Genetic analyses identify evidence for a causal relationship between Ewing sarcoma and hernias

Summary

Knowledge of Ewing sarcoma (EWS) risk factors is exceedingly limited; however, multiple small, independent studies have suggested a possible connection between hernia and EWS. By leveraging hernia summary statistics from the UK Biobank and a recently published genome-wide association study of EWS (733 EWS cases and 1,346 controls), we conducted a genetic investigation of the relationship of 5 hernia types (diaphragmatic, inguinal, umbilical, femoral, and ventral) and EWS. We discovered a positive causal relationship between inguinal hernia and EWS (OR 1.27, 95% confidence interval [CI] 1.01–1.59, and p = 0.041) through Mendelian randomization analysis. Further analyses suggested shared pathways through three genes: HMGA2, LOX, and FBXW7. Diaphragmatic hernia showed a stronger causal relationship with EWS among all of the hernia types (OR 2.26, 95% CI 1.30–3.95, p = 0.004), but no statistically significant local correlation pattern was observed. No evidence of a causal or genetic relationship was observed between EWS and the other three hernia types, including umbilical hernia, despite a previous report indicating an OR as high as 3.3. The finding of our genetic analysis provided additional support to the hypothesis that EWS and hernias may share a common origin.

We studied the genetic link between five hernia types and Ewing sarcoma (EWS), using two large, genome-wide association datasets of European ancestry. Findings indicate causal relationships between EWS and diaphragmatic and inguinal hernias, but not other hernia types. Our work provides additional support to the long-standing association between hernias and EWS observed in the epidemiologic literature and suggests potential shared pathways.

Introduction

Ewing sarcoma (EWS) is a rare yet severe malignant bone disease in children, with several epidemiological case-control studies suggesting its positive association with hernia.^1,2,3 Dating back to 1992, the relationship between umbilical and inguinal hernia with EWS was observed in an EWS therapeutic study.⁴ The relationship between inguinal hernia and EWS was later confirmed by a study examining medical records from 306 patients seen at the National Institutes of Health between 1960 and 1992.² Furthermore, a pooled analysis of two additional studies^5,6 totaling 138 cases and 574 controls showed a statistically significant association between EWS and umbilical hernia and a suggestive relationship between EWS and inguinal hernia.¹ However, a recent population-based study using data from 122 cases and 4,038 controls in the Danish Cancer Registry found no relationship between hernia and EWS.³ Hernias are heterogeneous defects, but all are characterized by the intrusion of a tissue or organ into an adjacent, inappropriate space; many are also considered developmental defects. The disruption of normal embryological tissue development during pregnancy could explain the common origin of EWS and hernias; hence, studying the associations with hernias could expand the understanding of the etiology of EWS.

The genetic epidemiology of both EWS and hernia has grown since 2005. There is a surprising genetic architecture of EWS in which common variants with moderate to strong effect sizes have been identified through genome-wide association studies (GWASs).^7,8 Meanwhile, a large, recent GWAS in the UK Biobank identified more than 100 variants associated with 5 types of hernia.⁹ The types of hernia examined (inguinal, diaphragmatic, femoral, umbilical, and ventral hernia) demonstrated varying degrees of shared genetic underpinnings. Building upon these observations, we evaluated the genetic and causal relationships between EWS and each of the five hernia types among individuals of European ancestry using genetic epidemiology approaches.

Material and methods

Data sources

EWS GWAS data

We used summary statistics data from the largest GWAS study of EWS. It consisted of 733 EWS cases and 1,346 cancer-free controls.⁷ The results were meta-analyzed from three study sources (Institut Curie, the National Cancer Institute, and the Childhood Cancer Survivor Study) on cases and controls with >80% genetically estimated European ancestry. All of the EWS cases were confirmed by medical record review and controls were cancer-free adults who were ancestry matched with cases. Genotypes were imputed using the 1000 Genomes Phase 3 release as a reference panel, and SNPs with an imputation information score >0.3, minor allele frequency (MAF) >0.01 present in all three study sources, and p values for the Cochrane Q heterogeneity test >0.01 were included for the following analysis.

Hernia GWAS data

Fadista et al. conducted a GWAS of the five hernia types in a case-control design in which cases were determined by International Classification of Diseases, 10^th edition codes or interview data and controls were individuals without any form of hernias in the UK Biobank.⁹ There were 28,707 inguinal, 31,193 diaphragmatic, 6,402 umbilical, 1,049 femoral, and 4,644 ventral hernia cases, and 65,492 controls of European ancestry. SAIGE (version 0.29.4) was used to perform the association test, which adjusted for case-control imbalance and relatedness.¹⁰ Summary statistics for each hernia type were downloaded from the Danmarks Nationale Biobank website. We excluded all of the nonbiallelic SNPs and SNPs with MAF <0.01 from the analysis.

The 1000 Genomes project (1000G)

The 1000G provides a comprehensive description of common human genetic variation.¹¹ It was used as our reference panel to estimate the linkage disequilibrium (LD) pattern and LD blocks. We used the European samples from the 1000G to match the EWS and hernia GWAS. We restricted all analyses to SNPs shared across EWS GWAS, hernia GWAS, and the 1000G project. SNPs on the sex chromosomes and SNPs with more than 0.2 in the absolute difference of MAF between EWS and 1000G were excluded.

Statistical analyses

Global genetic correlation

We used LD score (LDSC) regression to estimate the global genetic correlation between the hernia types and EWS trait.¹² As recommended by Bulik-Sullivan et al.,¹³ we retained the SNPs to the HapMap 3 reference panel. Because 15 pairs of global genetic correlation were examined, we set the threshold for statistical significance to 0.05/15 = 0.003 based on a Bonferroni adjustment correcting for multiple testing.

Mendelian randomization (MR) analyses

We performed MR analyses to investigate the causal relationship between the five different types of hernia and EWS. We selected instrument variables by clumping genome-wide significant SNPs in the hernia datasets such that SNPs with R² > 0.01 were removed. Palindromic variants with effect allele frequency between 0.42 and 0.58 and multiallelic variants were excluded before clumping. We evaluated the quality of the instruments by using (1) a directional test for each instrument—a SNP with a small p value of the Steiger filtering test suggests a causal direction from hernia to EWS¹⁴—and (2) the heterogeneity by the Cochran Q test. We excluded instrument variables with p values of the Steiger filtering test <0.05 after Bonferroni correction. Our main analysis was based on a random effects version of the inverse-variance weighted (IVW) method, which provides an unbiased estimate when none or horizontal pleiotropy is present. We examined whether there was any evidence of directional pleiotropy by the MR-Egger intercept test. MR-Egger, weighted mean, simple mode, and weighted-mode MR were performed as sensitivity analyses.

Polygenic risk score (PRS)

We used the multitrait analysis of GWAS (MTAG) algorithm that leveraged the genetic correlation among different hernias to improve the estimation of the GWAS estimates for each hernia type.¹⁵ MTAG has a lower genome-wide mean-square error than the original GWAS estimates, which is helpful for constructing PRSs. Since femoral hernia could lead to unstable MTAG results because its mean ${\chi }^2$ is <1.02, we left it out of the analysis and performed MTAG on the four remaining hernias. We constructed the PRS based on the summary statistics derived from MTAG for all hernias besides femoral hernia and from the original GWAS summary statistics for femoral hernia. We used the SNPs that reached the genome-wide significance threshold at 5 $\times$ 10⁻⁸ as the main analysis and those with p values less than 5 $\times$ 10⁻⁵ in a sensitivity analysis. SNPs were clumped to have R² < 0.01 based on the 1000G reference panel. The weighted PRS was constructed using the plink --score sum function in PLINK 1.9 and then deployed to evaluate its relationship with EWS in each GWAS study of EWS via logistic regression. Analyses by each study were adjusted by sex and genetic ancestry captured by the top three principal components. Study-specific results were then combined by METAL¹⁶ to produce meta-analyzed results for each hernia type. Heterogeneity by study (Q value, I² and H² statistics) was evaluated before the meta-analysis.

Colocalization and local genetic correlation

We used the COLOC¹⁷ algorithm on the regions 10,000 bp upstream and downstream of the SNPs (i.e., rs11366169, rs7742053, rs10822056, rs2412476, rs6047482, and rs6106336) reported to be associated with EWS⁷ and the genome-wide significant lead SNPs in the hernia study based on the MTAG estimates (MTAG p <5 $\times$ 10⁻⁸). We used the original summary statistics of EWS GWAS and hernia in the COLOC test. Approximate Bayes factor computations were used to estimate the posterior probabilities of whether a single SNP affects both EWS and each of the five hernia types in the region. In addition, we quantified the local genetic similarity between each pair of hernia type and EWS trait on 2,353 approximately independent LD blocks.¹⁸ We used a SUPERGNOVA algorithm for estimation¹⁹ based on the original summary statistics. The statistical significance threshold for testing the local genetic covariance was set at 0.05/2,353 = 2.12 $\times$ 10⁻⁵ after Bonferroni correction.

Results

Positive genetic correlation between EWS and hernia

The five hernia types were significantly correlated with one another on a global level, except that the pair between femoral and diaphragmatic hernia and the pair between femoral and umbilical hernia were only marginally significant (Figure 1). Although the sample size for ventral hernia is relatively small, we observed that its genetic correlation was strong with multiple hernias. We noted that some correlation estimates were quite different from what was previously reported—for example, the reported genetic correlation between inguinal hernia and ventral hernia was 0.16;²⁰ however, it was estimated to be 0.53 in our analyses. The difference could be driven by the larger number of cases in our analyses. None of the hernia types had a statistically significant genetic correlation with EWS, possibly due to a lack of statistical power (Table S1). However, all of the estimates suggested a positive global relationship, among which ventral hernia had the strongest estimated genetic correlation with EWS (genetic correlation = 0.26), diaphragmatic hernia had the second strongest (genetic correlation = 0.17), and inguinal hernia had the weakest (genetic correlation = 0.02).

Figure 1.

Genetic correlation among five hernia types and EWS

∗∗ indicates statistical significance after Bonferroni correction; ∗ indicates marginal significance with p < 0.05.

MR analyses suggest a causal relationship between diaphragmatic hernia and EWS

To examine the validity of the instrumental variables for the MR analysis, we tested assumptions on effect direction, heterogeneity due to balanced horizontal pleiotropy, and unbalanced pleiotropy. For all five hernias, there was no indication of a violation of assumptions for the random effect IVW method. In addition, all of the chosen IVs, being genome-wide significant, inherently satisfied F-statistics >10. This indicated sufficient instrument strength, ensuring that the IVW method was not vulnerable to the weak instrument.²¹ Because no violation of the assumptions for IVW method was identified, we reported the MR results of the IVW method below and summarized them in Table 1.

Table 1.

MR results for IVW method between each hernia type and EWS

Hernia type	nIV	F-statistics	R2 (exposure)	R2 (outcome)	Cochran’s Q	Egger-intercept p	OR	95% CI	p
Inguinal	70	60.2	0.045	0.033	0.64	0.91	1.27	1.01–1.59	0.041
Diaphragmatic	24	39.2	0.010	0.009	0.98	0.33	2.27	1.30–3.95	0.004
Umbilical	10	52.6	0.007	0.002	0.86	0.63	1.07	0.78–1.48	0.66
Femoral	3	43.3	0.002	0.001	0.35	0.74	0.95	0.73–1.24	0.69
Ventral	4	42.3	0.002	0.001	0.59	0.36	0.96	0.59–1.55	0.86

CI, confidence interval; EWS, Ewing sarcoma; IVW, inverse-variance weighted; MR, Mendelian randomization; nIV, number of instrumental variables; OR, odds ratio.

As shown in Table 1, the risk of EWS was 2.26 times higher among those with diaphragmatic hernia compared to those without (odds ratio [OR] = 2.26, 95% confidence interval [CI] = 1.30–3.95, and p = 0.004). In addition, we observed a marginally significant causal relationship between inguinal hernia and EWS (OR = 1.27, 95% CI = 1.01–1.59, and p = 0.041). One instrument rs12810758 from the Steiger test had a p = 0.01. Because it did not reach the significance threshold after multiple testing corrections, we included this SNP in the main analyses. However, we performed an additional sensitivity analysis excluding this SNP, resulting in a slight change in the OR estimate and the p values: OR = 1.22, 95% CI = 0.97–1.54, and p = 0.089. The MR analysis on single SNPs suggests that the significance was mainly driven by two instruments: rs12810758 and rs10519694. Forest and scatterplots of diaphragmatic and inguinal hernia are present in Figures 2 and 3, respectively. None of the umbilical, femoral, and ventral hernias showed a statistically significant association with EWS (Tables 1 and S2). We observed no significant association between hernias and EWS across various MR tests, including MR-Egger, weighted mean, simple mode, and weighted mode (Figures 2 and 3; Table S2). Given that the IVW method typically is the most powerful test when its model assumptions are satisfied, it was anticipated that the result of the IVW method demonstrated smaller p values than others for inguinal and diaphragmatic hernias.²²

Figure 2.

MR plots for diaphragmatic hernia

(A) Scatterplot for different MR methods, (B) Forest plot for relationship between diaphragmatic hernia and EWS for each single instrumental variable. Each black point and error bars in A/B represent point estimate of a single instrumental variable and standard error/95% confidence interval, respectively. Slope of lines in (A) corresponds to the estimated causal effect from different MR methods. Red points in (B) represents the combined causal estimate using all SNPs together using MR-Egger and IVW methods.

Figure 3.

MR plots for inguinal hernia

(A) Scatterplot for different MR methods, (B) Forest plot for relationship between diaphragmatic hernia and EWS for each single instrumental variable. Each black point and error bars in A/B represent point estimate of a single instrumental variable and standard error/95% confidence interval, respectively. Slope of lines in (A) corresponds to the estimated causal effect from different MR methods. Red points in (B) represents the combined causal estimate using all SNPs together using MR-Egger and IVW methods.

Evaluation of MTAG results and PRS

As shown in our study and some others,²⁰ the hernia traits were correlated at varying degrees; thus, PRS constructed from MTAG results is expected to perform better than those from the single-trait GWAS analysis. Figure S1 presents the Manhattan plots of the GWAS for four hernia types before and after leveraging correlation, indicating that more GWAS signals are present after leveraging. The number of lead SNPs, defined by clumping with an R² threshold of 0.01, is reported in Table S3 and the resulting GWAS-equivalent sample sizes in the expected ${\chi }^2$ are present in Table S4. As expected, the correlation between the MTAG Z score and the original Z score was high for inguinal and diaphragmatic hernia because they had the largest sample sizes. However, umbilical and ventral hernias demonstrated improvements not only in their effective sample size but also in the number of lead SNPs, which may be due to their high level of correlation with other hernias. Based on the MTAG results, we had 81, 34, 20, 3, and 42 SNPs for PRS based on a p value threshold of 5 $\times$ 10⁻⁸; 335, 214, 157, 57, and 191 SNPs based on a p value threshold of 5 $\times$ 10⁻⁵ for inguinal, diaphragmatic, umbilical, femoral, and ventral hernias, respectively. We observed a consistent positive relationship between inguinal/diaphragmatic hernia and EWS across different p value thresholds in the meta-analysis (Table 2), although statistical significance was not achieved. No indication of study heterogeneity was observed in the meta-analysis (not presented), and the study-specific effects are present in Tables S5 and S6.

Table 2.

Meta-analyzed results for each hernia PRS and EWS across three study sources

Hernia	p thresholds, OR (95% CI)
	5 × 10−8	5 × 10−5
Inguinal	1.14 (0.65–2.00)	1.16 (0.75–1.79)
Diaphragmatic	1.05 (0.40–2.80)	1.35 (0.67–2.29)
Umbilical	1.05 (0.15–7.59)	0.90 (0.36–2.27)
Femoral	0.94 (0.73–1.20)	0.98 (0.94–1.09)
Ventral	0.96 (0.20–2.96)	1.30 (0.48– 3.54)

PRS, polygenic risk score.

Two loci of inguinal hernia colocalize with EWS

Among the 6 SNPs associated with EWS and 120 SNPs associated with any of the hernia types in MTAG, we identified 1 region that surrounded rs12810758 (chromosome 12: 66318027–66338027, GRCh37) with a posterior probability of sharing the same SNP between EWS and inguinal hernia as high as 0.80. A total of 23 SNPs were included in the analysis and located within the DNA helicase G (HMGA2) gene on chromosome 12. An additional sensitivity analysis limited to 0.5 kb around rs12810758 showed a posterior probability of 0.81 with 3 SNPs. The top SNP rs12810758 (ref/alt:C/T) was positively associated with inguinal hernia (p = 8.9 $\times$ 10⁻¹¹, MTAG p = 4.5 $\times$ 10⁻¹¹) and positively associated with EWS (p = 0.00089). Another region surrounding rs7692200 (chromosome 4: 153110982–153130982) had weaker evidence of being linked to EWS (i.e., the posterior probability of sharing the same SNP between EWS and the inguinal hernia was as high as 0.57). This region, covering 16 SNPs, is located closest to the F box protein gene FBXW7,²³ a critical tumor suppressor of human cancer. The SNP rs7692200 (ref/alt:C/T) was positively associated with inguinal hernia (p = 9.3 $\times$ 10⁻⁸, MTAG p = 3.6 $\times$ 10⁻⁸) and positively associated with EWS (p = 0.003). Although no LD blocks achieved statistical significance in the local genetic correlation analysis for inguinal hernia, we observed a high positive local correlation among the two LD blocks covering the two top loci. The LD block located at chromosome 12: 66318027–66338027 had a correlation of 0.65 (p = 0.19), whereas the LD block located at chromosome 4: 153110982–15313092 had a correlation of 0.72 (p = 0.47).

Discussion

The EWS family of tumors is a group of rare, small, round, blue cell cancers that appear in the bone or soft tissue and are most common in adolescence. Their etiology is obscure, but a few clues have emerged from limited investigations. In particular, a 2005 meta-analysis examined hernia and EWS in five published studies, which together suggested that EWS patients had a history of hernia three times as often as controls.¹ However, the existence of a causal relationship between EWS and hernias and the nature of their connection remain unclear.

In this work, we present a genetic investigation of five hernia types (diaphragmatic, inguinal, umbilical, femoral, and ventral) with EWS using two independent datasets, UK Biobank data (>60,000 individuals) and EWS GWAS data (733 EWS cases and 1,346 controls). We demonstrated, through genetic epidemiology methods on the two independent GWAS, evidence of causal positive relationships between EWS and two hernia types, diaphragmatic and inguinal hernias, and no evidence of the relationship between EWS and the other three hernia types among European populations. Interestingly, we observed a relatively strong relationship between diaphragmatic hernia and EWS in terms of OR and global genetic correlation, but no clear evidence of local genetic sharing through the colocalization test; we speculate that the shared genetic effects between diaphragmatic hernia and EWS are locally mild but accumulatively strong. Alternatively, the colocalization test could suffer from substantial statistical power loss when its assumptions (i.e., causal variant must be included and at most one association is present for EWS and hernia in the set) are violated.^17,24 Therefore, we could not rule out the possibility of the existence of local correlation, even though none was identified in our analysis. However, we found two strong regional genetic loci indicating sharing pathways between inguinal hernia and EWS.

MR analysis suggested a moderate relationship between inguinal hernia and EWS (OR = 1.27). Although the global genetic correlation was low, additional analyses suggested the sharing of three genetic loci (Table 3). More specifically, the colocalization test of the genetic loci around SNP rs12810758 suggested that a potential common biological pathway could exist between inguinal hernia and EWS through the HMGA2 gene. HMGA2 expression was detected in many embryonic and fetal tissues but was downregulated during postnatal life except for many benign and malignant tumors.²⁵ Earlier work showed that EWS-FLI-1 fusion protein, characterized by the chromosomal translocation in 85% of ESW cases, can alter microRNA expression, which regulated the tumor growth, and the regulation was mediated by HMGA2.²⁶ HMGA2 was reported to be the closest gene to an identified EWSR1-ETS binding site.²⁷ In addition to SNP rs12810758, SNP rs10519694, which drove the significance of MR analysis, is an expression quantitative locus for the gene lysyl oxidase (LOX).²⁸ LOX was shown to be downregulated by the EWS/FLI1 oncoprotein and to display tumor suppressor activities in EWS cells.²⁹ Lastly, the colocalization test suggested mild evidence of the involvement of FBXW7 in the shared pathway of inguinal hernia and EWS. The SNP rs7692200, the top signal in the LD region closest to FBXW7, was not included as an instrumental variable in the MR analysis because it was not statistically significant in the original GWAS study of inguinal hernia but was significant after the multitrait leveraging. FBXW7 was found to mediate the tumorigenesis of EWS.³⁰ The three germline SNPs may be associated with EWS but were overlooked in previous GWAS of EWS because none of them were genome-wide significant or in LD with the significant SNPs. We further evaluated the role of these three SNPs in 29 EWS tumor samples in PECAN.³¹ However, none of the SNPs showed statistical significance in their association with the normalized expression levels for the corresponding genes, possibly due to a lack of statistical power (Figure S2). Conversely, although a previous report suggested an OR as high as 3.3 between umbilical hernia and EWS, such a relationship was not observed in the MR analysis. One reason, as discussed in previous research,¹ is that the common risk factor, such as farming, confounded the relationship. MR analysis could mitigate the impact of such confounding variables not associated with the chosen instrumental variables.

Table 3.

Three genetic loci of interest based on MR and colocalization analyses

SNP	Location	Gene	p (inguinal hernia)	p (EWS)	Evidence
rs12810758	12q14.3	HMGA2	8.9 × 10−11	0.00089	COLOC, MR
rs10519694	5q23.1	LOX	3.4 × 10−8	0.024	MR
rs7692200	4q22.2	FBXW7	9.3 × 10−8	0.003	COLOC

The statistical power for MR analysis for binary outcome depends on the proportion of cases in the hernia study, the true effect size, and the proportion of variance of exposure explained by SNPs.³² Because the proportion of cases was relatively small for umbilical, femoral, and ventral hernias, these MR analyses may be underpowered. Although we used MTAG to boost the statistical power for hernia GWASs, we interpreted the MTAG results with caution. MTAG relies on a key assumption that all SNPs share the same variance-covariance matrix of effect sizes across traits, which can be easily violated for individual SNPs. MTAG drove more lead SNPs into the analysis, especially for umbilical and ventral hernias. However, we still did not find any statistically significant evidence of their association with EWS. Lastly, our study also did not explore the sex-specific genetic effects. Hernia has a higher prevalence in males than females. Studies have found a stronger contribution of genetic risk factors for inguinal hernia in women compared to men, suggesting sex-specific genetic effects.³³ Interestingly, EWS has a male:female ratio of 1.5:1.³⁴ Because the present study may encounter limited statistical power for a sex-stratified analysis, we leave such an evaluation for future larger-scale investigations.

In conclusion, we used the genetic analysis as an effective and orthogonal approach to follow up on the risk factors discovered by traditional epidemiology studies. Our work provides additional support to the purported positive association between hernias and EWS, especially for diaphragmatic and inguinal hernias.

Data and code availability

GWAS summary statistics for hernia: https://www.danishnationalbiobank.com/GWAS.

GWAS summary statistics for EWS was obtained from collaborators.

Tumor eQTL data for EWS was obtained from https://pecan.stjude.cloud.

We used publicly available software (see the above section) for analysis and code, which may be available from the corresponding author on request.

Web resources

MTAG: https://github.com/JonJala/mtag.

LDSC version 1.0.1: https://github.com/bulik/ldsc.

SUPERGENOVA: https://github.com/qlu-lab/SUPERGNOVA.

TwoSampleMR 0.5.6: https://github.com/MRCIEU/TwoSampleMR.

PheWeb version 1.3.15: https://pheweb.org/UKB-SAIGE/about.

Plink 1.9: https://www.cog-genomics.org/plink/.

COLOC 5.1.0: https://cran.r-project.org/web/packages/coloc/index.html.