity:

U
influence, causality and biophysical modeling. Neuroimage. 58:339-361.
AN
8. Bajaj S, Adhikari BM, Friston KJ, Dhamala M (2016): Bridging the Gap: Dynamic Causal
Modeling and Granger Causality Analysis of Resting State Functional Magnetic Resonance
Imaging. Brain Connect.
9. Friston K (2009): Causal modelling and brain connectivity in functional magnetic
M

resonance imaging. PLoS Biol. 7:e33.

10. Gilson M, Moreno-Bote R, Ponce-Alvarez A, Ritter P, Deco G (2016): Estimation of
D

directed effective connectivity from fMRI functional connectivity hints at asymmetries in the
cortical connectome. PLoS Comput Biol. 12:e1004762.
TE

11. Frassle S, Lomakina EI, Razi A, Friston KJ, Buhmann JM, Stephan KE (2017): Regression
DCM for fMRI. Neuroimage. 155:406-421.
12. Friston KJ, Kahan J, Biswal B, Razi A (2014): A DCM for resting state fMRI. Neuroimage.
EP

94:396-407.
13. Friston KJ (2002): Bayesian estimation of dynamical systems: an application to fMRI.
Neuroimage. 16:513-530.
C

14. Marreiros AC, Kiebel SJ, Friston KJ (2008): Dynamic causal modelling for fMRI: a two-
state model. Neuroimage. 39:269-278.
AC

15. Gilson M, Deco G, Friston K, Hagmann P, Mantini D, Betti V, et al. (2017): Effective
connectivity inferred from fMRI transition dynamics during movie viewing points to a
balanced reconfiguration of cortical interactions. bioRxiv.110015.
16. Mitra A, Snyder AZ, Tagliazucchi E, Laufs H, Raichle ME (2015): Propagated infra-slow
intrinsic brain activity reorganizes across wake and slow wave sleep. Elife. 4:e10781.
17. Friston KJ, Mechelli A, Turner R, Price CJ (2000): Nonlinear responses in fMRI: the
Balloon model, Volterra kernels, and other hemodynamics. Neuroimage. 12:466-477.
18. Rolls ET, Webb TJ, Deco G (2012): Communication before coherence. Eur J Neurosci.
36:2689-2709.
19. Rolls ET (2012): Invariant visual object and face recognition: neural and computational

21
Rolls et al. Supplement
ACCEPTED MANUSCRIPT
bases, and a model, VisNet. Front Comput Neurosci. 6, 35:1-70.
20. Rolls ET (2016): Cerebral Cortex: Principles of Operation. Oxford: Oxford University
Press.
21. Pandya DN, Seltzer B, Petrides M, Cipolloni PB (2015): Cerebral Cortex: Architecture,
Connections, and the Dual Origin Concept. New York: Oxford University Press.
22. Rolls ET, Joliot M, Tzourio-Mazoyer N (2015): Implementation of a new parcellation of
the orbitofrontal cortex in the automated anatomical labeling atlas. Neuroimage. 122:1-5.
23. Messe A, Rudrauf D, Benali H, Marrelec G (2014): Relating structure and function in the

PT
human brain: relative contributions of anatomy, stationary dynamics, and non-stationarities.
PLoS Comput Biol. 10:e1003530.
24. Hagmann P, Cammoun L, Gigandet X, Meuli R, Honey CJ, Wedeen VJ, et al. (2008):

RI
Mapping the structural core of human cerebral cortex. PLoS Biol. 6:e159.
25. Rolls ET (2016): A non-reward attractor theory of depression. Neurosci Biobehav Rev.
68:47-58.

SC
26. Barnes J, Ridgway GR, Bartlett J, Henley SM, Lehmann M, Hobbs N, et al. (2010): Head
size, age and gender adjustment in MRI studies: a necessary nuisance? Neuroimage. 53:1244-
1255.

U
27. Di Martino A, Yan CG, Li Q, Denio E, Castellanos FX, Alaerts K, et al. (2014): The autism
AN
brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture
in autism. Mol Psychiatry. 19:659-667.
28. Liptak T (1958): On the combination of independent tests. Magyar Tud Akad Mat Kutato
Int Kozl. 3:171-197.
M

29. Cheng W, Rolls ET, Gu H, Zhang J, Feng J (2015): Autism: reduced functional
connectivity between cortical areas involved in face expression, theory of mind, and the sense
D

of self. Brain. 138:1382-1393.

30. Cheng W, Palaniyappan L, Li M, Kendrick KM, Zhang J, Luo Q, et al. (2015): Voxel-
TE

based, brain-wide association study of aberrant functional connectivity in schizophrenia

implicates thalamocortical circuitry. NPJ Schizophr. 1:15016.
31. Stephan KE, Penny WD, Moran RJ, den Ouden HE, Daunizeau J, Friston KJ (2010): Ten
EP

simple rules for dynamic causal modeling. Neuroimage. 49:3099-3109.

32. Markov NT, Ercsey-Ravasz M, Van Essen DC, Knoblauch K, Toroczkai Z, Kennedy H
(2013): Cortical high-density counterstream architectures. Science. 342:1238406.
C

33. Markov NT, Ercsey-Ravasz MM, Ribeiro Gomes AR, Lamy C, Magrou L, Vezoli J, et al.
(2014): A weighted and directed interareal connectivity matrix for macaque cerebral cortex.
AC

Cereb Cortex. 24:17-36.

34. Rolls ET (2014): Emotion and Decision-Making Explained. Oxford: Oxford University
Press.
35. Deco G, Rolls ET (2005): Neurodynamics of biased competition and co-operation for
attention: a model with spiking neurons. J Neurophysiol. 94:295-313.
36. Deco G, Rolls ET (2005): Attention, short-term memory, and action selection: a unifying
theory. Prog Neurobiol. 76:236-256.
37. Deco G, Rolls ET (2004): A neurodynamical cortical model of visual attention and
invariant object recognition. Vision Res. 44:621-644.

22