1
Genetic influences on the intrinsic and extrinsic functional organizations of the
1
cerebral cortex
2
3
Running title: GWAS of cerebral cortex functions
4
5
Bingxin Zhao
1
, Tengfei Li
2,3
, Stephen M. Smith
4
, Di Xiong
5
, Yue Yang
5
, Xifeng Wang
5
,
6
Tianyou Luo
5
, Ziliang Zhu
5
, Yue Shan
5
, Zhenyi Wu
1
, Zirui Fan
1
, Heping Zhang
6
, Yun Li
5,7,8
,
7
Jason L. Stein
7,9
, and Hongtu Zhu
3,5
*
8
9
1
Department of Statistics, Purdue University, West Lafayette, IN 47907, USA.
10
2
Department of Radiology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA.
11
3
Biomedical Research Imaging Center, School of Medicine, University of North Carolina at Chapel Hill,
12
Chapel Hill, NC 27599, USA.
13
4
Wellcome Centre for Integrative Neuroimaging, FMRIB, Nuffield Department of Clinical
14
Neurosciences, University of Oxford, Oxford, UK.
15
5
Department of Biostatistics, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA.
16
6
Department of Biostatistics, Yale University, New Haven, CT 06511, USA.
17
7
Department of Genetics, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA.
18
8
Department of Computer Science, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599,
19
USA.
20
9
UNC Neuroscience Center, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA.
21
22
*Corresponding author:
23
Hongtu Zhu
24
3105C McGavran-Greenberg Hall, 135 Dauer Drive, Chapel Hill, NC 27599.
25
E-mail address: htzhu@email.unc.edu Phone: (919) 966-7250
26
27
28
29
30
31
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted July 30, 2021. ; https://doi.org/10.1101/2021.07.27.21261187doi: medRxiv preprint
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.
2
Abstract
1
The human cerebral cortex plays a crucial role in brain functions. However, genetic
2
influences on the human cortical functional organizations are not well understood. Using
3
a parcellation-based approach with resting-state and task-evoked functional magnetic
4
resonance imaging (fMRI) from 40,253 individuals, we identified 47 loci associated with
5
functional areas and networks at rest, 15 of which also affected the functional
6
connectivity during task performance. Heritability and locus-specific genetic effects
7
patterns were observed across different brain functional areas and networks. Specific
8
functional areas and networks were identified to share genetic influences with cognition,
9
mental health, and major brain disorders (such as Alzheimer's disease and schizophrenia).
10
For example, in both resting and task fMRI, the APOE ε4 locus strongly associated with
11
Alzheimer's disease was particularly associated with the visual cortex in the secondary
12
visual and default mode networks. In summary, by analyzing biobank-scale fMRI data in
13
high-resolution brain parcellation, this study substantially advances our understanding of
14
the genetic determinants of cerebral cortex functions, and the genetic links between
15
brain functions and complex brain traits and disorders.
16
17
Keywords: Brain disorders; Brain function; fMRI; GWAS; Mental health; UK Biobank.
18
19
20
21
22
23
24
25
26
27
28
29
30
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted July 30, 2021. ; https://doi.org/10.1101/2021.07.27.21261187doi: medRxiv preprint
3
The human cerebral cortex is the largest part of the human brain and controls complex
1
brain functions. Based on known functional and topographic specializations at different
2
scales, the cerebral cortex of the human brain can be divided into distinct areas and
3
networks, providing insight into the brain's functional architecture
1,2
. To define such brain
4
partitions, a few brain parcellations
1,3-8
have been developed over the past decade
9
. In
5
functional magnetic resonance imaging
10,11
(fMRI), cerebral cortex functions can be
6
evaluated by measuring functional connectivity, correlation of blood-oxygen-level
7
dependent (BOLD) activity, among multiple cortical areas along a given parcellation. In
8
particular, resting-state fMRI captures the intrinsic functional organization of the cortex
9
without any explicit stimuli, whereas task-evoked fMRI measures extrinsic cortical
10
interaction and temporal synchrony in response to a specific task
12,13
. A variety of clinical
11
applications of both task-evoked and resting-state fMRI have revealed changes in brain
12
function in multiple neurological and psychiatric disorders
14-16
, such as schizophrenia
17,18
,
13
Alzheimer’s disease
19
, Parkinson’s disease
20
, autism spectrum disorders
21
, and major
14
depressive disorder (MDD)
22
.
15
16
Twin studies have established that brain functional organizations characterized by resting
17
and task fMRI are moderately heritable
23-30
(heritability range was (0.2,0.6) in a recent
18
review
31
). The narrow sense single-nucleotide polymorphism (SNP) heritability of resting
19
fMRI traits was reported to be around 10% across the entire brain and higher than 30%
20
in some functional regions
32
. The heritability of brain functional traits was typically lower
21
than that of brain structural traits
32-35
. Nevertheless, brain functional traits could more
22
directly connect genetic variations to mechanisms underlying behavioral differences
36
. A
23
few genome-wide association study (GWAS)
32,34,37
have been recently conducted on
24
resting fMRI traits using a whole brain spatial independent component analysis (ICA)
38-40
25
approach. The whole brain ICA is a parcellation-free dimension reduction method that
26
estimates the functional brain regions (i.e., ICA components/regions) directly from the
27
fMRI data. Although ICA is a powerful and popular fMRI tool, it is a data-driven method,
28
which might limit its generalizability and interpretability
12
. For example, the ICA
29
components estimated from training data may form a sample-dependent functional
30
network and it is not clear if they can be generalized to independent datasets. Moreover,
31
it might be difficult to use ICA to compare intrinsic and extrinsic functional architectures,
32
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted July 30, 2021. ; https://doi.org/10.1101/2021.07.27.21261187doi: medRxiv preprint
4
since the ICA components estimated in resting and task fMRI may not be well-aligned.
1
Additionally, ICA attempts to capture major variations in the data. As a result, ICA regions
2
typically have large sizes, limiting their ability to capture high-resolution details of brain
3
functionality. For example, an earlier study
40
defined 55 ICA components in the UK
4
Biobank
41
(UKB) dataset, most of which are distributed across multiple areas and
5
networks
37
.
6
7
In this paper, we used a parcellation-based approach to provide fine-grained details about
8
the genetic architecture of cerebral cortex functional organizations. A recently developed
9
human brain parcellation
1
, which partitioned the cerebral cortex into 360 areas (referred
10
to as the Glasser360 atlas hereafter, Table S1), was used to analyze resting and task fMRI
11
data of 40,253 individuals in the UKB study. The task implemented in the UKB study was
12
an emotional processing task
42,43
, known to robustly activate amygdala and visual
13
systems. The Glasser360 atlas was constructed using high-quality multi-modality data
14
from the Human Connectome Project (HCP
44
) and greatly improved the neuroanatomical
15
resolution of human cerebral cortex annotations. The 360 cortical areas were grouped
16
into 12 functional networks
45
, including four well-known sensory networks (the primary
17
visual, secondary visual, auditory, and somatomotor), four cognitive networks (the
18
cingulo-opercular, default mode, dorsal attention, and frontoparietal), the language
19
network, and three recently identified networks (the posterior multimodal, ventral
20
multimodal, and orbito-affective) (Figs. 1A-B, Fig. S1A). In addition to pairwise functional
21
connectivity among areas, we developed a parcellation-based dimension reduction
22
procedure to generate network level fMRI traits via a combined principal component
23
analysis (PCA) and ICA methods
32
in a training-validation design (Fig. S1B, Methods). The
24
area level functional connectivity pairs within each network and between each pair of
25
networks were aggregated into network level traits. Genetic architectures were examined
26
at both area and network levels for brain functions using these functional connectivity
27
traits. Together, there were 8,531 area level traits and 1,066 network level traits for
28
resting fMRI and 8,531 area level traits and 919 network level traits for task fMRI.
29
Compared to the whole brain ICA-based GWAS
32,34,37
in the prior literature for resting
30
fMRI, the current parcellation-based study 1) enabled the comparison between intrinsic
31
and extrinsic functional architectures using both resting and task fMRI; and 2) uncovered
32
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted July 30, 2021. ; https://doi.org/10.1101/2021.07.27.21261187doi: medRxiv preprint
5
much more and finer detail on the genetic influences on specific functional areas and
1
networks and their genetic links with brain traits and disorders.
2
3
RESULTS
4
Consistency and reproducibility of the cerebral cortex functional organizations
5
In this section, we examined the consistency and reproducibility of functional connectivity
6
using annotations defined in the Glasser360 atlas in the UKB study. As in Glasser, et al.
1
,
7
we first compared the group means of two independent sets of UKB subjects: the UKB
8
phases 1 and 2 data (imaging data released up through 2018
46
, n = 17,374) and the UKB
9
phase 3 data (data released in early 2020, n = 16,852, removing the relatives of subjects
10
in early released data). Figures S2-S3 illustrate the consistent spatial patterns of
11
functional connectivity across the two independent groups. The group mean maps were
12
highly similar, with the correlation across the 64,620 (360 × 359/2) functional connectivity
13
being 0.996 in resting fMRI and 0.994 in task fMRI. These results may suggest that the
14
HCP-trained Glasser360 atlas can provide a set of well-defined and biologically
15
meaningful brain functional traits that are generalizable across datasets.
16
17
Furthermore, we evaluated the reproducibility of the Glasser360 atlas using the repeat
18
scans from the UKB repeat imaging visit (n = 2,771, average time between visits = 2 years).
19
We performed two analyses. The first analysis is to compare the group mean maps of the
20
original imaging visit to those of the repeat visit. Functional connectivity maps were highly
21
consistent between the two visits, with correlation of 0.997 and 0.994 for resting and task
22
fMRIs, respectively (Figs. S4-S5). The second analysis quantified individual-level
23
differences between the two visits. Specifically, we evaluated the reproducibility of each
24
functional connectivity by calculating the correlation between two observations from all
25
revisited individuals. Overall, the average reproducibility was 0.37 (standard error = 0.11)
26
for resting fMRI and 0.30 (standard error = 0.08) for task fMRI (Figs. S6A-B). A few patterns
27
were observed. For example, the reproducibility of within-network connectivity was high
28
in resting fMRI but decreased in task fMRI (Fig. 1C, mean = 0.46 vs. 0.32, P < 2.2 × 10
-16
).
29
During task fMRI, the connectivity within activated functional areas (defined by group-
30
level Z-statistic maps, Supplementary Note) showed higher reproducibility than that
31
within nonactivated areas (Fig. 1D, Fig. S7A, mean = 0.40 vs. 0.30, P < 2.2 × 10
-16
). The
32
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted July 30, 2021. ; https://doi.org/10.1101/2021.07.27.21261187doi: medRxiv preprint