Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2021 Jun 15:10:e65554.
doi: 10.7554/eLife.65554.

Genetic architecture of 11 organ traits derived from abdominal MRI using deep learning

Yi Liu  1 , Nicolas Basty  2 , Brandon Whitcher  2 , Jimmy D Bell  2 , Elena P Sorokin  1 , Nick van Bruggen  1 , E Louise Thomas #  2 , Madeleine Cule #  1
Affiliations

Genetic architecture of 11 organ traits derived from abdominal MRI using deep learning

Yi Liu et al. Elife. .

Abstract

Cardiometabolic diseases are an increasing global health burden. While socioeconomic, environmental, behavioural, and genetic risk factors have been identified, a better understanding of the underlying mechanisms is required to develop more effective interventions. Magnetic resonance imaging (MRI) has been used to assess organ health, but biobank-scale studies are still in their infancy. Using over 38,000 abdominal MRI scans in the UK Biobank, we used deep learning to quantify volume, fat, and iron in seven organs and tissues, and demonstrate that imaging-derived phenotypes reflect health status. We show that these traits have a substantial heritable component (8-44%) and identify 93 independent genome-wide significant associations, including four associations with liver traits that have not previously been reported. Our work demonstrates the tractability of deep learning to systematically quantify health parameters from high-throughput MRI across a range of organs and tissues, and use the largest-ever study of its kind to generate new insights into the genetic architecture of these traits.

Keywords: adiposity; genetics; genome-wide association study; genomics; human; magnetic resonance imaging; medicine.

PubMed Disclaimer

Conflict of interest statement

YL, Nv, MC Employee, Calico Life Sciences LLC. This work was funded by Calico Life Sciences LLC. NB, BW, JB, ET No competing interests declared, ES Employee, Calico Life Sciences LLC.This work was funded by Calico Life Sciences LLC.

Figures

Figure 1.
Figure 1.. Visualisation of studied IDPs.
(A) Example Dixon image before and after automated segmentation of ASAT, VAT, liver, lungs, left and right kidneys, and spleen. (B) Relationship between IDPs and age and sex within the UKBB. Each trait is standardised within sex, so that the y axis represents standard deviations, after adjustment for imaging centre and date. The trend is smoothed using a generalised additive model with smoothing splines for visualisation purposes. (C) Relationship between IDPs and scan time and sex within the UKBB. Each trait is standardised within sex, so that the y axis represents standard deviations, after adjustment for imaging centre and date. The trend is smoothed using a generalised additive model with smoothing splines for visualisation purposes. (D) Correlation between IDPs. Lower right triangle: Unadjusted correlation (except for imaging centre and date). Upper left triangle: Correlation after adjustment for age, sex, height, and BMI. (E-G) Histograms showing the distribution of the eleven IDPs in this study.
Figure 1—figure supplement 1.
Figure 1—figure supplement 1.. Correlation between multiple measurements of fat, iron and volume.
(A) Correlation between multiple measurements of liver fat, liver iron, ASAT volume, and VAT volume in the UK Biobank. (B) Scatter plots showing the relationship between multiple measurements of liver fat, liver iron, ASAT volume, and VAT volume in the UK Biobank. ‘Combined’ refers to a combined IDEAL/multiecho measurement as described in the ‘multiecho pipeline’ section of the supplementary information.
Figure 1—figure supplement 2.
Figure 1—figure supplement 2.. IDPs plotted across imaging centre and across scan date.
(A) Organ volume IDPs, split by imaging centre. (B) Fat IDPs, split by imaging centre. (C) Iron IDPs, split by imaging centre. (D) Relationship between scan date and IDPs.
Figure 2.
Figure 2.. Disease phenome-wide association study across all IDPs and 754 disease codes (PheCodes).
The x-axis gives the effect size per standard deviation, and the y-axis -log10(p-value). The top three associations for each phenotype are labelled. Horizontal lines at disease phenome-wide significance (dotted line, p=6.63e-05) and study-wide significance (dashed line, p=6.03e-06) after Bonferroni correction. Note that the PheCodes are not exclusive and have a hierarchical structure (for example, T1D and T2D are subtypes of Diabetes), so some diseases appear more than once in these plots. LL: Leukocytic leukaemia. CLL: Chronic leukocytic leakaemia. T1D: Type 1 diabetes. T2D: Type 2 diabetes. CKD: Chronic kidney disease.
Figure 2—figure supplement 1.
Figure 2—figure supplement 1.. Phenome-wide associations across all IDPs and 83 biomarkers.
The x-axis gives the effect size per standard deviation, and the y-axis -log10(p-value). The top three associations for each phenotype are labelled. Horizontal lines at phenome-wide significance (dotted line, p=2.7e-05) and study-wide significance (dashed line, p=2.48e-06) after Bonferroni correction for the total number of measures. SHBG: Sex hormone binding globulin. MSCV: Mean sphered cell volume. MCH: Mean corpuscular haemoglobin. RC: Reticulocyte count. PDW: Platelet distribution width. ALT: alanine transaminase. ALP: Alkaline phosphatase. HLSRC: High light scatter reticulocyte count. GGT: Gamma glutamyl transferase.
Figure 2—figure supplement 2.
Figure 2—figure supplement 2.. Phenome-wide associations across all IDPs and 199 lifestyle and history traits.
The x-axis gives the effect size per standard deviation, and the y-axis -log10(p-value). The top three associations for each phenotype are labelled. Horizontal lines at phenome-wide significance (dotted line, p=2.7e-05) and study-wide significance (dashed line, p=2.48e-06) after Bonferroni correction for the total number of measures.
Figure 2—figure supplement 3.
Figure 2—figure supplement 3.. Phenome-wide associations across all IDPs and 770 medical history traits.
The x-axis gives the effect size per standard deviation, and the y-axis -log10(p-value). The top three associations for each phenotype are labelled. Horizontal lines at phenome-wide significance (dotted line, p=2.7e-05) and study-wide significance (dashed line, p=2.48e-06) after Bonferroni correction for the total number of measures.
Figure 2—figure supplement 4.
Figure 2—figure supplement 4.. Phenome-wide associations across all IDPs and 444 traits measured in online follow-up.
The x-axis gives the effect size per standard deviation, and the y-axis -log10(p-value). The top three associations for each phenotype are labelled. Horizontal lines at phenome-wide significance (dotted line, p=2.7e-05) and study-wide significance (dashed line, p=2.48e-06) after Bonferroni correction for the total number of measures.
Figure 2—figure supplement 5.
Figure 2—figure supplement 5.. Phenome-wide associations across all IDPs and 335 physical measures.
The x-axis gives the effect size per standard deviation, and the y-axis -log10(p-value). The top three associations for each phenotype are labelled. Horizontal lines at phenome-wide significance (dotted line, p=2.7e-05) and study-wide significance (dashed line, p=2.48e-06) after Bonferroni correction for the total number of measures. FVC forced vital capacity. FEV1 Forced expiratory volume in 1 s. FF fat-free.
Figure 3.
Figure 3.. Genetic architecture of all IDPs.
(A) Heritability (point estimate and 95% confidence interval) for each IDP estimated using the BOLT-REML model. Y-axis: Adjusted for height and BMI. X-axis: Not adjusted for height and BMI. The three panels show volumes, fat, and iron respectively. (B) Genetic correlation between IDPs estimated using bivariate LD score regression. The size of the points is given by -log10(p), where p is the p-value of the genetic correlation between the traits. Upper left triangle: Adjusted for height and BMI. Lower right triangle: Not adjusted for height and BMI. (C) Manhattan plots showing genome-wide signals for all IDPs for volume (top panel), fat (middle panel), and iron concentration (lower panel). Horizontal lines at 5e-8 (blue dashed line, genome-wide significant association for a single trait) and 4.5e-9 (red dashed line, study-wide significant association). P-values are capped at 10e-50 for ease of display. The genes with closest transcription start site are labelled.
Figure 3—figure supplement 1.
Figure 3—figure supplement 1.. Rare association studies in the subcohort with both exome sequence data and imaging-derived quantitative phenotypes.
Left: Manhattan plot shows the association between each gene organised by genomic coordinates. Right: QQ-plot showing calibration of SKAT-O test statistics. λGC: Genomic control parameter for each trait. Blue dashed line indicates Bonferroni significance threshold genome-wide (p=7.4e-06). Red dashed line indicates overall study significance threshold (p=6.7e-07). (A) Volume of visceral fat (n = 11,069 samples) .(B) Volume of spleen (n = 11,134). (C) Volume of the lungs (n = 11,134). (D) Liver volume (n = 11,134). (E) Kidney volume (n = 11,134). (F) Abdominal subcutaneous fat (n = 11,134). One gene achieved genome-wide significance but not study wide significance (RRNAD1: pSKAT-O = 6.5e-06; betaburden = −0.08). (G) Pancreas volume (n = 11,093) (H) pancreas iron level (n = 5,525) (I) liver iron (n = 11,069) (J) pancreatic fat (n = 5525) (K) liver fat (n = 11,069).
Figure 3—figure supplement 2.
Figure 3—figure supplement 2.. Genetic correlation between IDPs and complex traits.
Only IDPs and traits with statistically significant genetic correlation (p<1.61e-05 after Bonferroni correction for multiple testing) are shown.
Figure 3—figure supplement 3.
Figure 3—figure supplement 3.. Heritability enrichment in tissues and cell types for annotations based on gene expression (see Materials and methods).
The top three enrichments for each phenotype passing a trait-wide significance threshold are labelled. Horizontal lines and trait-wide (dotted line) and study-wide (dashed line) significance after Bonferroni correction.
Figure 3—figure supplement 4.
Figure 3—figure supplement 4.. Heritability enrichment in tissues and cell types for annotations based on chromatin accessibility (see Materials and methods).
The top three enrichments for each phenotype passing a trait-wide significance threshold are labelled. Horizontal lines and trait-wide (dotted line) and study-wide (dashed line) significance after Bonferroni correction.
Figure 3—figure supplement 5.
Figure 3—figure supplement 5.. Heritability enrichment in tissues and cell types in immune cell types (see Materials and methods).
The top three enrichments for each phenotype passing a trait-wide significance threshold are labelled. Horizontal lines and trait-wide (dotted line) and study-wide (dashed line) significance after Bonferroni correction.
Figure 3—figure supplement 6.
Figure 3—figure supplement 6.. QQ plots calculated based on a set approximately 500,000 LD-pruned, genotyped SNPs per trait.
See this image and copyright information in PMC

References

    1. Altshuler DM, Gibbs RA, Peltonen L, Altshuler DM, Gibbs RA, Peltonen L, Dermitzakis E, Schaffner SF, Yu F, Peltonen L, Dermitzakis E, Bonnen PE, Altshuler DM, Gibbs RA, de Bakker PI, Deloukas P, Gabriel SB, Gwilliam R, Hunt S, Inouye M, Jia X, Palotie A, Parkin M, Whittaker P, Yu F, Chang K, Hawes A, Lewis LR, Ren Y, Wheeler D, Gibbs RA, Muzny DM, Barnes C, Darvishi K, Hurles M, Korn JM, Kristiansson K, Lee C, McCarrol SA, Nemesh J, Dermitzakis E, Keinan A, Montgomery SB, Pollack S, Price AL, Soranzo N, Bonnen PE, Gibbs RA, Gonzaga-Jauregui C, Keinan A, Price AL, Yu F, Anttila V, Brodeur W, Daly MJ, Leslie S, McVean G, Moutsianas L, Nguyen H, Schaffner SF, Zhang Q, Ghori MJ, McGinnis R, McLaren W, Pollack S, Price AL, Schaffner SF, Takeuchi F, Grossman SR, Shlyakhter I, Hostetter EB, Sabeti PC, Adebamowo CA, Foster MW, Gordon DR, Licinio J, Manca MC, Marshall PA, Matsuda I, Ngare D, Wang VO, Reddy D, Rotimi CN, Royal CD, Sharp RR, Zeng C, Brooks LD, McEwen JE, International HapMap 3 Consortium Integrating common and rare genetic variation in diverse human populations. Nature. 2010;467:52–58. doi: 10.1038/nature09298. - DOI - PMC - PubMed
    1. Anderson ER, Shah YM. Iron homeostasis in the liver. Comprehensive Physiology. 2013;3:315–330. doi: 10.1002/cphy.c120016. - DOI - PMC - PubMed
    1. Basty N, Liu Y, Cule M, Thomas EL, Bell JD, Whitcher B. Automated measurement of pancreatic fat and iron concentration using Multi-Echo and T1-Weighted MRI data. 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI); 2020. pp. 345–348. - DOI
    1. Buch S, Stickel F, Trépo E, Way M, Herrmann A, Nischalke HD, Brosch M, Rosendahl J, Berg T, Ridinger M, Rietschel M, McQuillin A, Frank J, Kiefer F, Schreiber S, Lieb W, Soyka M, Semmo N, Aigner E, Datz C, Schmelz R, Brückner S, Zeissig S, Stephan AM, Wodarz N, Devière J, Clumeck N, Sarrazin C, Lammert F, Gustot T, Deltenre P, Völzke H, Lerch MM, Mayerle J, Eyer F, Schafmayer C, Cichon S, Nöthen MM, Nothnagel M, Ellinghaus D, Huse K, Franke A, Zopf S, Hellerbrand C, Moreno C, Franchimont D, Morgan MY, Hampe J. A genome-wide association study confirms PNPLA3 and identifies TM6SF2 and MBOAT7 as risk loci for alcohol-related cirrhosis. Nature Genetics. 2015;47:1443–1448. doi: 10.1038/ng.3417. - DOI - PubMed
    1. Bugianesi E, Bizzarri C, Rosso C, Mosca A, Panera N, Veraldi S, Dotta A, Giannone G, Raponi M, Cappa M, Alisi A, Nobili V. Low birthweight increases the likelihood of severe steatosis in pediatric Non-Alcoholic fatty liver disease. American Journal of Gastroenterology. 2017;112:1277–1286. doi: 10.1038/ajg.2017.140. - DOI - PubMed

Publication types