Abstract

   Abdominal aortic aneurysm has a high heritability and often co-occurs
   with other cardiometabolic disorders, suggesting shared genetic
   susceptibility. We investigate this commonality leveraging recent GWAS
   studies of abdominal aortic aneurysm and 32 cardiometabolic traits. We
   find significant genetic correlations between abdominal aortic aneurysm
   and 21 of the cardiometabolic traits investigated, including causal
   relationships with coronary artery disease, hypertension, lipid traits,
   and blood pressure. For each trait pair, we identify shared causal
   variants, genes, and pathways, revealing that cholesterol metabolism
   and inflammation are shared most prominently. Additionally, we show the
   tissue and cell type specificity in the shared signals, with strong
   enrichment across traits in the liver, arteries, adipose tissues,
   macrophages, adipocytes, and fibroblasts. Finally, we leverage
   drug-gene databases to identify several lipid-lowering drugs and
   antioxidants with high potential to treat abdominal aortic aneurysm
   with comorbidities. Our study provides insight into the shared genetic
   mechanism between abdominal aortic aneurysm and cardiometabolic traits,
   and identifies potential targets for pharmacological intervention.

   Subject terms: Cardiovascular genetics, Genome-wide association
   studies, Aneurysm
     __________________________________________________________________

   Here, the authors identify SNPs, genes, pathways, cells and tissues
   shared between abdominal aortic aneurysm and 21 cardiometabolic traits,
   generating a reference for common mechanisms and identifying drugs for
   comorbid conditions.

Introduction

   Abdominal aortic aneurysm (AAA), defined as focal dilation of the
   abdominal aorta by 50% or reaching ≥ 30 mm in diameter, is a complex
   vascular disease with an estimated global prevalence of 0.92%^[28]1. It
   is asymptomatic in early disease stages, with most AAA discovered by
   incidental imaging or screening protocols. Once reaching 55 mm, the
   five-year cumulative rupture rate is 25-40%^[29]2. Among ruptured
   patients, a mortality rate as high as 80% was observed^[30]3, rendering
   AAA a leading cause of death.

   AAA is characterized by remodeling and degradation of the extracellular
   matrix, apoptosis of smooth muscle cells, luminal thrombosis, and
   chronic inflammation^[31]4,[32]5. Plaques consisting of lipids, blood
   cells and other plasma substances accumulate around the lesion sites,
   with abundant infiltration of innate and adaptive immune cells both in
   the thrombus and the arterial wall^[33]6. Meanwhile, metabolic
   homeostasis can be perturbed, resulting in enhanced glycolysis in the
   aortic wall^[34]7 and altered serum levels of amino acids and
   lipids^[35]8–[36]10. Often, circulating total cholesterol, low-density
   lipoprotein cholesterol (LDL-C), triglycerides, and sulfur amino acids
   are elevated, whereas high-density lipoprotein cholesterol (HDL-C) and
   phosphatidylcholines are reduced. These changes resemble numerous other
   cardiovascular diseases (CVDs), such as coronary artery disease (CAD),
   myocardial infarction (MI), and peripheral arterial disease^[37]11.
   Indeed, atherosclerosis occurs in 25–55% AAA patients^[38]12, and known
   risk factors of AAA including male sex, age, smoking,
   hypercholesterolemia, hyperlipidemia, and hypertension^[39]13, are
   widely shared among CVDs.

   AAA is highly heritable, with an estimated 70% heritability by family
   and twin studies^[40]14,[41]15. In fact, high heritability is generally
   observed in cardiometabolic disorders^[42]16,[43]17, rendering genetic
   studies a valuable tool to decipher the disease mechanisms^[44]18.
   Genome-wide association studies (GWAS), particularly those performed in
   recent years with large sample sizes, have uncovered single nucleotide
   variants (SNVs) associated with many complex diseases^[45]19. A recent
   meta-GWAS of AAA examined 39,221 cases and 1,086,107 controls,
   resulting in 141 susceptible loci^[46]20, a several-fold increase in
   disease loci compared to earlier studies^[47]21,[48]22. Similarly,
   recent GWAS provided comprehensive variant profiles for dozens of
   cardiometabolic traits (CMTs), which have greatly enhanced our
   understanding of these diseases.

   In this study, we leverage these large GWAS data to identify genetic
   factors shared by AAA and CMTs. We aim to identify shared SNVs and
   genes, as well as the enriched pathways, cell types, and tissues.
   Importantly, these results offer valuable information for prioritizing
   drugs that target shared genes for treating AAA with comorbid
   conditions.

Results

GWAS datasets

   We obtained GWAS summary statistical data for 18 cardiometabolic
   diseases (CMDs) including AAA, 15 metabolic traits, and 6 immune cell
   traits (Fig. [49]1A). These traits are distributed over a broad
   spectrum of cardiac and metabolic functions, including heart functions,
   vascular circulation, glucose metabolism, lipid metabolism, and
   immunity. Most of the CMDs were studied in more than 10,000 case
   samples, whereas metabolic traits and immune cells were measured in a
   minimum of 560,000 individuals. Although European ancestry was
   dominant, many studies included various ancestral groups. Furthermore,
   the number of interrogated genotypes ranged between 4.5–52 million, and
   the significant SNVs (P < 5 × 10^−8) were ample (Supplementary
   Data [50]1). Overall, these datasets present a state-of-the-art
   discovery power for common SNVs-based genetic susceptibility to
   cardiometabolic disorders. Around these datasets, we designed analysis
   modules to elucidate the shared genetic architecture of AAA and CMTs,
   including shared SNVs, genes, pathways, tissues, and cell types
   (Fig. [51]1B). Coherent signals from various analyses are found and
   presented below.

Fig. 1. Interrogation of genetic components between AAA and the traits
related to cardiometabolism.

   [52]Fig. 1
   [53]Open in a new tab

   A Traits and diseases in this study include 18 cardiometabolic
   diseases, 15 metabolic traits, and 6 immune cell traits. This graph was
   created via [54]https://www.biorender.com/. B Analysis modules included
   computing genome-wide genetic correlations, inferring causality between
   AAA and the traits by bidirectional Mendelian randomization,
   identifying shared causal variants, genes and pathways, discovering
   tissues and cell being impacted the most by the shared signals, and
   prioritizing drugs for treating AAA comorbidities. CMD: cardiometabolic
   diseases, MT: metabolic traits.

Genetic correlation

   Genome-wide correlations computed by LDSC^[55]23 suggest positive
   correlations between AAA and 20 CMTs (Fig. [56]2A). The highest
   correlated traits are aortic aneurysms, followed by numerous diseases
   including MI, CAD, peripheral artery disease, subarachnoid hemorrhage,
   and heart failure (r[g] >= 0.3, P < 1 × 10^−10). Compared to the
   disorders, the physiological traits display weaker correlations, with
   lipids, adiposity, blood pressure, and glucose traits in descending
   order. Only HDL-C presented a negative correlation with AAA
   (r[g] = −0.25, P = 7.61 × 10^−32). Immune cell counts and percentages
   did not correlate with AAA, and thus were excluded from subsequent
   analyses. We also computed genetic correlation by functional elements.
   Repressors, enhancers and promoters tend to have the strongest
   correlations across traits (Supplementary Fig. [57]1), suggesting
   transcriptional regulation is genetically shared.

Fig. 2. Genetic correlation and causal inference between AAA and CMTs.

   [58]Fig. 2
   [59]Open in a new tab

   A The heatmap presents the genetic correlation r[g] calculated in LDSC,
   with the color scale indicating the strength of the correlation, and
   the r[g] value displayed next to the heatmap. The * marks the
   statistical significance: *: P < 0.05; **: P < 0.0016
   (Bonferroni-corrected P value threshold). B Causal inference by
   two-sample Mendelian Randomization with five methods. Odds ratios are
   shown as dots, the color bars present +/− 95% confidence intervals, and
   P values are depicted above the bars. CMD: cardiometabolic diseases,
   MT: metabolic traits. IMC: immune cell traits. All reported P values
   are two-sided, unless stated otherwise. Source data are provided with
   this paper.

Causal Inference

   Many cardiometabolic disorders share risk factors, rendering genetic
   correlation a result of complex pleiotropic effects. Mendelian
   Randomization (MR) overcomes the confounding factor issue and provides
   causal inference. We conducted bidirectional MR using several models
   and found a mutual causality between AAA and CAD (Fig. [60]2B).
   Furthermore, AAA was suggested as causal to MI. Reversely, 10 traits
   were inferred as causal to AAA, including hypertension (OR = 2.01,
   P = 3.36 × 10^−4), lipid and adiposity traits (OR = 1.46-1.73, P < 1.24
   × 10^−12), CAD (OR = 1.23, P = 2 .34 × 10^−5), and diastolic blood
   pressure (OR = 1.05, P = 1.13 × 10^−11). Conversely, HDL-C (OR = 0.65,
   P = 2.28 × 10^−21) and pulse pressure (OR = 0.97, P = 2.65 × 10^−8)
   were causally protective against AAA. Note that no apparent horizontal
   pleiotropy was detected as the intercept of MR-Egger did not
   significantly deviate from zero (Supplementary Table [61]1).

Cross-trait loci and causal variants

   Through cross-trait meta-analysis by MTAG (Multi-Trait Analysis of
   GWAS)^[62]24 and CPASSOC (Cross-Phenotype Association Analysis)^[63]25,
   we identified 203 SNVs collectively shared by the 21 trait pairs
   (Supplementary Data [64]2). Overall, AAA shares the largest number of
   SNVs with CAD (N = 46), followed by lipid traits (about 20−40 SNVs)
   (Supplementary Fig. [65]2). Next, to derive shared causal SNVs, we
   first fine-mapped the SNVs with FM-summary^[66]26 for a 99% credible
   set, and then colocalized these SNVs across traits by Coloc^[67]27. As
   such, a total of 177 causal variants shared by two traits were derived
   (Supplementary Data [68]3). We also applied HyPrColoc^[69]28 and
   derived 47 causal variants shared by multiple traits (Fig. [70]3A).
   Among the 47 shared causal variants, only four had the smallest GWAS P
   values (Fig. [71]3B), reinforcing that local lead SNVs in GWAS may only
   tag the causal SNVs^[72]26.

Fig. 3. The overall landscape of the pleiotropic associations across AAA and
CMTs.

   [73]Fig. 3
   [74]Open in a new tab

   A 47 causal variants are shared by multiple traits, as identified by
   HyPrColoc. B LocusZoom plots of four causal variants for AAA and
   multiple other CMTs. These variants are also the lead SNVs in the
   interrogated regions. P values from original GWAS  studies are
   presented. C KEGG pathway enrichment of the shared genes between AAA
   and CMTs, categorized by biological mechanisms. Only the top 15
   enriched pathways passing hypergeometric test P < 0.05 in each trait
   pair were included. Source data are provided in this paper.

   We observed the shared SNVs, both causal and non-casual, clustered
   proximal to lipid-related genes (Supplementary Data [75]3). For
   example, LPA was the closest gene for 9 SNVs shared by AAA and 11 other
   traits, among which rs10455872^[76]29 was causal to 4 trait pairs, and
   rs140570886^[77]30, rs76735376, and rs6905073 were shared by at least 3
   trait pairs. Similarly, CDKN2B-AS1 was annotated to 8 SNVs shared by 10
   trait pairs, including rs1537371^[78]31 which was causal to 3 trait
   pairs. We also rediscovered rs12740374^[79]32 on CELSR2 and
   rs11591147^[80]33 on PCSK9. Lastly, several shared causal SNVs were
   proximal to CETP, BUD13, TRIB1, LPL, and APOE, all of which encode
   lipid regulators and have been associated with CMDs^[81]34–[82]38.

Shared genes and pathways

   Annotating GWAS variants to genes solely by proximity is oversimplified
   and may not account for pleiotropy. We therefore adopted four
   approaches, TWAS-Fusion^[83]39, SMR^[84]40, MAGMA^[85]41, and
   GCTA-fastBAT^[86]42 to infer shared genes (Supplementary Fig. [87]3).
   Among these methods, the first two leverage expression quantitative
   trait loci (eQTL), and the latter two mainly utilize proximity for gene
   burden tests. We define disease genes as reported by all four methods
   and thus derived 405 genes (Supplementary Data [88]4), of which 109
   genes were linked to minimally three AAA-trait pairs (Supplementary
   Fig. [89]4). Notably, CELSR2, PSRC1, LRP1, and NOC3L were each shared
   among 14 AAA-trait pairs or more. Such broad distribution suggests
   their essential roles in cardiac and metabolic functions.
   Interestingly, all four genes participate in lipid metabolism;
   furthermore, all but NOC3L have been reported in
   inflammation^[90]43,[91]44.

   Pooling genes from any of the four methods for an overview of
   biological pathways, we discovered that their functions were enriched
   in lipoprotein organization, cholesterol transport, and acylglycerol
   homeostasis (Supplementary Fig. [92]5A). Strikingly, cholesterol
   metabolism was the most enriched pathway across all 21 trait pairs
   (Supplementary Fig. [93]5B). When classifying by etiological
   mechanisms^[94]20, the most prominent enrichments appeared in
   cholesterol metabolism, PPAR pathway in lipid metabolism, TGF-β pathway
   in inflammation, and ECM-receptor interaction in extracellular matrix
   dysregulation (Fig. [95]3C).

   Summarizing the shared SNVs and genes, we construct the comorbidity
   network for AAA, detailing the shared variants and genes for each trait
   pair (Fig. [96]4).

Fig. 4. Circular dendrograms presenting shared loci for AAA and CMTs.

   [97]Fig. 4
   [98]Open in a new tab

   The inner circle presents independent variants shared between AAA-trait
   pairs, with 177 shared causal variants marked in asterisks (posterior
   probability of H4 [PP.H4] > 0.7). The outer circle presents the genes
   inferred by Annovar for the shared variants. Genes are highlighted by
   colors to indicate overlap with the four gene identification methods:
   GCTA-fastBAT, MAGMA, TWAS, and SMR, with gray color for those not
   identified by any method, black color for those identified by at least
   one method, and red color for those identified by all four methods.
   Source data are provided in this paper.

Tissue and cell-type specificity

   The shared genes may function in certain tissues and cell types more
   specifically. We examined it from gene expression in GTEx^[99]45 and
   single-cell transcriptome, as well as heritability in tissue-specific
   genes and cell type-specific enhancers in CATLAS^[100]46. Combing both
   approaches, we discovered that liver, artery, and adipose tissue
   (Supplementary Fig. [101]6), and adipocytes, hepatocytes, fibroblasts,
   vascular smooth muscle cells, macrophages, and myeloid cells
   (Supplementary Fig. [102]7) were significantly enriched across many
   AAA-trait pairs, suggesting them as hubs for cardiac and metabolic
   functions (Fig. [103]5). Unique sharing is observed too. For example,
   muscle is only enriched by AAA and atrial fibrillation, the pituitary
   and brain are only enriched by AAA and BMI, and the pancreas is only
   enriched by AAA and HDL-C. While fibroblasts are broadly shared across
   traits, macrophages, and hepatocytes are more specific to AAA and lipid
   traits. Overall, these results align with the genes and pathways,
   highlighting lipid metabolism and immunity over and again.

Fig. 5. Tissue and cell-type specificity inferred from the shared signals
between AAA and CMTs.

   [104]Fig. 5
   [105]Open in a new tab

   A Enriched tissue types by the heritability or expression of the
   tissue-specific genes derived from GTEx. B Enriched cell types by the
   heritability of the cell type-specific enhancers derived from CATLAS,
   or expression of the cell type-specific genes in 11 single-cell
   transcriptome datasets. Source data are provided in this paper.

   We additionally used SMR^[106]40 and TWAS^[107]39 to pinpoint
   gene-tissue effects for each CMT. Collectively, 116 genes were inferred
   for their directions of effect in tissues (Supplementary
   Fig. [108]8A-B). Here we highlight four most broadly shared genes:
   CELSR2, PSRC1, LRP1, and NOC3L. Both methods detected a negative
   relationship between CELSR2 expression in the liver with AAA and five
   other CMTs (Supplementary Fig. [109]8C). Negative relationships were
   found for NOC3L expression in the skeletal muscle, and PSRC1 expression
   in the liver, whole blood, and esophagus mucosa, with AAA and numerous
   other CMTs. Meanwhile, LRP1 expression in the tibial artery was
   suggested for a positive relationship with AAA but a negative
   relationship with CAD.

Drug for AAA with comorbid conditions

   Collectively we identified 405 disease genes shared by AAA and various
   CMTs. As cardiometabolic disorders often coexist, we used these genes
   to identify drugs for treating AAA with comorbidities. As such, we
   utilized a pathway paring score approach developed in our earlier
   study^[110]47 to identify the best matching drugs and disease genes.
   Briefly, we computed the pathological pathways for each trait pair
   based on their shared genes, and the pharmacological pathways for each
   candidate drug based on their affected genes recorded in large
   drug-gene databases, e.g., DrugCentral^[111]48, DGIdb^[112]49, and
   PharmGKB^[113]50. The candidate drugs were mainly derived from
   screening cardiovascular compounds that targeted any of the 405 disease
   genes. We also supplemented the list with those compounds used in
   clinical practice or clinical trials for treating AAA. Collectively, 33
   candidate drugs distributed in 6 functional classes were examined,
   namely antihypertension (11 drugs), lipid-lowering (8 drugs),
   glucose-lowering (3 drugs), antiarrhythmics (1 drug), antithrombosis (4
   drugs), and antioxidant (6 drugs). Most of these drugs have been
   approved to treat various cardiovascular diseases (Supplementary
   Data [114]5).

   The best-matching drugs were defined with pairing scores >= 0.5
   (Fig. [115]6). Close to half drugs, which were distributed in 4
   functional categories, were suggested to treat AAA with hypertension.
   Therein amlopidine has the highest pairing score, followed by several
   antioxidants. Lipid-lowering drugs obtained high pairing scores for
   various trait pairs. Particularly, simvastatin and lovastatin both
   achieved high scores for AAA comorbid with CMDs, such as hypertension,
   MI, subarachnoid hemorrhage, transient ischemic attack, venous
   thromboembolism, or peripheral artery disease. Interestingly, other
   lipid-lowering drugs are suggested for AAA with metabolic traits. For
   example, fenofibrate and gemfibrozil achieved high scores for AAA
   comorbid with LDL-C, nonHDL-C, triglycerides, or total cholesterol.

Fig. 6. Matching between disease pathological pathways, inferred from shared
genes for each trait pair, and drug pharmacological pathways.

   [116]Fig. 6
   [117]Open in a new tab

   Matching scores greater than 0.5 are labeled.

   Notably, several herb-based antioxidants achieved high scores for
   various trait pairs too, including resveratrol, a stilbenoid polyphenol
   naturally enriched in red grapes; tanshinone I, a terpenoid exacted
   from the dry root of Salvia miltiorrhiza (Danshen); and quercetin, a
   flavonol found in many plants. Most of these herb products are in phase
   3 clinical trials (Supplementary Data [118]5) and have shown potential
   in preventing and treating CVDs, including AAA^[119]51–[120]54. Our
   analysis supports their extended application in treating comorbid
   conditions in AAA.

Discussion

   In this study, we discover extensive genetic associations between AAA
   and CMTs from GWAS summary statistics. Further analyses highlight the
   pleiotropic variants and genes, the biological pathways, and the types
   of cells and tissues that are shared by the trait pairs. All these
   findings help to elucidate the common genetic etiology between AAA and
   cardiometabolic disorders.

   We discovered that among all CMDs outside of the aortic aneurysm family
   (i.e., AAA, TAA, and AA), CAD displays a consistently strong
   relationship with AAA. For example, it has the second highest
   genome-wide association (r[g] = 0.34) and is the only trait with mutual
   causality with AAA (OR[AAA->CAD] = 1.10, OR[CAD->AAA] = 1.23).
   Epidemiological studies reported many risk factors common to AAA and
   CAD^[121]55, and the two diseases tend to co-occur^[122]56,[123]57. In
   our study, CAD shares the largest number of SNVs (N = 46), causal SNVs
   (N = 30), and disease genes (N = 50) with AAA. These shared signals
   were enriched for artery and liver tissues, reflecting their common
   malfunctions in artery and lipid metabolism. Artery-related diseases
   including peripheral artery disease (r[g ]= 0.33) and subarachnoid
   hemorrhage (r[g] = 0.32), and cardiac-function-related diseases such as
   MI (r[g] = 0.38) and heart failure (r[g] = 0.30), also displayed top
   genome-wide associations with AAA, although no causal relationship was
   found, suggesting other risk factors may have confounded the
   associations.

   Included in this study are the metabolic traits of lipids, adiposity,
   blood pressure, and glucose. By all levels of our inspection, lipid
   metabolism is most prominently shared. First, lipid traits rank as
   strong causal factors for AAA (OR = 1.46–1.73), next to hypertension,
   an established risk factor for AAA. Second, we observed clustering of
   the shared variants around lipid-related genes, including LPA,
   CDKN2B-AS1 and others. Third, the most broadly shared genes between AAA
   and CMTs, i.e., LRP1, PSRC1, CELSR2, and NOC3L, are all lipid related
   and their up or down regulation was associated with CMTs (Supplementary
   Fig. [124]8). Fourth, cholesterol metabolism appeared as the most
   significantly enriched biological pathway. Fifth, liver, adipose
   tissue, hepatocytes, and adipocytes are most broadly and significantly
   enriched among the AAA-CMT trait pairs. These tissues and cell types
   are important players in lipid metabolism and regulation. Lastly,
   lipid-reducing drugs were strong candidates to treat AAA with various
   comorbid CMTs. These results reinforce the notion that predisposition
   to lipid malfunction is a strong feature in CMTs^[125]58. Our study
   also supports a higher burden of inherited dyslipidaemia in patients of
   AAA, and lowering LDL-C serves as a therapeutic strategy for preventing
   and managing AAA^[126]59–[127]62. In comparison, glucose traits
   demonstrate neither genetic correlation nor causality to AAA. The
   relationship between glucose and AAA has been paradoxical. While
   epidemiological studies have reported an inverse correlation between
   the risk and growth rate of AAA and diabetic traits (e.g., HbA1c
   level^[128]63, fasting glucose level^[129]64, and T2D
   diagnosis^[130]65), no genetic correlation has been reported thus
   far^[131]66, inviting further investigations. Lastly, among the blood
   pressure traits, only diastolic blood pressure displays a mild
   correlation (r[g] = 0.16) and a weak causality (OR[DBP→AAA] = 1.05).

   Several genes appeared repetitively in our analyses. LPA encodes
   lipoprotein(a), which is pro-atherosclerotic, pro-inflammatory,
   pro-thrombotic, and anti-fibrinolytic. Substantial evidence suggest
   that elevated lipoprotein(a) promotes CAD, MI, atherosclerosis, and
   aortic valve stenosis^[132]67,[133]68. CDK2B-AS1 encodes a long
   non-coding RNA that participates in inflammation as well as metabolism
   of lipids and carbohydrates, and has been linked to numerous CMDs and
   immune diseases^[134]69,[135]70. LRP1 encodes LDL receptor-related
   protein and plays diverse roles in lipoprotein metabolism, endocytosis,
   cell growth, cell migration, inflammation, and apoptosis^[136]44.
   Furthermore, CELSR2 and PSRC1, together with SORT1, form a
   PRSC1-CELSR2-SORT1 axis which has been implicated in various
   CVDs^[137]43,[138]71. SORT1 encodes sortilin 1 that functions in lipid
   metabolism and immune responses, such as V-LDL secretion, LDL-C
   metabolism, PCSK9 secretion, inflammation, and formation of foam
   cells^[139]72. Finally, NOC3L is involved in adipocyte differentiation
   and glucose metabolism, and its decreased expression is associated with
   islet dysfunction^[140]73.

   We note that various disease genes in lipid metabolism are involved in
   immune responses too. Indeed, LPA is pro-inflammatory^[141]74;
   CDK2B-AS1 is not only associated with numerous CMDs but also with
   immune diseases, such as idiopathic pulmonary fibrosis and inflammatory
   bowel disease^[142]69,[143]70,[144]75. Interestingly, statins, other
   than lowering lipids, are found to inhibit inflammation in AAA^[145]76.

   Indeed, there are abundant immune signals in our results. For example,
   IL-6 is an important cytokine in CVDs including AAA^[146]77. Enhanced
   IL-6 signaling will over-activate the JAK-STAT pathway, a critical
   pathway that affects many aspects of the mammalian immune
   system^[147]78. rs6734238 was reported to be associated with elevated
   circulating IL-6^[148]79, whereas our analysis inferred this SNV as
   causal to AAA, LDL-C, total cholesterol, and triglycerides
   (Fig. [149]3B). We also identified two SNVs in the intronic regions of
   IL6R, rs4129267, and rs12126142, to be shared by AAA with atrial
   fibrillation and CAD, respectively. Furthermore, our pathway enrichment
   highlights the TGF-β signaling, which was shared by AAA and 12 CMTs
   (Fig. [150]3C). TGF-β regulates the differentiation and function of
   leukocytes and controls the type and scope of immune response^[151]80.
   Numerous studies have uncovered its importance in vascular smooth
   muscle cells (SMCs) and macrophages in the aneurysm
   development^[152]81,[153]82. SMCs can transdifferentiate to foam cells,
   a crucial step in atherosclerosis^[154]83. In our analysis, both
   vascular SMCs and macrophages were enriched by several AAA-trait pairs.
   Indeed, various single-cell RNA-sequencing studies suggested them as
   essential cell types for AAA^[155]84,[156]85. Our analysis reveals this
   close relationship also in genetic predisposition.

   Overall, many of our results recapitulate the relationships of AAA with
   its risk factors and known disease markers, indicating our results
   captured the main components of AAA genetics. To confirm this, we
   analyzed the AAA single-trait GWAS loci. Many of the shared
   lipid-related genes are reproduced, and genes in various pathological
   mechanisms are connected (Supplementary Fig. [157]9–[158]10). The most
   significantly enriched terms are lipid processes and cholesterol
   metabolism. The most enriched tissues are liver and blood vessels, and
   the most enriched cell types are fibroblasts, with a few others showing
   marginal enrichment, including endothelial cells, stromal cells,
   mesenchymal stem cells, macrophages, neutrophils, and monocytes.
   Therefore, the shared signals are the main signals in AAA genetics.

   Finally, the shared disease genes are transformed into treatment
   proposals for treating AAA with comorbid conditions. Most drug
   candidates we discovered have been used to treat CVDs, and some are in
   clinical trials for repurposing to treat AAA, including atorvastatin,
   curcumin, metformin, and five other drugs (Supplementary Data [159]6).
   Note that numerous drugs with high matching scores in our analysis are
   not in clinical tests, inviting future studies for their therapeutic
   potential.

   There are several limitations of this study. First, the GWAS data type
   enables analysis of common SNVs, but omits other variant types such as
   rare variants, the short insertions and deletions, and structural
   variants. Indeed, our previous whole-genome study identified a list of
   rare variants with strong predictability to AAA^[160]18. Second, we
   focus on genetics in this study as CMTs harbor high heritability in
   general, however, other factors such as epigenetics can also play
   important roles. As an example, smoking, as an established risk factor
   of AAA, is known to cause vast epigenetic changes^[161]86. Third, CMTs
   cover a plethora of diseases and physiological traits, and those
   included in our analysis are only representative. Fourth, our inference
   of molecular and cellular mechanisms may be limited by the reference
   knowledgebases and databases. For example, we only deciphered the
   directions of effects for a quarter of the disease genes, due to the
   lack of variant-gene expression models in GTEx. With future
   improvements in the data types and references, we will gain further