Development of human functional and
structural brain networks in adolescence
and its relevance to psychiatric disorders

Anna-Lena Dorfschmidt

Supervisors: Prof. Edward T. Bullmore &
Prof. Petra E. Vértes

Department of Psychiatry
University of Cambridge

This dissertation is submitted for the degree of
Doctor of Philosophy

Darwin College March 2023





Declaration

I hereby declare that except where specific reference is made to the work of others, the
contents of this dissertation are original and have not been submitted in whole or in part
for consideration for any other degree or qualification in this, or any other university. This
dissertation is my own work and contains nothing which is the outcome of work done in
collaboration with others, except as specified in the text and Preamble. This dissertation
contains fewer than 60,000 words excluding figures, tables, appendices and bibliography.

Anna-Lena Dorfschmidt
March 2023





Development of human functional and structural brain
networks in adolescence and its relevance to psychiatric

disorders

Anna-Lena Dorfschmidt

The human brain undergoes various phases of active development during the lifes-
pan. While these neurodevelopmental processes are fundamental to the emergence of new
cognitive and social capacities, they also coincide with a period of increased risk of neu-
ropsychiatric disorders, which generally have their highest rates of clinical incidence in the
first two decades. Since many neuropsychiatric disorders display sex differences in both
prevalence or clinical presentation, this raises the question of whether there are underlying
sex differences in processes of adolescent brain development. In this thesis, functional and
structural magnetic resonance imaging (MRI) is used to map normative brain development, in
adolescence and later life, which might differentially predispose men and women to different
levels of risk for adolescent and adult mental illness.

First, Chapter 1 reviews relevant research on understanding developmental changes in
the brain during adolescence, focusing on prior studies of normative sexual differentiation of
neurodevelopmental trajectories, and vulnerabilities associated with developmental changes.

Chapter 2 investigates whether there are sex differences in normative adolescent develop-
ment of functional connectivity networks, using an accelerated longitudinal cohort of healthy
adolescents aged 14-25 years (N=298), comprising 2 or 3 repeated scans on most participants.
Sexually divergent development of functional connectivity was identified in the default mode
network, limbic cortex, and subcortical structures. In these regions, females were shown
to have a more “disruptive” pattern of development, whereby weak functional connectivity
at age 14 became stronger during adolescence, specifically in a cortico-subcortical system
including many areas of the default mode network. Using open data on whole genome tran-
scription at multiple sites in adult post mortem brains (provided by the Allen Brain Institute),
this fMRI-derived map of sexually divergent brain network development was found to be
spatially co-located with brain regions where transcription was enriched for genes on the X
chromosome and neurodevelopmentally relevant genes.

Chapter 3 starts from the hypothesis that the known sex difference in the prevalence of
major depressive disorder (MDD), with increased rates of diagnosis in adolescent females
compared to males, could be the psychological or clinical representation of underlying sex



vi

differences in adolescent brain network development. To test this hypothesis, the sexually
differentiated fMRI network identified in the previous chapter was further contextualized.
The fMRI-derived map of sexually divergent brain network development was found to be co-
located with prior loci of reward-related brain activation; a map of functional dysconnectivity
in major depressive disorder derived from a prior, independent case-control study of adult
MDD; and an adult brain gene transcriptional profile enriched for MDD risk genes, as defined
by prior genome-wide association studies of MDD. These results collectively suggested
that normative sexual divergence in adolescent development of a cortico-subcortical brain
functional network was psychologically, anatomically and genetically relevant to depression.

Chapter 4 reviews literature on similarity-based structural brain networks. Subsequently,
Chapter 5 investigates adolescent changes in structural brain network development using
morphometric similarity networks derived from the same accelerated longitudinal cohort of
healthy adolescents previously used for analysis of functional network development. Mor-
phometric similarity was found to increase during adolescence in insula and limbic regions
and to decrease elsewhere in the brain. This profile of decreasing morphometric similarity, or
increasing dissimilarity, was associated with the well-known adolescent process of cortical
shrinkage, i.e., reduced macro-structural measures of cortical thickness, and with increased
magnetization transfer, a micro-structural measure of intra-cortical myelination. Regional
nodes of the morphometric similarity networks that became more dissimilar, putatively
more differentiated in terms of their cyto- and myelo-architectonics during adolescence,
were also found to de-couple from brain functional connectivity, suggesting that increasing
morphometric dissimilarity may reflect adolescent development of functional independence.

In an effort to move from group level to subject-specific analyses, and acknowledging
that brain development is not restricted to adolescence but is a continuous process throughout
life, in Chapter 6 a total of 41 prior studies, including a total of 90,000 structural MRI scans,
were aggregated to estimate lifespan trajectories of normative subcortical development from
180 days post conception to 100 years of age. This analysis identified novel milestones
of subcortical volume development; in particular a set of subcortical regions was defined
that reached peak grey matter volume during adolescence. Furthermore, subject-specific
deviations from normative, non-linear neurodevelopmental trajectories? were derived and
used to estimate case-control differences in subcortical volume across the lifespan in mul-
tiple neuropsychiatric disorders, demonstrating the potential clinical applications of these
normative subcortical growth charts.

In Chapter 7, these new experimental results on adolescent and life-span development
of functional and structural brain networks, and subcortical grey matter volume were sum-



vii

marised and drawn together, highlighting how these insights are aligned with each other and
with the existing scientific literature on brain development, sexual differentiation and risk of
psychiatric disorders.





Preamble

Work presented in Chapter 2 and Chapter 3 is adapted from material previously published
in a peer-reviewed journal. Further, work presented in Chapter 4 - 6 is being prepared
for publication in peer reviewed journals, and has been presented or will be presented at
scientific conferences. Analyses conducted within these studies, were principally designed
and carried out by me. Contributions of all co-authors are listed below each reference.

Chapters 2-3

Dorfschmidt, L., Bethlehem, R. A., Seidlitz, J., Váša, F., White, S. R., Romero-
Garcìa, R., Kitzbichler, M. G., Aruldass, A. R., Morgan, S. E., Goodyer, I. M., Fonagy,
P., Jones, P. B., Dolan, R. J., NSPN Consortium, Harrison, N. A., Vértes, P. E., &
Bullmore, E. T. (2022). Sexually divergent development of depression-related brain
networks during healthy human adolescence. Science advances, 8(21), eabm7825.
https://doi.org/10.1126/sciadv.abm7825

Data analysis: L.D. Research design: L.D., P.E.V., and E.T.B. NSPN MRI data prepro-
cessing: F.V., R.R-G., M.G.K., P.E.V. COBRE data preprocessing: S.E.M. BioDep data
preprocessing: A.R.A, M.G.K. Contribution of analytical tools: J.S., R.A.B., S.E.M.,
M.G.K., and A.R.A. Advice on statistical methods: S.R.W. NSPN study design: E.T.B.,
P.B.J., R.J.D., I.M.G., and P.F. BIODEP study design: E.T.B. and N.A.H. Manuscript
writing: L.D., E.T.B., and P.E.V.

Chapter 5

Dorfschmidt, L., R.A.I. Bethlehem, J. Seidlitz, F. Váša, S.R. White, R. Romero-
García, P.E. Vértes, E.T. Bullmore. Adolescent morphometric similarity development.
2022. Poster presentation at the Annual Meeting of the Organization for Human Brain
Mapping



x

Data analysis: L.D. MRI data preprocessing: F.V., R.R-G., P.E.V. Research design:
L.D., R.A.B., J.S, P.E.V., and E.T.B. Advice on statistical methods: S.R.W. NSPN study
design: E.T.B.

Chapter 6

Dorfschmidt, L., P.E. Vértes, E.T. Bullmore, A. Alexander-Bloch, R.A.I. Bethlehem,
J. Seidlitz. Lifespan development of subcortical regions. 2023. Annual Meeting of the
Organization for Human Brain Mapping

Data analysis: L.D. MRI data preprocessing: L.D., R.A.B., J.S. Code development:
L.D., S.R.W., R.A.B. Research design: L.D., R.A.B., J.S, A.A.B, P.E.V., and E.T.B.

Due to these contributions, work presented in Chapters 2-6 is presented in the plural
form of a first-person narrative (“we”), while the Introduction and Summary are presented
in the singular form (“I”). The above contributions do not affect the status of this thesis as
being principally my own work.



Acknowledgements

This thesis would not have been possible without the support of many colleagues, friends
and family.

First and foremost, the journey of this PhD had its start far from Cambridge: I would
like to thank my parents, who, when 19-year-old-me told them that she wanted to move to
Guatemala instead of going to med school, simply said: “If that is what your heart says, you
need to do it.” You showed me an approach to decision making that was as wholesome as
anyone could ever dream of.

Thank you also, to Johann Kruschwitz, Henrik Walter and Jon Clayden who showed me
the ropes of science and encouraged me to embark into the adventure that doctoral research
is – I would not have dared to take this path without your support.

I am immensely grateful to both my supervisors, Ed Bullmore and Petra Vértes, for their
scientific guidance. Ed – thank you for mentorship, your scientific rigour and the clarity you
can bring into a project. Petra – I am immensely grateful for the dedication with which you
supervise not only scientifically, but also personally. It was a pleasure to work with such a
power-woman who acts as a role-model for what science can look like. Thank you both.

My deepest gratitude goes to Richard Bethlehem, Jakob Seidlitz and František Váša –
you showed me the beauty of collaborative science. I could not have been luckier than to
have your scientific and moral guidance throughout this PhD, that simply would not have
been finished without you.

To past and present BMU friends and colleagues – Rafa, Jordi, Sarah, Manfred, Ray, Isaac,
Will, Sofia, Athina, Matt, Chats, Ricky, Elise – thank you for your support and your ideas.
A particular thanks to Eva-Maria Stauffer – without you, the start of this PhD (and some
middle bits) would have been incredibly lonely. To Saashi Bedford - colleague-housemate,
for your delicious bread, but more importantly for your time and thoughts when things were
not going well. And lastly to Simon White – for making sure every science catch-up started
with a where-are-you-emotionally.



xii

To the science friends I made along the way, in particular Jakub Vohryzek, Romy Lorenz,
and Katja Schüler – I have such respect for you and am truly grateful for your friendship.

To my friends in Cambridge and beyond – thank you for being there when I needed some
time off science. To the climbing crew, in particular Julia, Alex H., Jane Darby, Rob, Edo
and Will – life outdoors is good with you.

An meine Freunde, nah und (oft) fern: danke für Telefonate, Umarmungen und Eure
Nähe, die mich trägt.

Finally, to Alex – I cannot wait for the after.



Table of Contents

List of Figures xix

List of Tables xxiii

List of Abbreviations xxv

1 Introduction 1

1.1 Brain networks in the context of maturation and disease . . . . . . . . . . . 1

1.1.1 The brain as a complex network . . . . . . . . . . . . . . . . . . . 1

1.1.2 Magnetic Resonance Imaging . . . . . . . . . . . . . . . . . . . . 2

1.1.3 MRI-derived subject-specific brain network construction . . . . . . 4

1.1.4 Brain network topology and analysis . . . . . . . . . . . . . . . . . 7

1.2 Brain development throughout the lifespan . . . . . . . . . . . . . . . . . . 9

1.3 Adolescent brain development . . . . . . . . . . . . . . . . . . . . . . . . 10

1.3.1 Structural brain development . . . . . . . . . . . . . . . . . . . . . 11

1.3.2 Functional brain development . . . . . . . . . . . . . . . . . . . . 12

1.4 Sex differences in brain development . . . . . . . . . . . . . . . . . . . . . 13

1.4.1 Sex differences in structural MRI . . . . . . . . . . . . . . . . . . 14

1.4.2 Sex differences in resting state functional magnetic resonance imag-
ing (rsfMRI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.5 Vulnerabilities during development . . . . . . . . . . . . . . . . . . . . . . 17

1.6 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20



xiv Table of Contents

2 Sex differences in adolescent development of functional connectivity 23

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.2.1 Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.2.2 Residual motion correction . . . . . . . . . . . . . . . . . . . . . . 31

2.2.3 Sex stratified analysis of developmental parameters . . . . . . . . . 32

2.2.4 Age ⇥ sex interaction model . . . . . . . . . . . . . . . . . . . . . 36

2.2.5 Spatial auto-correlation (spin-tests) . . . . . . . . . . . . . . . . . 37

2.2.6 Sensitivity analyses . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.2.7 Gene enrichment analyses . . . . . . . . . . . . . . . . . . . . . . 43

2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.3.1 Sex differences in parameters of adolescent development of global
mean degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.3.2 Sex differences in maturational index . . . . . . . . . . . . . . . . 49

2.3.3 X-chromosome and developmental gene enrichment . . . . . . . . 52

2.3.4 Non-linear effects of age . . . . . . . . . . . . . . . . . . . . . . . 54

2.3.5 Sensitivity to alternative modelling strategies . . . . . . . . . . . . 56

2.3.6 Sensitivity to alternative motion-correction strategies . . . . . . . . 57

2.3.7 Sensitivity of results to intra-cranial volume (ICV) and global func-
tional connectivity (FC) . . . . . . . . . . . . . . . . . . . . . . . 59

2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3 Co-location of adolescent sex differences with major depression 65

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.1.1 Adolescent depression . . . . . . . . . . . . . . . . . . . . . . . . 65

3.1.2 Summary and hypotheses . . . . . . . . . . . . . . . . . . . . . . 67

3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.2.1 Adolescent functional connectivity sex difference . . . . . . . . . . 68



Table of Contents xv

3.2.2 Psychological co-location with depression . . . . . . . . . . . . . . 68

3.2.3 Anatomical co-location with depression . . . . . . . . . . . . . . . 69

3.2.4 Gene enrichment analysis . . . . . . . . . . . . . . . . . . . . . . 71

3.2.5 Diagnostic specificity . . . . . . . . . . . . . . . . . . . . . . . . . 73

3.2.6 Robustness of results to alternative pre-processing strategies . . . . 75

3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.3.1 Anatomical and psychological co-location with depression . . . . . 75

3.3.2 Celltype-specific and major depressive disorder (MDD) risk gene
enrichment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.3.3 Diagnostic specificity . . . . . . . . . . . . . . . . . . . . . . . . . 80

3.3.4 Robustness of results to alternative processing strategies . . . . . . 82

3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.4.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4 A review of cortical similarity and connectivity networks 87

4.1 From histological stains to whole-brain connectomics . . . . . . . . . . . . 87

4.2 Approaches to structural brain network construction . . . . . . . . . . . . . 88

4.2.1 Diffusion weighted imaging . . . . . . . . . . . . . . . . . . . . . 88

4.2.2 Inter-subject structural covariation . . . . . . . . . . . . . . . . . . 88

4.2.3 Intra-subject structural similarity . . . . . . . . . . . . . . . . . . . 91

4.3 Mapping to cytoarchitechtonics and tractography . . . . . . . . . . . . . . 93

4.4 Predicting cognitive and behavioral outcomes . . . . . . . . . . . . . . . . 94

4.5 Bridging from micro to macro scales . . . . . . . . . . . . . . . . . . . . . 94

4.6 Identifying neuroanatomical differences in health and disease . . . . . . . . 95

4.7 Clinical applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.8 Leveraging multimodal imaging . . . . . . . . . . . . . . . . . . . . . . . 96

4.9 Tracking developmental changes in brain anatomy . . . . . . . . . . . . . . 98

4.10 Future directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99



xvi Table of Contents

5 Adolescent morphometric similarity development 103

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5.1.1 Adolescent brain development . . . . . . . . . . . . . . . . . . . . 103

5.1.2 Structural network studies of adolescent development using MRI . . 104

5.1.3 Functional and metabolic correlates of structural network development105

5.1.4 Summary and Hypotheses . . . . . . . . . . . . . . . . . . . . . . 106

5.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.2.1 Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.2.2 Structural MRI acquisition and pre-processing . . . . . . . . . . . 107

5.2.3 Functional magnetic resonance imaging (FMRI) acquisiton and pre-
processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

5.2.4 Morphometric feature estimation and quality control . . . . . . . . 108

5.2.5 Modeling of developmental change in morphometric features . . . 110

5.2.6 Adolescent changes in morphometric similarity . . . . . . . . . . . 110

5.2.7 Co-location with adolescent changes in functional diversity . . . . . 112

5.2.8 Adolescent changes in structure-function coupling . . . . . . . . . 113

5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.3.1 Analyzable sample . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.3.2 Adolescent changes in global and regional MRI metrics . . . . . . 115

5.3.3 Adolescent change in morphometric similarity . . . . . . . . . . . 119

5.3.4 Neurobiological and psychological context of adolescent changes in
anatomical connectomes . . . . . . . . . . . . . . . . . . . . . . . 124

5.3.5 Adolescent development of structure-function coupling . . . . . . . 126

5.3.6 Exploratory analysis of lifespan changes in morphometric similarity 130

5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

6 Beyond human adolescence: lifespan trajectories 137

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137



Table of Contents xvii

6.1.1 Normative modelling . . . . . . . . . . . . . . . . . . . . . . . . . 138

6.1.2 Lifespan development of subcortical regions . . . . . . . . . . . . 138

6.1.3 Subcortical volume differences in atypical development . . . . . . 139

6.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

6.2.1 Aggregated dataset . . . . . . . . . . . . . . . . . . . . . . . . . . 140

6.2.2 MRI pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . 141

6.2.3 Lifespan trajectories . . . . . . . . . . . . . . . . . . . . . . . . . 142

6.2.4 GAMLSS models . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

6.2.5 Developmental milestones . . . . . . . . . . . . . . . . . . . . . . 145

6.2.6 Centile score estimation . . . . . . . . . . . . . . . . . . . . . . . 146

6.2.7 Case-control differences in centile scores . . . . . . . . . . . . . . 146

6.2.8 Quality control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

6.2.9 Leave-one-study-out . . . . . . . . . . . . . . . . . . . . . . . . . 148

6.2.10 Bootstrap analyses . . . . . . . . . . . . . . . . . . . . . . . . . . 149

6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

6.3.1 Lifespan development of subcortical structures . . . . . . . . . . . 149

6.3.2 Lifespan cerebellar and corpus callosum development . . . . . . . 152

6.3.3 Sensitivity of results to image quality . . . . . . . . . . . . . . . . 154

6.3.4 Sensitivity of trajectories to specific studies . . . . . . . . . . . . . 156

6.3.5 Bootstrapping studies of reliability and stability . . . . . . . . . . . 158

6.3.6 Case-control differences in subcortical volumes . . . . . . . . . . . 160

6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

7 Summary and concluding remarks 169

7.1 Summary of findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

7.2 Convergent themes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

7.2.1 Brain development during adolescence - and beyond . . . . . . . . 170

7.2.2 Sex differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171



xviii Table of Contents

7.2.3 Cortex vs subcortex . . . . . . . . . . . . . . . . . . . . . . . . . . 172

7.2.4 Vulnerabilities during development . . . . . . . . . . . . . . . . . 173

7.3 A note on open science . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

7.4 Future directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

References 179

Appendix A Supplementary Tables 221

Appendix B Supplementary Figures 237

Appendix C Supplementary Text 239

C.1 Primary datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

C.2 GAMLSS models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

C.3 Data, code, and image availability . . . . . . . . . . . . . . . . . . . . . . 253



List of Figures

1.1 Complex networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Brain network estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3 Basic graph metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.4 Microscopic sex differences . . . . . . . . . . . . . . . . . . . . . . . . . 15

1.5 Diagnosis of neuropsychiatric disorders . . . . . . . . . . . . . . . . . . . 19

1.6 The value of longitudinal data . . . . . . . . . . . . . . . . . . . . . . . . 20

2.1 Two modes of adolescent FC development . . . . . . . . . . . . . . . . . . 24

2.2 Modes of adolescent development . . . . . . . . . . . . . . . . . . . . . . 26

2.3 Dropout regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.4 Sex differences in framewise displacement . . . . . . . . . . . . . . . . . . 31

2.5 Effect of head motion (FD) on functional connectivity (FC) . . . . . . . . . 32

2.6 Distribution of random effects . . . . . . . . . . . . . . . . . . . . . . . . 36

2.7 FD regression of functional connectivity by sex sample . . . . . . . . . . . 40

2.8 Simplified model of partial least squares analysis . . . . . . . . . . . . . . 44

2.9 All baseline connectivity FC14 plots . . . . . . . . . . . . . . . . . . . . . 46

2.10 All baseline connectivity FC14 plots . . . . . . . . . . . . . . . . . . . . . 47

2.11 Significance of sex difference in adolescent rate of change FC14�26 . . . . . 48

2.12 Sex-specific maturational index . . . . . . . . . . . . . . . . . . . . . . . . 49

2.13 Trends in disruptive and conservative development of connectivity . . . . . 50

2.14 Significance of sex difference in maturational index (MI) . . . . . . . . . . 51



xx List of Figures

2.16 Relationships between gene expression profiles and sexually divergent ado-
lescent brain development . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2.17 Enrichment for X-chromosome and developmental genes . . . . . . . . . . 54

2.18 Non-linear effects of age on FC . . . . . . . . . . . . . . . . . . . . . . . . 55

2.19 Age ⇥ sex interaction model . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.20 Sensitivity of key findings to motion . . . . . . . . . . . . . . . . . . . . . 58

2.21 Sensitivity of key findings to alternative modeling strategies . . . . . . . . 60

3.1 Anatomical and psychological co-location with depression . . . . . . . . . 76

3.2 Neurosynth analysis of positive DMI regions . . . . . . . . . . . . . . . . . 77

3.3 Sexually divergent, disruptive brain systems are co-located with brain tissue
transcripts enriched for cell type-specific and MDD risk-related genes . . . 79

3.4 Illustrative correlations between DMI and three genes (somatostatin (SST),
neuropeptide Y (NPY) and cortistatin (CORT)): . . . . . . . . . . . . . . . 80

3.5 Diagnostic specificity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3.6 Robustness of MDD co-location results . . . . . . . . . . . . . . . . . . . 83

4.1 Cortical similarity network estimation . . . . . . . . . . . . . . . . . . . . 90

4.2 Cytoarchitectonic similarity predicts projection density . . . . . . . . . . . 92

4.3 Cortical depth profiles for magnetization transfer (MT) . . . . . . . . . . . 93

5.1 Global morphometric outliers . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.2 Local morphometric outliers . . . . . . . . . . . . . . . . . . . . . . . . . 110

5.3 Estimation of age effects on morphometric similarity . . . . . . . . . . . . 111

5.4 Structure-function coupling estimation . . . . . . . . . . . . . . . . . . . . 113

5.5 Adolescent changes in global macro-structural and micro-structural magnetic
resonance imaging (MRI) metrics . . . . . . . . . . . . . . . . . . . . . . 116

5.6 Adolescent changes in regional macro-structural and micro-structural MRI
metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118



List of Figures xxi

5.7 Adolescent changes in macro-structural and micro-structural MRI metrics
corrected for global effects . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5.8 Adolescent changes in morphometric similarity . . . . . . . . . . . . . . . 121

5.9 Cytoarchitectonic class-specific changes in morphometric similarity . . . . 122

5.10 Sex effects on adolescent changes in morphometric similarity . . . . . . . . 123

5.11 Divergent profile of morphometric similarity . . . . . . . . . . . . . . . . . 124

5.12 Neurobiological relevance of adolescent changes in morphometric similarity
network (MSN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.13 Psychological relevance of adolescent changes in MSN . . . . . . . . . . . 126

5.14 Morphometric dissimilarity was associated with functional participation . . 127

5.15 Adolescent development of structure-function coupling . . . . . . . . . . . 128

5.16 Structure-function coupling in relation to adolescent changes in MSN . . . 129

5.17 Morphometric dissimilarity partially explains adolescent age-related changes
in functional participation . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

5.18 Morphometric similarity network development over the life-cycle . . . . . 132

6.1 Aggregated lifespan MRI dataset: . . . . . . . . . . . . . . . . . . . . . . 140

6.2 Dataset demographics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

6.3 Normative modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

6.4 Lifespan development of subcortical structures: . . . . . . . . . . . . . . . 151

6.5 Lifespan development of cerebellum and corpus callosum: . . . . . . . . . 153

6.6 Developmental milestones: . . . . . . . . . . . . . . . . . . . . . . . . . . 154

6.7 Age-related variation in image quality: . . . . . . . . . . . . . . . . . . . . 155

6.8 Associations between centile scores and MRI scan quality: . . . . . . . . . 156

6.9 Leave-one-study-out sensitivity analysis: . . . . . . . . . . . . . . . . . . . 157

6.10 Quality control - study offsets: . . . . . . . . . . . . . . . . . . . . . . . . 159

6.11 Case-control differences in centile scores of subcortical volumes for multiple
neuropsychiatric disorders: . . . . . . . . . . . . . . . . . . . . . . . . . . 161

6.12 Psychopathology centile score differences by subcortical structure: . . . . . 162



xxii List of Figures

B.1 Effect of FD on FC in the motion-matched sample . . . . . . . . . . . . . . 238



List of Tables

2.1 Sample of healthy adolescent participants with fMRI data from the NSPN
cohort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.2 NSPN fMRI motion-matched sample overview . . . . . . . . . . . . . . . 39

3.1 MDD case-control sample charateristics . . . . . . . . . . . . . . . . . . . 71

4.1 Previously published studies using morphometric similarity networks. The
studies were identified by a pubmed search in January 2022 using the search
terms morphometric similarity network or MSN. . . . . . . . . . . . . . . 101

5.1 Neuroscience in Psychiatry Network (NSPN) structural MRI data sample
overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

5.2 Age and sex effects on individual morphometric features . . . . . . . . . . 117

6.1 Subject numbers for diagnostic groups of cases and controls . . . . . . . . 147

A.1 Sex differences in fMRI literature . . . . . . . . . . . . . . . . . . . . . . 225

A.3 ROIs with significantly different DMI . . . . . . . . . . . . . . . . . . . . 232

A.4 Significant regional effects of age on morphometric similarity . . . . . . . 233

A.5 Processing pipelines by dataset . . . . . . . . . . . . . . . . . . . . . . . . 235

A.2 Human Connectome Project (HCP) parcellation regions . . . . . . . . . . . 236





List of Abbreviations

ABCD adolescent brain cognitive development study

ACC anterior cingulate

AD Alzheimer’s disease

ADHD attention deficit hyperactivity disorder

AFNI Analysis of Functional NeuroImages

AHBA Allen Human Brain Atlas

AIC Akaike information criterion

ANOVA analysis of variance

ANX anxiety and phobia

ASD autism spectrum disorder

Ast astrocytes

BIC Bayesian information criterion

BIDS Brain Imaging Data Format

BioDep Biomarkers in Depression study

BOLD blood-oxygen-level-dependent

CN healthy control

CNV copy number variation

CORT cortistatin



xxvi List of Abbreviations

CRP C-reactive protein

CSEA cell-specific enrichment analysis

CT cortical thickness

DF degrees of freedom

DK Desikan-Killiany anatomical atlas

DLPFC dorsolateral prefrontal cortex

DMN default mode network

DSM-5 Diagnostic and Statistical Manual of Mental Disorders fifth edition

DTI diffusion tensor imaging

DWI diffusion weighted imaging

EI Euler index

End endothelial cells

EPI echo-planar imaging

Ex exitatory neurons

FA fractional anisotropy

FC functional connectivity

FD framewise displacement

FDA U.S. Food and Drug Administration

FDR false discovery rate

FI folding index

fMRI functional magnetic resonance imaging

GAMLSS generalized additive models for location scale and shape

FCglobal global functional connectivity

GM grey matter volume



xxvii

Gran cerebellar granule cells

GSR global signal regression

GWAS genome wide association study

HAM-D Hamilton Rating Scale for Depression

HARDI high-angular resolution diffusion-weighted image

HCP Human Connectome Project

IC intrinsic curvature

ICV intra-cranial volume

In inhibitory neurons

IQ intelligence quotient

LME linear mixed effects model

LOSO leave-one-study-out

MAD median absolute deviation

MC mean curvature

MCI mild cognitive impairment

MD mean diffusivity

MDD major depressive disorder

ME-ICA multi echo independent component analysis

MI maturational index

Mic microglia

MPM multi-parametric mapping

MRI magnetic resonance imaging

MSN morphometric similarity network

MT magnetization transfer



xxviii List of Abbreviations

NIH U.S. National Institute of Health

NODDI neurite orientation dispersion and density imaging

NPY neuropeptide Y

NSPN Neuroscience in Psychiatry Network

OCD obsessive compulsive disorder

Oli oligodendrocytes

OPC oligodendrocyte precursor cells

PCW post conception week

Per pericytes

PLS partial least squares regression

PLS1 first partial least squares regression component

PMA post menstrual age

PRS polygenic risk score

Purk purkinje cells

RF radio frequency

ROI region of interest

rsfMRI resting state functional magnetic resonance imaging

SA surface area

SCID structured clinical interview for DSM-V

SCN structural covariance network

SCZ schizophrenia

SES socio-economic status

sMRI structural magnetic resonance imaging

SNIP single nucleotid polymorphism



xxix

SST somatostatin

TE echo time

TR repitition time

tWM total white matter volume

UKB UK Biobank

VMPFC ventromedial prefrontal cortex

WHO World Health Organization

WM white matter volume





Chapter 1

Introduction

1.1 Brain networks in the context of maturation and disease

Over the course of the lifespan, the human brain undergoes various periods of increased
plasticity, during which it changes in both structure and function. These maturational periods
are often also a time of elevated incidence of a variety of neuropsychiatric disorders, which
are increasingly understood to arise in the context of atypical brain development (Paus
et al., 2008). One such period of neurodevelopmental activity is adolescence, a time of vast
social and cognitive development, during which many psychiatric disorders are first diag-
nosed (Kessler et al., 2005; Paus et al., 2008). Many such disorders display sex differences,
both in prevalence, as well as clinical expression. This thesis assesses maturational changes
in brain structure and function during adolescence in particular, and during the lifespan more
broadly, in relation to atypical development associated with mental health disorders. It further
explores whether observed sex differences in psychiatric disorders may be the expression
of underlying sex differences in brain maturation. A promising avenue for exploring brain
changes in structure and function comes from the field of graph theory and network science,
which has allowed researchers to derive “brain networks” from neuroimaging data and
analyse changes in their topology during development and in disease.

1.1.1 The brain as a complex network

The brain can be thought of as a network, organized across multiple spatio-temporal scales,
from genes and molecules, through microscopic neuronal and other cellular circuits, to
macroscopic brain regions and whole-brain systems or connectomes (Betzel and Bassett,



2 Introduction

2017; Bullmore and Sporns, 2009; Fornito et al., 2016). Networks have been used to represent
complex interactions between relative features of a system in a multitude of fields, including
psychology (Epskamp et al., 2018; Schueler et al., 2021), molecular biology (Szklarczyk
et al., 2015), and genetics (Zhang and Horvath, 2005). In general, networks consist of
multiple nodes that are connected via edges (Fig. 1.1B). Every network can be represented
as a matrix A, the adjacency matrix, which in the context of brain networks is often called
a “connectivity matrix” (Fig. 1.1A). Each of the entries, Ai, j, in the matrix describes a link
or edge between two nodes i and j (Fig. 1.1B). A network can be unweighted, or binary, in
which case each edge indicates the presence of a connection between two nodes (Fig. 1.1B);
or weighted, in which case each edge has a continuous variable that describes the strength
of connection between two nodes. As an example, in spatially embedded networks like the
brain, edges can be weighted by the physical distance between nodes, which may act as a
proxy measure of the “wiring cost” of the connnection between them (Fig. 1.1C); equally,
edges may be weighted by the strength of the co-fluctuation between two nodes’ functional
activation, providing a measure of co-activation (see below for details). Lastly, networks can
also be directed (Fig. 1.1D), meaning that edges (and their weights) have a direction, such
that Ai, j is not necessarily equal to A j,i.

Fig. 1.1 Complex networks: Introduction to networks. (A) An adjacency matrix A describes
the edge weights between a set of nodes, such that Ai, j is the edge weight between node i
and node j. (B) Networks consist of nodes which are connected to each other by edges. In
an unweighted network, the edges indicate the presence of a connection between two nodes.
(C) A network can be weighted, in which case edge weights indicate the strength of the
connection betwen two nodes. (D) In directed networks, edges are directional, such that the
edge weight Ai, j is not necessarily the same as A j,i.

1.1.2 Magnetic Resonance Imaging

So far, there is only one organism for which a brain network has been fully mapped on the
neuronal level - the nematode worm C. elegans with its 302 neurons (White et al., 1986).



1.1 Brain networks in the context of maturation and disease 3

Fully mapping the vast complexity of the human brain, consisting of about 80 billion neurons,
is currently not possible at equivalent microscopic scale, thus motivating research into
mapping whole brain networks or connectomes, resolved at the macroscopic scale of cortical
areas and subcortical nuclei, which can be resolved by non-invasive imaging techniques in
humans (and other species). Magnetic resonance imaging (MRI) is a non-invasive approach
to mapping brain structure and function that allows researchers to study complex network
organization in the brain.

Multiple imaging modalities, so-called sequences, exist to capture different aspects of
brain anatomy and function (Bernstein et al., 2004). In general, MRI images exploit the
differing magnetic properties of different tissue types and states in the brain. Put simply, the
scanner generates a large magnetic field which leads protons in the body to align with the
magnetic field . During scanning, a radio frequency pulse is applied which throws protons
out of alignment with the main magnetic field. It is possible to measure the time is takes
them to re-align with the magnetic field.

Structural MRI images are typically derived using a T1-weighted or T2-weighted se-
quence, which measure the longitudinal relaxation time of tissue following a radio frequency
pulse that aligns protons to the transverse plane (Bernstein et al., 2004). The time it takes
protons to re-align is effected by the density of fat and water in the tissue, thus the signal
strength can be exploited to differentiate between grey and white matter in the brain. Image
processing pipelines rely on this difference in contrast to segment grey from white matter,
and generate cortical surface meshes. Subsequently, pipelines derive cortical morphometric
features, including for example cortical thickness (CT) (estimated as the distance between
two corresponding points on the pial and grey/white surfaces), surface area (SA) (estimated
as the local area of a triangle, or vertex, on the surface mesh), and grey matter volume (GM)
(a combined estimate of thickness and area).

In recent years, functional MRI networks have evolved as a powerful tool to investigate
intrinsic brain activity, that is spontaneous fluctuations in brain activity independent of a
cognitive or sensory stimulus (van den Heuvel and Hulshoff Pol, 2010), in health and disease.
The blood-oxygen-level-dependent (BOLD) contrast measured by resting state functional
magnetic resonance imaging (rsfMRI) is reflective of local blood oxygenation changes
coupled to neuronal activity (Logothetis and Wandell, 2004). Commonly, the BOLD signal
is bandpass filtered and only low-frequency oscillations at between 0.01 Hz and 0.1 Hz are
retained (Achard et al., 2006; Biswal et al., 1995), since (i) network fluctuation are thought
to be maximally observed at low frequencies, thus filtering should increase statistical power;
(ii) filtering may help to reduce the influence of noise on the retained signal, which is thought



4 Introduction

to occur in higher-frequency oscillations; and lastly (iii) low-frequency drifts due to scanner
noise are removed through the high-pass component of the filtering process. It is worth
mentioning that a growing body of literature is investigating high-frequency contributions
to functional connectivity measures, in particular providing evidence for relevant signal in
higher frequencies during tasks (Craig et al., 2018). For the time being, in this thesis, which
uses resting state fMRI, not task data, the traditional approach is employed.

1.1.3 MRI-derived subject-specific brain network construction

The nodes of MRI-derived brain networks are typically locations in the brain, and the edges
are measures of connectivity between two distant locations. Different methods exist for
defining nodes and edges. While diffusion weighted imaging (DWI) (Hagmann et al., 2006)
and structural magnetic resonance imaging (sMRI) (Lerch et al., 2006) data can be used to
derive networks of anatomically connected areas and nuclei, functional magnetic resonance
imaging (fMRI) (Salvador et al., 2005) images can be used to estimate functional connectivity
between nodes and thus the resulting adjacency matrix is often described as a functional
connectivity matrix.

Here, I provide a brief overview of structural brain networks in the context of brain
network analysis. Chapter 4 will return to this topic again and introduce similarity-based
structural brain networks, and in particular morphometric similarity networks, in more depth.

In general, the process of constructing a brain network involves (i) the acquisition of
MRI data, (ii) the definition of network nodes, or region of interest (ROI) using a parcel-
lation scheme, and (iii) the estimation of the strength and sign of anatomical or functional
connectivity between those nodes, thus defining the values of the edges in the adjacency
matrix.

First, the nodes of a brain network are typically defined by a parcellation template
or atlas which is used to demarcate multiple macroscopic cortical areas and subcortical
nuclei of the brain, as previously defined by cytoarchitecture (von Economo and Koskinas,
1925), anatomical boundaries (Desikan et al., 2006), functional activation (Yeo et al., 2011),
or a combination thereof (Glasser et al., 2016b). Parcellation allows us to statistically
compare anatomical or functional connectivity measures between individuals using a common
reference atlas, but in doing so it trades off the ability to fully capture the individual variability
in brain organization. A critical consideration in choosing a parcellation template is the
number of nodes, often between 100 and 1000, to balance anatomical specificity with



1.1 Brain networks in the context of maturation and disease 5

computational feasibility and statistical power. The edges of the network are defined based
on the imaging modality chosen, as follow below.

Fig. 1.2 Brain network estimation: Brain networks can be constructed using sMRI, or
fMRI data, or DWI data. Typically, the nodes in these networks are regions of grey matter
defined a priori by a parcellation template. (A) Structural brain networks can be constructed
by estimating the pair-wise correlations between regional morphometric feature vectors to
constitute a morphometric similarity matrix . (B) Computational tractography methods can
be used to derive DTI networks, where the edges are weighted by the streamline count,
indicative of the strength of white matter tracts connecting spatially distributed brain regions.
(C) fMRI networks are typically derived by estimating the pairwise correlations between
resting state fMRI time series averaged over all voxels in each of all possible pairs of two
regions defined by the parcellation template.

Historically, diffusion tensor imaging (DTI) have been used to measure anatomical
connectivity. These networks can be constructed from DWI data which generate contrast by
exploiting the diffusion of water molecules through brain tissue. Computational tractography
is used to identify large-scale white matter tracts mediating connections between pre-defined
grey matter brain regions. This technique estimates the trajectories of white matter axonal
pathways using estimates of diffusivity orientation. Brain networks are computed from
these data by weighting the inter-regional connections by their streamline count (Fig. 1.2B,
bottom).

Structural covariance network (SCN) were later proposed as an alternative to DTI net-
works (Alexander-Bloch et al., 2013). These networks are constructed on the basis of a group
of scans from multiple subjects. SCN are estimated by correlating a single regional mor-



6 Introduction

phometric feature, e.g., cortical thickness, over multiple subjects, resulting in a group-level
network, where each edge describes the inter-regional correlation of a single morphometric
feature across subjects. These networks suffer from a number of limitations: group level
networks lack the ability to easily map changes over time, even though sliding-window ap-
proaches have been suggested as a mitigating measure (Váša et al., 2018); and they only make
use of a single morphometric feature at a time, thus failing to leverage the growing capacity
of multi-modal MRI to extract multiple morphometric features from different modalities of
MRI data (Lerch et al., 2017). In response to these concerns, recent work has focused on
the construction of subject-specific structural brain networks (Seidlitz et al., 2018), such as
morphometric similarity networks (MSNs), which consist of regions defined by a parcellation,
and edges, estimated as the correlation between each possible regional pair of standardized
MRI feature vectors (Fig. 1.2A, bottom). Morphometric similarity networks are based on
the idea that similarity of regional MRI feature vectors is a proxy measure of the similarity
of two regions in terms of their cytoarchitectonic and myeloarchitectonic organization; and
axo-synaptic connectivity is known to be stronger between architectonically similar brain
regions compared to cytoarchitectonically distinct or differentiated areas (Goulas et al., 2016,
2017).

MSNs can be estimated using vectors of morphometric feature values estimated at each
region. Thus T1-weighted MRI scans can be used to extract multiple macro-structural
features (Fig. 1.2A, top): for example, GM, the regional volume of each parcel; SA, the
surface area of the “inflated” cortical sheet; CT, the depth of the cortical sheet; and several
curvature measures can all be measured for each region and compiled in a feature vector
used to estimate the morphometric similarity, a proxy for anatomical connectivity, between
regions. Further, regional mean values of DWI-derived micro-structural MRI features can
also be included as features in analysis of morphometric similarity, i.e. the degree of
anisotropy, termed fractional anisotropy (FA), or the average diffusivity along the axonal
tracts connecting two regions, termed mean diffusivity (MD), can be estimated at each voxel
and averaged over voxels within each regional node. The estimation of MSNs is therefore
possible for a single subject, either based only on a T1-weighted image or also including
DWI data collected from the same subject.

Functional brain networks, are constructed from rsfMRI data. The nodes of these
networks are anatomical brain regions, i.e. regions of interest defined by a parcellation
template, whereas the edge weights are estimates of the functional connectivity (FC) between
each pair of regional nodes, typically measured in terms of the correlation between each pair
of regionally averaged rsfMRI time series (Biswal et al. (1995); Fig. 1.2C, bottom).



1.1 Brain networks in the context of maturation and disease 7

It is worth noting that while this thesis focuses on the analysis of MRI-derived brain
networks, other imaging methods can be used to construct whole-brain networks, including
electro-encephalography and magneto-encephalography (Van Diessen et al., 2015).

1.1.4 Brain network topology and analysis

Once a brain network has been constructed, its topology can be analysed using graph-
theoretical methods (Bullmore and Sporns, 2009; Fornito et al., 2016). A wide range of
network measures are available to characterise brain networks.

Possibly the simplest graph theoretical measure applied in brain network analysis is the
nodal degree (Fig. 1.3A). In an unweighted network, or binary graph, for each node i the
degree ki is calculated simply as the sum of the non-zero edges ei, j connecting it to the rest
of the brain:

ki =
N

Â
j=1; j 6=i

ei, j (1.1)

where ki is the degree of node i, N is the number of nodes in the network, and ei, j indicates
the presence of an edge between node i and an arbitrary node j. The sum is taken over all
edges ei, j ( j 6= 1,2,3, . . .N).

In a weighted graph, it is likewise possible to estimate the mean weighted degree (Fig.
1.3B), or node strength, as the average of the weights of all edges connecting the index node
to the rest of the brain network:

si =
N

Â
j=1; j 6=i

wi, j (1.2)

where si is the mean weighted degree of node i, N is the number of nodes in the network,
and wi, j is the weight of the edge between node i and an arbitrary node j. The sum is taken
over all edges wi, j ( j 6= 1,2,3, . . .N).

Both the binary degree, as well as the weighted degree, are measures of how well
connected a given node is to the rest of the network, providing one important measure of its
topological centrality and likely its functional importance in the context of the connectome
as a whole. To illustrate the concept with a real life example: even an observer that has
never seen a map of the UK, would likely be able to pin point the largest cities in the



8 Introduction

country by looking at a map of train lines. The observer will notice that a number of train
stations appear to have large numbers of train lines connecting them to the rest of the train
network. If we think of train stations as as network nodes and train lines as edges, then train
stations like London and Manchester have a large node degree, compared to the small train
station at Iverness in northern Scotland. In fact, many real life networks have highly skewed
distributions of node degree, such that a small number of nodes are highly connected to the
rest of the network and function as a relay between different parts of the network while most
other nodes are only directly connected to a small number of other nodes. These nodes highly
connected nodes are often termed “hubs” and they serve the role of integrating information
across the network, i.e. one may take a train to from Southampton to London in order to
travel on to Glasgow, illustrating the integrative role of the London station in transporting
passengers from the south to the north of the country.

An array of other graph-theoretical properties can be used to characterise a network’s
structure, for example: the shortest path length, a graph measure that describes the number
of steps that have to be taken to connect any given node in a network to another (Fig. 1.3C);
the clustering coefficient, a measure of the degree to which nodes in a graph tend to cluster
together (Fig. 1.3D); the participation coefficient, which is a measure of the degree to
which a node integrates between modules, measured as a nodes ratio of inter-modular to
intra-modular connections (Fig. 1.3E); modules, which are subsets of regions that are more
strongly connected to one another than to regions in other modules (Fig. 1.3F); and hubs
which are particularly highly connected nodes (Fig. 1.3F).

Brain networks are thought to be constrained by two major driving forces which promote
different network attributes: the minimization of cost (e.g. wiring volume, energy use), and
the maximization of efficiency (e.g. speed of communication, information flow) (Bullmore
and Sporns, 2012). It is believed that these constraints are balanced through a modular
network organization: sets of nodes within the same module are densely intra-connected,
but only sparsely inter-connected to nodes in other modules (Oldham and Fornito (2018);
Fig. 1.3F,G). The segregation achieved by modularity is balanced by so-called connector hubs,
which mediate integrative connections between regional nodes in different modules (Oldham
and Fornito, 2018). The human brain network’s balance of topological segregation and
integration is also evidenced by its core-periphery organization where a set of strongly
inter-connected nodes, or hubs, act as intermediaries between modules, forming a so-called
“rich club”. This topological organization allows for great robustness and adaptivity since in
case of failure of individual nodes, distributive property of the core network is still retained.



1.2 Brain development throughout the lifespan 9

Fig. 1.3 Basic graph metrics: (A) In a binary graph, the node degree indicates the number of
edges connecting a given node with all other nodes in the network. The red node has a lower
degree than the red one. (B) In a weighted network, the mean weighted degree is estimated
as the average weight over all edges connecting the index node to the rest of the brain. The
blue node here may have higher node strength compared to the red node. (C) The shortest
path between two nodes is the minimum number of steps, or edges, it takes to connect them.
(D) The clustering coefficient is estimated as the the number of edges between a node’s
neighbours divided by the number of edges that could possibly exist between them. The red
node’s clustering coefficient is high, the blue one’s is low. (E) Brain networks are typically
modular, meaning that subsets of regions, comprising each of several modules, are more
densely interconnected with each other than with nodes that are affiliated to different modules.
So-called hubs are nodes with high degree that often mediate information between modules.
(F) The participation coefficient is measured as the ration between a node’s intra-modular
degree (edges connecting to other nodes in the same module) and its inter-modular degree
(edges connecting to other nodes in other modules). (G) Brain networks tend to segregate by
strengthening within-module connections and forming a smaller number of long-distance
connections between mediating hubs that integrate information between the modules.

1.2 Brain development throughout the lifespan

Throughout the course of life, from conception to old age, the human brain undergoes
extraordinary changes in structure (Mills et al., 2014; Sowell et al., 2003; Váša et al., 2018;
Whitaker et al., 2016b) and function (Stevens, 2016; Váša et al., 2020). During the prenatal
period, the brain initially undergoes a phase of neurogenesis, which is largely completed
by 20 weeks post-conception, at which time axons start growing and synapses are formed.
Indeed, magnetic resonance imaging of prematurely born infants, born after 24 weeks post-



10 Introduction

conception, has demonstrated increases in grey and white matter volume in mid to late fetal
periods (Bethlehem et al., 2022). At birth, the brain has reached around 30% of its total
grey matter volume (Bethlehem et al., 2022; Gilmore et al., 2012). Recent work on the
largest existing MRI dataset (N ⇠ 120,000 brain scans) has confirmed that trajectories of
grey and white matter development in the cortex and subcortex can be mapped over the
lifespan (Bethlehem et al., 2022). From mid gestation onwards, cortical GM volume increases
rapidly, peaking in childhood at 5.9 years of age, then declining over the rest of the lifespan;
white matter volume (WM) volume also rapidly increases until early adulthood, at 28.7 years
old (yo), before declining gradually throughout adult life, with subsequently accelerated
decline in old age; and subcortical grey matter volume follows an intermediate growth
trajectory, peaking in adolescence at 14.4 years (Bethlehem et al., 2022). These normative
trajectories of cortical development suggest an early post-natal period of differentiation
between grey and white matter, which sees a switch from grey matter volume to white
matter volume being the proportionally dominant tissue type in the brain after around 3 years.
These and other MRI phenotypes of developmental changes in brain macro-structure are
thought to represent underlying microscopic neurodevelopmental processes, e.g., synaptic
proliferation and axonal myelination, that continue throughout adolescence and into early
adult life (Bethlehem et al., 2022; Miller et al., 2012; Petanjek et al., 2011). It is worth
noting that all these results were based on univariate models of structural development and
no network estimates have been investigated in a comparable sample.

To date, there is less certainty about lifespan changes in fMRI. A widely reported finding
is increasing within-network FC until early adulthood, and decreases thereafter (Betzel et al.,
2014; Váša et al., 2020). However, no single study has mapped functional connectivity
changes over the course of the lifespan in sufficiently large samples (Betzel et al., 2014; Fjell
et al., 2017; Ma et al., 2021; Wang et al., 2012) to provide a level of certainty anywhere
near that reached using univariate models of structural brain development as described
above (Bethlehem et al., 2022).

1.3 Adolescent brain development

Understanding structural (Raznahan et al., 2011; Sowell et al., 2004; Váša et al., 2018;
Whitaker et al., 2016a) and functional (Kundu et al., 2018; Váša et al., 2020) development
of brain networks during adolescence has been of particular interest to the neuroscientific
community. This is because adolescence is well-known to be a time of fundamental changes
in cognition and behaviour, and it is also a time of increasing incidence of a variety of



1.3 Adolescent brain development 11

psychiatric illnesses (Costello et al., 2003), including in particular mood disorders. The
pathophysiology of these disorders is increasingly understood in connection with atypical
maturational changes that occur in the adolescent brain (Paus et al., 2008). Thus, under-
standing normative adolescent brain development is expected to further our understanding
of atypical developmental trajectories on the pathway to mental health disorders, including
depression, in young people.

1.3.1 Structural brain development

A large body of work on structural brain development during childhood and adolescence
has focused on estimating maturational trajectories of individual (global or regional) mor-
phometric features, e.g., prototypically, cortical thickness. A prominent pattern that has
emerged from this work is that from the age of about 3 years, grey matter volume in the
brain steadily decreases (Bethlehem et al., 2022). Adolescent structural brain development
is shaped by a pattern of continued cortical grey matter volume decreases, largely driven
by cortical thinning, with relatively smaller decreases in surface area (Tamnes et al., 2017;
Whitaker et al., 2016b), while white matter volume and intra-cortical myelination both show
continued increase, albeit at a slower rate than during the first decade (Mills et al., 2016).
Further, it has been suggested that there is a difference in timing between subcortical and
association cortical adolescent maturation (Mills et al., 2014), with subcortical areas maturing
first, followed by later prefrontal maturation.

Studies of structural brain network development during this period have so far largely
focused on structural covariance networks and DWI-derived connectomes. Cross-species
work has highlighted that structural network hubs are established early in life, i.e., studies in
C. elegans showed that hub neurons are born early in development (Towlson et al., 2013).
Work on human subjects demonstrated that the integrative function of hubs as connectors
between modules is only fully established during adolescence (Oldham and Fornito, 2018).
Further, the above described process of cortical thinning and increasing myelination has been
shown to be topologically focused on association cortical hubs in adolescence, consolidating
topologically central components of the adult brain network that are less myelinated at
the beginning of adolescence (14 yo) and then see faster rates of myelination and cortical
shrinkage over the course of adolescence (Whitaker et al., 2016b). Association cortical areas
have been shown to support higher cognitive functions and play a topologically important role
in the network, such that these findings have been hypothesized to represent an adolescent
re-organization of the structural connectome relevant both to normal cognitive and behavioral
changes (Whitaker et al., 2016b). Further research has highlighted a process of consolidating



12 Introduction

anatomical connectivity between frontal cortex and the rest of the connectome (Váša et al.,
2018) during adolescence, possibly representative of an increasing relevance of prefrontal
cortex and its central role in cognitive control functions that emerge during adolescence.
In summary, these findings suggest that well-known processes of cortical thinning and
myelination support the reorganization of structural brain networks during adolescence, in
particular shaping the topological importance of association cortical and prefrontal areas, to
support adult behavior and cognition

1.3.2 Functional brain development

Two findings in particular have been widely reported by initial studies of developmen-
tal changes in functional connectivity during adolescence: (i) an increase in strength of
long-range connections; and (ii) an decrease in the strength of short-range connections,
hypothesized to represent a shift from localized to distributed networks during adoles-
cence (Dosenbach et al., 2010; Fair et al., 2007). Since most long-range axonal projections
start or finish in association cortical areas, the later emergence of long-range functional
connections has previously been associated with the idea that primary sensory and motor
areas mature during childhood, while association areas undergo changes during late adoles-
cence (Mills et al., 2014; Váša et al., 2020; Whitaker et al., 2016a). Recent work, however,
has reported that a large number of developmental results may have been confounded by
within-scanner head motion (Power et al., 2012; Satterthwaite et al., 2013). It has been found
that head motion both inflates age effects in general, as well as having a heterogeneous effect
on distance-dependent fMRI connectivity, such that head motion has a greater effect on long-
distance, compared to short-distance connections (Power et al., 2012; Satterthwaite et al.,
2013). This may be due to regionally specific effects of motion that inflate the BOLD signal
in one region, and decrease it in distant regions on the same axis, leading to anti-correlations
between distant regions and increased correlations between locally adjacent regions. These
findings are highly relevant for developmental studies since younger subjects tend to move
more. Consequently, the effect of head motion in younger subjects may decrease the strength
of long distance connectivity in younger subjects leading to a relatively higher long distance
connectivity in older subjects who move less (Power et al., 2012; Satterthwaite et al., 2013)
putting into question the distance-dependent changes in functional connectivity reported
in early developmental studies. This idea is supported by a more recent study that found
no distance-dependent effects of age on functional connectivity when applying advanced
motion-correction methods (Marek et al., 2015).



1.4 Sex differences in brain development 13

Several other aspects of functional connectivity development beyond the controversial
question of distance-dependent functional connectivity development have been investigated.
It has been suggested that cross-network integration increases with age (Marek et al., 2015).
Further, hub regions have been reported to refine their connectivity during adolescence, in
particular frontal hubs appear to increase their connectivity with the subcortex early during
adolescence, followed by strengthening of connectivity between cerebellar hubs and the
cortex (Hwang et al., 2013). Given the above-mentioned finding of timing-differences in
subcortical compared to prefrontal structural brain maturation, a last point of focus has
been functional connectivity development between subcortical and cortical regions during
adolescence (Van Duijvenvoorde et al., 2019; Váša et al., 2020). It has been suggested
that subcortico-cortical connectivity develops more heterogenously during adolescence
than cortico-cortical connectivity (Váša et al., 2020). Subcortical programms of selectively
strengthening some connections and weakening others may be representative of a a functional
reorganization of subcortico-cortical systems, in particular involving reward-related circuits
(Van Duijvenvoorde et al., 2019; Váša et al., 2020).

1.4 Sex differences in brain development

Microscopic sex differences, both in terms of gonadal sex steroids, as well as sex chromo-
somes, are known to shape physiological differences between males and females and have
been linked to sexual differentiation of the animal brain (McCarthy et al., 2012), suggesting
the existence of similar effects in humans (Raznahan and Disteche, 2021).

Experimental manipulation of gonadal hormones in animal models has been shown
to directly affect brain structure and function (Corre et al., 2016). In humans, more in-
direct methods of linking gonadal hormones to sexual differences in brain structure and
function have been employed: by studying subjects displaying longitudinal variation in
hormonal levels, either long-term variation, due to developmental phases like adolescence
or menopause (Mosconi et al., 2021), or short-term variation, due to the menstrual cy-
cle (Pritschet et al., 2020), or natural variation in testosterone levels over the course of
the day (Grotzinger et al., 2022); or by comparing healthy controls to cases of endocrine
disorders affecting sex hormone production (Tauber and Hoybye, 2021); or by studying
variation due to gender-affirming hormone treatment (Kranz et al., 2020). These studies have
demonstrated that sex hormones do indeed affect brain structure and function, both in the
short term, i.e. functional brain networks reorganize during the menstrual cycle (Pritschet



14 Introduction

et al., 2020), as well as long-term, i.e. menopause appears to effect both grey and white
matter volumes (Mosconi et al., 2021).

In the late 1950s, sex chromosomes were first shown to affect sex differences in mam-
malian brain organization (Phoenix et al., 1959b). Since then, a growing body of literature
has suggested that sex chromosomes affect brain organization and may also contribute to
phenotypic diversity of the human brain (Arnold, 2012; Raznahan and Disteche, 2021). There
are a number of reasons why sex chromosomes are likely to contribute to sex differences in
the brain. First, sex chromosomes play a special role in the rapid fixation of mutations and
evolution of genes, due to the fact that the X and Y chromosomes are haploid in males. In
males recessive mutations on one sex chromosome cannot be masked by the dominant allele
from the other chromosome copy. Thus, the recessive allele will be expressed and, when
advantageous, it has a higher chance of being passed on to offspring (Fig. 1.4A). This process
may have contributed to the importance of sex chromosomes for traits advantageous to males,
for example, as evidenced by an accumulation of genes relevant for male fertility on both the
X and Y-chromosomes. Second, in females, one of the X chromosomes is randomly inacti-
vated, balancing out the fact that males only have one X chromosome. However, about 15%
of genes escape X chromosome inactivation (Fig. 1.4B). These genes are thus upregulated in
females compared to males and may provide a likely source of sexually differentiated pheno-
typic expression (Disteche, 2016; Oliva et al., 2020). Third, not only is the X chromosome
enriched for genes expressed in the brain, X chromosome genes are also heterogeneously
expressed across the brain, thus suggesting effects on anatomically patterned and functionally
specialised brain systems (Fig. 1.4C). And finally, in both sexes, X chromosome expression
is upregulated, to ensure that its expression is relatively balanced compared to autosomal
gene expression (Fig. 1.4D), a process that happens prior to X-inactivation and thereby leads
to higher expression of X-linked genes versus autosomal genes in females compared to males
cells during embyronic development (DeCasien et al., 2022).

Together, these microscopic sex differences in gonadal hormones and sex chromosomes
suggest possible mechanisms for sex differences on the macroscopic level in the form of
sexually differentiated brain anatomy and function as measured using MRI.

1.4.1 Sex differences in structural MRI

A range of previous work has investigated sex differences in brain phenotypes, largely
focusing on structural MRI and task-activated MRI. The most obvious sex difference observed
is that, on average, male brains tend to be larger than female brains, a differences which



1.4 Sex differences in brain development 15

Fig. 1.4 Microscopic sex differences: (A) Sex chromosomes are diploid in females and
haploid in males. Recessive mutations in females can be masked by dominant alleles on
the second copy of the X chromosome. In males, however, such masking cannot occur
due to the chromosomes being haploid. When traits are male advantageous, they are more
likely to be passed on to offspring, thus leading to an accumulation of male-advantageous
genes on sex chromosomes. (B) In females, one X chromosome is randomly inactivated,
however, a number of genes escape this inactivation. (C) The inactivation being random
leads to a spatially diverse pattern of cells in which the maternal or paternal X chromosome
is deactivated. (D) In both males and females, X chromosomes are upregulated such as to
avoid a dosage equilibrium between sex chromosomes and autosomes.

is likely due at least partially to a difference in body size (Ruigrok et al., 2014). This sex
difference in brain size has been shown to be present at birth, with male brains estimated to
be approximately 8% larger than females (Gilmore et al., 2007; Knickmeyer et al., 2017),
and persistent throughout life.

Recent work has demonstrated effects of brain size on regional grey matter volume (Eikenes
et al., 2022; Warling et al., 2021), white matter tracts (Reardon et al., 2018; Sanchis-Segura
et al., 2020), and brain-behavior relationships (Dhamala et al., 2022). The effect sizes of
regional sex differences across different structural imaging phenotypes are attenuated when
correcting for total brain volume. There has been a long-standing interest in whether there are
sex differences in regional brain anatomy above and beyond sex differences in total brain size.
A number of MRI studies have reported sex differences in grey matter volume in multiple
regions, with effect sizes ranging from small to medium (Liu et al., 2020). Conversely,



16 Introduction

a meta-synthesis (i.e. a meta-analysis of meta-analyses) of a large number of studies has
reported a lack of coherence in prior findings, and suggested sex differences in grey matter
volume may after all be very small (Eliot et al., 2021). This idea stands in contrast to results
from recent large-scale studies that find consistent sex differences in grey matter volume
across most brain regions (Lotze et al., 2019; Williams et al., 2021). It has since been
suggested that a number of factors, including inconsistent approaches to correcting for brain
size, and variable sample sizes, may have contributed to the (perceived) lack of consistency
in studies of sex differences in grey matter volume (DeCasien et al., 2022), particularly since
meta-analyses do not correct for methodological discrepancies between studies. Thus while
studies of sex differences in grey matter volume need to be evaluated carefully with respect
to sample size and brain size correction methods (Sanchis-Segura et al., 2020), large-scale
neuroimaging studies appear to converge on a consistent picture of small to medium-sized
sex differences in volume across most brain regions (DeCasien et al., 2022; Williams et al.,
2021). However, it is worth noting that within-sex variability in imaging phenotypes is
large and also scales with head size (Eliot et al., 2021), and sex differences are statistical
differences in the mean of two overlapping distributions.

1.4.2 Sex differences in rsfMRI

In the past, research on sex differences in fMRI have largely focused on task-activated
fMRI, often investigating the “brain basis” for assumed sex differences in behavior and
cognition. Many such studies have suffered from small sizes and meta analyses have
found little overlap between findings (Eliot et al., 2021). It is not yet clear how resting
state functional connectivity differs between males and females, either during adolescence
or adulthood. One widely reported sex difference is increased functional connectivity
of the default mode network (DMN) in females (Allen et al., 2011; Biswal et al., 2010;
Bluhm et al., 2008; Filippi et al., 2013; Tomasi and Volkow, 2012). Female-increased
connectivity has also been reported in subcortical nuclei and limbic areas (cingulate gyrus,
amygdala, hippocampus) (Scheinost et al., 2015); whereas male-increased connectivity has
been reported for sensorimotor areas (Biswal et al., 2010; Filippi et al., 2013; Scheinost et al.,
2015). However, these effects are not consistently found across studies (Allen et al., 2011;
Tomasi and Volkow, 2012; Weissman-Fogel et al., 2010). Importantly, most research on
sex differences has focused on pre-selected regions, often including the amygdala (Alarcón
et al., 2015; Kilpatrick et al., 2006), with few studies having investigated sex differences
comprehensively over all brain regions (Biswal et al., 2010; Casanova et al., 2012; Filippi
et al., 2013; Zhang et al., 2016, 2018). While these regionally focused approaches increase



1.5 Vulnerabilities during development 17

statistical power, they fail to map global patterns. Finally, at least one study failed to observe
any sex effects at all (Weissman-Fogel et al., 2010).

It is important to note that almost all studies mentioned here were cross-sectional studies,
using either age-balanced, usually adult, samples of males and females, or covering a
very limited age range. Most prior rsfMRI studies of brain development have focused on
estimating "average" effects of age across both sexes, e.g., by including sex as a covariate in
the statistical model for estimation of developmental parameters. Few studies have reported
age-by-sex interactions or the conditioning of developmental parameters by sex (Scheinost
et al., 2015; Zhang et al., 2018). The lack of longitudinal data may have contributed to the fact
that few studies found convincing effects of age-by-sex interaction on functional connectivity.
Some interaction effects have been reported in several networks, including the default mode
network, the fronto-parietal, visual and auditory networks (Scheinost et al., 2015; Zhang et al.,
2016); but often these findings did not survive correction for multiple comparisons. While
cross sectional studies can make claims about male-female group differences, no within-
subject age-related changes in functional connectivity can be inferred. Thus cross-sectional
studies do not allow to determine whether observed differences in functional connectivity
are a result of sex, (atypical) maturational trajectories, or a combination of both (Mills
et al., 2014). Therefore modelling subject-specific trajectories over time is crucial for our
understanding of how sex might intersect with brain development.

Taken together, the current heterogeneity of results concerning the spatial locations and
sign of sex differences in brain structure and function suggests a need for further investigation,
in particular using large, longitudinal samples, with appropriate correction for motion-related
artifacts.

1.5 Vulnerabilities during development

Topological analysis of MRI-derived brain networks has furthered our understanding of
structural and functional brain development in health. Additionally, contrasting normative
results with patient data from multiple neuropsychiatric disorders has provided insight into
atypical deviations of network organization associated with disease. It has been found
that even very basic graph theoretical measures, such as degree, can highlight case-control
differences in network structure (Morgan et al., 2019; Váša et al., 2018). Contrasting
brain networks between healthy controls and cases of neuropsychiatric disorders has also
demonstrated how central several topological features are for the healthy functioning of the
brain. For example, it has been found that many disorders appear to disrupt the modular



18 Introduction

community structure of brain networks, leading to more segregated organization of the
connectome (Crossley et al., 2014).

It is notable, that many developmental disorders, e.g., autism spectrum disorder (ASD),
are first diagnosed during early to late childhood, whereas psychiatric disorders, in particular
mood disorders, e.g., major depressive disorder (MDD) or anxiety disorders, are typically
incident during adolescence (Fig. 1.5A; (Kessler et al., 2005; Paus et al., 2008)). Both these
periods are well-known for being neurodevelopmental phases of major reconfiguration or
rewiring of brain networks (Morgan et al., 2018). In line with this coincident timing, it can
be argued that atypical trajectories of developmental rewiring may lead to vulnerabilities
to disease, i.e. “moving parts get broken” (Paus et al., 2008). For example, it has been
suggested that in patients with schizophrenia an “exaggeration of typical adolescent changes”
may have occurred (Keshavan et al., 1994).

As mentioned above, many disorders display sex differences in their prevalence or clinical
expression profile, including ASD, which is four times as likely to be diagnosed in males
than in females, and MDD, which is twice as likely to be diagnosed in females (Fig. 1.5B). It
is worth acknowledging that socio-cultural as well as structural factors may contribute to this
sex difference in diagnosis (Sharma et al., 2021). There is undeniably a gendered influence
on health, diagnostic criteria may be sex-biased, and cultural expectations may contribute to a
discrepancy in seeking medical help (Phillips, 2005). For example, males often exhibit lower
help-seeking behavior, potentially contributing to the sex difference in incidence rates for
mood disorders (Galdas et al., 2005). On the other hand, current diagnostic criteria for ASD
may lead to under-diagnoses in females, who exhibit more camouflaging behavior (Fusar-Poli
et al., 2022), i.e. they show a greater tendency to mask disease-associated behavior either by
avoiding some types of behaviors, or conversely by explicitly performing behavior considered
to be more neurotypical. However, the concentrated emergence of multiple neuropsychiatric
disorders during neurodevelopmentally active periods of the lifespan suggests that sex
differences in brain development contribute at least in part to the pathogenesis of these
conditions. This is underlined by gene expression studies, i.e. research on postmortem brain
tissue suggests that gene expression in ASD is correlated with normative sex differences in
gene expression (Kissel and Werling, 2022).

Neuropsychiatric disorders may be associated with alterations in both the timing and/or
the shape of developmental trajectories (Di Martino et al., 2014a). For example, attention
deficit hyperactivity disorder (ADHD) has been associated with delayed brain maturation,
whereas ASD has been associated with an early acceleration of brain development (Shaw
et al., 2010). However, to date, neuroimaging studies of atypical brain development have



1.5 Vulnerabilities during development 19

Fig. 1.5 Diagnosis of neuropsychiatric disorders: (A) Many neuropsychiatric disorder
show a sex difference in prevalence. (B) Age at first diagnosis of psychiatric disorders
according to the literature (Solmi et al., 2022).

largely been cross-sectional. While this allows for estimating case-control group differences,
usually corrected for age, it also leads to several fundamental shortcomings (Di Martino
et al., 2014a). First, the lack of availability of longitudinal patient data contributes to a lack
of understanding of atypical developmental trajectories. Second, when only a single time
point is available for an individual, it is impossible to know anything about the shape of this
individual’s developmental trajectory. Fig. 1.6 illustrates this issue: while a single timepoint
for each subject can reveal their deviation from the norm, it is unclear which trajectory of
development they are on. Only a second measure could bring clarity. While the increasing
number of longitudinal neuroimaging studies focusing on normative brain development is
encouraging, future studies should also attempt to collect longitudinal patient data to allow
for mapping atypical trajectories.

A key factor motivating research into brain network changes in disease is that it is likely
that symptoms of neuropsychiatric disorder appear after the onset of atypical development,
raising the hope that, with advances in normative modelling, neuroimaging may be used
to track changes in brain development before symptom onset thus expediting diagnosis or
creating opportunities for prevention. It should be pointed out that currently the limited
regional availability (i.e. MRI scanners being concentrated in better hospitals, cities, and
high-income countries) and the high cost of neuroimaging limit its usefullness in harder-to-



20 Introduction

reach populations. A recent development, however, provides hope for the future: portable,
low-cost MRI scanners may facilitate reaching more remote populations in the future (Cho,
2023).

One in five adolescents have a mental illness that will persist into adulthood (Kessler et al.,
2005), and depression, schizophrenia and addiction are among the top ten leading causes of
medical disability worldwide, with no evidence of global reduction in disease burden (Col-
laborators et al., 2022). Research into the neurodevelopmental basis of neuropsychiatric
disorder should thus be of the utmost importance.

Fig. 1.6 The value of longitudinal data: Atypical development can alter both the timing of
development (delayed or precocious development) or alter its shape (failure to mature, halted
development, ectopic development). A single (cross-sectional) measure for each individual
cannot determine the shape of the individual’s atypical trajectory.

1.6 Thesis structure

Overall, the findings described above demonstrate that the human brain undergoes various
phases of active development during the lifespan. While brain network development is
fundamental to the emergence of new cognitive and social capacities, periods of rewiring
also expose individuals to an increased risk of neuropsychiatric disorders, highlighting the
relevance of understanding normative brain development. Further, many neuropsychiatric
disorders are known to display sex differences both in prevalence and clinical presentation,
which may be linked to sex differences in brain structure and function. However, while we
know that there are sex difference in brain physiology on the microscopic level (Arnold,
2012; Raznahan and Disteche, 2021), less is known about sex differences in macroscopic
brain development.



1.6 Thesis structure 21

This thesis maps normative functional and structural brain development in adolescence
and later life using magnetic resonance imaging. Chapters 2-3 and Chapters 5 estimate
adolescent changes in functional and structural brain network development in an accelerated
longitudinal cohort of healthy adolescents aged 14-25 years (N=298), each scanned between
one and three times, with a total of 520 scans. Chapter 2 asks the question: “Does adolescent
functional brain development differ between males and females?”. Chapter 3 builds on the
fMRI-derived map of sex differences in adolescent brain development from the previous
chapter to ask: “Are sex differences in adolescent functional connectivity maturation related
to major depression?”. Chapter 5 moves from functional to structural data, asking: “Are
there adolescent changes in morphometric similarity networks?”.

Finally, in an effort to move from group level to subject-specific analyses, and acknowl-
edging that brain development is not restricted to adolescence but is a continuous process
throughout life, Chapter 6 aggregates 90,000 scans from 41 prior studies ranging from mid
gestation to old age. This allows me to address the question: “How do subcortical regions
develop over the course of the lifespan in health, and how do individuals deviate from these
normative trajectories in association with disease?”.

Finally, Chapter 7 summarizes the experimental results on adolescent and lifespan
development of functional and structural brain networks, and subcortical grey matter volume.
It identifies convergent themes and aligns them with the existing scientific literature on brain
development, sexual differentiation and risk of psychiatric disorders.





Chapter 2

Sex differences in adolescent development
of functional connectivity

2.1 Introduction

As outlined in Chapter 1, adolescence is a period of large-scale functional reorganization
of the brain (Marek et al., 2015; Sowell et al., 2004; Váša et al., 2020) that coincides with
changes in cognition and behaviour. Adolescence is also a period of increased risk to
psychiatric disorders many of which show sex differences in prevalence and expression
profile (Kessler et al., 2005), raising as the question whether there may be underlying sex
differences in brain development. Sex differences on the microscopic scale, in the form of
gonadal hormones and sex chromosomes, are known to impact macroscopic measures of
brain structure and function (Arnold, 2012; Raznahan and Disteche, 2021). However, to date,
little is known about whether, and how, adolescent changes in functional connectivity may
differ between males and females. Here, we start from the position that there may indeed be
sex differences in adolescent processes of brain maturation.

Recent advances in developmental neurogimaging have produced a number of longitudi-
nal datasets covering the period from late childhood to early adulthood (Kiddle et al., 2017;
Satterthwaite et al., 2016), allowing the field estimate age-related changes longitudinally,
rather than cross-sectionally. Further, with the newly-gained awareness that head motion
differentially affects long-distance connections, previously reported findings suggesting
distance-dependent changes in functional connectivity during adolescence were put into
question. Thus recent work has endeavored to shed further light on adolescent functional con-



24 Sex differences in adolescent development of functional connectivity

nectivity development, using rigorous motion-controlling strategies and explicitly modelling
longitudinal changes during this period.

One such study estimated adolescent changes in regional functional connectivity weighted
degree, i.e. the average connectivity across all of a nodes edges to the rest of the brain (Váša
et al., 2020). They estimated the baseline connectivity at the beginning of adolescence, and the
rate of change in connectivity over the course of adolescence (Váša et al., 2020) for each node
(Fig. 2.1A) and found that regional functional connectivity weighted degree, was particularly
strong in primary motor and sensory cortical areas at the beginning of adolescence, and
cortico-cortical connectivity generally increased over the course of adolescence. However,
subcortico-cortical connectivity had a varied anatomical distribution, with particularly strong
functional connectivity increases between subcortical regions and association cortical areas,
and some decreases in connectivity between a number of subcortical regions and primary
motor and sensory cortical areas. These findings highlight a special role of subcortico-cortical
connectivity changes during adolescence.

Fig. 2.1 Two modes of adolescent FC development: From a linear model of age effects
on FC, two parameters of adolescent development are extracted. (A) First, regionally, the
weighted degree of FC of cortical regions and subcortical nuclei is estimated at baseline (14
years), FC14, and the rate of change in connectivity over the course of adolescence, FC14�26.
(B) The same parameters can be estimated for each edge. (C) The maturational index is
estimated as the correlation between edgewise baseline FC14 and the rate of change FC14�26.
(D) Visualization of two examplary regions, displaying (left) conservative development
(MI > 0), where edges that are strong at baseline, become stronger over the course of
adolescence, and (right) disruptive development (MI < 0), where edges that are strong at
baseline decrease in strength over the course of adolescence, and edges that are weak increase.
Adapted from Váša et al. (2020) under a CC BY 4.0 licence.

Moving from regional weighted degree to edge-wise connectivity (Fig. 2.1B), the au-
thors further developed a new network metric describing edge-wise adolescent functional
connectivity maturation. This metric suggested that adolescent development of functional



2.1 Introduction 25

connectivity can be described as occurring in two modes: a conservative mode of consol-
idating previous phases of development, and a disruptive mode of establishing functional
connectivity in brain systems that were not previously strongly connected. These develop-
mental modes are measured and differentiated using the maturational index (MI), which
describes a system level change in a node’s connectivity to the rest of the brain, reflecting
maturational changes across all of a node’s edges (Váša et al., 2020). Briefly, MI describes
how each of a node’s connections change during adolescence. It does so by examining the re-
lationship between edgewise functional connectivity strength at the beginning of adolescence,
e.g., 14 years old, denoted baseline connectivity or FC14; and the rate of change in functional
connectivity over the course of adolescence, e.g., 14-26 years old, denoted rate of change
or FC14�26 (Fig. 2.1B). More specifically, MI is estimated for each node by correlating the
baseline connectivity and rate of change for all edges connecting the index node to all other
nodes in the network (Fig. 2.1C). MI defines two distinct modes of adolescent development
of brain functional connectivity (Fig. 2.1D): (i) conservative development, indicated by a
positive MI, which is the result of a node’s strong (high FC) edges increasing in functional
connectivity over the course of adolescence, and its weak edges decreasing in strength; and
(ii) disruptive development, indicated by a negative MI, is the result of a nodes’ weak edges
gaining strength and its strong edges weakening, leading to a shuffling of a node’s ranked
edges. Thus, MI describes a “system level” change in a node’s wiring, reflecting maturational
changes across all of a node’s edges (Fig. 2.2).

Previous work demonstrated that conservative development was characteristic of primary
sensory and motor cortical areas, whereas disruptive development was mainly located in
association cortical and subcortical regions. Disruptive development has been suggested to
represent metabolically costly remodeling of cortical and subcortical systems to facilitate
emergence of adult cognitive and social behaviors (Váša et al., 2020).

Here, using fMRI data from a previously published (Váša et al., 2020) accelerated
longitudinal study (N=298; age range 14-26 years; 51% female; Table 2.1), stratified by age
and balanced for sex per age stratum (Kiddle et al., 2017), we estimated the effects of sex
on three parameters of adolescent development of resting-state functional connectivity: (i)
baseline connectivity at age 14, FC14; (ii) the adolescent rate of change, FC14�26, estimated
at nodal and edge-wise levels of analysis; and (iii) the maturational index for each node,
MI, which is the signed correlation coefficient between FC14 and FC14�26 across all edges
connecting a given node to the rest of the network. We hypothesized that (i) there may be sex
differences in parameters of adolescent brain development; and (ii) that these sex differences
may be co-located with expression of a weighted function of the whole genome enriched for
X chromosome and (iii) developmentally relevant genes.



26 Sex differences in adolescent development of functional connectivity

Fig. 2.2 Modes of adolescent development: In conservative development, a node’s strong
edges get stronger between 14 and 26 years (A; top). Thus, the MI, estimated by Spearman’s
correlation between baseline connectivity at age 14 (FC14) and adolescent rate of change of
connectivity (FC14�26), is positive (B, top). Conversely, in disruptive development, a node’s
weak edges get stronger over the course of adolescence, while its strong edges weaken (A,
bottom). Thus the MI, estimated by the correlation between FC14 and FC14�26), is negative.
(C) Cortical surface map of MI estimated at each regional node in the brain.

We found that there was a sex-related difference in adolescent brain network development:
females had significantly more disruptive development of functional connectivity in a default
mode cortical, limbic and subcortical network. Further, we found that this developmentally
divergent brain system was co-located with expression of a weighted function of the whole
genome enriched for X chromosome genes, and genes expressed during various phases of
brain development.

2.2 Methods

2.2.1 Sample

Data Collection

The data analysed in this chapter were collected as part of the Neuroscience in Psychiatry
Network (NSPN) consortium, a collaboration between the University of Cambridge and



2.2 Methods 27

University College London, with the aim of measuring cognitive and brain developmental
changes during adolescence in a healthy adolescent cohort representatively sampled from
the population of Greater London and Cambridgeshire (Kiddle et al., 2017). The analyzable
sample consisted of 2,402 participants, aged 14 to 26 years, that completed repeated self-
assessments of mental health status, with a subset also completing repeated functional and
structural magnetic resonance imaging assessments, as detailed below. Participants were
recruited in five agebins (14-15 years, 16-17 years, 18-19 years, 20-21 years, older than 22
years), with equal numbers of males and females in each agebin.

All participants aged 16 and older provided informed written consent for each aspect
of the study, and parental consent was obtained for those aged 14–15 years. The study was
ethically approved by the National Research Ethics Service and was conducted in accordance
with NHS research governance standards.

MRI sample

A sub-sample of 306 adolescents drawn from the NSPN cohort consented to complete
functional and structural MRI assessments. The age and sex stratification from the larger
cohort was maintained, such that each of the five agebins included about 30 males and females.
The exclusion criteria for this sample included: a current or past history of neurological
disorder or learning disability, and current treatment for psychiatric disorder or drug or
alcohol dependence. Each participant in the scanning sample was invited to provide magnetic
resonance imaging (MRI) data on at least two occasions: at baseline and at follow-up 12-
18 months later, with 29 participants additionally invited to attend follow-up scanning six
months after baseline. The fMRI scan was the first in a series of scans collected in each
scanning session which also included structural MRI using the multi-parametric mapping
(MPM) sequence (Weiskopf et al., 2013), and diffusion weighted imaging. Here, we present
results using the functional magnetic resonance imaging (fMRI) scans, whereas Chapter 5
will focus on complementary analysis of the structural MRI data. A total of 556 resting state
functional magnetic resonance imaging (rsfMRI) scans were available for analysis after after
completion of quality control procedures.

MRI data acquisition

Functional MRI data were acquired at three scanning centres (Wolfson Brain Imaging Centre,
University of Cambridge; University College London; and King’s College London), on
three identical 3T Siemens MRI scanners (Magnetom TIM Trio, VB17 software version)



28 Sex differences in adolescent development of functional connectivity

Sex # Scans # Scanned At Baseline # Subj./Agebin

1 2 3 µ Age s Age µ FD s FD 1 2 3 4 5

female 259 54 86 11 19.8 2.9 0.11 0.05 34 39 24 32 22
male 261 41 98 8 19.2 3.8 0.13 0.05 32 33 24 35 23

Table 2.1 Sample of healthy adolescent participants with fMRI data from the NSPN
cohort: The final sample after QC included data from N = 298 healthy young people who
participated in an accelerated longitudinal fMRI study. The recruitment was balanced for sex
in each of five age-defined strata. Subjects were scanned between 1 and 3 times with scans
taking place at baseline, 6 and/or 18 months later. The number of subjects who were scanned
1, 2 or 3 times respectively is listed under # Scans. Framewise displacement (FD), a measure
of head movement in mm, was significantly greater in males compared to females on average
over all ages, and in the youngest two age strata specifically (P < 0.05,uncorrected).

with standard 32-channel radio frequency (RF) receive head coils and RF body coils for
transmission, using a multi-echo echo-planar imaging (EPI) sequence (Barth et al., 1999) with
the following scanning parameters: repitition time (TR), 2.42s; GRAPPA with acceleration
factor 2; flip angle, 90o; matrix size, 64⇥ 64⇥ 34; field of view (FOV), 240⇥ 240 mm;
in-plane resolution, 3.75⇥3.75 mm; slice thickness, 3.75 mm with 10% gap, with sequential
slice acquisition of 34 oblique slices; bandwidth, 2368 Hz per voxel; echo time (TE), 13,
30.55 and 48.1 ms; and total scan time, 11 minutes.

MRI data pre-processing

The data e