Characterizing brain networks and their inter-individual variability by high-throughput imaging and computational modelling
Approved Research ID: 41655
Approval date: April 26th 2019
The human brain shows a marked level of neurobiological variability between individuals. Likewise, individuals also vary widely in aspects that may be considered causes (age, gender, general health, environmental factors) and effects (cognitive functions, socio-affective traits) of this neurobiological variability. However, the relationship between neurobiological and behavioural variability relate to each other, is still enigmatic. The goal of this project is to probe the relationship between brain features on one hand and a broad range of behaviours on the other hand in order to provide an overview on brain-behaviour and brain-environment relationships.
We propose to investigate the relationship between neuroimaging features on one hand and questionnaire & measurements on the other, across a broad range of features in order to establish principles of brain-behavior relationships and enhance existing brain atlases by information on behavioral associations. As a key aspect of the current proposal, we plan to use a priori information on both cortical segregation and task-relevant networks derived from our previous work on brain mapping (Eickhoff et al., 2016, Schaefer et al., 2017, Varikuti et al., 2018, etc) and meta-analytic definition of functional networks (e.g., Cieslik et al., 2016, Rottschy et al., 2012, Goodkind et al., 2015, Langner et al., 2013), respectively. This should provide the experimental probe of the conceptual models of functional inference by brain-behavior relationships as recently proposed by our group (Genon et al., 2018).
We would like to extend our scope to use genotype data for the following objectives:
- For genome-wide genotyping data based ancestry inference (Chen et al., 2013), which will be in turn added as a nuisance variable to the neuroimaging data to remove biases (Altmann & Mourao-Miranda, 2019) in the analyses described in the approved application.
- The same ancestry information will be utilized to replicate and extend the work described in (Altmann & Mourao-Miranda, 2019). Specifically, we will test whether ancestry can be predicted using different modalities of neuroimaging data using advanced machine learning techniques involving feature extraction, selection combined with SVM, random forests and depp neural networks.
- Additional polygenic risk scores will be calculated and used in a similar fashion as (1) and (2).
- We will use available information on UKB individuals including known genetic risk factors for RLS (risk allele of lead SNP in RLS top gene MEIS1, MES1 risk haplotype or polygenic risk score for RLS; cf ??(Schormair et al., 2017)?), insomnia complaints, and RLS promoting factors in life history and environment such as number of children, (brain) iron deficiency, and renal insufficiency to define proxies of RLS. Then we will analyze the fMRI in UKB individuals with and without RLS-proxy (and in individuals with and without insomnia) to seek association of multimodal neuroimaging data that differentiates between the risk and control groups using machine learning methods.
The following data will be returned to UKB:
- The ancestry inference scores and other polygenic risk scores.
- The risk characterization of the RLS.
- The code for analyses using the neuroimaging data will be made available.
We would like to extend our scope to include access to derivatives of the brain imaging data, a MRI modality, and cognitive assessment data. These data will be used for five different objectives:
1) Identifying general macroscale multimodal brain organizational principles using connectivity and gradient based analyses.
2) Identification of latent representations of the BCDG variables*, e.g. latent representation of cognitive functions.
3) Investigation of relationship between different modalities and their relationship with computational models and BCDG variables and their latent representations.
4) Identifying associations between the imaging data features, e.g. between regional gray mater volumes and freesurfer derived area-wise measurements with BCDG variables and their latent representations.
5) Multimodal associations between imaging modalities and BCDG variables and their latent representations.
* BCDG variables: behavior, cognition, demographics and genetic variables. We already have access to these data in the currently approved application.
The following data will be returned to UKB
1) Derivatives of the genotype data, e.g. polygenic risk score assessments.
2) The latent dimensions or the method and code to identify those will be made available.
3) The code for analyses using the neuroimaging data will be made available.
We would like to extend our scope to include access to data regarding first occurrences of mental, behavioral, and nervous system disorders and the dates of first in-patient diagnoses. These data will be used for the third of the following objectives:
1) Investigation of the relationship between age and the co-localization of brain activity at rest and neurotransmitter systems (Dukart et al., 2021) in a healthy population.
2) Investigation of the relationship between cognitive abilities and the relationship found in 1).
3) Examination of the effect of brain diseases (within F00-F99 and G00-G99) on the relationships found in 1) and 2). The new data will be used to form groups of patients with similar disease stages and may help to find stage-related dynamics in these relationships.
Dukart et al., 2021: https://doi.org/10.1002/hbm.25244
We will perform genome-wide association studies (GWAS) to investigate genetic architecture of human traits using single- and multiple-phenotype statistical tools such as MultiPhen and MOSTest as well as more traditional tools like PLINK . Our aim is to investigate the genetics behind neuroimaging markers based on structural, functional and diffusion data (as described in the main application) and their relationship with mental disorders and co-morbid diseases such as cardiovascular disease and metabolic disease as well as lifestyle factors. Based on the genetic variants identified by GWAS, we plan to identify candidate genes and pathways, study gene-gene interaction, and perform additional analyses such as Mendelian randomization, mediation, risk modeling. We will use summary statistics and results from other GWAS studies to compare our results and derive further insights.
 O'Reilly et al., MultiPhen: Joint Model of Multiple Phenotypes Can Increase Discovery in GWAS, PLoS One, 2012. doi: https://doi.org/10.1371/journal.pone.0034861
 van der Meer et al., Understanding the genetic determinants of the brain with MOSTest, Nat. Comm., 2020. doi: https://doi.org/10.1038/s41467-020-17368-1
 Purcell et al., PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses, Am J Hum Genet, 2007. doi: 10.1086/519795