% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Nicolaisen:903647,
author = {Nicolaisen, Eliana and Kharabian, Shahrzad and Mihalik,
Agoston and Ferreira, Fabio and Maleki Balajoo, Somayeh and
Eickhoff, Simon and Mourao-Miranda, Janaina and GENON, Sarah
and Thomas Yeo, B. T. and Valk, Sofie},
title = {{T}wo latent dimensions linking multi-featured brain
structure to behaviour in healthy adults},
reportid = {FZJ-2021-05296},
year = {2021},
abstract = {The development of solid theories of brain-behaviour
relationships is weakened by a replication crisis and
publication biases1,2. To overcome this, data-driven
multivariate approaches that integrate neurobiology and
behaviour have been promoted. For instance, Canonical
Correlation Analysis (CCA) can identify sets of behavioural
variables that correlate with sets of brain variables (i.e.,
latent dimensions). With such approach, studies in young
healthy adults found latent dimensions associated with mood
and cognitive-control/general-intelligence, but brain
patterns were relatively inconsistent3–5. More robust
results could be found with a regularized version of CCA
(RCCA) embedded in a machine learning framework, assessing
generalizability (evaluating how the model performs in new
data) and stability (if similar variables are selected in
different data splits)6. With this framework, we studied
robust latent dimensions linking behaviour and
multi-featured brain structure in young healthy adults.We
used T1 anatomical scans of the Human Connectome Project
Young Adult sample (S1200, n: 1047, age: 28.78 ±3.67 (mean
±sd); 560 females). Grey Matter Volume (GMV) was estimated
with CAT12, and Cortical Thickness (CT) and Surface Area
(SA) with FreeSurfer v5.3. Brain data were averaged by
regions (200 cortical, 73 subcortical) using the Schaefer
atlas7. Behavioural measures spanned cognition, emotion,
mental health and life outcome. RCCA linked the
multi-featured structural brain data (concatenating GMV, CT
and SA) with behaviour. In the main analysis, age and gender
were excluded of the model whilst in a supplementary
analysis both were regressed out from brain and behavioural
data. A multiple holdout machine learning framework6,8
was used for model selection and statistical inference. In
particular, generalizability and stability were used to find
the optimal regularization parameters. Significance of the
latent dimensions was assessed with 1000 permutations
respecting the family structure of the data.We found two
latent dimensions. The first dimension (r=0.43-0.55,
p=0.015) was positively associated with high cognitive
abilities (such as fluid intelligence, working memory and
executive functions) and self/goal-driven aspects (Fig. 1A).
This dimension was positively associated with anterior
temporal and medial prefrontal (pole) regions (GMV and SA),
amygdala and hippocampus (GMV), and insula (SA and CT) (Fig
1B). This dimension remained significant when regressing out
gender and age. The second dimension (r=0.27-0.41, p=0.015)
was positively associated with cognition (including
attention, language and episodic memory) and negatively
associated with negative social behaviour (such as rule
breaking and aggression) (Fig 2A). Brain associations with
this dimension were positive in visual regions (CT) and
putamen (GMV), and negative in the cerebellum (GMV),
superior frontal regions (CT), as well as Broca’s area and
visual, paralimbic and language regions (SA) (Fig 2B).
Importantly, this dimension did not remain significant when
regressing out gender and age.Two dimensions of structural
brain-behaviour interindividual variability were found. The
first dimension mainly reflected higher cognitive and
self-driven functions. Accordingly, it was associated with
brain structural variability in higher-hierarchical regions
of the self-centric and motivation network. The second
dimension appears to mainly reflect education-related
performance aspects and was negatively associated with
negative social behaviour. Importantly, this dimension did
not remain significant when removing variance related to
gender categories and age, suggesting the influence of
demographical factors on this brain-behaviour dimension.
Hence, future studies should investigate the influence of
genetic vs environmental factors on these dimensions, in
particular with heritability assessment. Better
characterizing these dimensions should help to prevent
pathological extremes.},
month = {Jun},
date = {2021-06-21},
organization = {27th Annual Meeting of the
Organization for Human Brain Mapping,
Virtual (Virtual), 21 Jun 2021 - 25 Jun
2021},
subtyp = {After Call},
cin = {INM-7},
cid = {I:(DE-Juel1)INM-7-20090406},
pnm = {5251 - Multilevel Brain Organization and Variability
(POF4-525) / 5253 - Neuroimaging (POF4-525)},
pid = {G:(DE-HGF)POF4-5251 / G:(DE-HGF)POF4-5253},
typ = {PUB:(DE-HGF)24},
url = {https://juser.fz-juelich.de/record/903647},
}