Fat Fraction and Iron Concentration in Lumbar Vertebral Bone Marrow in the UK Biobank

ABSTRACT

Background

Vertebral bone marrow (VBM) plays a critical role in bone homeostasis and metabolic health. Alterations in VBM fat and iron composition have been linked to age‐related metabolic and musculoskeletal disorders, yet remain underexplored in large population‐based studies.

Objectives

This study aimed to assess VBM adipose tissue and iron concentration in the UK Biobank imaging cohort (N = 26,524) using magnetic resonance imaging (MRI).

Methods

VBM adipose tissue using two approaches: fat fraction (FF) measured from 2‐point Dixon MRI images and proton density fat fraction (PDFF) from multi‐echo MRI scans, along with iron concentration from multi‐echo MRI images, were measured. Sex‐specific relationships between VBM measures, anthropometric and lifestyle factors as well as disease status were explored using correlation and linear regression analyses.

Results

VBM FF and PDFF were higher, whereas VBM iron concentration was lower in participants with osteoporosis and type‐2 diabetes (T2D; p < 0.00016). VBM FF and PDFF were positively associated with visceral adipose tissue and T2D and were inversely associated with spine bone mineral density (BMD) and total muscle (p < 0.00016) in both sexes; however, positive associations with smoking were observed only in women. VBM iron concentration was significantly positively associated with visceral adipose tissue, spine BMD, and alcohol intake, but negatively associated with T2D in men only.

Discussion

These findings enhance the understanding of VBM measures in metabolic health assessments, highlighting their role as potential indicators of metabolic and musculoskeletal health.

1. Introduction

Obesity, a chronic and multifactorial disease, is characterized not only by excess adiposity but also by dysfunctional fat distribution and ectopic fat accumulation [1]. Bone marrow adipose tissue (BMAT) accounts for approximately 8% of total body fat mass and plays a significant role in bone homeostasis, hemopoiesis and energy metabolism throughout the body [2]. Recent studies suggest that elevated BMAT is observed in a wide range of clinical conditions, including obesity, type 2 diabetes (T2D), osteoporosis and sarcopenia [3, 4, 5]; thus, understanding the pathophysiology and structural alterations that may accompany changes in the deposition of adipose tissue in the bone marrow is of great clinical importance, which collectively contributes to rising healthcare costs and diminished quality of life [3, 6].

Variations in bone marrow adipose tissue are closely related to age and body composition [4], and its distinct physiology may be linked to the development of metabolic disorders [7]. In contrast to white and brown adipose tissue depots and ectopic adipose tissue, bone marrow adiposity increases in states of caloric restriction, such as anorexia nervosa [8], and decreases upon weight recovery [9]. Despite this, the role of bone marrow fat and iron as indicators or mediators of obesity‐related pathology remains underexplored. As such, investigating VBM composition may provide novel insights into obesity‐related musculoskeletal decline and metabolic dysfunction.

The structural composition of vertebral bone marrow (VBM) is predominantly assessed using chemical shift encoding‐based water‐fat magnetic resonance imaging (MRI) techniques, which permits the calculation of proton density fat fraction (PDFF) [10, 11]. Assessment of bone marrow fat fraction has been employed in a number of smaller human cohort studies, revealing associations with various aspects of metabolic health [12]. Previous large‐scale studies in population imaging cohorts have emphasized the metabolic significance of bone marrow fat, including assessments across multiple skeletal regions [13].

Vertebrae bone marrow fat, whether measured as PDFF from multiecho sequences [14] or as fat fraction from 2‐point Dixon sequences [15], has been shown to strongly correlate with visceral adipose tissue and insulin resistance, indicating that the measurement may be used to screen subjects at risk of developing metabolic syndrome‐associated features. While both dual‐echo and multi‐echo MRI techniques allow estimation of vertebral bone marrow fat fraction [16], multi‐echo‐derived PDFF is considered more accurate as it accounts for confounding T2* effects and provides a more robust assessment of tissue fat content [17]. Furthermore, the deposition of adipose tissue in VBM is associated with the arthritic inflammatory condition of joints and ligaments of the spine and ankylosing spondylitis [18], and may be considered a marker for early development of these conditions [19]. Assessment of the VBM fat fraction has also been shown to improve fracture discrimination power when used in combination with bone mineral density (BMD) [20, 21].

Beyond fat fraction, VBM also serves as a reservoir for iron storage, with iron content increasingly recognized as a potential biomarker of metabolic dysfunction and bone health [22]. Developing methods for routine and large‐scale measurement of VBM iron may be useful for differentiating between types of anemia [23], monitoring iron overload and serving as a prognostic indicator in myelodysplastic syndromes [24]. Furthermore, iron overload is associated with bone weakening, characterized by reduced bone mass, osteopenia, osteoporosis, and bone fractures [25]. More broadly, iron accumulation is an important marker of organ health, with adipose tissue iron overload in [26] linked to increased inflammation, decreased insulin sensitivity, and alterations in adipokine release and energy homeostasis [27].

In this study, the vertebrae bone marrow fat fraction and iron concentration were investigated. While the primary focus is on VBM PDFF, from multi‐echo MRI scans, BMFF, from 2‐point Dixon MRI images, was also examined for consistency with prior work and to facilitate comparisons with the broader literature, along with iron concentration in men and women and assessed variations with regard to anthropometric, lifestyle, indices of frailty, sarcopenia, osteoporosis and T2D participants from the UK Biobank.

2. Methods

2.1. Data

The UK Biobank project is a population‐scale prospective cohort study of approximately half a million participants aged 40–69 years recruited between 2006 and 2010 from across the UK with an MRI scanning of the brain, cardiac and abdominal area in a sub‐cohort of 100,000 participants. Images were obtained using a Siemens Aera 1.5T scanner (Syngo MR D13; Siemens, Erlangen, Germany) with full details of the UK Biobank abdominal MRI protocol previously reported [28].

2.2. Phenotype Definitions

Anthropometric measurements, including age, weight (kg), height (cm), waist circumference (cm), and hip circumference (cm) were acquired at the UK Biobank imaging visit, and body mass index (BMI) and waist‐to‐hip ratio were calculated from these. Handgrip strength (HGS) was obtained from the self‐reported dominant hand [29]. Ethnicity was defined based on the self‐reported genetic ethnic background at the initial assessment visit, categorized as “Caucasian” from participants self‐identified as “White British” and having very similar genetic ancestry based on a principal component analysis of the genotypes.

Vigorous MET was used as a more reliable indicator of high‐intensity physical activity. All physical activity measures, alcohol intake frequency and smoking status were self‐reported at the UK Biobank imaging visit.

The Townsend deprivation index is a composite measurement derived from the UK biobank based on four variables: households without a car, overcrowded households, households not owner‐occupied and persons unemployed and was calculated immediately prior to participants joining the initial assessment visit. Insulin‐like growth factor 1 (IGF‐1) was measured in nanomole/L at the initial assessment visit. It is important to note that there was an average interval of 9 ± 1.7 years between the UK Biobank's initial assessment and the imaging visit.

Image‐derived phenotypes (IDPs) used include total skeletal muscle volume, mid‐thigh fat‐to‐muscle ratio, visceral adipose tissue (VAT) volume and abdominal subcutaneous adipose tissue (ASAT) volume as described in previous studies [29, 30].

2.3. Disease Definitions

Sarcopenia was defined as low muscle quality as recommended by the European Working Group on Sarcopenia in Older People 2 (EWGSOP2) [31], which combines both low muscle mass with a HGS < 27 kg in men and < 16 kg in women and low muscle strength with dual X‐ray absorptiometry (DXA)‐measured appendicular lean mass (ALM)/height²: 5.5/7.0 kg/m² (female/male participants). All DXA measurements were taken from the UK Biobank imaging visit. Frailty was defined using the criteria adopted by Hanlon et al. [32] for use with self‐reported UK Biobank questionnaire responses, which require the presence of three out of five indicators, including weight loss, exhaustion, no or only light physical activity in the last 4 weeks, slow walking speed, and low HGS [29]. Additional classifications included pre‐frailty (one or two of the five indicators), and not frail (none of the indicators). All frailty indicators were recorded during the UK Biobank imaging visit.

Type‐2 Diabetes (T2D) was defined based on the International Classification of Diseases 10th edition (ICD10) for type‐2 diabetes using E11 and self‐reported codes 1220 or 1223, respectively [33] or taking antidiabetic medication. Antidiabetic medication codes were selected from prescription records from a GP and the self‐reported medication fields 6153 and 6177, reporting as regularly taking “insulin” medication and self‐reported treatment field 20,003 [34] reporting insulin or antidiabetic drugs (BNF Sections 6.1.1 and 6.1.2; https://openprescribing.net/bnf/0601/). All self‐reported medication fields were taken from the UK Biobank imaging visit. In addition, bone mineral density (BMD) T‐score measurements were obtained from the UK Biobank imaging visit. These measurements were used to complete osteoporosis and osteopenia binary conditions, where osteoporosis was defined based on the ICD10 M80–M82 and based on WHO criteria to define osteoporosis (spine, femoral neck and hip T‐score ≤ −2.5) [13]. Back pain was defined based on self‐reported codes 6159 and 3571 as acute (if they had back pain for less than 3 months) and chronic back pain (if they had back pain for more than 3 months) [35].

2.4. Image Analysis

Details of the UK Biobank MR abdominal protocol utilized in this study have previously been published [30]. The IDPs included in this paper arise from the neck‐to‐knee 2‐point Dixon MRI acquisition and a separate single‐slice multi‐echo MRI acquisition, specific to the liver, for the Iterative Decomposition of Water and Fat with Echo Asymmetry and Least‐Squares Estimation (IDEAL). All MRI data were analyzed using a U‐Net model from a previously dedicated image‐processing pipelines for the neck‐to‐knee 2‐point Dixon and single‐slice multi‐echo MR acquisition, including the deep learning algorithms used to segment organs and tissue, as well as the estimation of PDFF and transverse relaxivity (R2*) values voxelwise from the single‐slice multi‐echo acquisition [30]. Specifically, a multi‐peak spectrum was constructed from the echo times in the acquisition protocol and used to perform nonlinear least squares under multiple regularization constraints that extend the IDEAL. Conversion from R2* to iron concentration using the formula: iron concentration = 0.202 + 0.0254 × R2* as previously described [36].

This pipeline generates more than 30 IDPs from the neck‐to‐knee Dixon data, including VBM compartments from T1 (the first thoracic vertebra) to S1 (the first sacral vertebra). The segmentation model used is described in more detail in the supplementary material. For training the deep learning segmentation model specifically on the VBM compartments, manual annotations from 120 participants were utilized. All annotations were carefully selected to be representative of the UK Biobank population, covering a broad range of anthropometric characteristics. This approach was designed to ensure the generalizability of the pipeline, and the segmentation quality was further confirmed through visual inspection at multiple stages by experienced analysts before use in the segmentation model. The VBM segmentations were projected into the single‐slice multi‐echo data to obtain a two‐dimensional region‐of‐interest (ROI) covering all voxels associated with VBM in the slice. VBM fat fraction (FF) was further computed to facilitate comparisons between FF and PDFF and to ensure consistency with previous studies computing FF in multi‐site bone marrow [13, 37]. The single‐slice VBM segmentation was mapped to the corresponding slice within the full‐volume 2‐point Dixon FF map, specifically aligned to the anatomical level of the single‐slice multi‐echo acquisition at the liver level. This enabled voxel‐wise quantification of the fat fraction within the VBM, calculated as fat/(fat + water), at the same axial location across both acquisitions.

Figure 1 illustrates the location of the single‐slice multi‐echo acquisition on the neck‐to‐knee Dixon scan, provides anatomical references of organs and tissue, and estimates of fat fraction and iron concentration in the vertebral bone marrow. Median FF values were calculated from the single‐slice VBM segmentation on the FF maps, while median PDFF and iron concentration were computed from the VBM segmentation on the single‐slice multi‐echo liver acquisitions. Additional code will be released in the future along with detailed documentation and validation metrics.

FIGURE 1.

Localization of vertebral bone marrow from reference MRI images. (A) Coronal slice of the neck‐to‐knee 2‐point (2‐pt) Dixon data showing the location of the single‐slice multi‐echo acquisition, (B) axial slice of the 2‐pt Dixon fat channel at the location with the VBM segmentation, (C) axial slice of the 2‐pt Dixon water channel at the location with the VBM segmentation, (D) the single‐slice multi‐echo data with the VBM segmentation, (E) voxelwise FF estimates (%) for the VBM segmentation, (F) voxelwise PDFF estimates (%) for the VBM segmentation, (G) voxelwise estimates of iron concentration (mg/g) for the VBM segmentation. MRI, magnetic resonance imaging; PDFF, proton density fat fraction; VBM, vertebral bone marrow.

2.5. Quality Control

MRI data from 44,433 UK Biobank participants were screened for inclusion. Participants with missing single‐slice multi‐echo data and single‐slice multi‐echo data not intersecting with VBM segmentation (n = 17,462), were excluded. Additionally, participants with segmentations of less than 64 voxels (1.9 cm² in area) were excluded from the analysis (n = 447). From the initial 44,433 participants, 26,524 were included in the analysis.

2.6. Statistical Analysis

All summary statistics, hypothesis tests, regression models and figures were performed using the R software environment version 4.5.0 for statistical computing and graphics [38]. Descriptive statistics are expressed as mean and standard deviation in all tables and in the text. IDPs were tested for normality using the Kolmogorov–Smirnov test. The null hypothesis was rejected in all cases, and log transformations were applied to the median vertebral bone marrow iron concentration. While the null hypothesis was rejected for VBM FF and PDFF, there was a limited deviation from normality when inspecting the quantile‐quantile plot. The Wilcoxon rank‐sum test was employed to compare means between two groups, whereas one‐way ANOVA was applied for comparisons across multiple groups. Monotonic relationships between IDPs and variables of interest were evaluated using Spearman's rank correlation coefficient.

The impact of vertebral bone marrow FF, PDFF, and iron concentration was evaluated on anthropometric, lifestyle and disease factors using linear regression models, adjusting for age, ethnicity, vigorous MET, the Townsend deprivation index, IGF‐1, spine BMD, alcohol, smoking, IDPs including total skeletal muscle volume, fat‐to‐muscle ratio in the mid‐thighs and VAT, as well as conditions including sarcopenia, frailty, osteoporosis, T2D and back pain. To evaluate the comparison between VBM PDFF and FF, sex‐specific linear regression models were fitted to predict VBM PDFF from FF, adjusting for all the relevant variables mentioned above. To account for the potential confounding effect of height on total skeletal muscle, ASAT and VAT volumes, these IDPs were indexed by dividing by height squared prior to inclusion in the linear regression model. Given the temporal gap between the time of initial study assessment (when biochemical markers were acquired) and the MRI visit, further adjustments for the time interval between the initial visit and the MRI visit date were included.

Summaries of the linear regression analysis are reported as regression coefficients (β) with 95% confidence intervals (CIs) shown in parentheses in the tables and as standardized regression coefficients with 95% CIs in the figure showing the linear regression analysis. The threshold for statistical significance of p‐values was adjusted for the number of formal hypothesis tests performed. The Bonferroni‐corrected threshold was 0.05/305 = 0.00016.

3. Results

3.1. Segmentation Performance

The deep learning model achieved a mean Dice similarity coefficient of 0.83 on a held‐out 20% test set. Visual quality control confirmed successful segmentation in the vast majority of cases across all organs and compartments. A small number of segmentations (< 2%) were excluded following quality assurance.

3.2. Study Population

Baseline characteristics for the 26,524 participants included in the study are shown in Table 1. Of the 26,524 participants, 84.7% were Caucasian, and 51.0% were women, with an average age of 64.14 ± 7.52 years for women and 65.53 ± 7.77 years for men (p < 0.00016). The average BMI for women and men were 25.90 ± 4.68 kg/m² and 26.89 ± 3.89 kg/m², respectively (p < 0.00016). Within the study cohort, 3458 participants were classified as having sarcopenia, 259 were classified as frail, 2500 had osteoporosis (including 878 diagnosed via ICD‐10 codes with an average duration of 9.8 ± 7.1 years from diagnosis to MRI), and 1440 had type 2 diabetes (1403 diagnosed via ICD‐10 and self‐reported codes, with an average duration of 8.9 ± 7.7 years). Back pain was reported by 5120 participants, of whom 66.3% had chronic symptoms.

TABLE 1.

Demographics and participant characteristics.

Characteristic, mean ± SD; n (%)	Female	Male	Total
N	13,523	13,001	26,524
Age (years)	64.14 ± 7.52	65.53 ± 7.77**	64.82 ± 7.68
Caucasian, n	11,333	11,140	22,473
Weight (kg)	68.65 ± 12.92	83.52 ± 13.32**	75.94 ± 15.08
Height (cm)	162.83 ± 6.23	176.16 ± 6.67**	169.36 ± 9.27
BMI (kg/m2)	25.90 ± 4.68	26.89 ± 3.89**	26.38 ± 4.34
Waist circumference	82.88 ± 11.90	94.57 ± 10.88**	88.62 ± 12.82
Hip circumference (cm)	100.65 ± 9.74	100.50 ± 7.37**	100.58 ± 8.66
WHR	0.82 ± 0.07	0.94 ± 0.07**	0.88 ± 0.09
Dominant hand grip strength (kg)	24.13 ± 6.07	38.87 ± 8.92**	31.39 ± 10.59
IGF‐1 (nmol/L)	21.61 ± 5.66	22.39 ± 5.26**	21.99 ± 5.48
Townsend deprivation index	−1.80 ± 2.76	−1.93 ± 2.74**	−1.86 ± 2.75
Vigorous MET (hours/week)	6.40 ± 7.90	7.89 ± 8.40**	7.13 ± 8.18
Alcohol consumption
Daily, n	1867	2670	4537
1–4 times per week, n	6945	7461	14,406
1–3 times per month, n	1816	1236	3052
Smoking status
Current, n	367	479	846
Previous, n	4158	4833	8991
Spine BMD (g/cm2)	1.00 ± 0.15	1.20 ± 0.16**	1.10 ± 0.19
Sarcopenia, n	1698	1760	3458
Frailty
Frail, n	151	108	259
Pre‐frail, n	4613	4701	9314
Type 2 diabetes, n	487	953	1440
Chronic back pain
Acute, n	745	980	1725
Chronic, n	1778	1617	3395
Osteoporosis	1840	660	2500
ASAT (L)	9.73 ± 4.35	6.98 ± 3.27**	8.38 ± 4.10
VAT (L)	2.79 ± 1.57	5.15 ± 2.38**	3.95 ± 2.33
Total skeletal muscle (L)	14.02 ± 1.93	21.77 ± 2.99**	17.82 ± 4.61
Mid‐thigh fat‐to‐muscle ratio (%)	2.46 ± 1.77	1.98 ± 1.82**	2.22 ± 1.81
VBM median FF (%)	52.53 ± 10.81	47.29 ± 10.94**	49.96 ± 11.19
VBM median PDFF (%)	39.97 ± 8.97	37.45 ± 8.74**	38.74 ± 8.95
VBM median iron concentration (mg/g)	1.82 ± 0.38	1.92 ± 0.41**	1.87 ± 0.40

Note: Significance refers to the p value for a Wilcoxon rank‐sums test, where the null hypothesis is the medians between the continuous variables from the two groups tested (female and male participants) being equal. *indicate statistically significant for p < 0.05, **indicate statistically significant after Bonferroni correction (p = 0.00016).

Abbreviations: ASAT, abdominal subcutaneous adipose tissue; BMD, bone mineral density; BMI, body mass index; FF, fat fraction; IGF, insulin‐like growth factor; MET, metabolic equivalent task; PDFF, proton density fat fraction; VAT, visceral adipose tissue, VBM, vertebral bone marrow.

As previous studies have primarily utilized FF to characterize bone marrow composition across multiple skeletal sites [13], both VBM FF and PDFF were measured to maintain consistency with existing literature and to examine the extent to which these two measures capture sex‐specific differences in VBM fat fraction. The results showed that median VBM FF was 52.53 ± 10.81% in women and 47.29 ± 10.94% in men, while median VBM PDFF was 39.97 ± 8.97% in women and 37.45 ± 8.74% in men (p < 0.00016), whereas median iron concentration was 1.82 ± 0.38 mg/g in women and 1.92 ± 0.41 mg/g in men (p < 0.00016).

The effect of disease on VBM measures by comparing distributions between groups was further investigated (Supporting Information S1: Tables S1–S5). Statistically significant higher VBM FF and PDFF were found in participants with sarcopenia, osteoporosis, and T2D (when compared to participants without sarcopenia, osteoporosis, and T2D, respectively) in both men and women. While VBM FF was statistically significantly higher in men with frailty when compared to not‐frail men, VBM FF did not differ with altered frailty in women. VBM iron concentration was statistically significantly higher in women with pre‐frailty when compared to not‐frail women, while it was lower in men with T2D and in both men and women with osteoporosis (when compared to non‐T2D and non‐osteoporosis, respectively) (p < 0.00016, Supporting Information S1: Table S3). After applying Bonferroni correction for multiple comparisons, no significant differences in the VBM measures were found between participants with and without chronic (Supporting Information S1: Table S5) or acute (results not shown) back pain.

3.3. Comparison of VBM PDFF and FF

VBM PDFF estimation showed high correlation with FF (r = 0.89, for women; r = 0.88, for men, p < 0.00016) with high R ² values from the unadjusted fitted linear regression models (R ² = 0.8, for women; R ² = 0.79, for men, p < 0.00016, Supporting Information S1: Figure S1). Sex‐specific linear regression models predicting VBM PDFF from FF adjusting for relevant variables along with 95% CIs were 0.73 (0.72, 0.74) for women and 0.71 (0.71, 0.72) in men, showing similar R ² values as the unadjusted VBM PDFF estimation (Supporting Information S1: Table S6).

3.4. Correlation for VBM FF, PDFF, and Iron Concentration

The sex‐specific correlation coefficients between variables of interest and median VBM PDFF and iron concentration are shown in Figure 2, Supporting Information S1: Figure S2. In summary, both VBM FF and PDFF were significantly positively correlated with age in both men and women (r = 0.29 for FF, r = 0.30 for PDFF in women; r = 0.24 for both FF and PDFF in men, p < 0.00016). Additionally, both VBM FF and PDFF showed statistically significant positive correlations with VAT index, mid‐thigh fat‐to‐muscle ratio, ASAT index, and BMI in both sexes. For these body composition measures, correlation coefficients were consistently below 0.2 (ranging from r = 0.10 to r = 0.19), though all associations reached statistical significance (p < 0.00016). There was an inverse correlation between VBM PDFF and total skeletal muscle index (r = −0.15 in women; r = −0.16 in men, p < 0.00016) with stronger correlations shown with VBM FF (r = −0.16 in women; r = −0.20 in men, p < 0.00016). There were weaker but significant negative correlations with IGF‐1, vigorous MET and HGS (r > −0.1, p < 0.00016) with both VBM FF and PDFF. Both VBM FF and PDFF were significantly negatively correlated with spine BMD in both men and women (r = −0.17 for FF, r = −0.15 for PDFF in women; r = −0.08 for FF, r = −0.06 for PDFF in men, p < 0.00016). All significant correlations with median VBM iron concentration were weak for all significant correlations, with the highest correlations shown for age, BMI, VAT, and spine BMD (r ≥ −0.1, p < 0.00016).

FIGURE 2.

Sex‐stratified coefficients between variables of interest and median vertebral bone marrow FF, PDFF, and iron concentration. Spearman’s correlation coefficients in red are statistically significant following correcting for multiple comparisons. Significant associations for p value below the Bonferroni correction (p = 0.00016) are shown in red, and non‐significant associations in gray. The intensity of red indicates the strength of significance, with darker shades representing smaller p values (converted to ${-\log }_{10}\left(p\right)$). ASAT, abdominal subcutaneous adipose tissue; BMD, bone mineral density; BMI, body mass index; FF, fat‐fraction; IGF, insulin‐like growth factor; MET, metabolic equivalent task; PDFF, proton density fat fraction; VAT, visceral adipose tissue; VBM, vertebral bone marrow.

3.5. Associations of VBM FF, PDFF, and Iron With Anthropometric Characteristics and Disease

Linear regression models were used to investigate the effect of median VBM FF, PDFF, and iron concentration on anthropometric, lifestyle factors and disease for men and women separately (Supporting Information S1: Table S7 and Figure 3). Median VBM FF and PDFF were significantly positively associated with age in both men and women, and current smoking in women (p < 0.00016). VBM FF and PDFF were also positively associated with VAT index (β = 4.86%/(L/m²) for FF, β = 3.64%/(L/m²) for PDFF in women and β = 2.98%/(L/m²) for FF, β = 1.77%/(L/m²) for PDFF in men p < 0.00016). In comparison, it showed a negative association with total skeletal muscle index (β = −2.57%/(L/m²) for FF, −1.92%/(L/m²) for PDFF in women and β = −2.20%/(L/m²) for FF and −1.32%/(L/m²) for PDFF in men p < 0.00016). Median VBM FF and PDFF were significantly negatively associated with spine BMD in both men and women (β = −12.13%/(g/cm²) for FF, β = −8.37%/(g/cm²) for PDFF in women; β = −6.83%/(g/cm²) for FF, β = −4.44%/(g/cm²) for PDFF in men, p < 0.00016). Osteoporosis was positively associated with VBM FF in men (β = 1.97%, p < 0.00016). A diagnosis of T2D was positively associated with VBM FF and PDFF, (β = 2.90% for FF, β = 2.40 for PDFF in women and β = 2.45% for FF, β = 1.74% for PDFF in men, p < 0.00016).

FIGURE 3.

Summary of linear regression coefficients for median vertebral bone marrow FF, PDFF, and iron concentration, representing the associations with the most relevant baseline characteristics in women and men separately. Standardized beta coefficients are displayed with 95% confidence intervals. Significant associations for p‐value below the Bonferroni correction (p = 0.00016) are shown in red for women and blue for men, and non‐significant associations in gray. BMD, bone mineral density; FF, fat fraction; IGF, insulin‐like growth factor; PDFF, proton density fat fraction; VAT, visceral adipose tissue; VBM, vertebral bone marrow.

Median VBM iron concentration was significantly positively associated with age and current smoking only in women and with daily alcohol intake, and spine BMD and VAT index in both men and women (p < 0.00016). At the same time, there was a negative association with T2D (β = −0.06 log(mg/g) in women and β = −0.05 log(mg/g) in men, p < 0.00016). No significant associations between VBM measures and sarcopenia, frailty, or back pain were found. It is worth noting that R ² values were low, reflecting a poor fit in the linear regression models for VBM PDFF and iron concentration for both men and women, with the highest R ² values only observed for VBM FF and PDFF in women (adjusted R ² = 0.17 for VBM FF and adjusted R ² = 0.16 for VBM PDFF; Supporting Information S1: Table S7).

4. Discussion

In this study, a comprehensive assessment of VBM FF, PDFF, and iron concentrations was performed in a large UK Biobank cohort. The primary aim was to evaluate associations between VBM fat fraction, iron concentrations and anthropometric as well as lifestyle characteristics highlighting significant sex‐specific associations that deepen the understanding of the variations in the regional fat accumulation, enabling greater insight into the anthropometric, lifestyle and disease factors. Fat fraction within VBM tissue and fat accumulation, measured as FF and PDFF, were examined to ensure consistency with existing literature [13] and to assess the extent to which these metrics capture sex‐specific associations in VBM fat composition.

VBM FF and PDFF were positively correlated with age and VAT in both men and women. While previous studies also reported associations between increased VBM PDFF and VAT index as well as ASAT index and BMI [39], in this study, only weak but statistically significant correlations with BMI and ASAT index were found (r ≤ 0.1). An inverse correlation between VBM fat fraction (both FF and PDFF) and total skeletal muscle index, IGF‐1, vigorous physical activity, and hand grip strength was found, where aging was associated with increased bone marrow adiposity, declining muscle function and physical activity, aligning with previous reports on vertebral bone marrow fat [39, 40, 41, 42]. Previous studies have also found positive associations between VBM fat and VAT and an inverse association with IGF‐1 [39], supporting the role of IGF‐1 as an important regulator of the fat and bone lineage; however, the linear regression models revealed significant associations with VAT index but not with IGF‐1. In obesity, excess fat deposition extends beyond traditional compartments, leading to fat accumulation in non‐adipose tissues such as the liver, kidneys, muscle, and bone marrow [1]. The observed associations between VBM fat and VAT, along with the inverse relationships with skeletal muscle and bone mineral density, highlight how obesity may contribute to musculoskeletal deterioration through ectopic fat infiltration. Thus, VBM fat composition may serve as an integrative marker reflecting both the presence and severity of obesity as a disease state.

Total muscle mass index was significantly associated with reduced VBM FF and PDFF, whereas no significant association was found with sarcopenia. Previous findings suggest that total muscle volume is more informative for sarcopenia, whereas muscle fat is more indicative of frailty [29]. Nevertheless, significant associations with frailty were not observed in the current models; however, the results showed that in men with frailty, VBM FF was higher when compared to non‐frail men. The results show that VBM fat accumulation is associated with reduced muscle volume; however, previous studies suggest that muscle mass and quality loss are generally influenced by multiple factors, including environmental conditions, diseases, inflammation, and mitochondrial function [43, 44]. Furthermore, despite previous studies reporting high L3 marrow fat in patients with back pain [45], the linear regression models revealed no significant associations with back pain, highlighting the importance of further research in this area to provide a better understanding of the impact of the fat accumulation in spinal bone marrow on back pain.

An inverse association between lumbar BMD and both VBM FF and PDFF was found. These results align with recent findings using deep learning techniques to facilitate multi‐site bone marrow FF analysis in the UK Biobank, revealing site‐ and sex‐specific characteristics crucial for understanding metabolic risk phenotypes [13, 37]. These studies also reported significant differences in spine BMFF between patients with osteoporosis and control groups, primarily in women [13]. In contrast, significant differences in VBM FF and PDFF were identified in both men and women with osteoporosis based on mean value comparisons. However, when examined using linear regression models, a significant positive association with VBM FF was observed only in men. Additionally, VBM PDFF showed a positive association with T2D, consistent with earlier studies indicating that bone marrow adipose tissue may act as a biomarker for glycemic regulation and potentially for late‐stage diabetic complications [46, 47].

Importantly, the findings must be interpreted in terms of technical factors influencing bone marrow adipose tissue estimation. A recent study by Haueise et al. [17] systematically compared 2‐point Dixon and multi‐echo techniques for quantifying fat fraction in the liver, paravertebral muscles, and vertebral bone marrow (Th8–L5) in healthy individuals. While they reported a strong correlation between the two methods, they also identified systematic differences in vertebral bone marrow fat fraction values. However, direct comparison with the findings in the current study is limited due to several methodological differences, including scanners, acquisition sequences, and sample size. Despite these differences, they demonstrated that sex‐specific linear regression corrections could reduce the bias between 2‐point Dixon and multi‐echo PDFF values; however, they observed discrepancies between single vertebral body fat fractions, suggesting that multi‐echo Dixon techniques may offer superior accuracy and are better suited for detailed analysis of regional bone marrow adiposity.

The results further demonstrate a significantly positive association between VBM iron concentration and factors such as alcohol intake, smoking, and VAT, with notable sex‐specific trends. VBM iron was also lower in participants with osteoporosis. Interestingly, VBM iron concentration was positively associated with age only in women, suggesting a potential sex‐dependent regulation of iron homeostasis in bone marrow that may be influenced by aging [48]. Furthermore, VBM iron concentration was positively associated with lumbar BMD, consistent with studies showing that higher dietary iron intake is linked to greater BMD [49] and lower risk of osteoporosis [50]. However, while dietary iron has been associated with improved bone density in some studies, others suggest that iron overload may reduce bone quality [51], indicating a complex relationship between iron levels and bone health. These findings underscore the need for further research to explore the complex relationship between iron homeostasis and bone fragility in greater depth.

This study benefits from several strengths, including its large sample size, balanced sex distribution, and advanced deep learning segmentation models to accurately measure VBM FF, PDFF, and iron concentrations. However, it is crucial to recognize its limitations. The UK Biobank, a substantial cross‐sectional study, may be influenced by selection bias as it represents a “healthier” group compared to the broader UK population and excludes younger individuals and cases of severe disease. Moreover, as the UK Biobank population is predominantly White, future work is needed to investigate diverse ethnic population data. Another potential limitation of this study is that the blood markers used were reported approximately 9 years before the MRI acquisition. Finally, the relatively low R ² values suggest that other factors may contribute to VBM measures. For instance, relevant chronic inflammatory diseases in the spine, such as ankylosing spondylitis, were diagnosed in only 83 participants (0.3%). Relevant markers, such as spine bone mineral density, were only available in 75% of the imaging cohort, which reduced the number of observations in this analysis. Additional cross‐sectional and longitudinal measurements will be required to assess age‐related changes in disease cohorts.

5. Conclusion

These findings contribute to the broader understanding of how metabolic risk phenotypes extend beyond traditional measures by incorporating novel indicators such as VBM measures into multi‐faceted risk models that can be associated with anthropometric, lifestyle characteristics, and disease. These results underscore the utility of vertebral bone marrow adiposity measures as potential biomarkers not only of musculoskeletal decline but also of systemic adiposity associated with obesity as a chronic disease. The data further underscore the relevance of ectopic fat depots, suggesting that VBM FF, PDFF, and iron may serve as complementary markers for comprehensive assessments of metabolic health.

6. Declarations

Fully anonymized images and participant metadata were obtained through the UK Biobank Access Application number 44584. The UK Biobank has approval from the North West Multi‐Center Research Ethics Committee (REC reference: 11/NW/0382) and obtained written informed consent from all participants before the study. All methods were performed in accordance with the relevant guidelines and regulations as presented by the appropriate authorities, including the Declaration of Helsinki.

Author Contributions

J.D.B., E.L.T., J.R.P., and M.T. conceived the study. J.D.B., B.W., E.L.T., J.R.P., and M.T. designed the study. M.T., B.W., J.R.P., and N.B. implemented the methods. M.T. and B.W. performed the data and statistical analysis. E.L.T., B.W., M.T., J.D.B., J.R.P., and N.B. drafted the manuscript. All authors read and approved the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Parkinson, James R. , Thanaj Marjola, Basty Nicolas, Whitcher Brandon, Thomas E. L., and Bell Jimmy D.. 2025. “Fat Fraction and Iron Concentration in Lumbar Vertebral Bone Marrow in the UK Biobank.” Obesity Science & Practice: e70088. 10.1002/osp4.70088.

Data Availability Statement

Our research was conducted using UK Biobank data. Under the standard UK Biobank data sharing agreement, we (and other researchers) cannot directly share raw data obtained or derived from the UK Biobank. However, under this agreement, all of the data generated and methodologies used in this paper are returned by us to the UK Biobank, where they will be fully available. Access is obtained directly from the UK Biobank to all bona fide researchers upon submitting a health‐related research proposal to the UK Biobank https://www.ukbiobank.ac.uk.