Multi-organ genetic causal connections inferred from imaging and clinical data through Mendelian randomization

Genetic causality between brain imaging and multi-organ diseases

In this section, we examined the causal relationship between brain imaging biomarkers and multi-organ diseases. At the Bonferroni significance level (P < 5.18ξ10⁻⁶, multiple testing adjustment for both directions), MR suggested 127 significant genetic causal effects on 58 brain imaging biomarkers from 20 diseases in 8 major categories, including mental and behavioral disorders, diseases of the nervous system, diseases of the circulatory system, cardiometabolic endpoints, interstitial lung disease endpoints, diseases marked as autoimmune origin, diseases of the eye and adnexa, and diseases of the genitourinary system (Fig. S2 and Table S3). Among all the diseases, heart-related diseases were the most frequent (66/127) and were predominantly associated with DTI parameters and a smaller number of regional brain volumes. No significant causal effects were observed from heart diseases to fMRI traits. The top three heart diseases that exhibited causal genetic impacts on brain structures included peripheral artery disease (15/66), hypertension (14/66), and hypertensive diseases (12/66) (Figs. S2-S4). For example, peripheral artery disease played a causal role in altering the white matter microstructure within the anterior limb of the internal capsule (ALIC, |β|>0.15, P < 3.04ξ10⁻⁸), the body of corpus callosum tract (BCC, |β|>0.14, P < 8.44ξ10⁻⁷), and the genu of corpus callosum tract (GCC, |β|>0.13, P < 4.24ξ10⁻⁶). Hypertension and hypertensive diseases were causally associated with the superior corona radiata (SCR, |β|>0.07, P < 9.47ξ10⁻⁸). In addition to DTI parameters, hypertension also negatively affected the total grey matter volume (|β|>0.02, P < 1.26ξ10⁻⁸) (Figs. 2C and S2-S3). In addition to heart- related diseases, several other non-neurological clinical endpoints also influenced brain health. For example, negative causal effects of asthma were found on the SCR (|β|>0.05, P < 7.26ξ10⁻⁷) and volume of the right inferior lateral ventricle (|β|>0.09, P < 8.66ξ10⁻⁷) (Fig. S5).

• Download figure
• Open in new tab

Fig. 2 Selected genetic causal effects of clinical outcomes on brain imaging biomarkers.

We illustrated selected significant (P < 5.18ξ10⁻⁶) causal genetic links from clinical endpoints (Exposure) to brain imaging biomarkers (Outcome) after adjusting for multiple testing using the Bonferroni procedure. (A). The causal effect of Alzheimer’s disease on brain imaging biomarkers. (B). The causal effect of dementia on brain imaging biomarkers. (C). The causal effect of hypertension on brain imaging biomarkers. IDP Category, category of imaging biomarkers; #IVs, the number of genetic variants used as instrumental variables. Different MR methods and their regression coefficients are labeled with different colors. See Table S1 for data resources of clinical endpoints and Table S2 for data resources of imaging biomarkers.

Brain disorders also causally affected the brain imaging biomarkers (49/127), with dementia and Alzheimer’s disease being the most common brain-related diseases (Fig. S2). Interestingly, brain disorders were primarily associated with fMRI traits. For example, Alzheimer’s disease was consistently found to be causally related to decreased functional activity in the dorsal attention (|β|>0.04, P < 5.07ξ10⁻⁸), frontoparietal, and secondary visual network (|β|>0.04, P < 4.70ξ10^-8), as well as DTI parameters of the SCR (|β|>0.03, P < 1.85ξ10⁻⁶) (Figs. 2A and S6). Both functional MRI and DTI parameters have been extensively used to study Alzheimer’s disease⁴⁹. Abnormalities in white matter, such as those in the left SCR, as well as decreased functional connectivity in attention-related networks, have been identified in patients with Alzheimer’s disease^{50, 51}. Similar to Alzheimer’s disease, dementia demonstrated negative causal genetic effects on functional activity in multiple networks, including the cingulo-opercular, default mode, dorsal attention, frontoparietal, language, posterior multimodal, and secondary visual networks (|β|>0.12, P < 4.74ξ10⁻⁸) (Figs. 2B and S7). In addition, mood disorders affected brain volume traits, such as the left and right putamen (|β|>0.04, P < 5.64ξ10⁻⁷) (Fig. S8). More results on causal genetic links from clinical endpoints to brain imaging traits were summarized in Figures S3-S9.

Brain and other organ diseases may also be caused by structural or functional changes in the brain. We investigated this direction by using brain imaging traits as exposure variables and clinical endpoints as the outcome variables. At the Bonferroni significance level (P < 5.18ξ10⁻⁶), we found 85 significant pairs between 22 brain imaging biomarkers and 20 clinical endpoints (Fig. S10 and Table S3). Most of the significant imaging-disease pairs were related to fMRI traits. Specifically, 66 of the 85 pairs were associated with fMRI traits, 10 with DTI parameters, and 9 with regional brain volumes. The majority of the significant findings were related to brain diseases (65/85), with a minor proportion linked to heart diseases (13/85), autoimmune diseases (4/85), COPD and related endpoints (1/85), diseases of the eye and adnexa (1/85), and diseases of the genitourinary system (1/85). For example, decreased activity in multiple functional networks was related to a higher risk of Alzheimer’s disease, such as the default mode and dorsal-attention networks (|β|>0.5, P < 5.65ξ10⁻¹³) (Figs. 3A and S11). We also identified genetic causal effects from DTI parameters on Alzheimer’s disease, such as the BCC and SCR (|β|>0.49, P < 6.79ξ10⁻⁷) (Figs. 3A and S11).

• Download figure
• Open in new tab

Fig. 3 Selected genetic causal effects of brain imaging biomarkers on clinical endpoints.

We illustrated selected significant (P < 5.18ξ10⁻⁶) causal genetic links from brain imaging biomarkers (Exposure) to clinical endpoints (Outcome) after adjusting for multiple testing using the Bonferroni procedure. (A). The causal effect of brain imaging biomarkers on Alzheimer’s diseases. (B). The causal effect of brain imaging biomarkers on dementia. IDP Category, category of imaging biomarkers; #IVs, the number of genetic variants used as instrumental variables. Different MR methods and their regression coefficients are labeled with different colors. See Table S1 for data resources of clinical endpoints and Table S2 for data resources of imaging biomarkers.

Dementia exhibited a similar pattern to Alzheimer’s disease, being causally influenced by decreased activity in multiple networks, such as the default mode, dorsal-attention, and secondary visual network (|β|>0.002, P < 2.77ξ10⁻⁷) (Fig. 3B and S12). fMRI traits, including the functional activity of the secondary visual network and functional connectivity of the default mode network, were also causally linked to other brain diseases, such as neuropsychiatric disorders (|β|>0.09, P<8.88ξ10⁻¹⁶) and neurological diseases (|β|>0.06, P < 1.72ξ10⁻⁶) (Figs. S13 and S14). Finally, we found that brain structural alterations may also influence other non-neurological diseases. For example, the left basal forebrain posed a negative causal effect on hypertensive diseases (|β|>0.12, P < 8.47ξ10⁻⁹) and hypertension (|β|>0.15, P < 6.19ξ10⁻¹¹). The left lingual negatively affected female genital prolapse (|β|>0.75, P < 8.68ξ10⁻⁷).

Causal genetic relationships between CMR traits and clinical outcomes

We first examined the causal effects from clinical endpoints to CMR measures of heart structure and function. We identified 111 significant results at the Bonferroni significance level (P < 6.85ξ10⁻⁶), covering 41 unique CMR traits of the ascending aorta (AAo), descending aorta (DAo), left atrium (LA), and left ventricle (LV). Significant causal effects were found from 13 unique clinical endpoints in three categories: diseases of the circulatory system (8/13), cardiometabolic endpoints (4/13), as well as COPD and related endpoints (1/13). The majority of significant findings were related to heart-related diseases, with 66 out of 111 being diseases of the circulatory system and 44 out of 111 being cardiometabolic endpoints (Fig. S15 and Table S4).

The most frequently observed genetic effects were related to hypertensive diseases and hypertension (Figs. 4A and S16). Specifically, hypertensive diseases had negative causal effects on AAo and DAo distensibility (|β|>0.10, P < 1.26ξ10⁻⁷), whose genetic associations have been found in previous studies^52–54. Hypertensive diseases also affected LV and LA traits, such as the global radial strain (|β|>0.06, P < 4.47ξ10⁻⁹), LA stroke volume (|β|>0.07, P < 3.04ξ10⁻⁸), and LV myocardial mass (|β|>0.11, P < 3.04ξ10⁻¹⁵). These findings were consistent with previous results derived from genetic association studies^{55, 56}. Hypertension exhibited a similar pattern to hypertensive diseases, having a causal impact on various AAo and DAo traits, such as DAo distensibility, along with LA and LV traits, such as the LA minimum volume (LA_min volume, |β|>0.03, P < 7.11ξ10^-6).

• Download figure
• Open in new tab

Fig. 4 Selected genetic causal effects between heart imaging biomarkers and clinical endpoints.

We illustrated selected significant (P < 6.85ξ10⁻⁶) causal genetic links from (A) clinical endpoints (Exposure) to heart imaging biomarkers (Outcome) and (B) heart imaging biomarkers (Exposure) to clinical endpoints (Outcome) after adjusting for multiple testing using the Bonferroni procedure. IDP Category, category of imaging biomarkers; #IVs, the number of genetic variants used as instrumental variables. Different MR methods and their regression coefficients are labeled with different colors. See Table S1 for data resources of clinical endpoints and Table S2 for data resources of imaging biomarkers.

In addition, angina pectoris causally influenced AAo maximum and minimum areas (AAo_max and AAo_min areas, |β|>0.71, P < 1.44ξ10⁻⁷). The aortic aneurysm had a positive causal effect on DAo maximum and minimum areas (DAo_max and DAo_min areas, |β|>0.71, P < 1.63ξ10⁻²⁴) and AAo_max and AAo_min areas (|β|>0.16, P < 2.22ξ10⁻⁶). These results align with clinical observations. Atrial fibrillation and flutter mainly affected LA traits, such as the LA ejection fraction (|β|>0.07, P < 5.88ξ10⁻⁸), LA maximum volume (LA_max volume, |β|>0.07, P < 1.62ξ10⁻⁶), and LA_min volume (|β|>0.08, P < 8.26ξ10⁻⁸). Atrial fibrillation is considered to result in a decrease in ejection fraction as well as an increase in LA volumes^{57, 58}. In addition to heart-related diseases, COPD and related endpoints were

found to influence CMR traits, such as the negative effect of COPD on DAo_max area (|β|>0.10, P < 4.45ξ10⁻¹¹, and Fig. S16). Emphysema, a form of COPD distinguished by the degradation of lung tissue, may contribute to the dilatation of the thoracic aorta⁵⁹. This could be attributed to emphysema’s involvement in the degradation of elastic fibers within the lungs, potentially triggering alterations in the aortic wall’s elasticity⁶⁰. Such changes may precipitate the dilatation or ballooning of the thoracic aorta, thereby escalating the risk of both aortic aneurysm and abdominal aortic abnormalities⁵⁹. This elucidates the potential biological mechanisms driving these causal relationships.

On the other hand, structural and functional irregularities of the heart may increase the risk of multi-organ diseases, given that the heart pumps blood to all other organs to maintain their functions⁶¹. We tested this direction by treating CMR traits as exposure variables and clinical endpoints as the outcomes. After Bonferroni adjustment (P < 6.85ξ10⁻⁶), we found 27 significant causal pairs, 25 for heart-related diseases and 2 for autoimmune diseases (Fig. S17 and Table S4). For example, global peak circumferential strain was positively linked to heart failure and antihypertensive medication (|β|>0.51, P < 8.49ξ10⁻⁸), while LV ejection fraction had a negative causal effect on these conditions (|β|>0.55, P < 1.12ξ10⁻⁶). In addition to heart-related diseases, heart structural changes affected diseases marked as autoimmune origin. For example, right ventricular end- systolic volume had a negative causal effect on the autoimmune diseases defined by Finngen²⁷ (|β|>0.18, P < 3.10ξ10⁻⁶) (Figs. 4B and S18). In summary, we discovered causal relationships between CMR traits and heart-related diseases, which were typically bidirectional. Additionally, we revealed the inter-organ causal relationships between the heart and other organs.

Causal genetic links between abdominal imaging biomarkers and clinical outcomes

We first examined the effects of multi-organ diseases on abdominal imaging biomarkers, such as the volume or iron content of the spleen, kidney, liver, lung, and pancreas⁸. At the Bonferroni significance level (P < 6.69ξ10⁻⁵), we discovered 51 significant causal pairs from multi-organ diseases to abdominal imaging biomarkers, with liver imaging traits being the most impacted (26/51). Brain-related diseases were the most prevalent among all significant findings (35/51), followed by heart-related diseases (7/51), rheumatoid endpoints (4/51), diseases of the eye and adnexa (3/51), and autoimmune diseases (2/51) (Fig. S19 and Table S5). These findings were in line with ongoing research on the interplay between the brain and abdominal organs, such as the brain-gut connection^62–64, brain- kidney connection^{65, 66}, and brain-liver connection⁶⁷.

Alzheimer’s disease and dementia were consistently found to be causally linked with various abdominal imaging biomarkers, such as the percent liver fat (|β|>0.37, P < 4.96ξ10⁻⁶), liver volume (|β|>0.03, P < 5.88ξ10⁻⁵), and adipose tissue measurement (|β|>0.06, P < 3.13ξ10⁻⁵). In addition to Alzheimer’s disease and dementia, there were multiple other brain-related diseases that may affect abdominal organs, including mental and behavioral disorders due to alcohol and psychoactive substance use, as well as sleep apnoea. For example, sleep apnoea influenced several abdominal traits, such as the liver volume (|β|>0.11, P < 2.80ξ10⁻⁵) and kidney volume (|β|>0.23, P < 2.57ξ10⁻⁶). Sleep apnea can lead to renal damage caused by ischemic stress, hemodynamic changes, or intermediary conditions such as hypertension, which can result in early chronic kidney disease^{68, 69}. Multiple heart-related diseases also genetically impacted abdominal organs. For example, heart failure and antihypertensive medication can lead to larger spleen volume (|β|>0.004, P < 2.66ξ10⁻⁶). It has been found that heart splenic enlargement often results from blood stasis and right heart disease is often accompanied by splenomegaly⁷⁰. Atherosclerosis was causally linked to the pancreas iron content, which was also supported by clinical evidence^{71, 72}.

In addition to brain and heart-related diseases, we also observed causal effects from other diseases, such as autoimmune diseases on spleen volume and liver iron content. The spleen, as the largest immune organ in the body, can become enlarged because of various rheumatic and immune system diseases, such as systemic lupus erythematosus, Felty’s syndrome, sarcoidosis, and autoimmune hepatitis^73–76. Liver iron content causally related to both autoimmune and rheumatological diseases (|β|>0.10, P < 3.02ξ10⁻⁵). It has been found that excessive deposition of iron ions exists in the affected tissues of autoimmune diseases, such as brain tissues of multiple sclerosis patients and synovial fluid of patients with rheumatoid arthritis⁷⁷. Disorders of the choroid and retina, as well as the eye and adnexa diseases, were also found to be causally linked to liver iron content (|β|>0.10, P < 6.48ξ10⁻⁵, Figs. 5A and S20). Metal tends to accumulate in human ocular tissues, particularly in the choroid and retinal pigment epithelium⁷⁸.

• Download figure
• Open in new tab

Fig. 5 Selected genetic causal effects between abdominal imaging biomarkers and clinical endpoints.

We illustrated selected significant (P < 6.69ξ10⁻⁵) causal genetic links from (A) clinical endpoints (Exposure) to abdominal imaging biomarkers (Outcome) and (B) abdominal imaging biomarkers (Exposure) to clinical endpoints (Outcome) after adjusting for multiple testing using the Bonferroni procedure. IDP Category, category of imaging biomarkers; #IVs, the number of genetic variants used as instrumental variables. Different MR methods and their regression coefficients are labeled with different colors. See Table S1 for data resources of clinical endpoints and Table S2 for data resources of imaging biomarkers.

Next, we tested the opposite direction that abdominal imaging biomarkers being the exposure variables and multi-organ diseases being the outcomes. At the Bonferroni significance level (P < 6.69ξ10⁻⁵), we identified 55 significant pairs, with heart-related diseases being the most prevalent (34/55), followed by brain-related diseases (11/55), diseases marked as autoimmune origin (2/55), diseases of the eye and adnexa (5/55), and diseases of the genitourinary system (3/55) (Fig. S21 and Table S5). For example, pancreas fat causally affected the deep vein thrombosis of lower extremities and pulmonary embolism (|β|>0.38, P < 1.04ξ10⁻¹²). A reduction in pancreatic fat content may directly improve cellular function and insulin secretion rate, affecting triglyceride levels and blood flow⁷⁹.

Larger liver, spleen, and kidney volumes were all causally linked to heart-related diseases (Fig. 5B and S22). Specifically, a larger liver volume was causally linked to hypertensive diseases and hypertension (|β|>0.13, P < 1.19ξ10⁻¹⁰); a larger kidney volume was causally related to a higher risk of stroke (|β|>0.25, P < 3.33ξ10⁻⁵); and a larger spleen volume had causal effects on various heart-related diseases, including coronary angioplasty (|β|>0.14, P < 4.94ξ10⁻⁷), coronary atherosclerosis (|β|>0.07, P < 6.94ξ10⁻¹⁰), and peripheral artery diseases (|β|>0.21, P < 4.98ξ10⁻⁵). In nephrotic syndrome, platelet over- activation and the use of diuretics and glucocorticoids can aggravate hypercoagulability. Therefore, kidney diseases, especially nephrotic syndrome, are prone to thrombotic and embolic complications, which can lead to stroke^80–82. For brain-related disorders, we detected causal effects from liver volume to alcohol use disorder, as well as mental and behavioral disorders due to alcohol and psychoactive substance use. In addition, percent liver fat had causal links with Alzheimer’s disease, dementia, and psychiatric diseases. Previous studies have reported that non-alcoholic fatty liver disease contributes to neurological conditions like cognitive impairment and memory loss via insulin resistance and inflammation, along with excessive cytokine secretion^83–86. In summary, we found that brain-related disorders result in alterations in abdominal organs. Furthermore, bidirectional relationships are observed in both neurodegenerative and psychiatric disorders. On the other hand, multiple abdominal organs are causally linked to heart- related diseases.

Causal genetic links between skeleton DXA traits and clinical outcomes

We first identified the causal effects of multiple organ diseases on DXA-derived skeleton traits³. At the Bonferroni significance level (P < 3.39ξ10⁻⁵), we found strong evidence that multi-organ diseases affected the human skeleton health, where heart-related diseases (6/12) and diseases of the nervous system (4/12) were the majority, as well as rheumatic disease (1/12) and diseases of the genitourinary system (1/12) (Fig. S23 and Table S6). Carpal tunnel syndrome (|β|>0.0007, P < 6.25ξ10⁻⁸) and sleep apnoea (|β|>0.001, P < 6.35ξ10⁻⁸) were causally related to higher average forearm length. Carpal tunnel syndrome is when the median nerve (nerve from the forearm to the palm of the hand) becomes pressed or squeezed, which affects the wrist-to-forearm ratio^{87, 88}. It has been observed that oral appliance therapy⁸⁹, which is an effective treatment of sleep apnea, is associated with skeletal changes⁹⁰. Furthermore, the nerve, nerve root, and plexus disorders were causally linked with higher average tibia length (|β|>0.001, P < 2.78ξ10^{- 5}), which was consistent with the previous finding that a specific type of plexus disorder, lumbosacral plexus disorder, is associated with lower leg⁹¹. Heart-related diseases also had causal effects on several DXA traits. For example, coronary heart disease had a negative causal effect on the average tibia length (|β|>0.0002, P < 2.50ξ10⁻⁵) and a positive causal effect on the hip width (|β|>0.0004, P < 1.68ξ10⁻⁵). Additionally, gonarthrosis affected the average tibia length (|β|>0.0005, P < 1.69ξ10⁻⁵) (Figs. 6A and S24).

• Download figure
• Open in new tab

Fig. 6 Selected genetic causal effects between skeleton imaging biomarkers and clinical endpoints.

We illustrated selected significant (P < 3.39ξ10⁻⁵) causal genetic links from (A) clinical endpoints to (Exposure) to skeleton imaging biomarkers (Outcome) and (B) skeleton imaging biomarkers (Exposure) to clinical endpoints (Outcome) after adjusting for multiple testing using the Bonferroni procedure. IDP Category, category of imaging biomarkers; #IVs, the number of genetic variants used as instrumental variables. Different MR methods and their regression coefficients are labeled with different colors. See Table S1 for data resources of clinical endpoints and Table S2 for data resources of imaging biomarkers.

The skeletal system serves as the foundational support for the human body, and therefore, skeletal abnormalities may potentially contribute to risk of multi-organ diseases. We identified 17 causal pairs at the Bonferroni significance level (P < 3.39ξ10⁻⁵). More than half (10/17) of these results were related to the heart, and the rest were rheumatic disease (3/17), diseases of the eye and adnexa (2/17), interstitial lung diseases (1/17), and diseases of the genitourinary system (1/17) (Fig. S23 and Table S6). Average tibia length was causally linked to coxarthrosis (|β|>30.03, P < 3.11ξ10⁻⁹), gonarthrosis (|β|>19.39, P < 3.18ξ10⁻⁷), and other rheumatological endpoints (|β|>13.12, P < 2.74ξ10⁻⁵). It has been found that the leg length discrepancy could lead to lower limb biomechanics, such as gonarthrosis, coxarthrosis, and other lower limb symptoms^{92, 93}. Torso length causally affected heart-related issues, such as coronary angioplasty (|β|>38.85, P < 9.45ξ10⁻⁵), coronary atherosclerosis (|β|>22.94, P < 1.22ξ10⁻⁵), hard cardiovascular diseases (|β|>18.12, P<1.30ξ10⁻⁵), and ischemic heart diseases (|β|>30.99, P < 1.33ξ10⁻⁵). Previous studies examining the relationship between skeletal length and heart diseases have primarily focused on leg length or overall body height, generally reporting negative associations^{94, 95}. Consistent with these associations, we found that a long torso can lead to a higher risk of heart disease. We also observed higher average tibia length to be causally linked to a lower risk of hypertension (|β|>6.46, P < 1.27ξ10⁻⁷), which was in line with previous findings⁹⁶ (Figs. 6B and S25). In conclusion, we found that rheumatoid endpoints (such as gonarthrosis) and diseases of the nervous system (such as nerve, nerve root, and plexus disorders) had a significant impact on bone health. Conversely, skeletal traits, like torso length, demonstrated a causal link with heart diseases.

Intra-brain causal connections

Variations in brain structure and function were closely linked with brain disorders, with parts of these relationships appearing to be bidirectional. We consistently found causal links between brain imaging biomarkers and multiple psychiatric disorders or neurological diseases, such as Alzheimer’s disease, dementia, mood disorder, and sleep apnea. For example, Alzheimer’s disease and dementia had bidirectional causal links with fMRI traits and DTI parameters. Previous studies have consistently shown that resting fMRI connectivity patterns are altered in patients with Alzheimer’s disease^{101, 102}, especially in brain regions involved in memory and cognitive function^{103, 104}.

Brain-heart causal connections

While association studies have been investigating the brain-heart interaction^{39, 105}, the causal genetic links within these heart-brain systems remain largely unexplored. We discovered causal connections from several heart-related diseases such as hypertension, hypertensive diseases, heart failure, and peripheral artery disease to DTI parameters in white matter tracts such as the SCR, ALIC, BCC, GCC, the splenium of corpus callosum, and the retrolenticular part of the internal capsule (RLIC). Additionally, these diseases were also linked to regional brain volumes, such as grey matter and left amygdala. Hypertension can lead to damage of the blood vessels in the brain¹⁰⁶, which can in turn lead to a reduction in the volume of grey matter in certain brain regions¹⁰⁷. This may result in cognitive impairment and an increased risk of developing dementia. Therefore, effective management of hypertension through lifestyle changes and medication can help reduce the risk of these negative effects on the brain. On the other hand, alterations in brain structure, such as deformations in the left ventral DC and left basal forebrain, were found to contribute to heart-related diseases like hypertension. Similarly, changes in the left superior temporal region were linked to heart failure. These could be attributed to the brain’s essential function in controlling blood pressure via a sophisticated network that involves multiple regions and pathways^{108, 109}.

Bidirectional connections between the brain and abdominal organs

Brain abnormalities affected multiple abdominal organs and the skeletal system. For example, Alzheimer’s disease and dementia causally affected the percent liver fat, and neurological diseases (defined by FinnGen²⁷) had a positive causal effect on lung volume. It has been found that neurological diseases, such as multiple sclerosis, Parkinson’s disease, amyotrophic lateral sclerosis, and Huntington’s disease, can cause respiratory muscle weakness^{110, 111}, which can affect lung volume and function. We also found sleep apnoea may lead to larger spleen and kidney volume. This may be due to the increased workload on the spleen to filter blood and remove damaged red blood cells. Additionally, the low oxygen levels associated with sleep apnea can lead to an increase in the number of red blood cells in the body, which can also contribute to splenomegaly and a change in kidney volume. Sleep apnea was causally associated with increased pancreas iron content, potentially due to the decreased oxygen levels that accompany sleep apnea, resulting in increased iron absorption in the body. The excess iron in the pancreas can lead to oxidative stress and inflammation^{112, 113}, which can contribute to the development of pancreatic damage and dysfunction^114–116.

In addition to the aforementioned neurological diseases, we also discovered that mental and behavioral disorders attributed to alcohol can lead to an increase in percent liver fat. It has been found that people with alcohol use disorder are more likely to develop alcoholic fatty liver disease¹¹⁷. Besides brain-related diseases, brain structural alternations could also lead to a higher risk of diseases of other organs. For example, right postcentral was causally linked to COPD. Some studies have suggested that chronic stress and anxiety, which are associated with changes in brain structure, may contribute to the development or worsening of respiratory conditions such as COPD^{118, 119}.

On the other hand, brain imaging biomarkers or disorders were causally affected by several diseases of other organs or systems. A high percent liver fat resulted in a lower risk of Alzheimer’s disease and dementia. Previous studies^{120, 121} have reported associations between the two and our results aligned with the most recent study¹²². More work is needed to understand the underlying pathophysiological mechanism. Brain imaging biomarkers were affected by multi-organ diseases, but some of them may affect the brain indirectly, such as through mediating effects of anxiety and depression. For example, diseases of the genitourinary system (ovarian cyst and menorrhagia) causally affected brain structural features. Ovarian cysts can cause hormonal imbalances due to the production of hormones by the cysts themselves¹²³. These hormonal imbalances can cause a range of symptoms, such as mood swings, anxiety, and depression¹²⁴, which can affect brain function and emotional regulation. Diseases of the eye and adnexa (conjunctivitis) had genetic causal effects on functional connectivity traits. Conjunctivitis can be caused by a viral or bacterial infection, which can potentially lead to cause systemic inflammation in the body¹²⁵. Systemic inflammation has been linked to changes in brain function and structure and may affect brain fMRI traits^{126, 127}.

In addition, asthma influenced regional brain volumes. One prevalent way in which asthma impacts the brain is via the emotional and psychological stress associated with managing a chronic illness. Anxiety, stress, and depression, often faced by individuals with asthma, can induce alterations in brain structure. Previous studies have shown that individuals with asthma may have reduced cognitive function, including impaired memory and attention, and changes in brain activity patterns during cognitive tasks^{128, 129}. Last but not least, autoimmune diseases (defined by FinnGen²⁷) affected brain imaging biomarkers, such as DTI parameters of the RLIC and SCR. Multiple sclerosis is an autoimmune disease that affects the central nervous system, and the damage to the myelin sheath that surrounds axons can occur in various regions of the brain^{130, 131}, including the internal capsule. The damage can cause disruptions in the neural connections passing through the anterior limb, leading to symptoms such as weakness, spasticity, and difficulty with balance and coordination. In some rare autoimmune diseases, such as neuromyelitis optica¹³² and autoimmune encephalitis, inflammation and damage can occur in the brain.

The resulting neurological symptoms can vary depending on the severity and location of the damage^{133, 134}. Furthermore, autoimmune diseases that cause systemic inflammation, such as rheumatoid arthritis and lupus, which can potentially affect the brain¹³⁵ and white matter tracts^{136, 137}. Chronic inflammation can lead to changes in the microstructure of white matter tracts, which can result in alterations in neural connectivity and function^{138, 139}.

Intra-heart causal connections

Bidirectional causal relationships were identified between heart-related diseases and CMR traits. Hypertension and hypertensive diseases were found to causally influence several CMR traits across various heart chambers and aorta regions. Conversely, variations in CMR traits were observed to potentially lead to heart diseases. These findings are in accordance with existing clinical evidence. For example, hypertension can cause the LA to enlarge, a condition known as left atrial hypertrophy^{140, 141}. This enlargement can lead to several complications, including atrial fibrillation, heart failure, and stroke¹⁴². As for atrial fibrillation, the electrical signals that control the heartbeat become chaotic, causing the heart to beat irregularly and often too fast. Over time, the constant irregularity of the heartbeat can also cause the LA to enlarge and weaken¹⁴³.

Bidirectional connections between the heart and abdominal organs

Heart diseases and various multi-organ imaging biomarkers were causally related. For instance, heart failure was found to cause an increase in spleen volume. When the heart is not able to pump blood effectively, it can cause an increase in the pressure from the veins to the spleen. This increased pressure can cause the spleen to enlarge, a condition known as splenomegaly¹⁴⁴. In addition, the backup of blood in the liver that can occur with heart failure can also contribute to the development of splenomegaly. Larger spleen volume can conversely lead to heart failure. An enlarged spleen can increase the workload on the heart, leading to further worsening of heart failure symptoms.

Pancreas was also found to have a causal effect on the heart. For example, excess pancreas fat was found to cause a higher risk of developing deep vein thrombosis of lower extremities and pulmonary embolism. Pancreatic steatosis is a condition where fat accumulates in the pancreas. This is associated with a number of metabolic abnormalities, including insulin resistance and inflammation, which can contribute to the development of cardiovascular disease^{145, 146}. The inflammatory and procoagulant effects of excess pancreatic fat could potentially contribute to an increased risk of deep vein thrombosis.

The heart and lungs are closely connected and work together as part of the cardiovascular system. The lungs are responsible for taking in oxygen from the air we breathe and transferring it into the bloodstream, while the heart pumps the oxygen-rich blood throughout the body to nourish cells and tissues. We found strong evidence of causal links from COPD and CMR traits of DAo. The degradation of elastic fibers through proteolysis is a characteristic of emphysema¹⁴⁷, which can potentially lead to the enlargement of the thoracic aorta.

Spleen, as the largest immune organ, is closely related with diseases marked as autoimmune origin. We found strong evidence showing the genetic causal relationship between spleen volume and autoimmune diseases (defined by Finngen²⁷), such as rheumatoid arthritis, systemic lupus erythematosus, and systemic sclerosis. It has been reported that autoimmune diseases can affect the spleen. For instance, conditions like lupus or rheumatoid arthritis can lead to splenomegaly, a condition often triggered by inflammation or the accumulation of abnormal immune cells in the spleen^{148, 149}.

Skeleton DXA traits

We also found genetic causal links between skeleton DXA traits and multiple organ diseases. Skeleton traits were causally affected by diseases of the nervous system, rheumatic disease, as well as nerve, nerve root and plexus disorders. Heart diseases may also impact the skeletal system by influencing bone health. It has been observed that individuals with heart disease, especially those with heart failure, have an increased risk of osteoporosis and bone fractures^{149, 150}. This may be due to a variety of factors. For example, some medications used to treat heart diseases, such as diuretics and steroids, can also increase the risk of osteoporosis¹⁵¹. Additionally, individuals with heart disease may have reduced mobility and physical activity, which can lead to decreased bone density and strength¹⁵². Skeleton problems could also inversely contribute to numerous organ diseases, with heart conditions being the most prevalent in our analysis. A long torso may lead to a high risk of coronary heart disease. We also noted that a higher average tibia length was causally associated with a lower risk of hypertension, a finding that aligns with clinical observations⁹⁶.

Limitations and conclusions

Our study has several limitations. First, we collected GWAS summary statistics from publicly available databases, meaning that we were not able to evaluate the impact of unobserved confounders (such as population stratification) on our results. Second, one common limitation of most existing MR methods is that they require several model assumptions, and thus may suffer from model misspecifications and data heterogeneity issues when integrating data from different data resources¹⁵³. We have systematically applied quality control measures and conducted sensitivity analyses in our study. Future research implementing more advanced MR methods may relax some of the model assumptions the current MR methods have made^{154, 155}. Furthermore, MR studies are designed to examine the effects of lifetime exposure factors on outcomes, not interventions within a specified period. As such, our findings may have different interpretations from rigorous results obtained from randomized controlled trials. Therefore, any clinical interventions based on these MR findings should be undertaken with caution.

In conclusion, we used two-sample bidirectional MR analyses to comprehensively explore the multi-organ causal connections between 88 clinical outcomes and 402 image-derived phenotypes of various organ systems. Our results revealed robust genetic evidence supporting causal connections within and across multiple organs. This will aid in unraveling complex pathogenic mechanisms and will contribute to the early prediction and prevention of multi-organ diseases from a whole body perspective.

Multi-organ imaging biomarkers

The imaging data were sourced from the UK Biobank (UKB) study, which enrolled approximately 500,000 individuals aged between 40 and 69 from 2006 to 2010 (https://www.ukbiobank.ac.uk/). These multi-organ imaging data were collected from the ongoing UKB imaging study project (https://www.ukbiobank.ac.uk/explore-your-participation/contribute-further/imaging-study), which aims to collect brain, heart, and abdomen scans from 100,000 participants. Ethical approval for the UKB study was secured from the North West Multicentre Research Ethics Committee (approval number: 11/NW/0382).

Studies of brain and heart diseases usually rely on magnetic resonance imaging (MRI) scans, which are well-established clinical endophenotypes. Cardiovascular magnetic resonance imaging (CMR) is a set of MRI techniques that are designed to assess ventricular function, cardiovascular morphology, myocardial perfusion, and other cardiac functional and structural features^{156, 157}. They have been frequently used to reveal heart- related issues clinically. The CMR traits used in the paper were originally generated from the raw short-axis, long-axis, and aortic cine images using the state-of-the-art heart imaging segmentation and feature representation framework^{38, 158, 159}. We divided the generated 82 CMR traits into 6 categories. The first two are aortic sections, namely ascending aorta (AAo) and descending aorta (DAo), which serve as the main ‘pipe’ in supplying blood to the entire body. The other four are the global measures of 4 cardiac chambers, including the left ventricle (LV), right ventricle (RV), left atrium (LA), and right atrium (RA), which altogether manage the heartbeat and blood flow. There are also some other traits, such as regional phenotypes of the left ventricle myocardial-wall thickness and strain (Table S1). The summary-level GWAS data of these 82 CMR traits were obtained from Zhao, et al. ³⁹.

Brain MRI provides detailed information about brain structure and function¹⁶⁰, such as abnormal growth, healthy aging, white matter diseases, structural issues, and functional abnormalities. In this paper, the summary-level GWAS data were collected from recent multi-modal image genetic studies, including regional brain volumes from structural MRI^{21, 161} (sMRI), diffusion tensor imaging (DTI) parameters from diffusion MRI^{23, 162} (dMRI), and functional activity (that is, amplitude³⁷) and functional connectivity phenotypes from resting functional MRI²⁵ (resting fMRI). In sMRI, we used ANTs¹⁶³ to generate regional brain volumes for cortical and subcortical regions and global brain volume measures. In dMRI, we used the ENIGMA-DTI pipelines^{164, 165} to generate tract-averaged parameters for fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, and mode of anisotropy in major white matter tracts and across the whole brain. For resting fMRI, we extracted phenotypes from brain parcellation-based analysis. We used the Glasser360 atlas¹⁶⁶, which divided the cerebral cortex into 360 regions in 12 functional networks¹⁶⁷. We considered 90 network-level resting fMRI phenotypes that evaluated interactions and spontaneous neural activity at rest.

The 11 imaging biomarkers from abdominal MRI were derived by Liu., et al⁸ using deep learning methods in terms of volume, fat, and iron in several organs and tissues, such as the liver, spleen, kidney, lung, pancreas, and adipose tissue. Skeleton DXA traits, including all long bone lengths as well as hip and shoulder width, were derived by Kun., et al³ using deep learning methods on whole-body dual-energy X-ray absorptiometry (DXA) images.

All eight skeleton traits have been controlled for height. The heritability of the above imaging biomarkers can be found in Supplementary Note.

FinnGen clinical endpoints

We used 88 clinical endpoints collected by the FinnGen project, which were selected from the R7 release and with more than 10,000 cases for most of the clinical endpoints (https://www.finngen.fi/en/access_results). As for some important diseases, such as Alzheimer’s disease, we set the cutoff of the number of cases to be 6,000. The 88 clinical endpoints covered diseases from various categories, namely, mental and behavioral disorders, diseases of the nervous system, diseases of the eye and adnexa, diseases of the genitourinary system, diseases of the circulatory systems, cardiometabolic endpoints, diseases marked as autoimmune origin, rheuma endpoints, interstitial lung diseases, COPD and related endpoints, as well as some unclassified endpoints. The definations can be found at https://risteys.finregistry.fi/. The FinnGen data used in our study was obtained from separate cohorts than those supplying imaging traits, which were derived from the UKB study, thus ensuring there was no sample overlap. Detailed information of these 88 clinical variables can be found in Table S2.

Mendelian randomization analysis

We examined the genetic causal relationships between the 402 imaging traits (101 brain regional volume traits, 110 brain DTI parameters, 90 network-level fMRI phenotypes, 82 CMR traits, 11 abdominal traits, and 8 skeleton DXA traits) and 88 clinical endpoints. Prior to conducting the Mendelian randomization (MR) analysis, we conducted standard preprocessing and quality control procedures. First, we selected genetic variants based on a significance threshold of 5×10⁻⁸ in the exposure GWAS data. To ensure the independence of the genetic variants used in MR, we implemented LD clumping with a window size of 10,000 and an r² threshold of 0.01, using the 1000 Genomes European ancestry data as a reference panel. We used the TwoSampleMR package (https://mrcieu.github.io/TwoSampleMR/) for harmonization, which enabled us to accurately align alleles between the selected variants in the exposure and the reported effect on the outcome.

We assessed the performance of 8 MR methods, which included Inverse variance weighted (fixed effect), Inverse variance weighted (multiplicative random effect), MR- Egger, Simple Median, Weighted Median, Weighted Mode, DIVW, GRAPPLE, and MR- RAPS^{40, 41, 43–48, 168}, where MR Egger was used as the pleiotropy test. To ensure the reliability of our results, we implemented several quality control procedures. We excluded causal estimates that relied on fewer than 6 genetic variants, as a larger number of genetic variants increases the statistical power of MR analysis^{46, 47}. We retained causal pairs that were significant in at least two out of the eight methods. We also screened for pleiotropy by using the MR-Egger intercept, the most used method for testing the pleiotropy assumption. If a causal estimate failed the MR-Egger intercept test, we required that it have significant results in at least one of the robust MR methods, such as Weighted Median, Weighted Model, MR-RAPS, or GRAPPLE. Out of 488 significant findings, 81 causal estimates failed the MR-Egger intercept test. However, when we interpreted the results, we focused on the ones that passed the MR-Egger intercept test.

Code availability

We made use of publicly available software and tools. Our analysis code will be made freely available at Zenodo.

Data availability

We used summary-level GWAS data in this study, which can be obtained from the FinnGen project (https://www.finngen.fi/en/access_results), BIG-KP (https://bigkp.org/), Heart-KP (https://heartkp.org/), and project-specific resources detailed in Liu., et al⁸ and Kun., et al³. Our multi-organ MR results can be explored at https://mr4mo.org/.