PreprintPDF Available

Data-driven brain-types and their cognitive consequences

Authors:

Abstract and Figures

The canonical approach to exploring brain-behaviour relationships is to group individuals according to a phenotype of interest, and then explore the neural correlates of this grouping. A limitation of this approach is that multiple aetiological pathways could result in a similar phenotype, so the role of any one brain mechanism may be substantially underestimated. Building on advances in network analysis, we used a data-driven community-clustering algorithm to identify robust subgroups based on white-matter microstructure in childhood and adolescence (total N=313, mean age: 11.24 years). The algorithm indicated the presence of two equal-size groups that show a critical difference in FA of the left and right cingulum. These different ‘brain types’ had profoundly different cognitive abilities with higher performance in the higher FA group. Further, a connectomics analysis indicated reduced structural connectivity in the low FA subgroup that was strongly related to reduced functional activation of the default mode network. Graphical abstract
Content may be subject to copyright.
Data-driven brain-types and their cognitive
consequences
Joe Bathelt1*, Amy Johnson1, Mengya Zhang1, the CALM team1 & Duncan E. Astle1
1 MRC Cognition & Brain Sciences Unit, University of Cambridge
* corresponding author
Address for correspondence:
Dr Joe Bathelt
MRC Cognition & Brain Sciences Unit
University of Cambridge
Cambridge CB2 7EF
joe.bathelt@mrc-cbu.cam.ac.uk
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Abstract
The canonical approach to exploring brain-behaviour relationships is to group
individuals according to a phenotype of interest, and then explore the neural
correlates of this grouping. A limitation of this approach is that multiple aetiological
pathways could result in the phenotype of interest, so the role of any one brain
mechanism may be substantially underestimated. We show that recent advances in
network analysis make it possible to group individuals at a neural level - to identify
subgroups of individuals with similarly organized brains. Across three independent
samples (total N = 313, mean age: 11.24 years, range: 5-21 years) we used a data-
driven community clustering algorithm to identify robust subgroups using white-
matter microstructure organization in childhood and adolescence. The algorithm
indicated the presence of two groups of roughly equal sizes. A critical organizational
difference between the groups was the role of the left and right cingulum. These
different ‘brain types’ had profoundly different cognitive abilities: Groups with
higher FA in the cingulum performed better across assessments of fluid
intelligence, vocabulary, verbal and visuospatial short-term and working memory,
and longer-term memory. We next explored the potential mechanistic role of the
cingulum. A connectomics analysis indicated reduced anterior-posterior structural
connectivity in the low cingulum FA subgroup. We then used resting-state
functional data from the same individuals and showed that cingulum FA was
strongly related to activation of the default mode network. In summary, inter-
individual differences in cingulum microstructural organisation allowed for
biologically-based grouping, which has a dramatic effect on cognition and the
functional activation of the default mode network. Using this new approach, we
propose that the cingulum plays a key role in the integration of cortical areas,
which is pivotal for cognitive ability in children and young people.
Keywords: adolescence, brain development, childhood, cingulum, cognitive
development, nosology, white matter
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Graphical abstract
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Introduction
Differential psychology is an influential strand of modern psychology, concerned
with identifying dimensions upon which individuals differ. This approach has been
applied at different points across development, from childhood to old adulthood, and
has been central to our understanding of typical and atypical behaviour and
psychopathology (Lubinski 2000; Cronbach 1957). Our understanding of the brain
basis mechanisms associated with these differences is based almost entirely on
mapping them via correlations with brain differences. This has established many
consistent and significant brain-behaviour relationships in health and disease. But
understanding neurobiology only through our prior understanding of cognition or
behaviour has drawbacks. Firstly, our understanding of the neurobiology is entirely
constrained by the choice of cognitive measures. Secondly, brain-behaviour
relationships established using this logic are difficult to replicate (Poldrack and
Yarkoni 2016; Uttal 2001). This partly reflects this dependency on task selection,
which differs across research groups and may only partially tap the dimension of
interest. But more critically, multiple aetiological pathways could lead to disorders
with superficially similar phenotypes (Stevens et al. 2017; Jones et al. 2014; Fried
and Kievit 2015). In short, grouping purely by cognitive or behavioural phenotype
is no guarantee of common underlying neurobiology.
At present, there are few alternatives to complement this standard approach. The
current study shows that it is possible to group individuals by brain organization
itself, rather than by cognitive or behavioural phenotype. That is, we identified
stable subgroups of individuals with similarly organized brains, which are distinct
from members of other subgroups. In doing so we hoped to identify the elements of
neurobiology most critical for dividing individuals into their respective subgroups
and their wider functional consequences. Identifying mechanisms of brain
difference a priori has the potential to highlight key organizational principles that
we may never capture by only looking for neural correlates of messy pre-defined
phenotypes. This was made possible by combining recent advances in network
analysis and community-clustering algorithms, with measures of microstructural
white matter organization.
The importance of white matter in cognitive
development
White matter makes up around half of the human brain and plays a critical role as
the main conductor of neural signalling. White matter also shows prolonged post-
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
natal changes that extend into the third decade of life (Lebel, Treit, and Beaulieu
2017) with pronounced development during mid-childhood and adolescence (Westlye
et al. 2009). In humans, the method of choice for measuring differences in white
matter in vivo is diffusion-weighted imaging (DWI). It quantifies the degree to
which diffusion of water molecules is restricted by the tightly-packed parallel axons
that make up white matter. Differences in DWI-derived measures have been found
to relate to individual differences across a range of cognitive domains. For instance,
language processing is related to the maturation of the arcuate fasciculus, a white
matter tract that connects the frontal and superior temporal lobe (Skeide, Brauer,
and Friederici 2015); working memory performance is associated with the integrity
of the superior longitudinal fasciculus, a tract connecting frontal and parietal
regions (Burzynska et al. 2011); executive control is linked to the integrity of frontal
and parietal connections alongside additional connections to motor control
regions (Chaddock-Heyman et al. 2013). Differences in white matter organisation
have also been identified in various neurodevelopmental disorders. For instance,
dyslexia is associated with a reduced organisation in white matter pathways along
the left dorsal and ventral language pathways (Zhao et al. 2016); children with
Attention Deficit Hyperactivity Disorder (ADHD) show reduced white matter
organisation of the corpus callosum and major tracts of the right hemisphere (Wu et
al. 2016); and reduced integrity of connections between the limbic system and
frontal and temporal cortex has been reported for autism spectrum disorder
(ASD) (Ameis and Catani 2015). In summary, DWI measures of white matter
organisation are sensitive to typical variation in cognitive abilities and show
differences in common neurodevelopmental disorders.
Using community detection to identify subgroups with
similar white matter organisation
Network science is the study of complex networks, which represent relationships
between data as a network of nodes connected by edges. This methodological
approach provides a mathematical tool for quantifying the organisation of networks
and the relationships between the nodes within them (Bullmore and Sporns 2009).
Defining subdivisions of highly-connected nodes within a network, so-called
communities, is an area of network science that has received considerable attention
as it applies to many real-world problems (Barabasi 2016). In our case, the network
represents the similarity of white matter organization in 20 major tracts cross
individuals. Community detection makes it possible to define subgroups of
participants that are most similar while being as distinct as possible from other
subgroups. This approach has been successfully applied to distinguish differences in
behaviour within heterogeneous groups, such as subgroups of neuropsychological
function in typically-developing children (Fair et al. 2012) and subgroups of
executive function-related behavioural problems in children who struggle in
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
school (Bathelt, Holmes, and Astle 2017). In summary, community-clustering
provides an ideal method to identify subgroups of individuals with similar
characteristics.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Participants & Methods
Participants
The participants were drawn from three separate studies of development. The
participants included in each study are characterised in the following sections
The Nathan-Kline Institute Rockland Sample (NKI-RS): The enhanced
Nathan Kline Institute-Rockland Sample (NKI-RS) is an ongoing, institutionally-
centred endeavour aimed at creating a large-scale community sample of
participants across the lifespan. Details about the sample are described on the NKI
website (http://fcon_1000.projects.nitrc.org/indi/enhanced/). NKI sample data is
available to researchers upon request. For the current study, data from all
participants who had structural imaging (T1, diffusion-weighted) and age
information available was requested (date of data access: 18 July 2017). The study
was approved by the NKI institutional review board and all adult and child subjects
provided informed consent (Nooner et al. 2012)
Centre for Attention, Learning, and Memory (CALM): For this sample,
children aged between 5 and 18 years were recruited on the basis of ongoing
problems in attention, learning, language and memory, as identified by
professionals working in schools or specialist children’s services in the community.
Following an initial referral, the CALM staff then contacted referrers to discuss the
nature of the children’s problems. If difficulties in one or more areas of attention,
learning, language or memory were indicated by the referrer, the family were
invited to the CALM clinic at the MRC Cognition and Brain Sciences Unit in
Cambridge for a 3-hour assessment. This assessment included the cognitive
assessments reported here. Exclusion criteria for referrals were significant or severe
known problems in vision or hearing that were not corrected or having a native
language other than English. Written parental consent was obtained and children
provided verbal assent. Families were also invited to participate in MRI scanning
on a separate visit. Participation in the MRI part of the study was optional and
required separate parental consent and child assent. Contra-indications for MRI
were metal implants, claustrophobia, or distress during a practice session with a
realistic mock MRI scanner. This study was approved by the local NHS research
ethics committee (Reference: 13/EE/0157).
Attention and Cognition in Education (ACE): This sample was collected for a
study investigating the neural, cognitive, and environmental markers of risk and
resilience in children. Children between 7 and 12 years attending mainstream
school in the UK, with normal or corrected-to-normal vision or hearing, and no
history of brain injury were recruited via local schools and through advertisement
in public places (childcare and community centres, libraries). Participating families
were invited to the MRC Cognition and Brain Sciences Unit for a 2-hour
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
assessment, which included the cognitive assessments reported here, and structural
MRI scanning. Participants received monetary compensation for taking part in the
study. This study was approved by the Psychology Research Ethics Committee at
the University of Cambridge (Reference: Pre.2015.11). Parents provided written
informed consent and children verbal assent.
Cognitive assessments
Procedure: All children for whom we have cognitive data were tested on a one-to-
one basis with a researcher in a dedicated child-friendly testing room at the MRC
CBU. The battery included a wide range of standardized assessments of learning
and cognition. Regular breaks were included throughout the session. Testing was
split into two sessions for children who struggled to complete the assessments in
one sitting. Measures relating a cognitive performance across different domains are
included in this analysis. Tasks that were based on reaction times were not
included in this analysis due to their different psychometric properties compared to
the included tasks that were based on performance measures.
Fluid intelligence: Fluid intelligence was assessed on the Reasoning task of the
Wechsler Abbreviated Scale of Intelligence, 2nd edition (Wechsler 2011). Both
children in the CALM and ACE sample completed this assessment.
Working Memory: The Digit Recall, Backward Digit Recall, Dot Matrix, and Mr X
task of the Automatic Working Memory Assessment (AWMA) (Alloway et al. 2008)
were administered individually. In Digit Recall, children repeat sequences of single-
digit numbers presented in an audio format. In Backward Digit Recall, children
repeat the sequence in backwards order. These tasks were selected to engage verbal
short-term and working memory, respectively. For the Dot Matrix task, the child
was shown the position of a red dot for 2 seconds in a series of four by four matrices
and had to recall this position by tapping the squares on the computer screen. In
the Mr X task, the child retains spatial locations whilst performing interleaved
mental rotation decisions. These tasks were selected to engage visual short-term
and working memory, respectively. These assessments were the same in the CALM
and ACE sample.
Vocabulary: For the CALM sample, the Peabody Picture Vocabulary Test, Fourth
Edition (PPVT-4) (Dunn and Dunn 2007) was used to assess receptive vocabulary
knowledge. Children were required to select one of four pictures showing the
meaning of a spoken word. In the ACE sample, the Vocabulary subtest of the
Wechsler Abbreviated Scale of Intelligence, 2nd edition (Wechsler 2011), was used.
For this task, children had to define words that were presented verbally and
visually, and correct definitions were scored.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Long-term memory: For the CALM sample, the Stories task of the Children’s
Memory Scale (Cohen 1997) was used to assess long-term memory. Children were
read two short stories and were asked to recall these stories after a delay of 10 min.
No long-term memory task was used in the ACE sample.
MRI data acquisition
NKI-RS: Subjects in the NKI sample underwent a scan session using a Siemens
TrioTM 3.0 T MRI scanner. T1-weighted images were acquired a magnetization-
prepared rapid gradient echo (MPRAGE) sequence with 1mm isotropic resolution.
Diffusion scans were acquired with an isotopic set of gradients with 64 directions
using a weighting factor of b=1000s*mm-2 and an isotropic resolution of 2mm.
Details about the scan sequences are described
elsewhere (http://fcon_1000.projects.nitrc.org/indi/enhanced/mri_protocol.html).
CALM & ACE: Magnetic resonance imaging data were acquired at the MRC
Cognition and Brain Sciences Unit, Cambridge U.K. All scans were obtained on the
Siemens 3 T Tim Trio system (Siemens Healthcare, Erlangen, Germany), using a
32-channel quadrature head coil. For ACE, the imaging protocol consisted of two
sequences: T1-weighted MRI and a diffusion-weighted sequence. For CALM, the
imaging protocol included an additional resting-state functional MRI (rs-fMRI)
sequence. T1-weighted volume scans were acquired using a whole brain coverage
3D Magnetisation Prepared Rapid Acquisition Gradient Echo (MP-RAGE) sequence
acquired using 1mm isometric image resolution. Echo time was 2.98 ms, and
repetition time was 2250 ms. Diffusion scans were acquired using echo-planar
diffusion-weighted images with a set of 60 non-collinear directions, using a
weighting factor of b=1000s*mm-2, interleaved with a T2-weighted (b=0) volume.
Whole brain coverage was obtained with 60 contiguous axial slices and isometric
image resolution of 2mm. Echo time was 90 ms and repetition time was 8400 ms.
For functional scans, a total of 270 T2*-weighted whole-brain echo planar images
(EPIs) were acquired (time repetition[TR] = 2 s; time echo [TE] = 30 ms; flip angle =
78 degrees, 3x3x3mm). During this scan, participants were instructed to lie still
with their eyes closed and not fall asleep.
Overview of sample characteristics in the NKI, ACE, and CALM sample
n: useable MRI
(total) Gender:
male-female Age: mean (SD) Age range
NKI 74 (117) 43/31 13.93 (3.164) 6.97-21.58
ACE 74 (86) 35/39 9.99 (1.525) 6.92-12.65
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
CALM
165 (206) 109/56 9.81 (1.191) 5.92-17.92
Microstructural integrity of major white matter tracts
This analysis aimed to identify white matter tracts that show robust inter-
individual differences during development (see Figure 1A). Diffusion-weighted
images were pre-processed to create a brain mask based on the b0-weighted image
(FSL BET) (Smith 2002) and to correct for movement and eddy current-induced
distortions (eddy) (Graham, Drobnjak, and Zhang 2016). Subsequently, the diffusion
tensor model was fitted and fractional anisotropy (FA) maps were calculated (dtifit).
Images with a between-image displacement great than 3mm as indicated by FSL
eddy were excluded from further analysis. All steps were carried out with FSL
v5.0.9 and were implemented in a pipeline using NiPyPe v0.13.0 (Gorgolewski et al.
2011) . To extract FA values for major white matter tracts, FA images were
registered to the FMRIB58 FA template in MNI space using symmetric
diffeomorphic image registration (SyN) as implemented in ANTS v1.9 (Avants et al.
2008). Visual inspection indicated good image registration for all participants.
Subsequently, binary masks from a white matter atlas in MNI space (Mori, Oishi,
and Faria 2009) were applied to extract FA values for 20 major white matter tracts.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Figure 1 A Illustration of processing steps to obtain estimates of similarity between
individuals across major white matter tracts B
Overview of processing steps to derive
structural connectivity of the cingulum C Illustration of the ROI used for
tractography of the cingulum. The ROI is marked with a dotted line on consecutive
FA maps coloured according to the direction of diffusion. The left side of each figure
shows the FA values.
Community detection
Communit
y
detection is an optimisation clustering method. Networks in the curre
nt
analysis represented the child-by-
c
hild correlations across the FA values in
20
w
hite matter tracts. The community algorithm starts with each network node, i
.e.
child, in a separate
ommunity and then iteratively parcellates the network in
c
ommunities to increase the quality index (Q), which represents the segrati
on
b
etween communities, until a maximum is reached. The current study used t
he
algorithm described by Rubinov and Sporns
(
Rubinov and Sporns 2010)
as
implemented in the Brain Connectivity Toolbox (https://sites.google.com/site/bctnet/
)
v
ersion of August 2017, which is an extension of the method described by Blondel
et
al. (Blondel et al. 2008)
t
o networks with positive and negative edges. Th
is
ve
nt
20
.e.
to
on
he
as
)
et
is
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
algorithm is not deterministic and may yield different solutions at each run. In
order to reach a stable community assignment, we applied the consensus clustering
method described by Lancichinetti and Fortunato (Lancichinetti and Fortunato
2012). In short, an average community assignment over 100 iterations was
generated. The community assignment was then repeated further until the
community assignment did not change between successive iterations.
In order to test the reliability of the community detection algorithm under varying
conditions, random networks with known community structure were created. The
networks consisted of 100 nodes with 4 modules. The connection likelihood within
and between clusters was systematically varied between 0.1 and 0.9. The quality
index of the community structure was calculated at each combination of between-
and within-cluster connection likelihood. The results indicated a high-quality index
for networks with higher within-cluster than outside-cluster connection likelihood
(see Figure 2A). High connection density outside of clusters had a large influence,
even when the connection likelihood within modules was very high. We further
tested the robustness of the community assignment by adding increasing
percentages of random Gaussian noise (mu=0, sigma=1) to the network matrix and
repeated the consensus clustering procedure (see Figure 2B). The quality index
indicated a good separation of the clusters between 5 and 30% noise (Q between
0.62 and 0.65). No stable assignment could be reached at 35% of noise and above. In
short, these results indicate that the community assignment is robust to a
considerable amount of noise. Further, the Keringhan-Lin algorithm was used to
test the grouping with an alternative method that can integrate negative edge
weights (Kernighan and Lin 1970). The results indicated nearly identical results
with this alternative method (see Results section).
In order to test if the clustering identified in one sample generalized to a sample
that was not used to inform the algorithm, we compared the within versus-
between-cluster summed correlation to a permutation sample with random cluster
assignment over 1000 repetitions. If the clustering provides a good account of the
data, then the summed within-cluster correlation should be considerably higher
than the between-sample correlation and both should be significantly different from
random cluster assignments. This provides a good way of validating the initial
community clustering solution with the subsequent independent datasets.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Figure 2 Results of robustness testing using networks with a known community
structure. A Quality index in networks with different probabilities of within- and
between-community connections B Quality index in networks with increasing ratios
of Gaussian noise.
Tractography of the cingulum
To foreshadow our results, the clustering algorithm i
d
entified the cingulum
as
c
ritical for determining subgroup membership. We subsequently followed this
up
with tractography. For the tractography, a virtual in-
v
ivo dissection approach w
as
applied (Catani and de Schotten 2008) (see Figure 1B for an overvie
w
of t
he
processing steps). For this purpose,
M
RI scans were converted from the nati
ve
DICOM to compressed NIfTI-
1
format using the dcm2nii tool developed at t
he
McCauseland Centre for Neuroimaging ([
h
ttp://www.mccauslandcenter.sc.ed
u/
mricro/mricron/dcm2nii.html
]
). Subsequently, the images were submitted to t
he
DiPy v0.8.0 implementation (Garyfallidis et al. 2014) of a non-local means de
-
noising algorithm (Coupe et al. 2008) to boost signal-to-
n
oise ratio. Next, a bra
in
m
ask of the b0 image was created using the brain extraction tool (BET) of t
he
F
MRIB Software Library (FSL) v5.0.8. Motion and eddy current correction w
as
applied to the masked images using FSL routines. Finally,
f
ractional anisotro
py
m
aps were created based on a diffusion tensor model fitted through t
he
FSL dtifit algorithm (Behrens et al. 2003).
A
constant solid angle (CSA) was fitt
ed
to the 60-gradient-direction diffusion-
w
eighted images using a maximum harmon
ic
order of 8 using DiPy. Next, probabilistic whole-
b
rain tractography was perform
ed
b
ased on the CSA model with 8 seeds in any voxel with a General FA value high
er
as
up
as
he
ve
he
u/
he
-
in
he
as
py
he
ed
ic
ed
er
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
than 0.1. The step size was set to 0.5 and the maximum number of crossing fibres
per voxel to 2. The cingulum in the left and right hemisphere was reconstructed in
the native space of each participant by drawing a single ROI on 10-15 consecutive
axial slices that followed the anatomy of the cingulum as described in (Catani and
de Schotten 2008) (see Figure1C). To check the reliability of the cingulum
tractography, two researchers performed tractography independently on 40
datasets. The spatial correlation between the density maps indicated very good
inter-rater agreement (left: mean=0.95, SE=0.005, range=0.86-0.990; right:
mean=0.95, SE=0.005, range=0.87-0.990).
Structural connectivity of the cingulum
To estimate the structural connectivity of the cingulum, streamlines included in the
cingulum tractography that intersected with cortical and subcortical ROIs were
counted. For ROI definition, T1-weighted images were preprocessed by adjusting
the field of view using FSL’s robustfov, non-local means denoising in DiPy, deriving
a robust brain mask using the brain extraction algorithm of the Advanced
Normalization Tools (ANTs) v1.9 (Avants et al. 2011), and submitting the images to
recon-all pipeline in FreeSurfer v5.3 (http://surfer.nmr.mgh.harvard.edu). Regions
of interests (ROIs) were based on the Desikan-Killiany parcellation of the MNI
template (Desikan et al. 2006) with 34 cortical ROIs per hemisphere and 17
subcortical ROIs (brain stem, and bilateral cerebellum, thalamus, caudate,
putamen, pallidum, hippocampus, amygdala, nucleus accumbens). The surface
parcellation of the cortex was transformed to a volume using the aparc2aseg tool in
FreeSurfer. Further, the cortical parcellation was expanded by 2mm into the
subcortical white matter using in-house software. In order to move the parcellation
into diffusion space, a transformation based on the T1-weighted volume and the b0-
weighted image of the diffusion sequence was calculated using FreeSurfer’s
bbregister and applied to volume parcellation. For each pairwise combination of
ROIs, the number of streamlines intersecting both ROIs was estimated and
transformed into a density map. The intersection was symmetrically defined, i.e.
streamlines starting and ending in each ROI were averaged.
Default mode network activation
Resting-state functional MRI data were processed to obtain functional activation
within the default mode network. Only participants who completed the full resting-
state sequence had full coverage of the brain in the resting-state sequence, and also
had useable T1-weighted data were included in the analysis. For motion and eddy
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
current correction, volumes were co-registered to the middle volume
using mcflirt (Jenkinson et al. 2002). For quality control, the frame-wise
displacement was calculated as the weighted average of the rotation and translation
parameters (Power et al. 2012) using fsl_motion_outliers. Participants with a frame-
wise displacement above 0.5 on this measure were excluded from the analysis.
Complete data was available for 124 participants (78 male, Age: mean=10.00,
SE=0.196, group1: n=59, group2: n=58, no group: n=6). There was no significant
difference in age between the sample included in the rsfMRI analysis and the
sample included in the main analysis (t(284)=-0.74, p=0.458). The samples did not
differ on any of the cognitive measures ( MR: t=-0.85, p=0.396, Vocab: t=-
0.80, p=0.423; DR: t=-0.73, p=0.465; DM: t=-0.62, p=0.539; BR: t=-0.80, p=0.427;
MX: t=-0.23, p=0.820; CM: t=-0.74, p=0.461). Regarding data quality, there was no
difference in frame-wise displacement between the groups defined through data-
driven clustering (group 1: mean=0.01, SE=0.001; group2: mean=0.01, SE=0.004,
t(122)=-0.71, p=.0.480). The groups also did not differ in the number of motion
outliers (group 1: mean=18.05, SE=1.007; group 2: mean=17.08, SE=1.154,
t(122)=0.63, p=.0.529). There was no significant correlation between the framewise
displacement and any of the cognitive measures (Pearson correlation, MR: r=-
0.12, p=0.188; Vocab: r=0.09, p=0.349; DR: r=-0.00, p=0.986; DM: r=-0.11, p=0.226;
BR: r=0.07, p=0.456; MX: r=0.05, p=0.570; CM: r=0.11, p=0.207).
Functional connectivity was assessed using independent component analysis by
means of the multivariate exploratory linear decomposition into a fixed set of 25
independent components (MELODIC; (Beckmann et al. 2005; Beckmann and Smith
2004)). The correspondence between the independent components derived in the
data and the canonical networks described by Yeo et al. 2011 (Yeo et al. 2011) was
established by calculating the spatial correlation between the maps. Two maps
showed a high spatial correlation with the canonical default mode network (IC3:
r=0.303, IC9: r=0.493). Two additional component maps also had a high spatial
correlation with the default mode network map (r=0.286, r=0.276), but showed
activation within the white matter and CSF and were therefore dismissed as
artefacts. The default mode network in individual participants was calculated as
described by Supekar and colleagues (Supekar et al. 2010). In brief, ICA with
automatic dimensionality detection was performed for each participant. The groups
did not differ in the number of ICA components generated (C1: mean=100.70,
SE=1.821; C2: mean=102.19, SE=1.862; U=1549, p=0.254). Subsequently, a
goodness-of-fit measure was calculated based on the average z-scored activation
within the DMN-mask minus the activation outside of the DMN-mask and the
component with the largest score was selected. To compare the spatial extent of
activation within the individual DMN components between the groups, a
permutation procedure with 5,000 repetitions and cluster-free threshold
enhancement for correction of multiple comparisons as implemented in FSL
randomise was used. To compare functional connectivity between groups, the
average bandpass-filtered (0.01-0.1Hz) signal for a 8mm-sphere placed on the
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
posterior cingulate cortex (PCC, MNI: 0,-52,18), medial prefrontal cortex (mPFC,
MNI: 1, 50, -5), and the left and right temporoparietal junction (TPJ, MNI: -46, -68,
32; 46, -68, 32) and the partial correlation between mPFC and PCC was calculated
controlling for the signals in the left and right TPJ (Heuvel et al. 2008).
Statistical analysis
Shapiro-Wilk tests were used to test if data were normally distributed. If a
significant deviation from normality was indicated (p>0.05), non-parametric tests
were used. Specifically, Mann-Whitney U tests were used instead of independent t-
tests. The Bonferroni method was used to correct for multiple comparisons unless
otherwise stated.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Results
The analysis followed a three-step sequential logic:
The first aim was to determine subgroups of maximally similar white matter
organisation using community detection in a population-representative sample
of typical development (NKI sample).
The second aim was to investigate potential cognitive correlates of white matter
tracts that distinguished groups identified in the first part of the analysis. This
was carried in large developmental samples that had detailed cognitive
assessments and showed large variation in cognitive abilities (ACE & CALM
sample).
The third aim was to investigate the relationship between white matter
variation in more detail by mapping the particular connections related to
different aspects of cognitive ability and relating variation in white matter
microstructure to functional connectivity in a sample with detailed cognitive
assessments and multimodal neuroimaging data (CALM sample).
Community clustering indicates the presence of two
groups that differ in FA of the cingulum
Grouping children in the NKI sample according to their similarity of FA values
within major white matter tracts using consensus clustering indicated the presence
of two groups. The quality index indicated a good separation between groups
(Q=0.47, Figure 3 A and B). The groups defined through community clustering did
not differ in age (C1: n=34, mean=13.568, SE=0.59; C2: n=40, mean=14.23,
SE=0.470; t-test: t(72)=-0.89, p=0.375) and did also not differ from the whole sample
in proportion of males and females (C1: 24/10 male/female, chi=0.06, p=0.800; C2:
21/19, chi=0.05, p=0.825). Comparison of group assignment with an alternative
clustering method (Kernighan-Lin algorithm) indicated very high agreement with
an equal number of clusters using both methods and only 6 out 74 children being
assigned to a different cluster depending on the algorithm.
Next, FA values for all tracts were compared between the groups defined through
community clustering. The groups differed on the FA of the left and right anterior
cingulum (cingulate gyrus region) (left anterior cingulum: C1: mean=0.49,
SE=0.179, C2: mean=-0.42, SE=0.121, Man-Whitney: U=304, p<0.001, pcorrected<0.001;
right anterior cingulum: mean=0.50, SE=0.167, C2: mean=-0.42, SE=0.132 Man-
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Whitney: U=300, p<0.001, pcorrected<0.001). There were no significant differences
between the groups for any other tract (all other pcorrected>0.1, see Figure 3C).
The range of age-regressed, z-transformed values for the left and right cingulum
was used to group children in the CALM and ACE sample (see Figure 3D). A
majority of the samples fell within the range for C1 or C2 (ACE: C1: n=36 (42%),
C2: n=43 (51%), not categorized: n=5 (6%); CALM: C1: n=95 (49%), C2: n=87 (45%),
not categorized: n=11 (6%)) . The proportion of participants being assigned to C1 or
C2 was equal (ACE: chi2=0.64, p=0.423; CALM: chi2=0.01, p=0.943). Comparison of
the within-group versus between-group connection strength indicated that the
grouping based on the clusters identified in the NKI sample provided a good
account of the data for both the ACE and the CALM sample that significantly
differed from random group assignments (ACE: intra-cluster= 64.85, p<0.001; inter-
cluster: -58.95, p=0.001; CALM: intra-cluster=521.17, p<0.001; inter-
cluster=21.10, p<0.001).
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Figure 3 Overview of clustering based on fractional anisotropy (FA) in major white
matter tracts in the NKI sample. A: Correlation within groups defined through
community-based clustering of FA values in the NKI sample showing two distinct
groups with high correlation within the cluster. B: Spring-layout depiction of the
clusters identified through community clustering show good separation between the
clusters (thresholded at R0.3 visualisation purposes). C: Differences in FA for each
major white matter tract between the clusters identified through community
clustering. The error bars show one standard error, the middle of the bar indicates
the median. Significant differences between clusters were found for the left and right
t
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
anterior Cingulum. D: FA values in the left and right anterior cingulum in the ACE
and CALM with grouping according to the ranges identified in the NKI sample
clusters. The bottom figure show the percentage of participants in each cluster range
with orange for C1, yellow for C2, and grey for participants that fall outside of both
ranges. Abbreviations: IFOF: inferior fronto-occipital fasciculus, ILF: inferior
longitudinal fasciculus, SLF: superior longitudinal fasciculus
Brain-defined subgroups differ in cognitive
performance
Next, differences in cognitive performance between the clustering-defined groups
were investigated. For the CALM sample, children who fell within the C1 range
showed significantly higher performance across all cognitive assessments
(all pcorrected<0.05, see Figure 4A). For the ACE sample, C1 showed significantly lower
performance across all cognitive measures, apart from visuospatial working
memory (p=0.207, pcorrected>0.999, see Figure 4B).
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Figure 4 Comparison of cognitive scores between the groups defined based on
cingulum FA in the CALM sample (A) and the ACE sample (B
). Legend: Comparison
between C1 and C2; C0: not assigned to C1 or C2, *** p
corrected
<0.001, ** p
corrected
<0.01,
* p
corrected
<0.05
Subgroups show differences in fronto-parietal and
fronto-temporal connections of the cingulum
T
he groups were compared on the number of streamlines within the cingulum th
at
c
onnect cortical and subcortical regions. Streamlines associated with the cingulu
m
n
at
m
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
were densest around in the core area of the cingulum, but also extended into the
frontal, parietal, and temporal cortex in both hemispheres (see Figure 5A). Areas
that were consistently connected through streamlines of the cingulum included the
orbitofrontal and superior frontal cortex, insula, inferior and superior parietal
cortex, precuneus, and parahippocampal cortex (see Figure 5B). The groups showed
significant differences in some but not all of these connections. Differences between
the groups were found in left hemisphere connections between the superior frontal
cortex and the precuneus, the parahippocampal cortex and the superior parietal
lobe, the posterior cingulate cortex and the superior parietal cortex, and in right
hemisphere connections between the inferior frontal cortex and the inferior
temporal cortex, the precentral gyrus and the cuneus, the precentral gyrus and the
precuneus (all p<0.05). C2 had fewer streamlines in these connections compared to
C1. In addition, the relationship between the connection strength (number of
streamlines) and cognitive scores was investigated in a regression analysis with
continuous scores instead of the group comparison. There was a significant
relationship between the strength of the connection between the left precuneus and
the left superior frontal cortex with matrix reasoning scores (beta=0.286, p<0.001,
pcorrected=0.005). Other combinations of cognitive scores and connection strengths
were not significant after Bonferroni-correction for multiple comparisons.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Figure 5 Structural connectivity of the cingulum. A: Average streamline density of
the cingulum reconstruction. B: Connections of the cingulum. Grey lines indicate
connections that were present in at least 60% of the sample. Blues lines indicate
connections of the left cingulum that had fewer streamlines in group 2 compared to
group 1 and green lines indicate fewer connections in group2 compared to group 1 for
the right cingulum.
or
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Relationship between structural connectivity and
default mode network activation
The default mode network (DMN) could be identified in group-level ICA (see
Figure 6A). The groups did not differ in the number of components identified in
individual independent component decomposition (C1: mean=100.70, SE=1.821; C2:
mean=102.19, SE=1.862; U=1549, p=0.254). Comparison of ICA-derived maps
between C1 and C2 indicated that the spatial extent of the DMN was significantly
reduced in the posterior cingulate cortex for C1 (MNI peak coordinates: 0, -70, 35;
TCFE-corrected p<0.05; see Figure 6B). Comparison of the DMN activation
indicated reduced activation in C1 compared to C2 (C1: mean=-0.16, SE=0.137; C2:
mean=0.13, SE=0.126, U=1372, p=0.049, see Figure 6D). Next, the relationship
between variation in the microstructure of the cingulum and differences in
functional connectivity of the medial prefrontal cortex (mPFC) and posterior
cingulate cortex (PCC) was evaluated. While the clustering-defined groups did not
differ in absolute mPFC-PCC partial correlation (C1: mean=0.31, SE=0.020; C2:
mean=0.32, SE=0.020, U=1511, p=0.239), analysis of the relationship between
cingulum FA and mPFC-PCC correlation indicated a significant group interaction
(group x cingulum FA: t(118)=2.01, p=0.047). Follow-up analyses showed that
cingulum FA was significantly associated with the mPFC-PCC correlation in C1
(beta=0.38, p=0.005), but not in C2 (beta=0.03, p=0.811, see Figure 6E).
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Figure 6 A Default mode network (DMN) at the group level, B individual DMN
average in C1 and C2, C Results of a two-sample t-test contrasting the DMN in C1
and C2 showing a reduced spatial extent of C1 compared to C2 in the posterior
cingulate cortex. D DMN Activation in C1 and C2 E Relationship between the
partial correlation of the medial prefrontal cortex and posterior cingulate cortex and
FA of the cingulum in C1 and C2.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Discussion
The first aim of the study was to identify groups that show individual differences in
white matter organization. We employed data-driven community clustering to
create subgroups of participants that were maximally similar in white matter
organisation across 20 major white matter tracts. In theory, this could have
resulted in multiple groups with a highly complex differentiation of brain
organization. However, the subgrouping is dominated by the integrity of a pair of
large tracts within the brain: the results indicated the presence of two groups
primarily distinguished by FA of the left and right cingulum. Whilst this finding is
somewhat unexpected, the cingulum has been implicated in a broad range of
psychiatric and developmental disorders, including autism spectrum disorder (Ikuta
et al. 2014), attention deficit hyperactivity disorder (Cooper, Thapar, and Jones
2015), schizophrenia (Abdul-Rahman, Qiu, and Sim 2011), major
depression (Schermuly et al. 2010), mild cognitive impairment (Metzler-Baddeley et
al. 2012), and dementia (Kantarci et al. 2011). One reason why such substantial and
robust individual differences in the cingulum exist may be its prolonged
development. The cingulum is one of the only tracts that show extended
development with changes throughout childhood and adolescence, only reaching a
stable level in the third decade of life (Lebel, Treit, and Beaulieu 2017; Tamnes et
al. 2009). Similar tracts, also involved in long-range connectivity, like the superior
longitudinal fasciculus, inferior longitudinal fasciculus, and inferior fronto-occipital
fasciculus already plateau by the end of the second decade of life. Consequently,
individual differences may arise from differences in timing of cingulum maturation.
Alternatively, the cingulum may be particularly sensitive to environmental
influences given that diffusion properties of white matter can change in response to
environmental demands, e.g. in the context of motor or cognitive
training (Caeyenberghs et al. 2016; Scholz et al. 2009). Inter-individual differences
in cingulum FA could also, therefore, be a marker of accumulated differences in
experience between individuals.
The second aim of the study of the study was to investigate whether subgroups
formed on the basis of similarity of white-matter organization differ in cognitive
performance. Children with lower cingulum FA showed lower performance on
assessments of general intelligence, vocabulary, and short-term, long-term, and
working memory. This broad difference in cognitive performance clashes with
previous findings that indicated a comparatively narrow association between
cingulum FA and executive function, which included sustained and divided
attention (Takahashi et al. 2010), working memory (Golestani et al. 2014), and
planning (Kubicki et al. 2003). There are several possible reasons for this contrast:
First, the cognitive assessments in the current study were more comprehensive
than in many previous studies. Second, the current study had much higher
statistical power, while previous studies were likely to be underpowered with
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
sample sizes ranging from 16 (Kubicki et al. 2003) to 38 participants (Takahashi et
al. 2010). Finally and most importantly, the effect observed here is likely larger and
more widespread because individuals are grouped on the basis of brain
organization. Grouping according to phenotype likely incorporates a high degree of
heterogeneity in underlying neurobiology (Fair et al. 2012). The novel data-driven
approach taken here created groups that maximise within-group homogeneity in
neurobiology.
The prominent role of the cingulum in the brain-based grouping and the strong
cognitive differences between the groups implies that this tract plays a critical role
in the coordination of different brain regions. Indeed, the role of white matter
structures is often seen as an extension of the brain regions that it connects (Catani
et al. 2013). The cingulum contains strong connections between the medial
prefrontal cortex (mPFC) and posterior cingulate cortex (PCC) and is related to
their communication (Heuvel et al. 2008). We directly tested whether the functional
connection between these regions is constrained by the cingulum. While a strong
association between cingulum FA and mPFC-PCC was observed in a group of
children with low cingulum FA, there was no association in children with high
cingulum FA. A possible interpretation is that low cingulum FA limits the
communication between these regions, but only up to a critical value. Once this
critical value is reached – as in the case of the high FA group - further increases no
longer influence the strength of the functional connection.
Integration between parietal and prefrontal areas is central to some theories of
general intelligence (Deary, Penke, and Johnson 2010; Barbey 2017). According to
the parietal-prefrontal integration theory (P-FIT), general cognitive ability arises
from the interplay between parietal areas involved in the integration and
abstraction of sensory information, prefrontal areas involved in reasoning and
problem-solving, and the anterior cingulate cortex involved in response
selection (Jung and Haier 2007; Barbey et al. 2012; Colom et al. 2009). The role of
the cingulum may be to enable the efficient communication between these regions.
The connectivity mapping in the current study indicates that the cingulum plays a
broader role in structural connectivity beyond connections along the prefrontal-
parietal axis. The dense short-range and longer-range connections of the cingulum
may enable communication within multiple large-scale structural brain
networks. Indeed, the connection showing the strongest association with cognitive
performance involved the superior frontal cortex and the precuneus. Both regions
are part of a small number of so-called hub regions that are densely connected in
functional and structural brain networks (Heuvel and Sporns 2011; Power et al.
2013). The connectivity of these hub regions is thought to form the backbone of a
network architecture that enables the efficient transfer of information and the rapid
transition between different network states, both of which may be necessary for
cognitive performance in complex tasks (Barbey 2017).
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
In conclusion, this is the first study to investigate the relationship between
individual differences in white matter connections and cognitive performance with a
brain-first approach. The results indicate that consistent individual differences
exist in the microstructural organization of the left and right cingulum. Lower
microstructural organisation of the cingulum was related to lower cognitive
performance across a range of cognitive domains and was also linked to reduced
BOLD activation within the default mode network. These findings suggest that
cingulum connections play an important role in brain organisation and cognitive
performance in childhood and adolescence. The current study is an initial step
towards a bigger goal of understanding individual differences in neuroanatomy; it
illustrates the benefits of bringing together large samples, multi-modal
characterisation, and machine learning approaches like community clustering to
identify biologically-defined groups that are difficult to distinguish at a behavioural
level. In future, this new approach may enable our field to isolate key biological
mechanism that drive differences in brain organisation, allowing for more objective
identification of individual needs, and open new avenues for effective
interventions.
Acknowledgements
We want to thank Dr Edwin Dammaijer and Dr Rogier Kievit for helpful comments
on early drafts of the paper. The Centre for Attention Learning and Memory
(CALM) research clinic is based at and supported by funding from the MRC
Cognition and Brain Sciences Unit, University of Cambridge. The Principal
Investigators are Joni Holmes (Head of CALM), Susan Gathercole (Chair of CALM
Management Committee), Duncan Astle, Tom Manly and Rogier Kievit. Data
collection is assisted by a team of researchers and PhD students at the CBSU that
includes Sarah Bishop, Annie Bryant, Sally Butterfield, Fanchea Daily, Laura
Forde, Erin Hawkins, Sinead O’Brien, Cliodhna O’Leary, Joseph Rennie, and
Mengya Zhang. The authors wish to thank the many professionals working in
children’s services in the South-East and East of England for their support, and to
the children and their families for giving up their time to visit the clinic.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Conflicts of interest
The authors declare no conflict of interest
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
References
Abdul-Rahman, M. F., A. Qiu, and K. Sim. 2011. “Regionally Specific White Matter
Disruptions of Fornix and Cingulum in Schizophrenia”. Edited by Wang Zhan. PLoS
ONE 6 (4). Public Library of Science (PLoS): e18652.
doi:10.1371/journal.pone.0018652.
Alloway, T. P., S. E. Gathercole, H. Kirkwood, and J. Elliott. 2008. “Evaluating the
Validity of the Automated Working Memory Assessment”. Educational Psychology
28 (7). Informa UK Limited: 725–34. doi:10.1080/01443410802243828.
Ameis, S. H., and M. Catani. 2015. “Altered White Matter Connectivity as a Neural
Substrate for Social Impairment in Autism Spectrum Disorder”. Cortex 62
(January). Elsevier BV: 158–81. doi:10.1016/j.cortex.2014.10.014.
Avants, B., C. Epstein, M. Grossman, and J. Gee. 2008. “Symmetric Diffeomorphic
Image Registration with Cross-Correlation: Evaluating Automated Labeling of
Elderly and Neurodegenerative Brain”. Medical Image Analysis 12 (1). Elsevier BV:
26–41. doi:10.1016/j.media.2007.06.004.
Avants, B. B., N. J. Tustison, G. Song, P. A. Cook, A. Klein, and J. C. Gee. 2011. “A
Reproducible Evaluation of ANTs Similarity Metric Performance in Brain Image
Registration”. NeuroImage 54 (3). Elsevier BV: 2033–44.
doi:10.1016/j.neuroimage.2010.09.025.
Barabasi, A. L. 2016. Network Science. Cambridge University Press.
Barbey, A. K. 2017. “Network Neuroscience Theory of Human Intelligence”. Trends
in Cognitive Sciences, November. Elsevier BV. doi:10.1016/j.tics.2017.10.001.
Barbey, A. K., R. Colom, J. Solomon, F. Krueger, C. Forbes, and J. Grafman. 2012.
“An Integrative Architecture for General Intelligence and Executive Function
Revealed by Lesion Mapping”. Brain 135 (4). Oxford University Press (OUP): 1154–
64. doi:10.1093/brain/aws021.
Bathelt, J., J. Holmes, and D. E. Astle. 2017. “Data-Driven Subtyping of Executive
Function Related Behavioural Problems in Children”, bioRxiv. doi:10.1101/158949.
Beckmann, C. F., M. DeLuca, J. T. Devlin, and S. M. Smith. 2005. “Investigations
into Resting-State Connectivity Using Independent Component Analysis”.
Philosophical Transactions of the Royal Society B: Biological Sciences 360 (1457).
The Royal Society: 1001–13. doi:10.1098/rstb.2005.1634.
Beckmann, C.F., and S.M. Smith. 2004. “Probabilistic Independent Component
Analysis for Functional Magnetic Resonance Imaging”. IEEE Transactions on
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Medical Imaging 23 (2). Institute of Electrical and Electronics Engineers (IEEE):
137–52. doi:10.1109/tmi.2003.822821.
Behrens, T.E.J., M.W. Woolrich, M. Jenkinson, H. Johansen-Berg, R.G. Nunes, S.
Clare, P.M. Matthews, J.M. Brady, and S.M. Smith. 2003. “Characterization and
Propagation of Uncertainty in Diffusion-Weighted MR Imaging”. Magnetic
Resonance in Medicine 50 (5). Wiley-Blackwell: 1077–88. doi:10.1002/mrm.10609.
Blondel, V. D., J.-L. Guillaume, R. Lambiotte, and E. Lefebvre. 2008. “Fast
Unfolding of Communities in Large Networks”. Journal of Statistical Mechanics:
Theory and Experiment 2008 (10). IOP Publishing: P10008. doi:10.1088/1742-
5468/2008/10/p10008.
Bullmore, E., and O. Sporns. 2009. “Complex Brain Networks: Graph Theoretical
Analysis of Structural and Functional Systems”. Nature Reviews Neuroscience 10
(3). Springer Nature: 186–98. doi:10.1038/nrn2575.
Burzynska, A. Z., I. E. Nagel, C. Preuschhof, S.- C. Li, U. Lindenberger, L.
Backman, and H. R. Heekeren. 2011. “Microstructure of Frontoparietal Connections
Predicts Cortical Responsivity and Working Memory Performance”. Cerebral Cortex
21 (10). Oxford University Press (OUP): 2261–71. doi:10.1093/cercor/bhq293.
Caeyenberghs, K., C. Metzler-Baddeley, S. Foley, and D. K. Jones. 2016. “Dynamics
of the Human Structural Connectome Underlying Working Memory Training”.
Journal of Neuroscience 36 (14). Society for Neuroscience: 4056–66.
doi:10.1523/jneurosci.1973-15.2016.
Catani, M., and M. Thiebaut de Schotten. 2008. “A Diffusion Tensor Imaging
Tractography Atlas for Virtual in Vivo Dissections”. Cortex 44 (8). Elsevier BV:
1105–32. doi:10.1016/j.cortex.2008.05.004.
Catani, M., M. T. de Schotten, D. Slater, and F. DellAcqua. 2013. “Connectomic
Approaches before the Connectome”. NeuroImage 80 (October). Elsevier BV: 2–13.
doi:10.1016/j.neuroimage.2013.05.109.
Chaddock-Heyman, L., K. I. Erickson, M. W. Voss, J. P. Powers, A. M. Knecht, M. B.
Pontifex, E. S. Drollette, et al. 2013. “White Matter Microstructure Is Associated
with Cognitive Control in Children”. Biological Psychology 94 (1). Elsevier BV: 109
15. doi:10.1016/j.biopsycho.2013.05.008.
Cohen, M. J. 1997. Children’s Memory Scale. The Psychological Corporation.
Colom, R., R. J. Haier, K. Head, J. Alvarez-Linera, M. Angeles-Quiroga, P. C. Shih,
and R. E. Jung. 2009. “Gray Matter Correlates of Fluid Crystallized, and Spatial
Intelligence: Testing the P-FIT Model”. Intelligence 37 (2). Elsevier BV: 124–35.
doi:10.1016/j.intell.2008.07.007.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Cooper, M., A. Thapar, and D. K. Jones. 2015. “ADHD Severity Is Associated with
White Matter Microstructure in the Subgenual Cingulum”. NeuroImage: Clinical 7.
Elsevier BV: 653–60. doi:10.1016/j.nicl.2015.02.012.
Coupe, P., P. Yger, S. Prima, P. Hellier, C. Kervrann, and C. Barillot. 2008. “An
Optimized Blockwise Nonlocal Means Denoising Filter for 3-D Magnetic Resonance
Images”. IEEE Transactions on Medical Imaging 27 (4). Institute of Electrical and
Electronics Engineers (IEEE): 425–41. doi:10.1109/tmi.2007.906087.
Cronbach, L. J. 1957. “The Two Disciplines of Scientific Psychology.”. American
Psychologist 12 (11). American Psychological Association (APA): 671–84.
doi:10.1037/h0043943.
Deary, I. J., L. Penke, and W. Johnson. 2010. “The Neuroscience of Human
Intelligence Differences”. Nature Reviews Neuroscience, February. Springer Nature.
doi:10.1038/nrn2793.
Desikan, R. S., F. Segonne, B. Fischl, B. T. Quinn, B. C. Dickerson, D. Blacker, R. L.
Buckner, et al. 2006. “An Automated Labeling System for Subdividing the Human
Cerebral Cortex on MRI Scans into Gyral Based Regions of Interest”. NeuroImage
31 (3). Elsevier BV: 968–80. doi:10.1016/j.neuroimage.2006.01.021.
Dunn, L. M., and D. M. Dunn. 2007. Peabody Picture Vocabulary Test. 4Th ed.
Pearson Education: Minneapolis, MN, USA.
Fair, D. A., D. Bathula, M. A. Nikolas, and J. T. Nigg. 2012. “Distinct
Neuropsychological Subgroups in Typically Developing Youth Inform Heterogeneity
in Children with ADHD”. Proceedings of the National Academy of Sciences 109 (17).
Proceedings of the National Academy of Sciences: 6769–74.
doi:10.1073/pnas.1115365109.
Fried, E. I., and R. A. Kievit. 2015. “The Volumes of Subcortical Regions in
Depressed and Healthy Individuals Are Strikingly Similar: a Reinterpretation of
the Results by Schmaal Et Al.”. Molecular Psychiatry 21 (6). Springer Nature: 724–
25. doi:10.1038/mp.2015.199.
Garyfallidis, E., M. Brett, B. Amirbekian, A. Rokem, S. van der Walt, M.
Descoteaux, and I. Nimmo-Smith et al. 2014. “Dipy a Library for the Analysis of
Diffusion MRI Data”. Frontiers in Neuroinformatics 8 (February). Frontiers Media
SA. doi:10.3389/fninf.2014.00008.
Golestani, A. M., L. Miles, J. Babb, F. X. Castellanos, D. Malaspina, and M. Lazar.
2014. “Constrained by Our Connections: White Matters Key Role in Interindividual
Variability in Visual Working Memory Capacity”. Journal of Neuroscience 34 (45).
Society for Neuroscience: 14913–18. doi:10.1523/jneurosci.2317-14.2014.
Gorgolewski, K., C. D. Burns, C. Madison, D. Clark, Y. O. Halchenko, M. L.
Waskom, and S. S. Ghosh. 2011. “Nipype: A Flexible Lightweight and Extensible
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Neuroimaging Data Processing Framework in Python”. Frontiers in
Neuroinformatics 5. Frontiers Media SA. doi:10.3389/fninf.2011.00013.
Graham, M. S., I. Drobnjak, and H. Zhang. 2016. “Realistic Simulation of Artefacts
in Diffusion MRI for Validating Post-Processing Correction Techniques”.
NeuroImage 125 (January). Elsevier BV: 1079–94.
doi:10.1016/j.neuroimage.2015.11.006.
Heuvel, M. P. van den, R. Mandl, J. Luigjes, and H. Hulshoff Pol. 2008.
“Microstructural Organization of the Cingulum Tract and the Level of Default Mode
Functional Connectivity”. Journal of Neuroscience 28 (43). Society for Neuroscience:
10844–51. doi:10.1523/jneurosci.2964-08.2008.
Heuvel, M. P. van den, and O. Sporns. 2011. “Rich-Club Organization of the Human
Connectome”. Journal of Neuroscience 31 (44). Society for Neuroscience: 15775–86.
doi:10.1523/jneurosci.3539-11.2011.
Ikuta, T., K. M. Shafritz, J. Bregman, B. D. Peters, P. Gruner, A. K. Malhotra, and
P. R. Szeszko. 2014. “Abnormal Cingulum Bundle Development in Autism: A
Probabilistic Tractography Study”. Psychiatry Research: Neuroimaging 221 (1).
Elsevier BV: 63–68. doi:10.1016/j.pscychresns.2013.08.002.
Jenkinson, M., P. Bannister, M. Brady, and S. Smith. 2002. “Improved Optimization
for the Robust and Accurate Linear Registration and Motion Correction of Brain
Images”. NeuroImage 17 (2). Elsevier BV: 825–41. doi:10.1006/nimg.2002.1132.
Jones, E. J. H., T. Gliga, R. Bedford, T. Charman, and M. H. Johnson. 2014.
“Developmental Pathways to Autism: A Review of Prospective Studies of Infants at
Risk”. Neuroscience & Biobehavioral Reviews 39 (February). Elsevier BV: 1–33.
doi:10.1016/j.neubiorev.2013.12.001.
Jung, R. E., and R. J. Haier. 2007. “The Parieto-Frontal Integration Theory (P-FIT)
of Intelligence: Converging Neuroimaging Evidence”. Behavioral and Brain Sciences
30 (02). Cambridge University Press (CUP): 135. doi:10.1017/s0140525x07001185.
Kantarci, K., M. L. Senjem, R. Avula, B. Zhang, A. R. Samikoglu, S. D. Weigand, S.
A. Przybelski, et al. 2011. “Diffusion Tensor Imaging and Cognitive Function in
Older Adults with No Dementia”. Neurology 77 (1). Ovid Technologies (Wolters
Kluwer Health): 26–34. doi:10.1212/wnl.0b013e31822313dc.
Kernighan, B. W., and S. Lin. 1970. “An Efficient Heuristic Procedure for
Partitioning Graphs”. Bell System Technical Journal 49 (2). Institute of Electrical
and Electronics Engineers (IEEE): 291–307. doi:10.1002/j.1538-
7305.1970.tb01770.x.
Kubicki, M., C.-F. Westin, P. G. Nestor, C. G. Wible, M. Frumin, S. E. Maier, R.
Kikinis, F. A. Jolesz, R. W. McCarley, and M. E. Shenton. 2003. “Cingulate
Fasciculus Integrity Disruption in Schizophrenia: a Magnetic Resonance Diffusion
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Tensor Imaging Study”. Biological Psychiatry 54 (11). Elsevier BV: 1171–80.
doi:10.1016/s0006-3223(03)00419-0.
Lancichinetti, A., and S. Fortunato. 2012. “Consensus Clustering in Complex
Networks”. Scientific Reports 2 (1). Springer Nature. doi:10.1038/srep00336.
Lebel, C., S. Treit, and C. Beaulieu. 2017. “A Review of Diffusion MRI of Typical
White Matter Development from Early Childhood to Young Adulthood”. NMR in
Biomedicine, September. Wiley-Blackwell, e3778. doi:10.1002/nbm.3778.
Lubinski, D. 2000. “Scientific and Social Significance of Assessing Individual
Differences: Sinking Shafts at a Few Critical Points”. Annual Review of Psychology
51 (1). Annual Reviews: 405–44. doi:10.1146/annurev.psych.51.1.405.
Metzler-Baddeley, C., D. K. Jones, J. Steventon, L. Westacott, J. P. Aggleton, and
M. J. OSullivan. 2012. “Cingulum Microstructure Predicts Cognitive Control in
Older Age and Mild Cognitive Impairment”. Journal of Neuroscience 32 (49). Society
for Neuroscience: 17612–19. doi:10.1523/jneurosci.3299-12.2012.
Mori, S., K. Oishi, and A. V. Faria. 2009. “White Matter Atlases Based on Diffusion
Tensor Imaging”. Current Opinion in Neurology 22 (4). Ovid Technologies (Wolters
Kluwer Health): 362–69. doi:10.1097/wco.0b013e32832d954b.
Nooner, K. B., S. J. Colcombe, R. H. Tobe, M. Mennes, M. M. Benedict, A. L.
Moreno, L. J. Panek, et al. 2012. “The NKI-Rockland Sample: A Model for
Accelerating the Pace of Discovery Science in Psychiatry”. Frontiers in Neuroscience
6. Frontiers Media SA. doi:10.3389/fnins.2012.00152.
Poldrack, R. A., and T. Yarkoni. 2016. “From Brain Maps to Cognitive Ontologies:
Informatics and the Search for Mental Structure”. Annual Review of Psychology 67
(1). Annual Reviews: 587–612. doi:10.1146/annurev-psych-122414-033729.
Power, J. D., K. A. Barnes, A. Z. Snyder, B. L. Schlaggar, and S. E. Petersen. 2012.
“Spurious but Systematic Correlations in Functional Connectivity MRI Networks
Arise from Subject Motion”. NeuroImage 59 (3). Elsevier BV: 2142–54.
doi:10.1016/j.neuroimage.2011.10.018.
Power, J. D., B. L. Schlaggar, C. N. Lessov-Schlaggar, and S. E. Petersen. 2013.
“Evidence for Hubs in Human Functional Brain Networks”. Neuron 79 (4). Elsevier
BV: 798–813. doi:10.1016/j.neuron.2013.07.035.
Rubinov, M., and O. Sporns. 2010. “Complex Network Measures of Brain
Connectivity: Uses and Interpretations”. NeuroImage 52 (3). Elsevier BV: 1059–69.
doi:10.1016/j.neuroimage.2009.10.003.
Schermuly, I., A. Fellgiebel, S. Wagner, I. Yakushev, P. Stoeter, R. Schmitt, R.J.
Knickenberg, F. Bleichner, and M.E. Beutel. 2010. “Association between Cingulum
Bundle Structure and Cognitive Performance: An Observational Study in Major
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Depression”. European Psychiatry 25 (6). Elsevier BV: 355–60.
doi:10.1016/j.eurpsy.2010.05.001.
Scholz, J., M. C. Klein, T. E. J. Behrens, and H. Johansen-Berg. 2009. “Training
Induces Changes in White-Matter Architecture”. Nature Neuroscience 12 (11).
Springer Nature: 1370–71. doi:10.1038/nn.2412.
Skeide, M. A., J. Brauer, and A. D. Friederici. 2015. “Brain Functional and
Structural Predictors of Language Performance”. Cerebral Cortex 26 (5). Oxford
University Press (OUP): 2127–39. doi:10.1093/cercor/bhv042.
Smith, S. M. 2002. “Fast Robust Automated Brain Extraction”. Human Brain
Mapping 17 (3). Wiley-Blackwell: 143–55. doi:10.1002/hbm.10062.
Stevens, M. C., G. D. Pearlson, V. D. Calhoun, and K. L. Bessette. 2017. “Functional
Neuroimaging Evidence for Distinct Neurobiological Pathways in Attention-
Deficit/Hyperactivity Disorder”. Biological Psychiatry: Cognitive Neuroscience and
Neuroimaging, September. Elsevier BV. doi:10.1016/j.bpsc.2017.09.005.
Supekar, K., L. Q. Uddin, K. Prater, H. Amin, M. D. Greicius, and V. Menon. 2010.
“Development of Functional and Structural Connectivity within the Default Mode
Network in Young Children”. NeuroImage 52 (1). Elsevier BV: 290–301.
doi:10.1016/j.neuroimage.2010.04.009.
Takahashi, M., K. Iwamoto, H. Fukatsu, S. Naganawa, T. Iidaka, and N. Ozaki.
2010. “White Matter Microstructure of the Cingulum and Cerebellar Peduncle Is
Related to Sustained Attention and Working Memory: A Diffusion Tensor Imaging
Study”. Neuroscience Letters 477 (2). Elsevier BV: 72–76.
doi:10.1016/j.neulet.2010.04.031.
Tamnes, C. K., Y. Ostby, A. M. Fjell, L. T. Westlye, P. Due-Tonnessen, and K. B.
Walhovd. 2009. “Brain Maturation in Adolescence and Young Adulthood: Regional
Age-Related Changes in Cortical Thickness and White Matter Volume and
Microstructure”. Cerebral Cortex 20 (3). Oxford University Press (OUP): 534–48.
doi:10.1093/cercor/bhp118.
Uttal, W. 2001. The New Phrenology: The Limits of Localizing Cognitive Processes
in the Brain. MIT Press.
Wechsler, D. 2011. Wechsler Abbreviated Scale of Intelligence - Second Edition
(WASI-II). Pearson.
Westlye, L. T., K. B. Walhovd, A. M. Dale, A. Bjornerud, P. Due-Tonnessen, A.
Engvig, H. Grydeland, C. K. Tamnes, Y. Ostby, and A. M. Fjell. 2009. “Life-Span
Changes of the Human Brain White Matter: Diffusion Tensor Imaging (DTI) and
Volumetry”. Cerebral Cortex 20 (9). Oxford University Press (OUP): 2055–68.
doi:10.1093/cercor/bhp280.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
Wu, Z.-M., J. Bralten, Q.-J. Cao, M. Hoogman, M. P. Zwiers, L. An, L. Sun, et al.
2016. “White Matter Microstructural Alterations in Children with ADHD:
Categorical and Dimensional Perspectives”. Neuropsychopharmacology 42 (2).
Springer Nature: 572–80. doi:10.1038/npp.2016.223.
Yeo, B. T. Thomas, F. M. Krienen, J. Sepulcre, M. R. Sabuncu, D. Lashkari, M.
Hollinshead, J. L. Roffman, et al. 2011. “The Organization of the Human Cerebral
Cortex Estimated by Intrinsic Functional Connectivity”. Journal of Neurophysiology
106 (3). American Physiological Society: 1125–65. doi:10.1152/jn.00338.2011.
Zhao, J., M. T. de Schotten, I. Altarelli, J. Dubois, and F. Ramus. 2016. “Altered
Hemispheric Lateralization of White Matter Pathways in Developmental Dyslexia:
Evidence from Spherical Deconvolution Tractography”. Cortex 76 (March). Elsevier
BV: 51–62. doi:10.1016/j.cortex.2015.12.004.
.CC-BY 4.0 International licensepeer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/237859doi: bioRxiv preprint first posted online Dec. 21, 2017;
... Ongoing discussion centres on whether four [4][5][6] or five [7] groups exist, and what characterises each cluster. Other examples include the identification of protein communities involved in cancer metastasis [8], responder types to cancer treatment [9], Parkinson's disease subtypes [10], brain types [11], and behavioural phenotypes [12][13][14][15][16]. ...
Article
Full-text available
Background Cluster algorithms are gaining in popularity in biomedical research due to their compelling ability to identify discrete subgroups in data, and their increasing accessibility in mainstream software. While guidelines exist for algorithm selection and outcome evaluation, there are no firmly established ways of computing a priori statistical power for cluster analysis. Here, we estimated power and classification accuracy for common analysis pipelines through simulation. We systematically varied subgroup size, number, separation (effect size), and covariance structure. We then subjected generated datasets to dimensionality reduction approaches (none, multi-dimensional scaling, or uniform manifold approximation and projection) and cluster algorithms (k-means, agglomerative hierarchical clustering with Ward or average linkage and Euclidean or cosine distance, HDBSCAN). Finally, we directly compared the statistical power of discrete (k-means), “fuzzy” (c-means), and finite mixture modelling approaches (which include latent class analysis and latent profile analysis). Results We found that clustering outcomes were driven by large effect sizes or the accumulation of many smaller effects across features, and were mostly unaffected by differences in covariance structure. Sufficient statistical power was achieved with relatively small samples (N = 20 per subgroup), provided cluster separation is large (Δ = 4). Finally, we demonstrated that fuzzy clustering can provide a more parsimonious and powerful alternative for identifying separable multivariate normal distributions, particularly those with slightly lower centroid separation (Δ = 3). Conclusions Traditional intuitions about statistical power only partially apply to cluster analysis: increasing the number of participants above a sufficient sample size did not improve power, but effect size was crucial. Notably, for the popular dimensionality reduction and clustering algorithms tested here, power was only satisfactory for relatively large effect sizes (clear separation between subgroups). Fuzzy clustering provided higher power in multivariate normal distributions. Overall, we recommend that researchers (1) only apply cluster analysis when large subgroup separation is expected, (2) aim for sample sizes of N = 20 to N = 30 per expected subgroup, (3) use multi-dimensional scaling to improve cluster separation, and (4) use fuzzy clustering or mixture modelling approaches that are more powerful and more parsimonious with partially overlapping multivariate normal distributions.
... Ongoing discussion centres on whether four (4)(5)(6) or five (7) groups exist, and what characterises each cluster. Other examples include the identification of protein communities involved in cancer metastasis (8), responder types to cancer treatment (9), Parkinson's disease subtypes (10), brain types (11), and behavioural phenotypes (12)(13)(14)(15)(16). Particularly sharp increases can be seen for finite mixture modelling (which includes latent class and latent profile analysis) and k-means. Illustration generated using Bibliobanana (17). ...
Preprint
Full-text available
Background. Cluster algorithms are gaining in popularity in biomedical research due to their compelling ability to identify discrete subgroups in data, and their increasing accessibility in mainstream software. While guidelines exist for algorithm selection and outcome evaluation, there are no firmly established ways of computing a priori statistical power for cluster analysis. Here, we estimated power and classification accuracy for common analysis pipelines through simulation. We systematically varied subgroup size, number, separation (effect size), and covariance structure. We then subjected generated datasets to dimensionality reduction approaches (none, multi-dimensional scaling, or uniform manifold approximation and projection) and cluster algorithms (k-means, agglomerative hierarchical clustering with Ward or average linkage and Euclidean or cosine distance, HDBSCAN). Finally, we directly compared the statistical power of discrete (k-means), “fuzzy” (c-means), and finite mixture modelling approaches (which include latent class analysis and latent profile analysis). Results. We found that clustering outcomes were driven by large effect sizes or the accumulation of many smaller effects across features, and were mostly unaffected by differences in covariance structure. Sufficient statistical power was achieved with relatively small samples (N=20 per subgroup), provided cluster separation is large (Δ=4). Finally, we demonstrated that fuzzy clustering can provide a more parsimonious and powerful alternative for identifying separable multivariate normal distributions, particularly those with slightly lower centroid separation (Δ=3). Conclusions. Traditional intuitions about statistical power only partially apply to cluster analysis: increasing the number of participants above a sufficient sample size did not improve power, but effect size was crucial. Notably, for the popular dimensionality reduction and clustering algorithms tested here, power was only satisfactory for relatively large effect sizes (clear separation between subgroups). Fuzzy clustering provided higher power in multivariate normal distributions. Overall, we recommend that researchers 1) only apply cluster analysis when large subgroup separation is expected, 2) aim for sample sizes of N=20 to N=30 per expected subgroup, 3) use multi-dimensional scaling to improve cluster separation, and 4) use fuzzy clustering or mixture modelling approaches that are more powerful and more parsimonious with partially overlapping multivariate normal distributions.
Article
Full-text available
An enduring aim of research in the psychological and brain sciences is to understand the nature of individual differences in human intelligence, examining the stunning breadth and diversity of intellectual abilities and the remarkable neurobiological mechanisms from which they arise. In this Opinion article, we survey recent neuroscience evidence to elucidate how general intelligence (g) emerges from individual differences in the network architecture of the human brain. The reviewed findings motivate new insights about how network topology and dynamics account for individual differences in g, represented by the Network Neuroscience Theory. According to this framework, g emerges from the small-world topology of brain networks and the dynamic reorganization of its community structure in the service of system-wide flexibility and adaptation.
Article
Full-text available
Understanding typical, healthy brain development provides a baseline from which to detect and characterize brain anomalies associated with various neurological or psychiatric disorders and diseases. Diffusion MRI is well suited to study white matter development, as it can virtually extract individual tracts and yield parameters that may reflect alterations in the underlying neural micro-structure (e.g. myelination, axon density, fiber coherence), though it is limited by its lack of specificity and other methodological concerns. This review summarizes the last decade of diffusion imaging studies of healthy white matter development spanning childhood to early adulthood (4-35 years). Conclusions about anatomical location, rates, and timing of white matter development with age are discussed, as well as the influence of image acquisition, analysis, age range/sample size, and statistical model. Despite methodological variability between studies, some consistent findings have emerged from the literature. Specifically, diffusion studies of neurodevelopment overwhelmingly demonstrate regionally varying increases of fractional anisotropy and decreases of mean diffusivity during childhood and adolescence, some of which continue into adulthood. While most studies use linear fits to model age-related changes, studies with sufficient sample sizes and age range provide clear evidence that white matter development (as indicated by diffusion) is non-linear. Several studies further suggest that maturation in association tracts with frontal-temporal connections continues later than commissural and projection tracts. The emerging contributions of more advanced diffusion methods are also discussed, as they may reveal new aspects of white matter development. Although non-specific, diffusion changes may reflect increases of myelination, axonal packing, and/or coherence with age that may be associated with changes in cognition.
Preprint
Full-text available
Introduction Many developmental disorders are associated with deficits in controlling and regulating behaviour. These difficulties are frequently observed across multiple groups of children including children with diagnoses of attention deficit hyperactivity disorder (ADHD), specific learning difficulties, autistic spectrum disorder, or conduct disorder. The co-occurrence of these behavioural problems across disorders typically leads to comorbid diagnoses and can complicate intervention approaches. An alternative to classifying children on the basis of specific diagnostic criteria is to use a data-driven grouping that identifies dimensions of behaviour that meaningfully distinguish groups of children. Methods The sample consisted of 442 children identified by health and educational professionals as having difficulties in attention, learning and/or memory. The current study applied community clustering, a data-driven clustering algorithm, to group children by similarities across scales on a commonly used rating scale, the Conners-3 questionnaire. Further, the current study investigated if the groups identified by the data-driven algorithm could be identified by white matter connectivity using a structural connectomics approach combined with partial least squares analysis. Results The data-driven clustering yielded three distinct groups of children with symptoms of either: (1) elevated inattention and hyperactivity/impulsivity, and poor executive function, (2) learning problems, and (3) aggressive behaviour and problems with peer relationships. These groups were associated with significant inter-individual variation in white matter connectivity of the prefrontal and anterior cingulate cortex. Conclusion In sum, data-driven classification of executive function difficulties identifies stable groups of children, provides a good account of inter-individual differences, and aligns closely with underlying neurobiological substrates.
Article
Full-text available
Unlabelled: Brain region-specific changes have been demonstrated with a variety of cognitive training interventions. The effect of cognitive training on brain subnetworks in humans, however, remains largely unknown, with studies limited to functional networks. Here, we used a well-established working memory training program and state-of-the art neuroimaging methods in 40 healthy adults (21 females, mean age 26.5 years). Near and far-transfer training effects were assessed using computerized working memory and executive function tasks. Adaptive working memory training led to improvement on (non)trained working memory tasks and generalization to tasks of reasoning and inhibition. Graph theoretical analysis of the structural (white matter) network connectivity ("connectome") revealed increased global integration within a frontoparietal attention network following adaptive working memory training compared with the nonadaptive group. Furthermore, the impact on the outcome of graph theoretical analyses of different white matter metrics to infer "connection strength" was evaluated. Increased efficiency of the frontoparietal network was best captured when using connection strengths derived from MR metrics that are thought to be more sensitive to differences in myelination (putatively indexed by the [quantitative] longitudinal relaxation rate, R1) than previously used diffusion MRI metrics (fractional anisotropy or fiber-tracking recovered streamlines). Our findings emphasize the critical role of specific microstructural markers in providing important hints toward the mechanisms underpinning training-induced plasticity that may drive working memory improvement in clinical populations. Significance statement: This is the first study to explore training-induced changes in the structural connectome using a well-controlled design to examine cognitive training with up-to-date neuroimaging methods. We found changes in global integration based on white matter connectivity within a frontoparietal attention network following adaptive working memory training compared with a nonadaptive comparison group. Furthermore, the impact of different diffusion MR metrics and more specific markers of white matter on the graph theoretical findings was evaluated. An increase in network global efficiency following working memory training was best captured when connection strengths were weighted by MR relaxation rates (influenced by myelination). These results are important for the optimization of cognitive training programs for healthy individuals and people with brain disease.
Article
Full-text available
In their recent meta-analysis of magnetic resonance imaging data from 15 research samples worldwide, Schmaal et al. examined structural differences of nine subcortical brain volumes between 1728 patients with Major Depressive Disorder (MDD) and 7199 healthy participants. In the authors’ univariate analyses, none of the nine volumes was associated with depression severity, and only hippocampal volume was significantly decreased in MDD patients compared to controls, with the largest effect being observed in the recurrent MDD group (difference 1.4%, Cohen’s d of 0.17). The study is the result of a huge collaborative effort, and we commend the authors on their insightful manuscript. However, as the findings present the best empirical evidence currently available, an accurate interpretation of the results is paramount, especially considering the report’s goal to “robustly discriminate MDD patients from healthy controls” (p.1). We therefore add two observations and future research directions.
Article
Full-text available
In this paper we demonstrate a simulation framework that enables the direct and quantitative comparison of post-processing methods for diffusion weighted magnetic resonance (DW-MR) images. DW-MR datasets are employed in a range of techniques that enable estimates of local microstructure and global connectivity in the brain. These techniques require full alignment of images across the dataset, but this is rarely the case. Artefacts such as eddy-current (EC) distortion and motion lead to misalignment between images, which compromise the quality of the microstructural measures obtained from them. Numerous methods and software packages exist to correct these artefacts, some of which have become de-facto standards, but none have been subject to rigorous validation. In the literature, improved alignment is assessed using either qualitative visual measures or quantitative surrogate metrics. Here we introduce a simulation framework that allows for the direct, quantitative assessment of techniques, enabling objective comparisons of existing and future methods. DW-MR datasets are generated using a process that is based on the physics of MRI acquisition, which allows for the salient features of the images and their artefacts to be reproduced. We apply this framework in three ways. Firstly we assess the most commonly used method for artefact correction, FSL's eddy_correct, and compare it to a recently proposed alternative, eddy. We demonstrate quantitatively that using eddy_correct leads to significant errors in the corrected data, whilst eddy is able to provide much improved correction. Secondly we investigate the datasets required to achieve good correction with eddy, by looking at the minimum number of directions required and comparing the recommended full-sphere acquisitions to equivalent half-sphere protocols. Finally, we investigate the impact of correction quality by examining the fits from microstructure models to real and simulated data.
Article
Background A challenge facing clinical neuroscientists is how best to synthesize diverse and sometimes inconsistent evidence for neuropsychological deficits and brain system dysfunction found in psychiatric disorders into models that guide etiological and treatment research. Multiple-pathway models suggest that psychiatric symptoms might arise from pathophysiology in different neural systems. This study tested dual-pathway model predictions for attention-deficit/hyperactivity disorder (ADHD) that reward and executive function cognitive deficits should be related to abnormalities in corresponding functionally specialized neural systems. Methods Behavioral inhibition and preference for immediate rewards were assessed in N = 251 adolescent boys and girls ages 12 to 18 diagnosed with DSM-IV combined-subtype ADHD or non-ADHD control subjects. Following taxometric analyses of test performance, the resulting subgroups were compared on a functional magnetic resonance imaging monetary incentive delay task probing reward anticipation and go/no-go task of motor response inhibition. Results Three ADHD subgroups were identified consistent with different proposed pathways—ADHD with executive function/motor inhibition deficits, ADHD with both executive and reward deficits, and ADHD with relatively normal test performance. Each cognitive domain mapped to different ADHD brain dysfunction features as expected. However, no brain abnormalities were found common to all ADHD subgroups despite the fact they had nearly identical ADHD-related clinical characteristics. Conclusions The results suggest that combined-subtype ADHD is a collection of discrete disorders for which a comparable behavioral end point arises through different neurobiological pathways. The findings raise caution about applying common cause, single-deficit conceptual models to individual ADHD patients and should prompt researchers to consider biologically defined, multifactorial etiological models for other psychiatric diagnoses.
Article
Studies of brain alterations in children with attention-deficit/hyperactivity disorder (ADHD) have shown heterogeneous results. The aims of the current study were to investigate white matter microstructure in children using both categorical and dimensional definitions of ADHD and to determine the functional consequences of observed alterations. In a large single-site sample of children (aged 8-15 years) with ADHD (n=83) and healthy controls (n=122), we used tract-based spatial statistics on diffusion tensor imaging data to investigate whole-skeleton differences of fractional anisotropy (FA), mean, axial, and radial diffusivity (MD, AD, RD), and mode of anisotropy related to ADHD status (categorical) and symptom severity (dimensional). For categorical differences observed, we analyzed their association with cognitive functioning in working memory and inhibition. Compared to healthy controls, children with ADHD showed decreased FA and increased RD in widespread, overlapping brain regions, mainly in corpus callosum (CC) and major tracts in the left hemisphere. Decreased FA was associated with inhibition performance in the participants with ADHD. Using dimensional definitions, greater hyperactivity/impulsivity symptom severity was associated with higher FA also in widespread regions, mainly in CC and major tracts in the right hemisphere. Our study showed white matter alterations to be related to ADHD status and symptom severity in patients. The co-existence of decreased FA and increased RD in the absence of alterations in MD or AD might indicate altered myelination as a pathophysiological factor in ADHD.Neuropsychopharmacology accepted article preview online, 29 September 2016. doi:10.1038/npp.2016.223.