Study protocol: the Whitehall II imaging sub-study

The sub-study is funded by the Lifelong Health and Wellbeing Phase-3 programme grant “Predicting MRI abnormalities with longitudinal data of the Whitehall II sub-study” (MRC-G1001354; Ebmeier KP (PI), Geddes JR, Kivimäki M, Mackay CE, Singh-Manoux A, Smith SM), as well as the HDH Wills 1965 (English Charity Register: 1117747; Ebmeier KP (PI)), and the Gordon Edward Small Charitable (Scottish Charity Register: SC008962; Ebmeier KP (PI)) Trusts. Collection of blood and buccal mucosal samples for a characterisation of immune function and associated measures is funded by the UK Medical Research Council programme grant K013351 (“Adult determinants of late life depression, cognitive decline and physical functioning - The Whitehall II Ageing Study”, Kivimäki M (PI), Singh-Manoux A, Brunner E, Batty GD, Kumari M, Ebmeier KP, Hingorani A) and the ESRC professional fellowship scheme to Kivimäki.

Ethical approval was granted generically for the “Protocol for non-invasive magnetic resonance investigations in healthy volunteers” (MSD/IDREC/2010/P17.2) by the University of Oxford Central University / Medical Science Division Interdisciplinary Research Ethics Committee (CUREC/MSD-IDREC), who also approved the specific protocol: “Predicting MRI abnormalities with longitudinal data of the Whitehall II sub-study” (MSD-IDREC-C1-2011-71). The Health Research Authority NRES Committee South Central – Oxford B approved the Study: “The Whitehall II Immune Function Sub-study” (REC reference: 13/SC/0072, IRAS project ID: 120516).

The study follows the Medical Research Council (MRC) Policy on data sharing, i.e. images and other data will be available for analysis by other groups after completion of the study, as is the case with the Whitehall II study (see http://​www.​ucl.​ac.​uk/​whitehallII/​data-sharing[8, 9].

Each participant recruited for the WHII imaging sub-study undergoes a detailed clinical and cognitive assessment lasting up to two hours.

The clinical assessment consists of a (A) self-administered questionnaire, a (B) semi-structured clinical interview and (C) cognitive assessment (median = 56 minutes, interquartile range: 51–61 minutes).

Scanning is carried out at the Oxford Centre for Functional MRI of the Brain (FMRIB) using a 3 T Siemens Magnetom Verio (Erlangen, Germany) Scanner with a 32-channel receive head coil. The neuroimaging protocol comprises both structural and functional sequences and lasts approximately 50 minutes. MRI sequences include: a) high-resolution T1-weighted, b) diffusion MRI (dMRI), c) resting-state functional MRI (rfMRI), d) Fluid Attenuated Inversion Recovery (FLAIR) and e) T2*. A full description of the MRI parameters adopted in our sequences is provided in Table 2.

This sequence is primarily used to study grey matter (GM) structural macroscopic tissue in both cortical and subcortical brain regions. GM reductions have been widely associated with impending disease and age-related cognitive dysfunction [67‐69].

A Multi-Echo MPRAGE (MEMPR) with motion correction, developed at the Massachusetts General Hospital (MGH, Boston), was employed [70, 71]. This sequence has the advantage of combining the properties of the classical MPRAGE sequence, which has high contrast aiding cortical segmentation, with Multi-Echo FLASH, which improves segmentation of subcortical regions.

The pre-processing pipeline includes: a) re-orientating images to the standard (MNI) template, b) bias field correction, c) registration to the MNI template using both linear (FLIRT) and non-linear (FNIRT) registration tools and d) brain extraction. Brain tissues are segmented using FMRIB's Automated Segmentation Tool (FAST) that allows extracting measures of total GM, WM and cerebrospinal fluid (CSF). FIRST (http://​fsl.​fmrib.​ox.​ac.​uk/​fsl/​fslwiki/​FIRST) [72], an automated model-based segmentation/registration tool, is applied to extract subcortical grey matter structures. Brain tissues and subcortical regions are visually inspected to ensure an accurate segmentation (Figure 2A,B).

Diffusion MRI exploits the principles of traditional MRI to measure the random motion of water molecules and subsequently to 1) infer information about white matter (WM) microstructural properties and 2) delineate the gross axonal organisation of the brain [73]. WM is characterised by bundles of myelinated axons surrounded by myelin sheaths that are built up by layers of membranes. This restricts diffusion of free water molecules; i.e. the myelin layers and the axonal membrane cause a lower restriction along than across the axon and thus a higher anisotropy.

A number of strategies were used to minimise distortions caused by, for example, magnetic susceptibility, eddy-currents, and subject-motion. We employed monopolar diffusion encoding gradients with parallel imaging (GRAPPA) to minimise echo time, which increases the signal to noise ratio (SNR), at the cost of a small increase in eddy-current distortion. We used a recently developed dMRI correction strategy that takes advantage of the complementary information from pairs of diffusion images acquired with reversed phase-encoding (PE) directions to correct for susceptibility-induced distortions [74]. A single non-diffusion weighted (b-value = 0 s/mm²) volume with reversed PE was combined with the non-reversed dMRI data to estimate an off-resonance field, which is then applied to correct susceptibility distortions [75].

The Pre-processing pipeline uses a recently developed approach that simultaneously considers and corrects for susceptibility-induced distortions, eddy-currents and head motion (based on methods developed and applied to the Human Connectome Project (HCP) diffusion MRI data [74]). Briefly, a generative model approach is used to estimate all types of distortion and a single resampling step with spline interpolation is used to correct for all of them simultaneously. Fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD) maps are generated using DTIFit, part of FMRIB’s Diffusion Toolbox (http://​fsl.​fmrib.​ox.​ac.​uk/​fsl/​fdt), that fits a diffusion tensor model at each voxel [76, 77] (Figure 3, b). Subsequently, crossing fibre orientations can be estimated, and probabilistic tractography can be performed to reconstruct white matter bundles and assess structural connectivity [78].

rfMRI is used to investigate resting state networks (RSNs), which comprise brain regions sharing a common time-course of spontaneous fluctuations, and are thought to represent functionally-critical neuronal networks that reflect properties of functional brain organisation [79]. RSNs have been consistently observed across subjects, sessions and functional brain imaging modalities (fMRI, PET, EEG), and their presence has also been reported in studies when participants were asleep, and in anaesthetized monkeys and rats [79‐85]. Although potentially harder to interpret than task-based fMRI, the rfMRI approach has the considerable advantage of providing an assay of brain function without requiring subjects to undertake a specific task, particularly in cases where a subject may be less able to understand and/or respond to complex instructions.

We compared a recently developed Multiband MRI sequence [86, 87] with ‘standard’ EPI. Multiband provides a considerable improvement in temporal (Multiband: 1.3 seconds vs. Standard EPI: 3 seconds) and spatial (Multiband: 2 mm isotropic vs. Standard EPI: 3 mm isotropic) resolution, which allows: a) better definition of the spatial maps, b) wider frequency range exploration in time-series analyses and c) more detailed network analyses. To ensure that the new multiband sequence was robust in our older population, we acquired both sequences (standard and multiband) on a subset of participants (N = 76). Results of this comparison will be presented in the next section. In all cases subjects were instructed to lie in dimmed light with their eyes open, blink normally, but not to fall asleep.

rfMRI data pre-processing (motion correction, brain extraction, high-pass temporal filtering with a cut-off of 100 s, and field-map correction is carried out using MELODIC (Multivariate Exploratory Linear Optimized Decomposition into Independent Components, part of FSL http://​fsl.​fmrib.​ox.​ac.​uk/​fsl/​fslwiki/​melodic/​) [88, 89]. In order to reduce the presence of spatially and/or temporally-related artefacts a data-cleaning approach is applied. Single-subject independent component analysis (ICA) is followed by an automatic component classification with FMRIB's ICA-based X-noiseifier (FIX) to identify and regress out the “signal” of the artefactual components reflecting non-neuronal fluctuations [90, 91]. The pre-processed and “cleaned” functional data are registered to the individual's structural scan and standard space images using FNIRT, then optimized using boundary-based-registration approach [92], and finally spatially smoothed using an isotropic Gaussian kernel of 6 mm full width at half maximum (FWHM).

This project represents the first major application of the Multiband sequence in older adults: therefore, as part of the protocol development, the following were tested: a) whether it was possible to identify the previously reported set of “canonical” RSNs with the Multiband sequence [81, 88]; b) whether the Multiband sequence was effectively associated with a “better” signal relative to a Standard EPI sequence. For the first analysis MELODIC was used across the first 50 consecutive participants and was able to detect a set of RSNs that matched the “canonical” RSNs (Figure 4). For the second analysis a direct comparison was made between Multiband and Standard EPI sequences that had both been acquired on a subset (N =76) of participants. In order to be able to compare the two sequences with respect to spatial detail, spatial smoothing was not applied in this analysis. Additionally, an independent set of RSNs’ spatial maps, derived from data acquired for the Human Connectome Project (HCP), was used as a common template to extract time series and spatial maps from Multiband and Standard EPI sequence data using the dual regression approach [93, 94]. HCP templates are of higher resolution and therefore contain more information compared with the standard template maps. Group maps were then obtained performing a one-sample t-test on the subjects’ spatial maps (output of the second stage of dual regression) for each component, calculating the corresponding z map and applying a mixture model correction to ensure comparable null distributions [95]. Figure 5 shows the quantile-quantile plots (Q-Q plots) of the group maps against the normal distribution. The group maps’ z-statistic distribution was expected to follow a normal distribution around zero (random noise – the null part of the spatial maps), while the tails should deviate from it according to the RSNs’ identified signal. Therefore our results suggest that Multiband sequence allows stronger RSN signal detection (higher z-values) with respect to Standard EPI sequence.

This sequence is commonly used in clinical practice, for example to characterise periventricular lesions adjacent to the sulci, WM hyperintensities and WM lesions. Age-related increase in the number of hyperintensities and/or periventricular lesions has been widely reported [96, 97]. Moreover, it has been shown that WM changes are associated with subtle structural and functional brain changes [98], and with decline in essential cognitive abilities in healthy elderly people [99]. Clinical ratings of FLAIR are performed by trained neuroscientists based on visual rating scores. WM lesions and/or hyperintensities are visually identified on FLAIR images by three independent raters who make assessments blind to demographic details. All axial slices of each subject are visually inspected. Peri-ventricular and deep WM hyperintensities are rated separately using a 4-point ordinal scale from 0–3; the sum of these two ratings lead to the total Fazekas score (an integer from 0–6) [100] (Figures 6A and B).

The T2* sequence allows the identification of cerebral microbleeds, which reflect small deposits of the iron-storing protein hemosiderin in the brain and may be a sign of cerebral small-vessel disease [101, 102]. Microbleeds can be found all over the brain and have been shown to be associated with neurological dysfunction and both clinical and cognitive impairment [103, 104]. Their underlying mechanism is still under investigation but the deleterious effect, probably due to inflammatory effects, has been proved to affect neuron functionality and/or cortical cerebral activity [105]. All axial slices of T2* images are visually explored performed by trained neuroscientists and microbleeds identified (Figure 6C).

Following the MRI examination, blood samples (3 × 8 ml) and two buccal mucosal epithelial samples are taken from each participant. Blood samples are drawn using Vacutainer CPT tubes (Becton Dickinson) in a single venepuncture, inverted 8 times to mix anticoagulant additive with blood and processed within two hours of collection. Buccal mucosal epithelial cells are collected by nylon swabs (microbiome sample) and by twirling the brush against the epithelium and shaking the brush in RNAlater solution for 15 sec to remove the cells (RNA sample). CPT tubes are centrifuged in 1600 g for 30 min, allowing the separation of serum from white blood cells and from red blood cells. After the centrifugation, serum samples and peripheral blood mononuclear cells (PBMCs), i.e. T cells, B cells and NK cells are collected. PBMC cells are washed with phosphate-buffered saline (PBS), counted under microscope and cryopreserved using Cell Freezing Medium (5% DMSO/11%HSA in PBS). All samples are stored in −80°C for later analysis.

In order to investigate immune function in sample material, PBMC cells are stimulated in vitro with αCD3/αCD28 mAb and with TLR2/1 ligand Pam3CSK4 (Pam₃Cys) and TLR4 ligand LPS in RPMI at 37°C in 5% CO_2. Cells and supernatants are collected after 6 and 24 h of stimulation and stored for later analysis. RNA samples are investigated using RT-PCR or transcriptomics, while serum samples and cell supernatants are analysed with ELISA, Luminex or proteomics. Also, genomics/epigenomics and metabolomics techniques as well as advanced bioinformatics tools will be employed in the sample analysis.

We will examine the following overarching hypotheses in the sub-cohort recruited from WH II Phase 11:

(1) Early risk factors for cognitive dysfunction and late onset depression (such as cardio-vascular risk, see (i.) to (v.) below) will be associated with cerebral atrophy, reduced perfusion and impaired white matter integrity. Resulting impaired cognitive function (as measured by tests of episodic memory and executive function, see below) will be associated with specific brain volumes measured by advanced methods such as voxel based analysis and the automatic identification of specific structures, such as hippocampus, and also with reduced white matter integrity, quantified both globally [106], and homing in on fibres of interest (including measures of structural connectivity) [78];

(2) Resilience, measured in the sub-study directly [107] and by the absence of depression and cognitive impairment particularly in the presence of higher early risk (see above), is associated with high white matter integrity [108], and increased coherence of frontal regions in resting state networks (which provides a measure of functional connectivity) [109] or (compensatory) increased frontal BOLD signal in a memory encoding task [93]. The most important risk factors will be: (i) Antecedent vascular and metabolic risk trajectories and morbidity (adverse major blood lipids/apolipoproteins, hyperglycaemia and diabetes, adiposity, high blood pressure, smoking, chronic inflammation (C-reactive protein, interleukin 6), ECG abnormalities, angina, myocardial infarction, stroke; Framingham cardiovascular risk scores); (ii) low antecedent levels of physical and mental activity (measured by questionnaire [110‐112]); (iii) baseline cognitive performance levels and up to 15-year gradients of memory and executive function decrement; (iv) history of depressed mood (3 to 8 repeat measurements using the CESD [12] and the GHQ [10], (v) genotype (APOE4 plus >48 k single nucleotide polymorphisms relevant to cardiovascular, metabolic and inflammatory syndromes [113]). The primary clinical hypotheses which we will address are: a) The quality of frontal compensatory activity will be affected by vascular risk, hypertension, including absence of protective factors, such as physical fitness, cardio-vascular and cerebrovascular prophylaxis (aspirin, statins, antihypertensives), by mental activity (“use it or lose it”), and by frontal lobe atrophy (as observed in treatment resistant depression); b) Time-trajectories derived from the WHII data set, e.g. of cognitive function or depression scores, and vascular risks-factor trajectories, will account for more of the diversity of outcome than singular measurements during any one of the previous follow-phases. This may allow imputations about the natural history of brain changes, but from clinician’s point of view, the least labour and time-intensive predictive test will be the most attractive. An important clinical task would, therefore, be to determine the minimum data set to predict outcome. c) Overall resilience, as supported by the absence of current and past affective or cognitive symptoms and good performance [78], can be modelled longitudinally from observations antecedent to medial temporal atrophy and frontal compensatory activity and structural integrity (high FA in DTI scan) as described in (a) and (b). In addition, we will address the following specific hypotheses which are of independent interest but will also support the primary findings: (d) Global cognitive performance depends on both hippocampal and frontal integrity/connectivity and (compensatory) frontal activity measured relatively higher frontal coherence within executive resting state networks; (e) Clinical impairment (outcome) appears if functional frontal compensation does not keep up with the degree of hippocampal and frontal atrophy observed; (f) First onset of depression after the age of 65 compared with no depression ever (outcome), matched for age, sex, education, and potential causal factors, such as vascular risk, is associated with reduced frontal/executive network structural and functional connectivity; (g) Clinical impairment may present as impaired cognition and/or as major depressive syndrome depending on the frontal networks affected; (h) Greater hippocampal atrophy will be observed with increasing age and APOE ϵ4 alleles; the absence of antidepressant medication; and among those with a family history of dementia.