Autoimmune (AI) and immune-mediated (IM) diseases are a driving force behind disability in the United States with 5–8% of the population affected (2002). Currently there are upwards of 80 conditions that occur as a consequence of immunological attacks on the body's tissues and organs. As a whole, AI and IM diseases share many epidemiological and pathogenic similarities. Most notable is the occurrence of a skewed sex distribution with females more likely to be affected by an AI disease. The female sex ratio of cases is most extreme in Sjögren syndrome (9:1) that occurs as a result of immune-mediated attack on salivary and lacrimal glands in comparison to type 1 diabetes which is sex neutral (Gale and Gillespie, 2001). Over the last decade, genome-wide association studies have consistently identified genes within the major histocompatibility complex (MHC), located on the short arm of chromosome 6, strong regulators of disease risk in a number of AI diseases such as multiple sclerosis (MS) (International Multiple Sclerosis Genetics Consortium et al., 2007, 2011; Patsopoulos et al., 2011), rheumatoid arthritis (RA) (Plenge et al., 2007; Raychaudhuri et al., 2008; Okada et al., 2012), and vitiligo (Jin et al., 2010; Quan et al., 2010). Despite these commonalties, their underlying genetic and molecular profiles vary and are hindered by low case counts with most AI diseases rarely occurring in the general population (i.e., Addison's disease prevalence estimated at ~1 in 20,000). Elusive and overlapping symptoms further confound diagnosis of a condition, often leading patients on a long and costly diagnostic odyssey. Additionally, time to diagnosis, time between presentation of first symptoms to confirmed diagnosis, can approach a year in conditions such as amyotrophic lateral sclerosis (Paganoni et al., 2014).

Large-scale studies for AI and IM diseases typically require recruiting and consenting patients from specialty clinics and individual medical practices, all of which necessitates considerable financial investment and manpower. In recent years, researchers have begun to utilize electronic health records (EHR) for use in clinical and genetic association studies covering a range of conditions from type 2 diabetes (Ng et al., 2014) to response to drugs or treatment (Oetjens et al., 2014; Laper et al., 2016). Use of EHRs rapidly facilitates biomedical studies by providing researchers with access to a repository of extensive, longitudinal medical data. In particular, EHRs offer the opportunity to assess less common conditions such as MS by directly accessing data from specialty disease clinics within the framework of a health care organization. In conjunction with DNA repositories, it becomes feasible to combine genetic and phenotypic data in order to study the immunogenetic architecture of autoimmune disease (Goris and Liston, 2012), much of which is still unknown.

Individuals with one AI disease are at greater risk to develop another AI condition; RA and type 1 diabetes have been observed to occur jointly in patients from a United Kingdom cohort, while there is reduced comorbidity between RA and MS (Somers et al., 2009). Additionally, a GWAS study of inflammatory bowel disease, which includes CD and ulcerative colitis, found variants initially identified in GWAS studies of MS and RA (Liu et al., 2015). In order to delineate potential, common genetic pathways involved in AI and IM disease, we have performed a targeted genetic association study of SNPs known to be associated with AI (MS and RA) and IM (Crohn's disease or CD) in a European American population extracted from the Vanderbilt University Medical Center's (VUMC) DNA biorepository linked to de-identified EHRs. Based on previous evidence from GWAS (Sivakumaran et al., 2011) and candidate gene studies, we hypothesized that MS, RA, and CD share genetic factors within the MHC region, known to include many immune system genes, and outside the MHC region where less is known about genes that may play a role in autoimmunity. Elucidating the role that genetics plays in disease risk and progression is vital for future studies to incorporate risk of a comorbid disorder and lead to better screening and prevention strategies.

In order to determine whether common autoimmune and immune-mediated diseases share an underlying genetic etiology, we performed a targeted genetic association study of common variants in MS, RA, and CD which had been implicated in previous studies. Additionally, joint modeling of the three phenotypes and pathway analysis was performed which identified overlapping KEGG pathways involved in bacterial pathogenesis. Although this study was underpowered to detect associations with moderate effect sizes below 1.80, we replicated a number of strong, well-studied associations in MS and CD.

Studies of MS have historically been confounded by the clinical heterogeneity of the disease as well as the complex interaction of genetics and environment in disease initiation, progression, and severity (Davis et al., 2013; Davis and Haines, 2015). Despite these obstacles, genetic studies identified the MHC region and the HLA genes as major predictors of MS over 40 years ago (1972). The HLA-DRB1^*15 allele is a hallmark susceptibility locus for MS (Caillier et al., 2008), which is marked by rs3135388 (International Multiple Sclerosis Genetics Consortium et al., 2007). Our study replicated this association at p = 0.003 with an OR of 3.02 and risk allele frequency (RAF) of 25% in cases and 12% in controls, similar to another publication (OR = 2.7; RAF_cases = 29.8%; RAF_controls = 13.6%; Briggs et al., 2010).

While we replicated other associations between MS and variants in the HLA genes, perhaps more interesting were the inverse associations for SNPs that were identified in studies of RA. These inverse associations may represent false positive associations in the present study (a strong possibility underscoring the small sample sizes and limited power) or interesting complex pleiotropic or co-morbid relationships. SNPs rs3093024 and rs3093023 are both located in the CCR6 gene on chromosome 6. Originally identified in independent GWAS of RA in a Chinese and Japanese population, these variants contributed to a small increase in risk of RA within these populations (OR = 1.12–1.13; Kochi et al., 2010; Jiang et al., 2014). In the BioVU European American population, these variants were associated with MS in the opposite direction [OR = 0.39–0.43]. Both MS and RA are T helper type 1 cell (Th1)—mediated AI diseases that have inconsistently been observed to co-occur in individuals at a rate varying from the general populace. In a large population study utilizing the United Kingdom General Practice Research Database, an inverse rate of comorbidity was observed between RA and MS (Somers et al., 2009). The opposite was observed in U.S.-based multiplex families with MS in which RA occurred in 2% (Barcellos et al., 2006) of cases vs. ~1% in the general U.S. population (Helmick et al., 2008). Additionally, a Taiwanese population study of MS and comorbid AI diseases also found patients diagnosed with MS were at greater risk of developing RA (OR = 4.8; Kang et al., 2010). In chart reviews from the Vanderbilt University Medical Center's EHR linked to BioVU, we noted two MS patients out of 162 (1.2%) who were diagnosed with RA. While this rate falls within the range observed in previous studies, it should be noted that the present numbers from BioVU may be biased given that medical charts were not reviewed with the intent to identify additional overlapping AI diseases.

There were a number of factors that limited the strength of this study, particularly the small number of strict cases available with genetic data. Had we accepted a more lenient definition for cases, such as including patients whose medical records contained an ICD-9-CM code for another AI disease, we could have increased the number of cases for RA (n = 286). By utilizing a more lenient case definition, we are able to replicate previous RA associations for rs6910071 [(Stahl et al., 2010); OR = 2.41 (CI 1.72–3.39); p = 3.42 × 10⁻⁷] and rs660895 [(Plenge et al., 2007); OR = 2.21 (CI 1.60–3.03); p = 9.98 × 10⁻⁷] at a Bonferroni significance threshold (1.5 × 10⁻⁴).

The strict case definitions also may have limited our ability to more fully describe the shared genetic architecture for these AI and IM outcomes. That is, the case definitions explicitly excluded patients with two or more overlapping conditions (among MS, RA, and CD). Had this exclusion not been in place, the resulting cases would be a mix of bona fide cases with two or more AI or IM conditions and those who had a billing code for one AI or IM disease but were diagnosed with another. It is well known that many of these AI and IM patients experience a diagnostic odyssey. Multiple sclerosis patients, for example, typically receive an eventual differential diagnosis where laboratory tests are ordered to rule out other conditions. More complex data mining followed by manual review of EHRs would be required to identify true cases of AI or IM for study.

Another challenge unique to EHRs compared with epidemiologic cohorts is the patient-to-patient variability related to follow-up data. For example, Davis et al. (2013) noted that the median follow-up time for MS patients in BioVU was 4.5 years, which is similar to the observed median follow-up time for BioVU patients with normal electrocardiograms (5.0 years; Denny et al., 2010) and for BioVU patients with a heart transplant (5.5 years; Oetjens et al., 2014). The range of follow-up on a per patient basis, however, can be extreme (0–20 years for MS patients; Davis et al., 2013). The extreme differences in follow-up time per patient is also reflected on the number of clinic visits available in BioVU per patient. An analysis of ~15,000 patients in BioVU revealed an average of ~82 clinic visits per patients with a wide range of one to 1456 clinic visits per patient (Crawford et al., 2015). The paucity of clinical data for a proportion of patients in any given EHR is a known caveat for use of these data in a research setting (Hersh et al., 2013).

For patients with adequate follow-up data, access to the rich, longitudinal EHR data is a strength that enables identification of patients with complex diseases that might be difficult to extract from epidemiological cohorts. Additional work, however, is required to better define clinical populations. Indeed, EHRs in the United States recently adopted the 10th revision of the International Statistical Classification of Diseases and Related Health Problems codes (ICD-10) in response to meaningful use encouraged in part by the Health Information Technology for Economic and Clinical Health (HITECH) Act as part of the American Recovery and Reinvestment Act (ARRA). Compared with ICD-9-CM, the ICD-10 billing codes expanded from 13,000 to 68,000, offering much needed granularity for both clinical care and research relevant to precision medicine.

The most significant individual SNP associations in the joint model were HLA variants (Table 4) and ITLN1 (Barrett et al., 2008; Franke et al., 2010; Liu et al., 2015) and PTPN2 (Okada et al., 2012) genes. PTPN2 is a non-receptor type tyrosine-specific phosphatase that dephosphorylates receptor and non-receptor protein tyrosine kinases in a diverse range of signaling cascades that are responsible for hematopoiesis, cell proliferation, and inflammatory response. PTPN2 also negatively regulates a number of immune pathways responsible for T-cell differentiation (Spalinger et al., 2015) and Interleukin-6 (IL6; Table 4) cytokine signaling (Yamamoto et al., 2002).

Consistent evidence finds that autoimmune diseases co-occur at an increased rate in family-based and population studies (Barcellos et al., 2006; Eaton et al., 2007; Somers et al., 2009; Sardu et al., 2012) suggesting that these conditions share common genetic or environmental etiologies. Here we implicate bacterial pathogenesis (e.g., NOD-like receptor signaling and Shigellosis) as a common underlying susceptibility factor for MS, CD, and RA. Individual SNPs and genes in these pathways may not be significant at a multiple-correction threshold due to small sample size and clinical heterogeneity as described previously. Yet, pathway analysis takes into consideration modest effect sizes distributed throughout genes in a pathway and increases the likelihood of identifying relevant biological components. Infectious pathogens have been hypothesized to play a major role in the development of autoimmune diseases through a number of mechanisms. A couple of these mechanisms included epitope spreading and molecular mimicry (Ercolini and Miller, 2009). Molecular mimicry results due to a pathogen carrying proteins, amino acid sequence, or antigens that closely resemble host factors which lead to T or B cells triggering an autoimmune response (Lo et al., 2000; Paludan and Bowie, 2013).

In conclusion, pathway analysis of a joint model identified overlapping KEGG pathways with implications for autoimmunity driven by bacterial pathogenesis. These pathways have already been found in individual studies of AI/IM diseases. Future studies with a larger clinical population may allow for replication and fine-tuning of these results. Longitudinal clinical or epidemiological data will be necessary to determine whether acute or chronic infections indeed lead to susceptibility or progression of comorbid AI/IM diseases.

The listed authors provided substantial contributions to the conception or design of the work (NR and DC), the acquisition (DC, NR, and JM), analysis (NR and MB), or interpretation of the data (NR and DC) for the work; drafted the work (NR) or revised it critically for important intellectual context (NR, MB, JM, and DC); gave final approval of the version to be published (NR, MB, JM, and DC); and agreed to be accountable for all aspects of the work in ensuring that questions related to accuracy or integrity of any part of the work are appropriately investigated and resolved (NR, MB, JM, and DC).