Exploring the causal role of the immune response to varicella-zoster virus on multiple traits: a phenome-wide Mendelian randomization study

Varicella-zoster virus (VZV) is a highly communicable pathogen. It is widely present in the general population, with an estimated incidence of 4–4.5 per 1000 person-years [1]. Primary infections of VZV are commonly seen in children as chickenpox. The virus will then remain latent in the sensory dorsal root ganglion cells and can be reactivated throughout life. The reactivations of the virus may cause damage anywhere on the body, from the nerves to the skin [2]. The common manifestations are neurological symptoms, such as radicular pain, itching and unpleasant sensations. The virus can affect the cranial nerve and cause diseases (e.g. tooth exfoliation [3] and facial palsy [2]). Facial palsy could trigger other symptoms such as tinnitus, hearing loss and nausea. VZV reactivation may also result in retinal necrosis and severe ocular diseases [4]. The virus can directly enter the cerebrospinal fluid causing meningitis [5] or invade the spinal cord and produce myelopathy [6]. It is also possible that the virus travels to the arteries to induce vasculopathy like haemorrhagic stroke [7]. Recent studies have shown associations between herpes zoster and other diseases such as depression and anxiety [8]. Therefore, the infection of VZV can have a long-term impact on multiple health conditions.

Humoral immune responses to the same infectious agent vary greatly between individuals. Evidence shows that genetic factors may play a determining role in the individual antibody responses to a variety of viruses, especially the variants in the major histocompatibility complex (MHC) region [9]. Genes likely affect the humoral immune response by controlling the serum immunoglobulin levels, seroconversion rates and intensity of antigen-specific immune responses [10]. Therefore, once a host is exposed to the pathogen, their immune response will be affected by genetic factors. Studying infectious immunity-associated genes helps understand the pathologic mechanisms. Immunoglobulin G (IgG) antibody is the most common antibody in the blood and is persistent throughout a person’s life [11]. It can be used as a stable biomarker of lifetime exposure to viruses. Increased levels of anti-VZV IgG without the presence of herpes zosters implies that it correlates with VZV reactivations [12]. Previous genome-wide association studies (GWASs) investigated single nucleotide polymorphisms (SNPs) that influence the antibody responses against VZV infections. While Petars et.al found no SNPs significantly associated with anti-VZV IgG levels in 1000 healthy individuals [10], five SNPs (rs13197633, rs34073492, rs56401801, rs13204572, rs1048381) were identified in a larger cohort (N = 7595) from the UK Biobank [13].

Mendelian randomization (MR) is a well-established tool for inferring causal relationships. By using SNPs as instrumental variables (IVs), MR helps to avoid the issue of unobserved confounding in observational studies. The MHC region is a complex genomic region, in which genes are associated with the risks of many diseases [14]. We hypothesized that IVs in the MHC region may play an important role in MR analysis. However, using MHC IVs in MR has rarely been reported in the literature due to concerns about horizontal pleiotropy [15]. In fact, several robust MR approaches have emerged more recently with the aim to minimize the bias from horizontal pleiotropy and/or outliers. For example, MR-Egger estimation allows a genetic instrument to have a direct effect on the outcome, as long as the direct effect is independent of the instrument’s strength [16]. The weighted median method provides consistent causal estimates if at least 50% of the weights come from valid IVs [17]. MR-RAPS assumes that the pleiotropic effects are balanced and assigns low weights to outliers when estimating the causal effect [18]. The MR-PRESSO method detects the outliers which will then be removed to obtain robust estimations [19]. Nevertheless, each method has its own strengths and weaknesses and requires specific assumptions. It has been recommended that multiple MR methods can be performed to seek concordant evidence for a causal relationship [20].

This study aimed to explore the causal role that VZV-specific immune responses might play in the development of multiple traits, by performing phenome-wide MR studies (MR-PheWAS) between VZV titre measurement and 1370 independent traits using five different MR approaches. We also investigated the impact of the genetic instruments in the MHC region on MR analysis.

The outline of the study is shown in Fig. 1.

The mean anti-VZV IgG levels (log10-transformed) and standard deviation were 2.724 and 0.514, ranging from 0 to 4.021. As many as 1421 SNPs were found associated with anti-VZV IgG, of which 1205 (84.8%) were in the MHC region. The complete list of these SNPs was provided in Additional file 2: Table S1. Four of the SNPs (rs13197633, rs13204572, rs1048381 and rs56401801) were also associated with anti-VZV IgG in a previous UK biobank GWAS [13]. After clumping, 49 independent anti-VZV IgG-associated SNPs with F-statistic greater than 10 were selected as IVs (IV_full) for MR analysis, among which 5 SNPs were inside (IV_mhc) and 44 were outside the MHC region (IV_no.mhc) (Table 1). The average F-statistic for IV_mhc, IV_no.mhc and IV_full was 40, 19.1 and 16.7, respectively. The numbers of the IVs used in MR tests and the SNPs removed after Steiger filtering were listed in Additional file 2: Table S2. A total of 1370 traits (73 ordinal, 773 categorical, 459 binary and 65 continuous) were included in the MR analysis. Additional file 2: Table S2 listed all these results.

Among the traits that showed a positive causal association with anti-VZV IgG, six (Dupuytren disease, mononeuropathies of upper limb, sarcoidosis, coeliac disease, teeth problems and led to earlier onset of allergic rhinitis) were reported in the literature. In particular, the onset of herpes leading to tooth exfoliation was reported in many patients, as the virus could spread to the trigeminal nerve and cause damage to the tooth nerve [27, 28]. When the anti-VZV level was high, the increased risks of upper limp mononeuropathy could be explained by an uncommon complication of herpes zoster infections, zoster-associated mononeuropathy (ZAM) [29]. Although uncommon, herpes zoster directly leading to mononeuropathy was found in a few clinical cases. Patients having ZAM suffered from viral spreading to the peripheral nerves [30]. Dupuytren disease is a condition that affects the fascia and causes deformity to the hand, though the causes of itself remain unknown [31]. A positive association between the risk of herpes zoster and plantar fascial fibromatosis incidence was identified in the literature [32], but no causal evidence was ever reported. Coeliac disease, sarcoidosis and allergic rhinitis are the consequences of dysregulation of the immune system. Coeliac disease and sarcoidosis are caused by the immune system attacking the intestine and multiple systems, respectively. Allergic rhinitis often refers to the overreacts of the immune system in the upper respiratory tract. There has been extensive discussion about the correlation between the incidence of herpes zoster reactivations and autoimmune diseases [33, 34]. VZV reactivation can induce different types of sarcoidosis, such as scar sarcoidosis [35] and central nervous system sarcoidosis [36]. The incidence of herpes zoster was positively correlated with coeliac disease [37] and allergic rhinitis [38]. All the evidence from these studies supported our causal findings. However, the relationships between herpes infections and these autoimmune diseases might not be simply unidirectional. For example, the virus itself might increase the risk of flares in autoimmune diseases like systemic lupus erythematosus [39]. In addition, herpes zosters are more likely to reactivate in these patients as they might take immunosuppressive medications [40]. However, the immunosuppressive treatment also decreases the antibody levels in one’s body [41]. Therefore, interactions between immune responses to infections, autoimmune diseases and immunosuppressive treatments could be potentially considered in future studies.

Causal associations between VZV infections and eight traits (number of incorrect matches in pairs matching [42, 43], guilty/tense feeling [8, 44, 45], BMI [46, 47] and body/trunk fat mass and percentage) were inconclusive from previous studies. The direct causal relationships of four traits (carpal tunnel syndrome, FEV1, PEF, other specific joint disorders) and VZV were not reported before, but they may have plausible biological links [48‐51]. In addition, we discovered that inguinal hernia and diabetic eye disorder may be causally associated with VZV immune response. The definitions of other serious medical conditions/disabilities diagnosed by a doctor and long-standing illness, disability or infirmity were unclear and require further explorations. Our MR results provided new insight into the causal relationships between VZV immune response and these traits, which however need to be replicated in other independent studies.

The role of the MHC region plays in MR studies had largely been overlooked [52]. In some studies, IVs in the MHC region were simply removed to minimize the risk of bias due to the pleiotropy introduced by these SNPs [15, 53]. However, some MHC SNPs may be strongly associated with exposure, especially for immunity-related traits [13]. Here, we explored the impact of MHC instruments by performing three separate MR analyses using IV_full, IV_mhc and IV_no.mhc. For the traits showing a suggestive causal relationship with anti-VZV IgG levels, there were substantial differences in the results among the three IV sets. Evidence for a correlation of the estimated causal effects was shown between IV_full and IV_no.mhc and also between IV_full and IV_mhc, with approximately one-third of the traits showing statistically significant (P < 0.05) causal relationships with anti-VZV IgG and in the same directions. However, using the two subsets (IV_no.mhc and IV_mhc) cannot fully identify all the causal relationships found using IV_full. In fact, each subset had its specific causal findings. The number of traits that showed suggestive causal evidence with VZV immune response was almost four times larger using IVs from the MHC region than from non-MHC regions. As the IVs in the MHC regions have stronger associations with anti-VZV IgG, it was not unexpected that causal relationships were detected in a higher number of traits when using IV_mhc. Interestingly, there was a discrepancy in causal estimates from IV_mhc and IV_no.mhc with opposite causal directions shown in ten traits. This might indicate that IVs in and outside the MHC region drive different mechanisms or intermediaries behind anti-VZV IgG levels. For example, MHC SNPs might be more involved in the immune responses in one’s body [54] while non-MHC SNPs be more associated with general health conditions that also affect the antibody levels [55]. Taken together, MHC instruments might play a different but important role in affecting the antibody levels and subsequently the clinical outcomes. Partitioning the IVs by MHC region might lead to conclusions that are different from the MR results using all the IVs. It would be advisable to investigate this separately in future MR studies. Moreover, as MHC IVs can potentially have an impact on MR analysis, imputing the region using specialist software would enhance accuracy and resolution for causal findings [56].

It is worth noting that suggestive causal relationships between some traits and anti-VZV IgG shown in this study were already reported previously (e.g. stroke [57] and systemic lupus erythematosus [39]) (Additional file 2: Table S3). However, the evidence we found was not strong enough possibly due to the complicated causal mechanisms that weakened the marginal causal effect size and therefore reduced the power to detect them in a phenome-wide MR study. Previous GWAS identified five SNPs associated with anti-VZV IgG levels [13], four (rs13197633, rs13204572, rs1048381 and rs56401801) of which were also found associated with anti-VZV IgG levels in our study. Rs34073492, also identified in previous GWAS, was however removed from analysis post-QC (MAF > 0.02).

The anti-VZV IgG levels in our study population were not likely driven by vaccination. This is because the age group enrolled in the UK national programme for shingles vaccination is between 70 and 79 years [58], while the participants for the UK biobank cohort were aged 40–69 when first enrolled.

Multiple testing decreases type I error but increases type II error [59]. It would be excessively conservative to use a stringent statistical significance threshold in MR studies which are in general of low power [60], with a risk of not detecting potentially important causal relationships. Therefore, in this exploratory study, we also reported MR results with a P-value less than 0.05 in the supplementary tables to minimize the reporting bias and to enable researchers to take forward our findings for further investigations in their studies.

Our phenome-wide MR study found anti-VZV IgG to be a causal risk factor of 22 traits, which helped gain new insights into the causal roles of VZV-specific immunity in diseases. While some of the causal relationships have priori biological explanations, others require further investigations. When selecting IVs for MR analysis, separating SNPs from the MHC region may potentially provide information on different causal pathways, and help better understand the specific role of the MHC region in future MR studies.