Population-level imaging, genomic and phenotypic analyses to determine how bone marrow adiposity impacts human health

Lead Research Organisation: University of Edinburgh


Stop and think about your bones: what images come to mind? Perhaps a skull with grinning jaws, or the strong white limbs stretching out towards your fingers and toes. You might even think of the bone marrow within them, producing the blood that courses through your veins. But this is not the whole picture, for your skeleton hides a secret: it is full of fat, and no one knows why.

This unsolved mystery is surprising. Scientists first noticed that our bone marrow contains fat-storing cells, called adipocytes, over a century ago. Having fat in our bones might strike you as unusual, but it is not: in humans and other mammals, this bone marrow adipose tissue (BMAT) develops steadily after birth and accumulates rapidly during puberty. Indeed, by the time we reach adulthood BMAT can comprise up to 70% of bone marrow volume, representing almost 10% of our total fat stores. Therefore, it seems likely that BMAT plays some role in the normal functioning of our bodies.

Recent studies suggest that BMAT may also influence numerous diseases. One way of measuring BMAT is by a technique called magnetic resonance imaging (MRI). MRI scans have shown that BMAT further increases with ageing and in many diseases. For example, BMAT often increases in osteoporosis, suggesting that it might contribute to the bone fragility that defines this disease. Increased BMAT also occurs in obesity and type 2 diabetes, metabolic diseases that are placing a huge burden on our society. Therefore, excessive BMAT may lead to poor metabolic health. Finally, BMAT might enhance the growth of tumours within the bone, such as those that have spread from breast or prostate cancers, as well as leukaemia and other cancers that begin in the bone marrow.

Based on these findings, BMAT is now attracting considerable interest as a potential player in the development of numerous diseases. Unfortunately, study of BMAT has been relatively limited, and analysis of BMAT using MRI has never been done across large populations. Consequently, the roles of BMAT in normal physiology and disease remain poorly understood. So, what is the function of BMAT, and how might it impact human health?

Our team of scientists from the Universities of Edinburgh, Westminster and Dundee is now working to answer these key questions. To do so, we will use information being collected by the UK Biobank, a major study that is following the health and wellbeing of 500,000 volunteers from across the UK. The UK Biobank is doing MRI scans of 100,000 participants, which is estimated to be completed before the end of our 3-year research project. Using this MRI data, we will measure BMAT in each participant. This will be done by using artificial intelligence techniques to create computer software for automatic analysis of the MRI scans. These approaches will establish how the amount of BMAT varies across this very large population.

The power of the UK Biobank is that it has also collected DNA samples and health data on each participant. Therefore, once we have measured participants' BMAT we will be able to discover how this relates to other aspects of human health and disease. This includes physiological factors, such as age and sex; genetic factors, such as mutations in particular genes; and disease outcomes, such as heart disease, obesity, diabetes, osteoporosis, and many other conditions of ill health.

Together, this unprecedented, large-scale research project will help to unravel the mystery of BMAT whilst also establishing new methods for automated MRI analysis. The latter will be of huge help to the NHS, which currently faces a major backlog of unanalysed MRI scans. More broadly, understanding the impact of BMAT on human health has great potential to improve diagnoses and treatment of numerous diseases, including osteoporosis, diabetes, cardiovascular disease and several types of cancer. This will be vital if we are to reduce the public health impact of these worldwide health problems.

Technical Summary

Bone marrow adipose tissue (BMAT) accounts for up to 70% of total bone marrow volume and 10% of total fat mass in healthy humans. BMAT further increases with ageing and in diverse clinical conditions, such as osteoporosis, obesity, type 2 diabetes and radiotherapy. Unlike other adipose depots, BMAT also increases in states of caloric restriction, including anorexia nervosa. Thus, BMAT is a major feature of normal anatomy; is distinct to other types of adipose tissue; and is altered in many clinical contexts. However, study of BMAT has been extremely limited, and therefore how BMAT impacts human health and disease remains to be established.

This proposal will fill in this critical gap in knowledge by leveraging the power of the UK Biobank (UKBB). Magnetic resonance imaging (MRI) is emerging as a key tool to measure BMAT, but this has never been done on a population level. UKBB is now MRI scanning 100,000 of its participants; hence, using this data, Objective 1 is to measure BMAT of each participant by developing machine learning-based algorithms for automated MRI analysis. Establishing these new automated MRI analysis tools will yield extensive academic, clinical and commercial benefits. Objective 2 is to use GWAS and PheWAS to identify genetic and phenotypic factors associated with altered BMAT. No previous MRI study has analysed BMAT in more than 600 subjects; hence, our population-level BMAT analysis will provide unprecedented power to identify these associations. Finally, Objective 3 is to use Mendelian randomisation in UKBB and other large-scale cohorts to determine if SNPs associated with altered BMAT are causally linked to physiological and pathological outcomes. This will reveal if BMAT directly influences physiological traits and/or the aetiology of diverse diseases.

In summary, this work will yield cutting-edge approaches for automated MRI analysis while revealing fundamental new knowledge about the impact of BMAT on human health and disease.

Planned Impact

Our research will have multiple benefits across the following stakeholders and end users:

1) Researchers working on this project: This proposal will assemble a research team spanning the diverse fields of BMAT, biomedical imaging, genetic epidemiology and machine learning (ML). This interdisciplinarity provides extensive career development and research opportunities. PI Cawthorn will develop new skills in MRI, ML, population-level human studies and research team management. Co-Is MacGillivray & Semple will extend their existing imaging and ML skills by applying these to population-scale datasets; Bell & Thomas will gain new skills in MRI analysis of BMAT; and Theodoratou & Pearson will gain recognition for use of population-scale imaging data in their research. Finally, the postdoctoral scientist will be trained in MRI analysis, ML and genetic epidemiology; in turn, the PI and Co-Is will extend their supervision and mentorship skills. Given the vast potential of the UKBB imaging study, MRI automation, 'big data' and ML, these skills are highly valued in our research fields and many other employment sectors. We will realise these benefits within our project's 3-year timeframe.

2) Other researchers within and beyond our immediate fields, including biomedical research, ML and imaging; see the 'Academic Beneficiaries' section for further details.

3) The NHS and public health policy-makers: Our research will provide insights into whether BMAT influences multi-morbidities and chronic diseases, including osteoporosis, type 2 diabetes, obesity, cardiovascular disease, haematological disorders, bone metastases and iatrogenic conditions. This will improve our understanding of the pathogenesis of these diseases, thereby benefitting public health by revealing new approaches for prevention, diagnosis and treatment. In addition to the burden of these diseases, the NHS is also struggling with insufficient capacity to analyse MRI data. Indeed, in 2016 the NHS had to pay £88 million for out-of-hours reporting of radiology scans, while nearly 2/3 of vacant radiologist posts sat empty for 12 months or more. Automated methods to streamline MRI analysis would thus greatly benefit both healthcare providers and patients. Once our 3-year project has concluded, the above benefits could be realised within 5-10 years.

4) Commercial sector beneficiaries: The above possibilities would also benefit pharmaceutical and biotechnology firms in the commercial sector. For example, our research may identify BMAT as a new drug target for the above diseases, and/or a biomarker for efficacy during drug development. Algorthims for large-scale automated MRI analysis are desperately needed, representing a major commercial opportunity for firms such as Canon Medical Research. Many companies are also launching products based on new approaches for biomedical image analysis, including Perspectum Diagnostics (LiverMultiScan) and Resonance Health (Ferriscan). Members of our research team have previously worked with these and other companies, underscoring our track record in research translation and commercialisation. Therefore, we are well placed to exploit these opportunities. As above, many of these commercial impacts could be realised within 5-10 years of project completion.

5) The wider public: Most people are completely unaware that our bone marrow is full of fat. The fact that BMAT increases in starvation states is particularly intriguing. By revealing new insights into the health implications of adipose tissue, this research program will be of great interest to the general public and will therefore be widely disseminated through the appropriate media outlets. This will be done within our project's 3-year timeframe. In addition, new strategies to prevent, diagnose and treat multi-morbidities will greatly benefit the UK economy and the health of the wider public. Such benefits might be realised within 10-15 years of project completion.
Title OPTIMAT: Deep Learning algorithm for automated analysis of bone marrow from MRI data 
Description This is a new deep learning model for identifying volumes of interest (VOIs) from magnetic resonance imaging (MRI) data. It is especially accurate for the identification of small VOIs among large 3D imaging volumes, outperforming conventional deep learning models. We are using our new model to analyse the bone marrow fat fraction (BMFF) from MRI data in the UK Biobank. We have a draft manuscript (attached in the files section) describing the innovation, and its potential for establishing the utility of BMFF measurements as a novel clinical biomarker. 
Type Of Material Technology assay or reagent 
Year Produced 2022 
Provided To Others? Yes  
Impact This innovation allows accurate, rapid, high-throughput identification (segmentation) of small anatomical regions from MRI data. The ability to accurately segment very small VOIs from large 3D image volumes is something that conventional deep learning models have so far been unable to achieve. One direct application of our innovation is for analysis of large-scale MRI data available from the UK Biobank. In particular, we have focussed on segmentation of four bone marrow regions, followed by measurement of BMFF in these regions. The goal of this work is to determine the physiological and pathological factors associated with altered BMFF, which will establish the utility of BMFF as a clinical biomarker. However, our new model could also be applied to other small anatomical regions of interest. 
URL https://doi.org/10.1101/2022.12.06.22283151
Description Cherry collaboration (total-body PET) 
Organisation University of California, Davis
Country United States 
Sector Academic/University 
PI Contribution My lab is analysing total-body PET scans from Prof Cherry's lab, to determine glucose uptake into bone marrow adipose tissue (BMAT), red marrow and bone.
Collaborator Contribution Prof Cherry's group has led the development and application of total-body PET scanning. They have provided us with data from their initial cohorts of total-body PET scans using 18F-fluorodeoxyglucose.
Impact None so far.
Start Year 2020
Description Fowler collaboration 
Organisation University of Edinburgh
Department Centre for Discovery Brain Sciences
Country United Kingdom 
Sector Academic/University 
PI Contribution I co-supervised a Master's student in Dr Jill Fowler's lab (University of Edinburgh), with the research focusing on sex differences in the ability of caloric restriction (CR) to protect against cerebral ischaemia, and whether adiponectin knockout influences this effect of CR. Due to the COVID restrictions we had to modify the approach to this project, and therefore adapted it to use UK Biobank data to investigate the relationship between circulating adiponectin (a hormone produced from bone marrow adipose tissue) and stroke. In this project I provided expertise in use of the UK Biobank data, and also connections with my University of Edinburgh collaborators who have further expertise in these analyses.
Collaborator Contribution Dr Fowler secured funding for the Master's student and was lead supervisor on the initial project. Dr Fowler provided expertise in analysis of brain samples for pathways relating to ischaemic injury and neuroprotection.
Impact None so far. We are working on a manuscript reporting the impact of adiponectin on stroke.
Start Year 2019
Description Paccou collaboration 
Organisation Lille University Hospital
Country France 
Sector Hospitals 
PI Contribution My team is leading the analysis of bone marrow adiposity (BMA) in MRI data from the UK Biobank, including development of new machine learning methods to identify bone marrow regions of interest from these MRI data.
Collaborator Contribution Prof Julien Paccou and his colleague, Dr Sammy Badr, are experts in MRI analysis of BMA. They have assisted with identifying the bone marrow sites of interest for our analyses.
Impact No outputs so far.
Start Year 2020
Description Paccou collaboration 
Organisation University of Lille
Country France 
Sector Academic/University 
PI Contribution I initiated the UK Biobank studies to analyse bone marrow adiposity on a population scale. My team then implemented the methods to do this, including deep learning and MRI analyses.
Collaborator Contribution In 2020 we began collaborating with Prof Julien Paccou at the University of Lille and CHU Lille. Prof Paccou provided expertise in analysis of our UK Biobank MRI data, particularly ensuring that we analysed clinically relevant skeletal sites. He and his colleague, Dr Sammy Badr, are co-authors on our initial study.
Impact This preprint: https://www.medrxiv.org/content/10.1101/2022.12.06.22283151v1.full-text And also the invited seminars and other talks related to this work. The collaboration is multi-disciplinary, including radiology, artificial intelligence, musculoskeletal biology, metabolism and endocrinology, gerontology, genomics, epidemiology, immunology, and sex differences.
Start Year 2020
Description Falkirk Science Festival 2022 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Public/other audiences
Results and Impact I designed and led a public engagement event entitled "The Many Faces of Fat and Diabetes" at the 2022 Falkrik Science Festival, on 14/05/2022. The event delivered various interactive activities designed to increase the public's understanding about adipose tissue, metabolism, obesity and diabetes.The main questions that the researchers aimed to answer were: How does our body store fat? Is this always a bad thing? Is all fat the same? Is it just used to store energy, or does fat play other roles in our bodies? What are the differences between Type 1 and Type 2 diabetes? The festival visitors had a chance to measure the blood glucose levels, use a microscope to detect different types of fat and learn more about how different organs store fat tissue. Over the course of the day, we interacted with several hundred members of the public, who asked many questions and clearly learned a lot about the relationship between fat, diabetes and overall health.
Year(s) Of Engagement Activity 2022
URL https://www.ed.ac.uk/cardiovascular-science/public-involvement-resources/public-engagement-news/cvs-...