Multi-organ imaging demonstrates the heart-brain-liver axis in UK Biobank participants

Medical imaging provides numerous insights into the subclinical changes that precede serious diseases such as heart disease and dementia. However, most imaging research either describes a single organ system or draws on clinical cohorts with small sample sizes. In this study, we use state-of-the-art multi-organ magnetic resonance imaging phenotypes to investigate cross-sectional relationships across the heart-brain-liver axis in 30,444 UK Biobank participants. Despite controlling for an extensive range of demographic and clinical covariates, we find significant associations between imaging-derived phenotypes of the heart (left ventricular structure, function and aortic distensibility), brain (brain volumes, white matter hyperintensities and white matter microstructure), and liver (liver fat, liver iron and fibroinflammation). Simultaneous three-organ modelling identifies differentially important pathways across the heart-brain-liver axis with evidence of both direct and indirect associations. This study describes a potentially cumulative burden of multiple-organ dysfunction and provides essential insight into multi-organ disease prevention.


Supplementary
Supplementary Table 2: Entries are standardised beta coefficients, 95% confidence intervals and p-values from linear regression models. Each cell represents one model, with the brain variable placed as the outcome/response and the liver variable placed as the exposure/predictor. Models are adjusted by age, sex, diabetes, hypertension, BMI ≥ 30 kg/m 2 , high cholesterol, smoking, physical activity, alcohol consumption, deprivation, educational level, red blood count, total cholesterol, glycosylated haemoglobin, and imaging confounders (including head size, imaging site, scanner coordinates, date of scanning and interactions). P-values are from two-sided T-tests for each coefficient. An asterisk indicates significance after adjustment for multiple testing with a false discovery rate of 5%. PDFF = proton density fat fraction, cT1 = corrected T1 relaxation time, ICVF = intracellular volume fraction, ISOVF = isotropic volume fraction. Source data are provided as a Source Data file. Supplementary Table 3: Entries are standardised beta coefficients, 95% confidence intervals and p-values from linear regression models. Each cell represents one model, with the brain variable placed as the outcome/response and the heart variable placed as the exposure/predictor. Models are adjusted by age, sex, diabetes, hypertension, BMI ≥ 30 kg/m 2 , high cholesterol, smoking, physical activity, alcohol consumption, deprivation, educational level, red blood count, total cholesterol, glycosylated haemoglobin, and imaging confounders (including head size, imaging site, scanner coordinates, date of scanning and interactions). Pvalues are from two-sided T-tests for each coefficient. An asterisk indicates significance after adjustment for multiple testing with a false discovery rate of 5%. LVSVi = left ventricular stroke volume indexed to body surface area, LV GFI = left ventricular global function index, LVM/LVEDV = left ventricular mass to volume ratio (left ventricular mass / left ventricular end diastolic volume), AoD = aortic distensibility, ICVF = intracellular volume fraction, ISOVF = isotropic volume fraction. Source data are provided as a Source Data file. Supplementary Table 4: Entries are standardised beta coefficients and p-values from linear regression models. Each beta coefficient represents one model, with the heart variable placed as the outcome/response and the liver variable placed as the exposure/predictor. Models are adjusted by age, sex, diabetes, hypertension, BMI ≥ 30 kg/m 2 , high cholesterol, smoking, physical activity, alcohol consumption, deprivation, educational level, red blood count, total cholesterol, glycosylated haemoglobin. P-values are from two-sided T-tests for each coefficient. An asterisk indicates significance after adjustment for multiple testing with a false discovery rate of 5%. PDFF = proton density fat fraction, cT1 = corrected T1 relaxation time, LVSVi = left ventricular stroke volume indexed to body surface area, LV GFI = left ventricular global function index, LVM/LVEDV = left ventricular mass to volume ratio (left ventricular mass / left ventricular end diastolic volume), AoD = aortic distensibility. Source data are provided as a Source Data file. non-significant variables were removed one at a time to arrive at the final model. Liver variables were orthogonalized prior to modelling (see Supplementary Table 12 for details). P-values are from two-sided Ztests for each coefficient, and p-values were considered significant after adjustment for multiple testing with a 5% false discovery rate. Paths are adjusted by age, sex, diabetes, hypertension, BMI ≥ 30kg/m 2 , high cholesterol, smoking, physical activity, alcohol consumption, deprivation, educational level, red blood count, total cholesterol, glycosylated haemoglobin. PDFF = proton density fat fraction, cT1 = corrected T1 relaxation time, LVSV = left ventricular stroke volume, LV GFI = left ventricular global function index, LVM/LVEDV = left ventricular mass to volume ratio (left ventricular mass / left ventricular end diastolic volume). Source data are provided as a Source Data file.

Supplementary Table 6:
Final results from multi-organ path analysis for simultaneous liver/heart associations with grey matter volume, white matter hyperintensities and free-water fraction (ISOVF), in the subset with complete cases across three organs (N= 6,865).Path analysis models were fitted using the sem function in the lavaan package in R. All paths are adjusted by age, sex, height, diabetes, hypertension, high cholesterol, smoking, physical activity, alcohol intake frequency, Townsend deprivation score, education, systolic blood pressure, BMI ≥ 30 kg/m 2 , total cholesterol, glycated haemoglobin and red blood cell count. Paths featuring the brain (labelled Equation 1 above) are additionally adjusted by head size, imaging site, scanner coordinates and date of imaging. P-values are from two-sided Z-tests for each coefficient, and p-values were considered significant after adjustment for multiple testing with a 5% false discovery rate. The chi-square in model 3 is zero as this has reduced to a single fitted linear equation. Prior to simultaneous modelling, heart and liver predictors were orthogonalized to remove within-organ correlation (see Supplementary Table 12 for details). Source data are provided as a Source Data file.   19 Mojtahed    Method details [1] Target variable was present at the baseline measurement but missing at time of imaging. In these cases, a linear model was constructed based on all data (the full 500,000 UKB set) relating, for example:

Supplementary Table 7: Technical and clinical validity of liver metrics Evidence from mixed/healthy cohorts A: Measurement and description of liver features in UK Biobank
Weight_at_imaging ~ Weight at baseline + age + sex + age x sex + days between baseline date and imaging date.
Then the fitted linear estimate was used to impute the value where missing.
[2] MICE (multiple imputation by chained equations) was performed using the mice package in R, across the set of well-populated biochemistry fields, with the full set of UKB data, within a set of rows such that all variables were less than 6% missing, with the addition of age, sex, smoking, BMI, alcohol intake and Townsend deprivation score added to the multiple imputation equations. MICE was run with 7 iterations to produce a single data set.
[3] Same process as with method [1] but there were still a small number of residual missing values (missing at both baseline and imaging) which were replaced with the mean.
[4] A very small number of missing values were replaced with the mean.  Table 12: Principal components analysis was performed for each organ, with the number of retained components equal to the number of input variables. Set sizes was determined by complete rows available for each organ. Varimax rotation was applied. For easier interpretability, three original heart variables were reversed so that all heart/liver variables would share the same directionalitythese are: aortic stiffening = aortic distensibility * -1, reduced LV stroke volume = LV stroke volume * -1, and reduced LV GFI = LV GFI * -1. LV= left ventricular, SV = stroke volume, GFI = global function index, PDFF= proton density fat fraction, LVM/LVEDV = left ventricular mass to volume ratio. Source data are provided as a Source Data file.