The landscape of genetic susceptibility correlations among diseases and traits

Abstract

Objective: The aim of the study was to comprehensively explore the genetic susceptibility correlations among diseases and traits from large-scale individual genotype data.

Materials and Methods: Based on a knowledge base of genetic variants significantly (P < 5 × 10⁻⁸) linked with human phenotypes, genetic risk scores (GRSs) of diseases or traits were calculated for 2504 individuals with whole-genome sequencing data from the 1000 Genomes Project. Associations between diseases/traits were statistically evaluated by pairwise correlation analysis of GRSs. Overlaps between the genetic susceptibility correlations and disease comorbidity associations from hospital claims data in more than 30 million patients in United States were assessed.

Results: Correlation analysis of GRSs revealed 823 significant correlations among 78 diseases and 89 traits (false discovery rate adjusted P-value or Q-value < 0.01). It is noticeable that GRSs were correlated in 464 associations (56.4%) even if they were combinations of distinct sets of risk variants without chromosomal linkage, suggesting the presence of genetic interactions beyond chromosome position. When 312 significant genetic susceptibility correlations between diseases were compared to nationwide disease comorbidity correlations obtained from data from 32 million Medicare claims in the United States, 108 overlaps (34.6%) were found that had both genetic susceptibility and epidemiologic comorbid correlations.

Conclusion: The study suggests that common genetic background exists between diseases and traits with epidemiologic associations. The GRS correlation approach provides a rich source of candidate associations among diseases and traits from the genetic perspective, warranting further epidemiologic studies.

INTRODUCTION

Associations among diseases and traits has been the focus of medical research. These associations are either cause-and-effect relationships or simple co-occurrences. For example, diabetes mellitus causes retinopathy, nephropathy, and diabetic foot diseases, while patients with hypertension are more likely to have diabetes. Knowledge of the associations between diseases is clinically significant, because if a patient is diagnosed with a disease, we can predict what other diseases are likely to occur and prevent or manage them in the early stages.

Several studies have attempted to comprehensively understand the associations between diseases. Hidalgo et al.¹ built the “phenotypic disease network” based on comorbidity correlations between diseases in 30 million patients, and Bulik-Sullivan et al.² modeled the genetic correlations between diseases and traits by linkage-disequilibrium score regression.

Large-scale individual-level genotype data are available since completion of the 1000 Genomes Project, which reconstructed the whole genomes of 2504 individuals from 26 populations.³ Also, over recent decades, thousands of genome-wide association studies (GWASs) have uncovered genetic variants significantly associated with diseases and traits. It is now possible to estimate an individual’s genetic susceptibility to hundreds of diseases and traits.

In the current study, the genetic susceptibility of 2504 subjects from the 1000 Genomes Project were calculated for hundreds of disease and traits, based on the presence of single nucleotide variations known to be linked with human phenotypes by GWASs. The genetic susceptibility associations among diseases and traits were explored to find 823 significant correlations with common genetic backgrounds whose comorbid associations have already been reported or that are candidates for further epidemiologic studies.

METHODS AND MATERIALS

Individual genotype data

Genotype data of 2504 individuals were downloaded from the 1000 Genomes Project website in VCF format (ftp://ftp.1000genomes.ebi.ac.uk/Vol02405/ftp/release/20130502/). The 2504 individuals were of African (n = 661), American (n = 347), East Asian (n = 504), European (n = 503), and South Asian (n = 489) descent.

Genetic variant data associated with phenotypes

Trait- and disease-associated single-nucleotide polymorphisms (SNPs) were downloaded from GWASdb v2.0 (http://jjwanglab.org/gwasdb) on January 9, 2016.⁴ Diseases and traits were identified by their Experimental Factor Ontology (EFO) IDs.⁵ Genetic variants were extracted whose risk alleles were reported and P-values for association with phenotypes were <5 × 10⁻⁸, and only genetic loci on autosomal chromosomes were included. After excluding phenotypes with <5 associated genetic variants and further curation of data, 167 diseases and traits and 4486 associated genetic variants from 638 publications were included for analysis (Supplementary Table S1).

Calculation of genetic risk score

For each disease or trait, the total number of risk alleles was calculated for each individual, with positive or negative weights in case of quantitative traits, based on the direction of risk that each allele confers.⁶ Then the counts were normalized to z scores, with mean 0 and standard deviation 1 within the population each individual belonged to (Supplementary Table S2). The z score is defined as the genetic risk score (GRS).

Network construction

Pearson’s correlation coefficients of GRS between disease/trait pairs were calculated, and correlating P-values were calculated. P-values were adjusted for multiple hypothesis testing and transformed to q-values by controlling false discovery rates⁷ with the “qvalue” package of R. Networks were drawn with Cytoscape software (version 3.2.1; http://www.cytoscape.org/).

Linkage disequilibrium analysis

Information on linkage disequilibrium for pairs of genetic variants was retrieved from the SNAP database (https://www.broadinstitute.org/mpg/snap/index.php) in default setting (SNP dataset, 1000 Genomes Pilot 1; population panel, CEU; r² threshold, 0.8; distance limit, 500 base pairs).

Disease comorbidity data analysis

The disease comorbidity correlations were reported by Hilago et al. by analyzing data from 32 million Medicare claims pertaining to 13 million white individuals over 65 years old from the United States. Full datasets about disease correlations with 3-digit International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes were downloaded from the study website (http://hudine.neu.edu/). The dataset provides the relative risks and 99% confidence intervals between disease pairs, based on the prevalence and comorbid correlations as reported in the Medicare claims data. All 78 diseases in the present study were assigned 3-digit ICD-9-CM codes, and 312 significant genetic susceptibility correlations between diseases were mapped to the epidemiologic comorbidity correlations.

RESULTS

After adjusting for multiple hypothesis testing, 823 significant correlations among 167 diseases and traits were discovered at a q-value < 0.01 (Table 1; Supplementary Table S3). Five associations had absolute correlation coefficients >0.5: obesity vs body mass index (r = 0.70), atrial fibrillation vs ischemic stroke (r = 0.59), body weight vs body mass index (r = 0.58), coronary heart disease vs early-onset myocardial infarction (r = 0.57), and low-density lipoprotein (LDL) cholesterol vs total cholesterol (r = 0.54). Atrial fibrillation is a well-known risk factor for stroke, and anticoagulation is indicated for patients with atrial fibrillation to prevent stroke. But the strong positive correlation between genetic risk for atrial fibrillation and stroke suggests that they may have a common genetic background (Figure 1 A). Atrial fibrillation and ischemic stroke share a genetic risk variant, rs2200733 on chromosome 4. And rs17042171 and rs6817105, associated with atrial fibrillation, are in linkage disequilibrium with rs12646447 on chromosome 4 that predicts ischemic stroke (Supplementary Table S3). Two studies have reported that genetic risk variants on PITX2 (chromosome 4) and ZFHX3 (chromosome 16), which were initially associated with atrial fibrillation, are independent risk factors for ischemic stroke.^8,9

Table 1.

Representative polygenic susceptibility correlations between diseases/traits (q-value < 0.01)

Disease or trait A (EFO ID)	Disease or trait B (EFO ID)	Corr. coeff. (r)	Common SNPs	Common SNP (No.)	SNPs in Disease/trait A (No.)	SNPs in Disease/trait B (No.)	Share common SNPs or in LD
Atrial fibrillation (0000275)	Stroke, ischemic (0000712)	0.59	rs2200733	1	17	5	Yes
Malaria (0001068)	Venous thromboembolism (0004286)	0.40	rs505922, rs8176719	2	20	13	Yes
Type 1 diabetes (0001359)	Vitiligo (0004208)	0.31		0	26	29	Yes
Glaucoma (open angle) (0004190)	Coronary heart disease (0000537)	0.29	rs7865618	1	7	50	Yes
Waist circumference (0004342)	Menarche (age at onset) (0004703)	0.29		0	8	37	Yes
Self-reported allergy (0003785)	Helicobacter pylori serologic status (0005247)	−0.28	rs10004195, rs10024216, rs11096956a	41	450	60	Yes
Lung cancer (0001071)	Smoking behavior (0004318)	0.27	rs1051730	1	13	8	Yes
Multiple myeloma (0001378)	Telomere length (0004505)	0.27	rs1317082, rs10936599	2	42	8	Yes
Type 2 diabetes (0001360)	Low birth weight (0004344)	0.26	rs6931514	1	95	7	Yes
Coronary heart disease (0000537)	Intracranial aneurysm (0003870)	0.26	rs12413409	1	50	9	Yes
Asthma (0000270)	White blood cell count (0004308)	0.20		0	30	7	No
Rheumatoid arthritis (0000685)	Graves’ disease (0004237)	0.19	rs6457617	1	93	19	Yes
Colorectal cancer (0000365)	Fasting insulin (0004469)	0.18	rs174550	1	36	16	Yes
Age-related macular degeneration (0001365)	Circulating myeloperoxidase levels (serum) (0005243)	0.18		0	29	9	No
Inflammatory bowel disease (0003767)	Interstitial lung disease (0004244)	0.18		0	114	87	No
Interstitial lung disease (0004244)	Telomere length (0004505)	0.16	rs1317082, rs10936599, rs12696304	3	87	8	Yes
Psoriasis (0000676)	Ankylosing spondylitis (0003898)	0.15		0	26	13	No
Rheumatoid arthritis (0000685)	IgA nephropathy (0004194)	0.15		0	93	7	No
Venous thromboembolism (0004286)	Thyroid-stimulating hormone measurement (0004748)	0.14		0	13	23	Yes
Breast cancer (0000305)	Mean platelet volume (0004584)	0.14		0	73	44	No
Obesity (0001073)	Inflammatory bowel disease (0003767)	0.13		0	82	114	Yes
Blood pressure (0004325)	Magnesium levels (0004845)	−0.13		0	62	6	Yes
Schizophrenia (0000692)	Inflammatory bowel disease (0003767)	0.12		0	40	114	No
Waist-hip ratio (0004343)	Renal function-related traits (blood urea nitrogen) (0004741)	0.12	rs2074356	1	18	16	Yes
Bladder cancer (0000292)	Telomere length (0004505)	0.11	rs10936599	1	13	8	Yes
Obesity (0001073)	Menarche (age at onset) (0004703)	−0.11	rs633715	1	82	37	Yes
Ulcerative colitis (0000729)	Type 2 diabetes (0001360)	0.10		0	85	95	No

LD: linkage disequilibrium; ^asee Supplementary Table S3 for full list.

Figure 1.

Scatter plots of diseases/traits. (A) Atrial fibrillation vs ischemic stroke (r = 0.59); (B) venous thromboembolism vs susceptibility for severe malaria (r = 0.40); (C) type 1 diabetes vs vitiligo (r = 0.31); (D) coronary heart disease vs open-angle glaucoma (r = 0.29); (E) type 2 diabetes vs low birth weight (r = 0.26); (F) Helicobacter pylori infection vs self-reported allergy (r = −0.28).

Figure 2 is the network representation of 341 associations among 139 diseases and traits with q-value < 0.01 and absolute correlation coefficient >0.10. Diseases and traits form several clusters with distinct pathogenesis, such as metabolic disease, autoimmune disease, and cardiovascular, allergy, and coagulation disorders.

Figure 2.

Network representation of 341 associations among 139 diseases/traits with q-value < 0.01 and absolute correlation coefficient >0.10. Diseases/traits form several clusters with distinct pathogenesis, such as metabolic disease, immune system disease, and cardiovascular and coagulation disorders. Node colors represent upper-level disease categories in the Experimental Factor Ontology hierarchy (red, cardiovascular disease; yellow, cancer; green, metabolic disease; purple, immune system disease; pink, respiratory system disease; dark blue, infectious disease; gray, nervous system disease; light blue, traits). Blue edges are positive correlation coefficients and red edges are negative correlation coefficients. The thickness of edges is proportional to the absolute values of correlation coefficients. Edges with solid lines represent that the 2 diseases/traits have common risk variants or the genetic risk variants for each disease/trait are in linkage disequilibrium, while nodes connected with dotted lines do not have any genetic risk variants in common or in linkage disequilibrium. See online version of this figure for references to color.

Venous thromboembolism, which is the most representative disease in coagulation disorders, is positively associated with a susceptibility to severe malaria (r = 0.40, Figure 1B), end-stage coagulation (factors VIII, XIII, and von Willebrand factor levels, r = 0.43), thyroid-stimulating hormone measurement (r = 0.14), coronary heart disease (r = 0.15), and total cholesterol (r = 0.14). They are interconnected by variants associated with the ABO blood group (rs505922 and rs8176719, etc., Supplementary Table S3), and the non-O blood group has been reported to be associated with venous thromboembolism, susceptibility to severe malaria, and higher levels of factor VIII.^10,11 Autoimmune diseases constitute the largest part of the network (Figure 2). Examples include type 1 diabetes vs vitiligo (r = 0.31, Figure 1C),¹² Crohn’s disease vs ulcerative colitis (r = 0.23), rheumatoid arthritis vs systemic sclerosis (r = 0.20), and rheumatoid arthritis vs Graves’ disease (r = 0.19), whose epidemiologic associations have also been reported. Among cardiovascular diseases, coronary heart disease is positively correlated with early-onset myocardial infarction (r = 0.57), open-angle glaucoma (r = 0.29, Figure 1D),¹³ intracranial aneurysm (r = 0.26), lipoprotein-associated phospholipase A2 activity (r = 0.16), and LDL cholesterol (r = 0.13). Metabolic diseases, such as metabolic syndrome and type 2 diabetes, are closely linked with metabolic traits such as fasting blood glucose, insulin, glycated hemoglobin, high-density lipoprotein (HDL) cholesterol, and triglycerides (Figure 2). It is interesting that type 2 diabetes is associated with low birth weight (r = 0.26, Figure 1E) and rs6931514 variant, associated with CDKAL1, is shared by type 2 diabetes and low birth weight, suggesting that the 2 phenotypes have a common genetic background.¹⁴ In allergic disease, self-reported allergy (r = −0.28, Figure 1F) and allergic sensitization (r = −0.23) are negatively associated with Helicobacter pylori infection. This suggests that the previously suggested epidemiological association¹⁵ between allergy and H. pylori infection may be determined by the same genetic background.

With regard to cancers, colorectal cancer and breast cancer are linked with metabolic diseases, while viral hepatitis related to hepatocellular cancer and Hodgkin’s lymphoma is connected to autoimmune diseases, suggesting different pathogenic mechanisms according to the type of cancer (Figure 2).^16–18

Out of 823 significant correlations (q-value < 0.01), 464 genetic correlations (56.4%) between diseases and traits had no common risk variants or the constituting risk variants were in linkage disequilibrium. Even though the GRSs are combinations of disjoint sets of risk variants without chromosomal linkage, GRSs correlate, suggesting that genetic interactions exist beyond chromosome position. Supplementary Figure S1 shows 123 correlations with at least 0.1 correlation coefficient, with no risk variants in common or in linkage disequilibrium between traits or diseases. They are mainly autoimmune diseases, and it is possible that genetic interactions beyond chromosomal positions, such as epistasis, may underlie the genetic associations between immune system diseases.^19,20

Next to be tested was whether the presence of one disease increases the likelihood of the other when the GRSs of the disease pairs correlate. Disease comorbidity correlations from Medicare claims data from the United States were originally reported by Hidalgo et al.¹ Out of 312 significant polygenic susceptibility correlations between diseases with ICD-9-CM codes, 108 overlaps (34.6%) were found that had both genetic susceptibility and epidemiologic comorbid correlations (Supplementary Table S4). Examples included coronary heart disease vs intracranial aneurysm (r = 0.26 and relative risk, 1.57 [99% CI, 1.56–1.58]), asthma vs obesity (r = 0.07 and relative risk, 2.18 [99% CI, 2.14–2.22]), and asthma vs rheumatoid arthritis (r = 0.10 and relative risk, 1.36 [99% CI, 1.32–1.39]). Previous epidemiologic studies have reported an increased incidence of unruptured intracranial aneurysm among female patients with coronary heart disease²¹ and increased risk for asthma in patients with rheumatoid arthritis.²² Further epidemiologic studies demonstrating the comorbidity relationship between them and genomic studies investigating their common genetic background will enable more precise risk stratification and early screening strategies.

DISCUSSION

In summary, 823 genetic correlations across human diseases and traits were uncovered by investigating the correlation of GRSs in 2504 individuals based on individual genotype data and the database of genetic variants from GWASs. The associations between diseases and traits with epidemiological evidence may be explained by correlating genetic susceptibility, as validated by comparison to Medicare claims data. Also, half of the correlations of genetic susceptibility were not directed by overlapping or adjacent variants, especially among immune system diseases. The associations between diseases without shared genetic variants were novel findings from the current approach based on individual genotype data, which could not be addressed in previous studies based on the literature or chromosomal linkage.^2,23

Genetic variants for risk factors are used as instrumental variables to infer causal relationships with diseases or health conditions in so-called Mendelian randomization.²⁴ From that perspective, phenotypes with common genetic backgrounds are likely to have cause-and-effect relationships. For example, GRS for coronary heart disease is significantly correlated with GRS for LDL cholesterol (r = 0.13, q = 5.9 × 10⁻⁹) but not with C-reactive protein (r = −0.03, q = 0.35) or HDL cholesterol (r = −0.03, q = 0.35). In Mendelian randomization studies, LDL cholesterol has been confirmed as a true and causative risk factor for coronary heart disease, but C-reactive protein and HDL cholesterol have been suggested as mere biomarkers.²⁵ Diseases and traits with a common genetic background in the current study may be causally related and candidates for further study within a Mendelian randomization design.

The polygenic risk correlation approach provides a rich source of candidate associations between diseases and warrants future epidemiologic studies. Li et al.²³ reported 120 relationships between risk factors and diseases based on the presence of shared gene variants and validated by exploring electronic medical records in 5 cases. In the current study, 108 out of 312 correlations (34.6%) between diseases were found to co-occur significantly by Medicare claims data.

The limitation of the study is that the allele counting method to calculate genetic risk assumes that the contribution of each genetic variant is equal and does not consider the prior probability of having a variant. It is an arbitrary choice and might not be accurate, as it has not been documented as accurate or predictive. Also, the specific ethnicity in which each genetic variant was reported in GWASs was not taken into account. And the study could not incorporate phenotypic data on individuals because they are not available. Incorporating clinical and environmental data and the prior probability of having a disease into the polygenetic modeling framework will enable more accurate calculation of associations between diseases/traits.

CONCLUSIONS

This is the first study to comprehensively explore the associations between diseases and traits by polygenic risk score correlations based on large-scale genotype data. It is noticeable that half of the genetic risk score correlations between diseases are not directed by overlapping or adjacent variants, suggesting the presence of genetic interactions beyond chromosomal linkage. The polygenic risk correlation approach provides a comprehensive list of candidate associations between diseases or traits, and Medicare claims data analysis confirms the comorbidity associations between them. The clinical implication of the study is that our knowledge about disease associations from a genetic and epidemiologic perspective will enable early screening or intervention in patients diagnosed with specific diseases.

SIGNIFICANCE

The polygenic risk correlation approach provides a comprehensive list of candidate associations between diseases or traits, and Medicare claims data analysis confirms the comorbidity associations between them.

AVAILABILITY OF DATA AND MATERIALS

The datasets used and/or analyzed during the current study are available from the corresponding author on request.

Competing interest

The author declares that he has no competing interests.

Author contributions

JHO designed the study, performed the analyses, and wrote the manuscript.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Journal of the American Medical Informatics Association online.