Genetic association study of fatal pulmonary embolism

Pulmonary embolism (PE) is one manifestation of venous thromboembolism (VTE) and a potentially lethal complication of VTE’s other clinical presentation deep vein thrombosis (DVT). Approximately 65% of VTE patients display only DVT, whereas around 30% of VTE patients also manifest PE [1].

PE diagnosis is challenging due to its variable and often non-specific clinical presentation. Silent PE develops in around 50% of patients with DVT and autopsy reports demonstrated that only 30–45% of all PE cases were diagnosed prior to death. The annual incidence of PE is estimated to be about 50–100 per 100,000 individuals and increases with age [2, 3]. Males are more often affected than females. However, women above 75 years display an incidence of more than 500 cases per 100,000 individuals [4]. In forensic practice, pulmonary embolism is relatively common, either as sudden unexpected death or in the context of potential maltreatment. In our own praxis, we observe pulmonary embolism in ca. 4% of all medicolegal autopsies.

Pulmonary embolism is a complex multi-factorial disease which is caused by external as well as multiple genetic factors. However, most studies on risk factors were conducted analysing associations with VTE and not PE in particular [5]. External risk factors can be divided into permanent, patient-associated risk factors and, temporary circumstances like immobilisation [2]. In 20–50% of the cases, PE occurs in absence of these risk factors [6, 7]. Moreover, twin and family studies suggested a major genetic constituent risk to be related to VTE by demonstrating an increased risk for individuals with affected siblings and a strong heritability [8‐10]. The genetic predisposition to elevated risk of VTE is described by the term thrombophilia, which is often associated with gene variations of the physiological coagulation cascade [11]. Minor changes in the balance between the system of coagulation and fibrinolysis might cause thrombus formation. These changes can occur through mutations in genes coding for coagulation factors and inhibitors, which lead to altered levels or functional losses of associated proteins [12].

Until the beginning of 2000, only few gene loci were robustly identified to be associated with VTE. Most of these variations represent strong genetic risk factors increasing the risk of VTE in heterozygous carriers almost 10-fold, such as deficiencies of antithrombin or protein C [5, 13]. However, these genotypes occur in less than 1% of the population and thus represent only few patients suffering from VTE [14]. Subsequently, new loci, presenting weaker risk factors but higher frequencies, were identified through studies of candidate genes [15] and genome-wide association studies (GWAS) [14, 16‐18].

Since PE is usually regarded as a complication of DVT, their inherited risk factors are often regarded to be similar, but the prevalence and the risk associated with some risk factors such as Factor V Leiden or activated protein C resistance differ between DVT and PE [19, 20]. The Factor V Leiden paradox states that the rs6025 polymorphism increases the risk of DVT much more than the risk of PE [20]. These findings emphasize the necessity of considering DVT and PE as different phenotypes in VTE genetic studies. Only one study investigating a broad panel of genetic VTE risk factors in patients with fatal PE has been reported so far [21]. Brandimarti et al. conducted a study in an Italian population investigating associations between 12 thrombophilic risk factors and VTE in 44 individuals who died from PE. Strong associations between PE and the polymorphisms C677T (MTHFR), Pai-1 4G/5G and ITGB3 were found [21].

The present study aims to replicate the results of the aforementioned study using a considerably larger sample size. Furthermore, this study intends to evaluate the association of additional SNPs which were robustly associated with VTE or DVT in previous studies in subjects who died from PE.

Genotyping achieved call rates between 95.7% (rs8176592) and 99.1% (rs3813948) as well as an overall call rate of 98.1% for the 560 samples that were genotyped for 22 SNPs.

At a significance level of 0.05, eight of 22 SNPs presented nominally significant results in either the main or the eight subgroup analyses before the correction for multiple testing. After utilizing the multiple-comparison Bonferroni correction for the main analysis (p = 0.0017), no significantly different genotype distributions remained between cases and controls. However, p ≤ 0.05 was regarded in this study as an evidence of association (EOA) suggesting noteworthiness for the investigation of causes for PE. The results for those variants demonstrating p < 0.05 in the logistic regression analyses are depicted in Table 2. Three of 22 SNPs demonstrated EOAs in the main analysis (rs1800790, rs3813948, rs6025). The five additional polymorphisms rs169713, rs1801131, rs4524, rs5985 and rs8176592 presented EOA in subgroup analyses. Values ranged between p = 0.006 (rs169713 in “all cases without surgery, tumour or BMI below 30 kg/m² vs. all controls”; rs6025 in “all cases without tumour vs. all controls”) and p = 0.05 (rs1801131 in “cases without known risk factors vs. all controls”). EOAs were found for rs169713 in two subgroups, for rs1800790 in four, for rs1801131 in one, for rs3813948 in four, for rs4524 in one, for rs8176592 in one and for rs6025 in five subgroups (Table 2).When being compared to literature [33], the MAF of our control group presented deviations ranging from 0.043 (rs1801133) to 11.2% (rs169713) (online supplementary material: Table 1).

A meta-analysis of our results with previously published results from Brandimarti et al. [21] showed for some previously proposed SNPs opposing or strongly deviating effects, with significant heterogeneity between studies for rs8176592 (p_het 0.028) and rs5918 (p_het 0.001). For rs1801133 and rs10931292, both studies showed similar tendencies though the meta-analysis did not reach statistical significance (Table 3). From our own study, FGB (rs1800790, p_meta = 0.068) and F5 (rs6025, p_met = 0.064) remained with marginal evidence of association.

Clinical medicine as well as forensic post mortem investigations would benefit from well-defined genetic risk factors for PE. However, almost all studies on the association of gene loci with VTE up to now do not differentiate between PE and DVT [24, 27, 34]. There is however a big difference in consequences for a patient between DVT and the potentially deleterious complication of PE. Furthermore, it is well known that, fortunately, only a minority of patients with increased DVT risk, e.g. from Factor V Leiden, succumb from PE, a phenomenon described as Factor V Leiden paradox. On the other hand, studies on risk factors for PE are mostly focused on external risk factors and not on candidate genes from studies on VTE, which results in an insufficient knowledge concerning their influence on PE (e.g. [35]).

The only study that focuses on genetic variants in deaths from PE we are aware of is that by Brandimarti et al. [21] that found significant differences between a group of persons deceased from PE and a control group for SNPs in MTHFR*C6777T (rs1801133), SERPINC1C*4G/5G (rs397832688) and ITGB3*T196C (rs5918). However, the groups tested in this study were small: 44 persons who died from PE and 102 controls. While rs397832688 was not amenable to our SNPtype assay design, our study of 375 controls and 185 cases could not reproduce a significant difference in genotype frequencies for the two SNPs MTHFR*C6777T (rs1801133) and ITGB3*T196C (rs5918), respectively, between PE cases and controls reported by Brandimarti et al. [21]. These differences in results may be explained by the small number of PE cases and controls in the initial study [21]. Small sample sizes may lead to overestimations of the particular effect size [36].The overestimation of the effect size can, in turn, result in overestimations of the power of replication studies, which might have affected the a priori calculated power for a few variants of the present study [37]. For instance, the statistical power for rs5918 in our study design was determined a priori on the basis of the genotyping results of Brandimarti et al. [21] as being close to 100% to detect a significant difference at p < 0.05 with the given risk estimate and study size. However, the statistical power for the rs5918 polymorphism based on the truly obtained genotype frequencies amounts to 16%. The post hoc calculated power for the other SNPs tested in our study ranges between 1 (rs1801133) and 62% (rs1800790). This emphasizes that, according to Ziegler et al. [38], extensive sample sizes are necessary for a reliable detection of the small effects which are expected for the contributing factors in complex diseases like PE. Most of the variants that we analysed showed only small effects (see OR and their 95% CI in Table 2). In comparison to Brandimarti et al. [21], we used approximately the 4-fold number of cases as well as controls, but for a reliable determination and replication of the small effect sizes, the study size still would need to be strongly increased.

In our study, rs1800790, rs3813948 and rs6025 demonstrated EOA for the main logistic regression analysis. The Factor V Leiden mutation (rs6025) displayed similar effects in the subgroup analyses indicating a rather small effect of stratification. According to the FV Leiden paradox, the risk of DVT is substantially increased in carriers of the FV polymorphisms, whereas the risk for PE is only mildly increased [39]. The rs6025 was not found in any of the 44 cases studied by Brandimarti et al.; however, Van Stralen et al. [39] demonstrated in carriers of the Factor V Leiden mutation an approximately 8-fold increased risk of DVT (OR 7.7, 95% CI 3.9–15.3), whereas the risk of PE was small (OR 1.4, 95% CI 0.7‑2.7). Our study is consistent with an about 2-fold increased risk of PE in carriers of the mutation which supports a role of this mutation for PE but is well in line with the results from the aforementioned study.

The FGB polymorphism 455 G/A (rs1800790) has been associated with higher levels of fibrinogen, which in turn suggests an association of the variant with a hypercoagulable state as well as thrombosis [40]. Studies could not agree whether the polymorphism exerts a protecting role or none at all. Whereas some studies [40, 41] detected a lower VTE risk to be associated with the polymorphism, another [42] found no association between the mutation and PE. Brandimarti et al. [21] could not demonstrate significant differences, but our study demonstrated an EOA for the polymorphism in the main analysis and three subgroup analyses (see Table 2). The computed ORs indicate a protective effect of the rare allele.

The protein C4BPB binds to protein S and thus presents a main element of the coagulation and fibrinolysis cascade. A previous study demonstrated a slightly increased risk associated with the polymorphism rs3813948 located in the C4BPB gene [27]. However, the results from our study indicate the opposite effect, demonstrating a rather protective effect of the polymorphism. A possible explanation for the opposing results is that Buil et al. [27] analysed the association with VTE without differentiation in DVT and PE cases, whereas our study examined those polymorphisms only in PE deaths. In this case, the results would suggest strongly differing risks associated with the polymorphism for PE and DVT, which need to be examined in further studies.

The polymorphisms at rs169713, rs1801131, rs4524, rs5985 and rs8176592 presented EOA in the subgroup analyses that we performed in order to maximize the effect of the genetic variant by enriching the susceptible group through exclusion of externally influenced cases, as proposed [38].

Franco et al. [43] demonstrated a protective role of the rare SNP allele for the Factor XIII p.Val34Leu polymorphism (rs5985). The subjects who did not harbour the rare allele presented a 6-fold increased risk of developing DVT. Our study results suggest the rare allele of this polymorphism to exert a protective role not only for DVT but possibly also for PE.

Lincz et al. [44] examined the activity of the TFPI and frequencies of TFPI polymorphisms in patients presenting Factor V Leiden and a manifestation of VTE (rs8176592). They found a strongly increased VTE risk in carriers of the polymorphism in the presence of Factor V Leiden. However, most other studies demonstrated a small protective effect of the polymorphism on the VTE risk by increasing the TFPI plasma level [32]. In the present study, we replicated the protective effect of the T33C mutation. Individuals without environmental risk factors, which do not exhibit the cysteine variant, displayed an about 3-fold increased risk of dying of PE.

As the first VTE susceptibility locus outside the traditional coagulation cascade, the rs169713 polymorphism of the HIVEP1 locus was significantly associated with VTE risk in a study by Morange et al. [26]. HIVEP1 affects the transcriptional regulation of target genes associated with inflammatory processes [5]. By increasing the damage on endothelial cells and enhancing hypercoagulability, inflammation is considered to act as a contributor to the VTE risk [45]. The results of our study support the aforementioned result, by demonstrating notable differences in two subgroup analyses. Moreover, our results suggest that the rare allele is associated with a slightly higher risk of PE than VTE in general.

Associations of the A1298C (rs1801131) and C677T (rs1801133) variants of the MTHFR gene and an increased VTE risk were found by Liu et al. [46]. Brandimarti et al. [21] examined these two polymorphisms but their results did not demonstrate significant differences of the polymorphisms in cases and controls. Furthermore, a study on a Chinese population [47] demonstrated significant results for the C677T variant of the MTHFR gene and VTE. Our study demonstrated evidence for differences in genotype frequencies of cases and controls for the A1298C variant within the subgroup analysis of cases without known risk factors. The C-allele yielded a protective effect on the PE risk. Our results suggest that subjects without known risk factors that do not exhibit the C-allele may display an almost 3-fold increased risk of developing PE.

The rs4524 polymorphism of the Factor V gene was associated in a meta-analysis of more than 65.000 individuals with a slightly increased risk for VTE [22]. Moreover, the effect allele was repeatedly associated with an elevated risk of DVT in three Dutch studies conducted by Bezemer et al. [48]. Our results suggest a rather decreased risk, if any, of PE for carriers of the effect allele in male individuals. These results might be in line with the aforementioned Factor V Leiden paradox. Nevertheless, the paradox was only observed as well as analysed for the rs6025 polymorphism [39] and needs to be investigated for the rs4524 mutation.

The results of our study, the largest in terms of sample size ever conducted solely on deaths from PE, may indicate that the risk of death from PE is—at least in part—determined by a number of gene variants. As the evaluation of the impact of these variants is far from conclusively resolved, further analyses in even larger samples would be rewarding. We followed Ziegler et al. [38] who suggested that trying to simplify the respective disease might improve the outcome of association studies. The authors suggest reducing the diseases’ complexity by choosing subgroups of case individuals which are more homogenous. We already demonstrated that for some SNPs, EOAs were only achieved in subgroup analyses. Further increase of the case group would also lead to increased subgroups and thus more distinct results. Moreover, it would be useful to select preferably non-anonymous age-matched controls, who did not develop the disease despite strong external exposure to risk factors.

In summary, our relatively large case-control genetic association study provided supportive evidence for genetic differences at eight candidate risk loci between cases with death from PE and controls, whereas it has refuted some others. Genomic variation modulating Factor V, Factor XIII, FGB, TFPI or HIVEP1 should be worth to be followed in subsequent studies. The results demonstrate evidence for an association; however, more investigations are necessary to prove a causal relationship. Therefore, sample sizes could further be increased and the number of independent tests could be minimized by focusing on the genetic variants demonstrating EOAs. From the forensic point of view, as already mentioned by Brandimarti et al. [21], genetic testing might be especially discriminatory in cases of medical malpractice.