The following projects are linked to the main application that holds the raw data. Do you have a project that is missing or want to link to this application to share access to raw data then please get in touch.
- Current projects
- Investigating the brain structural and functional correlates of hallucination-like experiences
- Machine learning classification
- Polygenic risk score for obesity as a risk factor neurodegeneration in an ageing population
- Latent phenotypes of impulsivity (linked application: 43332)
- Detecting early imaging phenotypes of tau associated dementia in the general population (linked application: 46620)
- Neuroinflammatory biomarkers of depression and its neurobiological underpinnings (linked application: 48943)
- Peripheral immunity and its role in differentiating between resilience and depression in adults with a history of childhood maltreatment. (linked application: 54358)
- Multifactorial Integrative Network Inference of NeurodeGeneration (MINING) (linked application: 56956)
- How does substance misuse affect brain networks in ageing and its relation to impulsivity and negative emotion (linked application: 64044)
- Differentiation of dementia from late-life depression using brain imaging techniques (linked application: 58271)
- EXTERNAL:: Association of immune GWAS loci with brain structure, function, and mental health (linked application: 48096)
Graham Murray, John Suckling and Colleen Rollins
Hallucinations are perceptions of stimuli, such as hearing voices or seeing visions, that do not exist in the physical world (Woods et al., 2015). Though a core symptom of schizophrenia, hallucinations are transdiagnostic phenomena and occur among the general population at a rate of 5-13% (Maijer et al., 2018). A growing body of neuroimaging studies have implicated specific brain structural markers and alterations in the brain’s resting state networks in the pathophysiology of hallucinations (Garrison et al., 2015; Alderson-Day et al., 2015). However, many findings are inconsistent, reporting opposing directionalities, and interpretation of findings has been limited by small sample sizes and methodological heterogeneity. Here, I propose a multimodal approach to investigate the brain structural and functional correlates of hallucination-like experiences amongst a large sample of individuals with and without clinical diagnoses. Items assessing hearing un-real voices (Field ID 20463, 20465) and seeing un-real visions (Field ID 20471, 20473) will be correlated with metrics from T1 structural and resting state functional MRI data (Category ID 100) to elucidate the neurobiological basis of hallucinations. Field IDs relating to diagnosis, unusual and psychotic experiences, cognitive function, and medication may also be used as covariates or follow-up analyses.
References:
- Woods A, Jones N, Alderson-Day B, Callard F, Fernyhough C. Experiences of hearing voices: analysis of a novel phenomenological survey. The lancet Psychiatry 2015; 2(4): 323-31.
- Maijer K, Begemann MJH, Palmen S, Leucht S, Sommer IEC. Auditory hallucinations across the lifespan: a systematic review and meta-analysis. Psychological medicine 2018; 48(6): 879-88.
- Garrison JR, Fernyhough C, McCarthy-Jones S, Haggard M, Australian Schizophrenia Research B, Simons JS. Paracingulate sulcus morphology is associated with hallucinations in the human brain. Nat Commun 2015; 6: 8956.
- Alderson-Day B, McCarthy-Jones S, Fernyhough C. Hearing voices in the resting brain: A review of intrinsic functional connectivity research on auditory verbal hallucinations. Neuroscience and biobehavioral reviews 2015; 55: 78-87.
John Suckling & Matt Lemming
We plan to use the Biobank data for two different tasks. One will classify functional MRI data, using the derived functional connectome (for which the fMRI timeseries and the T1-weighted MRI are necessary), as a proof-of-concept of the effectiveness of machine learning models on large amounts of functional connectivity data. We would like to classify fMRI data by disease type, focusing on the most common neurological illnesses, such as dementia, schizophrenia, and Alzheimer's. However, if the sample sizes of patients with relevant diseases are insufficient, we will classify patients by covariates as a proof-of-concept of the machine learning model itself, using three commonly-recorded covariates: handedness, sex, and age at time of scan. The second task is to derive simple measurements (grey matter/white matter/CSF percentages, hippocampal volume, and others) from structural MRI data in an unsupervised fashion in order to inform a machine learning model, which will offer information about the risk of a certain patient to a specific disorder. Together, these tasks will investigate the possibility of a general-purpose, fully automated classifier of neurological and psychiatric conditions, based on MRI data.
And other information:
Deep learning models have seen high success in recent years in classifying MRIs based on diseases such as dementia, schizophrenia, and Alzheimer's. Recent advances in deep learning bring even more promise to this field. Nonetheless, deep learning in medical imaging uniquely suffers from low sample size with which to train the model, often forcing researchers to artificially augment data. The UK Biobank's large datasets, however, offer an opportunity to classify MRIs by disease type on a previously inaccessible scale.
For fMRI data, we seek to create a proof-of-concept machine learning model that can classify large amounts of connectivity data by common covariates, such as age, gender, and handedness, to assess which machine learning model is best-equipped to this task. This would elucidate which model is best-equipped to classify common psychiatric illnesses, and, if successful, would motivate acquisition of large amounts of fMRI data with which to train such a model. If the UK Biobank has sufficient numbers of subjects with a common mental or neurological illness, we will train a model based on those covariates, though this is primarily a methods study that is agnostic to any particular illness.
For a second project concerning structural data, we seek to build a control dataset of T1-weighted MRIs against which to compare in-house data. We will derive simple measurements from these scans, to use as a baseline against in-house data for an automated analysis system. We have developed a system using FSL tools to automatically derive measurements (white matter, gray matter, and cerebrospinal fluid percentages, as well as hippocampal volume) from T1-weighted MRIs. To compare our results to data in other datasets (such as ADNI), we would need to use our methods to derive volume information from the raw T1-weighted images directly.
Lisa Ronan
Structural brain change is a normal part of human aging and coincides with cognitive decline. In turn these are known risk factors for neurodegenerative diseases such as dementia and Alzheimer’s disease (Yanker 2000). At the same time, aging populations face the additional burden of increasing levels of obesity(Arterburn 2004). Significantly, obesity and cognitive decline are thought to be related. Indeed it has been reported that individuals who are obese in mid-life are 74% more likely to develop dementia than their lean counterparts (Whitmer 2005). It is speculated that obesity and cognitive decline are causally related(Debette 2011). For example, it has been proposed that adipose tissue itself may result in dysregulation of various endocrine functions critical to memory processing, neurogenesis, neuroprotection, as well as vascular health. At the same time, it is also suggested that because obesity and aging are similar at a cellular-pathway level, obesity may increase rates of neurodegeneration and by extension, raise the risk of cognitive decline and dementia.
However more recent studies on the genetics of obesity have raised another intriguing possibility, namely that the genes involved in obesity-risk are also involved in determining brain structure and cognition. In other words, the association between obesity and cognitive decline may be due to shared genetic action, i.e. the same genes influencing different traits (pleiotropy). Body mass index (BMI) itself is significantly influenced by genes, with heritability estimated at 70-80% (Stunkard 1986). Importantly, the genes related to obesity-risk are predominantly enriched in the CNS including cerebral tissue (Locke 2015), and are involved in basic functions such as synaptic function and glutamate signaling. Significantly, it has recently been reported that genes associated with obesity are related to variability in brain structure and various cognitive functions, suggesting a possible genetic mechanism by which obesity relates to neurodegeneration and by extension cognitive decline (Hagenaars 2016; Vainik 2018; Ronan et al., in preparation). Taken together these findings raise the possibility that a genetic predisposition to obesity may constitute a liability to brain structural changes that may in turn influence cognitive function independent of other factors related to obesity such as vascular risk or endocrine dysregulation as suggested elsewhere. This hypothesis is supported by other studies which demonstrate that such co-morbidities fail to fully mediate the relationship between BMI and brain structure (Bettcher 2013). The aim of the current project is to extend this work and to determine whether a genetic predisposition to obesity underpins the association between BMI and neurodegeneration in an aging population. The hypothesis that a genetic predisposition to obesity may be related to brain structural changes and by extension cognitive function through shared genetic action is a novel approach to investigating the epidemiological association between BMI and dementia risk.
Samuel Chamberlain
Impulsivity refers to a tendency towards premature, unduly hasty, risky behaviours, which lead to untoward functional consequences (e.g. Evenden, Psychopharm, 1999). Impulsivity manifests in extreme forms in mental disorders, notably attention deficit hyperactivity disorder (ADHD), and addiction related disorders. However, recent work indicates that impulsivity can be measured dimensionally, i.e. along a continuum, in the background population (e.g. Chamberlain et al., Psych Med, 2018). By going beyond the conventional 'case-control' study design to explore impulsivity, population cohorts are ideally suited to characterising neurobiological processes implicated in these disabling symptom manifestations. This proposal seeks to utilize the Biobank database to address two core aims: (i) Identify latent phenotypes of impulsivity using a rich set of measures; and (ii) Link these intermediate phenotypes identified in step (i) to polygenic risk scores (for ADHD, conduct disorder, antisocial personality disorder, derived from prior data papers; as well as those identified via GWAS for Biobank itself), brain structure/function (fronto-striatal circuitry involved in reward and habit learning), and inflammation (single nucleotide polymorphisms of relevant genes, and blood-based inflammatory markers).
Detecting early imaging phenotypes of tau associated dementia in the general population (linked application: 46620)
Timothy Rittman
There are currently no available disease modifying treatments for neurodegenerative diseases. One of the main challenges is to identify a population of people to treat early enough in the disease process when there are few or no symptoms of dementia. Using neuroimaging as an early diagnostic biomarker may help to address this problem, however no such biomarkers currently exist. Progressive Supranuclear Palsy (PSP) and Frontotemporal Dementia (FTD) are two forms of dementia both associated with the abnormal hyperphosphorylation and aggregation of the protein tau in neurons and glial cells. Clinically they are characterised by changes in cognition, behaviour and movement. We have identified neuroimaging fingerprints in PSP and FTD derived from structural and functional MRI. We will investigate the extent to which such disease state fingerprints exist among the cohort imaged within the UK Biobank in order to identify a group at high risk of developing these diseases. We will then use outcome data to provide insight into this high risk group. The UK biobank neuroimaging data will be used to identify a population who most strongly express the disease associated neuroimaging fingerprints. For this data we will use T1 structural neuroimaging data and resting state functional MRI data. In order to characterise the identified population and look for signs of early disease, we will use clinical measures that are relevant to dementia. This includes background demographic and basic health data, medical history and current medication use, cognitive measures, and recent health (including measures of mobility and falls). We expect to be able to identify an at-risk group of individuals who are associated with imaging changes of tau associated dementias. We will test the hypothesis that these people will show changes in cognition, mobility and a recent decline in their health status.
Neuroinflammatory biomarkers of depression and its neurobiological underpinnings (linked application: 48943)
Richard A.I. Bethlehem & Edward T. Bullmore
There is compelling evidence that inflammation is often associated with and can cause depression. Owing to the large heterogeneity within individuals with depression, future trials of anti-inflammatory drugs in major depressive disorder (MDD) will likely focus on subgroups of depressed patients enriched by peripheral biomarkers of immune system activation that are mechanistically predictive of response to treatment.
C-reactive protein (CRP) is one plausible biomarker for clinical samples in major depressive disorder. Elevated CRP levels have been strongly associated with depressive symptoms and MDD in several large population samples and clinical meta-analyses. There is evidence from post hoc analysis of clinical trial data that depressed patients with higher levels of CRP at randomization were more responsive to anti-depressant effects of an anti- TNF antibody (infliximab). Thus it is hypothetically plausible that MDD patients with high blood levels of CRP might be more responsive to anti-inflammatory anti-depressants.
It is currently unclear how peripheral immune states indexed by high CRP levels have effects on mood via an intermediate effect on brain structure and function more generally. The ideal peripheral biomarker would be mechanistically validated so that it could be linked to central brain states associated with mood disorder and predictive of a therapeutic (anti- depressant) response to an immunomodulatory drug. To date, however, there have been no studies systematically exploring the relationship between peripheral CRP levels and brain states measured by neuroimaging and other central biomarkers.
Furthermore, MDD has previously been associated with a raised peripheral neutrophil/lymphocyte ratio. More recently, our analyses of peripheral blood cell counts demonstrated that MDD is associated with increased neutrophils, helper (CD4+) T-cells and intermediate monocytes, and correlates with depression severity. This finding of immunological heterogeneity within the inflamed MDD cohort has implications for treatment stratification and requires replication and further investigation in a larger cohort.
Peripheral immunity and its role in differentiating between resilience and depression in adults with a history of childhood maltreatment. (linked application: 54358)
Anne-Laura van Harmelen, Richard A.I. Bethlehem & Sofia Orellana
Experiences of childhood maltreatment (CM) provide up to a fourfold risk of Major Depressive Disorder (MDD). However, a small subset of those affected by CM remain mentally healthy and are deemed resilient. The overall aim of this project is to elucidate some of the neurobiological mechanisms that differentiate between outcomes of psychopathology after CM, as these are poorly understood, by assessing the relationship between peripheral immunity and brain structure and function.
To this end, our first research question asks whether pro-inflammatory biomarkers, which have been shown to be elevated in adults with a history of CM, are lower in resilient individuals. The second, and main, research question asks which set of brain areas have different grey matter volume in resilient vs vulnerable individuals with a history of CM, and whether these differences are associated with their levels of pro-inflammatory biomarkers. Finally, we also ask the same question but for resting state fMRI connectivity within networks relevant for emotional cognition, instead of grey matter volume
Multifactorial Integrative Network Inference of NeurodeGeneration (MINING) (linked application: 56956)
Kamen Tsvetanov, Timothy Rittman & Richard A.I. Bethlehem
With the global demographic shift towards an older population and increasing burden of dementia in ageing societies, there is a pressing need to maintain mental wellbeing into late life, allowing people to work and live independently for longer. The global ambition for disease-modifying treatment for dementia, and stratified approaches to secondary prevention, is hindered by the lack of knowledge about how - and when - the molecular pathological substrates of pathological ageing change human brain function sufficiently to cause cognitive decline.
A mechanistic and integrative model of healthy and pathological ageing must incorporate models of the physiological and pathological effects of dementia and ageing. Our previous work has shown how cognitive function, and response to treatment, critically depends on brain connectivity, with separate contributions from structural and functional connectivity (including frequency- and network-specific interactions). Our studies show that (i) across the adult lifespan, cognitive function across multiple domains becomes increasingly dependent on the strength of integration in distributed brain networks (measured by MRI or MEG); (ii) there are genetic moderators of this connectivity; (iii) connectivity correlates with the progression of pathology across the brain (integrating MRI and MEG with PET); and (iv) connectivity changes correlate with cognitive impairment in patients with distinct genetic mutations.
Despite this apparent convergence of evidence, there are three unresolved issues.
First, how do structural versus functional brain-behaviour relationships differ in pre-symptomatic disease states?
Second, do changes in the structure and functional dynamics of multiple networks have independent or synergistic effects on cognitive function, at the pre-symptomatic and/or symptomatic stages of disease?
Third, what are the genetic influences on the emergence of disease-related changes in network dynamics, that in turn determine the onset of symptoms?
How does substance misuse affect brain networks in ageing and its relation to impulsivity and negative emotion (linked application: 64044)
Valerie Voon, Ying Zhao & Richard A.I. Bethlehem
Substance misuse can cause serious health issues. For example, smoking is one of the biggest causes of death and illness in the UK and alcohol misuse increases the risk of serious health problems (nhs.uk, 2018). Our research will start with the brain health of smoke/alcohol use groups and move on to other substance misuses afterwards. In brain imaging studies, evidence has shown differences in brain structure, activation and connectivity between smoke/alcohol use groups compared to individuals who do not (see reviews Voon et al., 2020 for alcohol use and Jasinska et al., 2014 for smoking). Resting-state functional connectivity in alcohol users shows a widespread pattern of decreased functional connectivity. For example, there is decreased connectivity in the precentral, visual and insular cortex which associated with reduced awareness, in the precuneus and supplementary motor regions which related to body control deficits (Vergara et al., 2017, 2018), and in subthalamic connectivity with ventral striatum and subgenual cingulate which associated with elevated premature responding (Morris et al., 2016). Besides, increased connectivity is shown between medial orbitofrontal and anterior cingulate cortex that related to reward processing and impulsivity (Cheng et al., 2019). The smoking group shows a more diverse pattern as short-term smoking has been reported to enhance cognition, while chronic use was linked to cognitive impairments and/or deterioration in midlife and old age (see review Conti et al., 2019). Such dissociation is reflected in enhanced functional connectivity from cingulate to frontal and parietal regions (in acute smoking), while chronic smoking related to weaker connectivity between dorsal anterior cingulate and ventral striatum (Hong et al., 2009 and Jasinska et al., 2014). Despite the comprehensive brain dysfunction studies related to smoking/alcohol use, less is known how smoking/alcohol use affects brain health, particularly brain networks, in ageing, impulsivity and negative emotion domains (e.g. depression and anxiety).
Differentiation of dementia from late-life depression using brain imaging techniques (linked application: 58271)
Zoe Kourtzi, Onno Kampman, Varun Warrier & Richard A.I. Bethlehem
We want to study how to differentiate dementia and late-life depression before clinical onset of either. This is relevant because they are often interlinked and misdiagnosed as the other.
We will look at several brain imaging modalities; DTI, T1, T2, and fMRI scans. Promising features include white matter hyperintensities, microbleeds, structural features, as well as features extracted from functional connectivity. We will jointly model dementia and depression, and look at higher level interplay between their predictive features. We will also extract features from resting state functional connectivity using graph theoretic approaches, which we hypothesise contain rich information about these two conditions. Many of these features are not used in clinical settings, and could provide much more nuanced insights for clinicians. In terms of modelling, we will start with more traditional classifier approaches such as random forests, as well as explore deep learning approaches. It will be important to understand what these models are learning, so that we gain insight in how these conditions differ.
We will also model the relationship between the genome and the brain using polygenic risk scores. These can be considered as independent validations, and an additional type of subject label.
EXTERNAL:: Association of immune GWAS loci with brain structure, function, and mental health (linked application: 48096)
Michael Gandal, Bogdan Pasaniuc, Aaron ALexander-Bloch & Richard A.I. Bethlehem
Aim 1. Examine how genome-wide polygenic risk for immunophenotypes relate to variability in brain connectivity, structure, and mental health.
Weighted polygenic risk scores (PRS) will be computed by summing the number of single nucleotide variants (SNPs) associated with inflammatory markers (e.g., CRP) or autoimmune disorders comorbid with neuropsychiatric disease (e.g., Crohn's disease). PRS will be related to MRI measures of structural and functional connectivity, cortical thickness, and mental health. We will also investigate whether exposure to stress (e.g., low income, social support) moderates the relationship between immune-related polygenic risk, neuroimaging measures, and mental health.
Aim 2. In a subset of individuals for whom mental health data is available both pre and post-MRI scan, assess how changes in mental health are predicted by environmental, neural, and genetic factors.
We will model the effects of PRS, stress, and brain structure on pre to post-scan changes in mental health. Pre-scan mental health data will be gathered from the baseline visit (2006-2010) and post-scan mental health data from hospital records, and online follow-up questionnaires for Healthy Minds and Thoughts and Feelings. Computational models will be created using a variety of methods (e.g., multilevel modeling, latent change score structural equation modeling).