Abstract
Background and Objectives Whether chronic autoimmune inflammatory diseases causally affect the risk of Alzheimer disease (AD) is controversial. We characterized the relationship between inflammatory diseases and risk of AD and explored the role of circulating inflammatory biomarkers in the relationships between inflammatory diseases and AD.
Methods We performed observational analyses for chronic autoimmune inflammatory diseases and risk of AD using data from 2,047,513 participants identified in the UK Clinical Practice Research Datalink (CPRD). Using data of a total of more than 1,100,000 individuals from 15 large-scale genome-wide association study data sets, we performed 2-sample Mendelian randomizations (MRs) to investigate the relationships between chronic autoimmune inflammatory diseases, circulating inflammatory biomarker levels, and risk of AD.
Results Cox regression models using CPRD data showed that the overall incidence of AD was higher among patients with inflammatory bowel disease (hazard ratio [HR] 1.17; 95% CI 1.15–1.19; p = 2.1 × 10−4), other inflammatory polyarthropathies and systematic connective tissue disorders (HR 1.13; 95% CI 1.12–1.14; p = 8.6 × 10−5), psoriasis (HR 1.13; 95% CI 1.10–1.16; p = 2.6 × 10−4), rheumatoid arthritis (HR 1.08; 95% CI 1.06–1.11; p = 4.0 × 10−4), and multiple sclerosis (HR 1.06; 95% CI 1.04–1.07; p = 2.8 × 10−4) compared with the age (±5 years) and sex-matched comparison groups free from all inflammatory diseases under investigation. Bidirectional MR analysis identified relationships between chronic autoimmune inflammatory diseases and circulating inflammatory biomarkers. Particularly, circulating monokine induced by gamma interferon (MIG) level was suggestively associated with a higher risk of AD (odds ratio from inverse variance weighted [ORIVW] 1.23; 95% CI 1.06–1.42; pIVW = 0.007) and lower risk of Crohn disease (ORIVW 0.73; 95% CI −0.62 to 0.86; pIVW = 1.3 × 10−4). Colocalization supported a common causal single nucleotide polymorphism for MIG and Crohn disease (posterior probability = 0.74), but not AD (posterior probability = 0.03). Using a 2-sample MR approach, genetically predicted risks of inflammatory diseases were not associated with higher AD risk.
Discussion Our data suggest that the association between inflammatory diseases and risk of AD is unlikely to be causal and may be a result of confounding. In support, although inflammatory biomarkers showed evidence for causal associations with inflammatory diseases, evidence was weak that they affected both inflammatory disease and AD.
Progressive cerebral neurodegeneration with extracellular β-amyloid (Aβ) plaques and intraneuronal neurofibrillary tangles are pathologic features distinguishing Alzheimer disease (AD), the most common cause of dementia. The brains of patients with AD show evidence for a sustained brain innate immune response.1 Moreover, higher levels of circulating inflammatory markers in the blood were observed in patients with AD compared with that of healthy controls.2 Observational studies and meta-analyses have reported a higher risk of dementia among patients with rheumatoid arthritis (RA),3 psoriasis,4 and inflammatory bowel disease (IBD),5 although contradictory results also have been reported.6,7 Whether chronic autoimmune inflammatory diseases causally affect the risk of AD is controversial.
The observed associations between these inflammatory diseases and AD may be because of common circulating inflammatory biomarkers, modulation of which could provide opportunities for the prevention and treatment of AD. However, previous studies showing inconsistent effects of anti-inflammatory drugs on AD have cast doubt on the potential for therapeutic modulation of chronic inflammation in AD.8,9 In this study, we investigated the relationship between different chronic autoimmune inflammatory diseases and risks of AD/dementia using a real-world observational analysis using large-scale population-based electronic health records (EHRs) and an instrumental variable analysis with genetic instruments known as Mendelian randomization (MR), which is less susceptible to confounding and reverse causation than conventional observational analysis.10 We performed bidirectional MR to further examine the role of circulating inflammatory proteins associated with or induced by autoimmune disease in the relationships between inflammatory diseases and AD. Identification of such inflammatory proteins could identify potential therapeutic targets for AD.
Results
Observational Analysis (CPRD)
The CPRD cohort included a total of 6,613,198 participants who met our quality checks (Figure 1). We included 85,147 patients with RA; 57,114 patients with IBD; 20,743 patients with MS; 220,729 patients with psoriasis; and 339,960 patients with OID in the analyses with a median clinical follow-up of 13 years. There were more obese patients and more patients with cardiovascular diseases and type 2 diabetes in patients diagnosed with inflammatory diseases, except for MS at baseline, than the nonexposed (Table 1).
CPRD = Clinical Practice Research Datalink; FRD = first registration date; TOD = transfer out date.
Baseline Characteristics of Eligible Participants Included in the Clinical Practice Research Datalink Observational Analysis
Cox regression models with different covariate adjustments showed consistent findings for each of the disease relationships with AD. The overall incidence of AD was significantly higher among patients with IBD (HR 1.17; 95% CI 1.15–1.19; p = 2.1 × 10−4), OID (HR 1.13; 95% CI 1.12–1.14; p = 8.6 × 10−5), psoriasis (HR 1.13; 95% CI 1.10–1.16; p = 2.6 × 10−4), RA (HR 1.08; 95% CI 1.06–1.11; p = 4.0 × 10−4), and MS (HR 1.06; 95% CI 1.04–1.07; p = 2.8 × 10−4) compared with the corresponding nonexposed group (model 3 in Table 2). These associations also were consistent with the analysis adjusted for propensity score. Effect sizes were similar in sensitivity analyses involving further restrictions of sampling: (1) restricting the analyses with inflammatory disease diagnosis from 2004 onward, (2) using linkage-eligible patients, (3) extending the outcome from AD to recording of any dementia diagnosis, (4) using patients with both an AD diagnosis in CPRD and HES/ONS plus at least 1 dementia drug prescription in CPRD, (5) removing AD cases that developed within 2 years of inflammatory disease diagnosis, and (6) removing AD cases that developed within 3 years of inflammatory disease diagnosis (eMethods, links.lww.com/WNL/C487 and eTables 9 and 10, links.lww.com/WNL/C486). Similar associations were also found in sensitivity analyses using dementia as the outcome with a slightly higher HR among patients with IBD (HR 1.22, 95% CI 1.20–1.23, p = 1.7 × 10−5). In the time-dependent CIF analysis, the 10-year risk of developing AD in the patients with RA was 1.12% (95% CI 1.06%–1.19%) compared with 1.07% (95% CI 1.02%–1.15%) among the non-RA participants (Figure 2B and eFigure 3) and 1.09% (95% CI 1.03%–1.14%) among the patients with MS compared with 1.02% (95% CI 0.07%–1.07%) among the non-MS participants (Figure 2D and eFigure 3). The estimated 10-year risk of developing AD was also higher in patients with IBD, psoriasis, and OID compared with participants without these inflammatory diseases, whereas the 10-year risk of death was lower in patients with IBD, psoriasis, or OID (Figure 2, A, C, E, F and eFigure 3). In the sensitivity analyses with dementia as the outcome, the 10-year risk of developing dementia in patients with one of the inflammatory disease diagnoses was consistent with the findings in the analyses of AD, whereas the 10-year risk of developing dementia was higher in patients with RA (3.76%; 95% CI 3.67%–3.85%) and MS (2.45%; 95% CI 2.37%–2.52%) (eFigures 4 and 5).
Comparisons Between the Associations of Inflammatory Diseases With AD Using Cox Regression Models (CPRD) and Using 2-Sample MR
AD = Alzheimer disease.
MR (Inflammatory Diseases and AD/Dementia)
Using the 2-sample MR approach, genetically predicted liability to inflammatory diseases was not associated with a higher risk of AD or dementia (Table 2 and eTable 11, links.lww.com/WNL/C486). After accounting for multiple comparisons, each doubling in odds of genetically predicted RA was associated with a lower risk of AD (ORIVW 0.96; 95% CI 0.93–0.99; pIVW = 0.006) and lower risk of dementia (ORIVW 0.95; 95% CI 0.94–0.97; pIVW = 3.8 × 10−8). However, this relationship was not sustained after excluding genetic instruments on chromosome 6 (RA→AD: ORIVW 1.02; 95% CI 0.97–1.08; pIVW = 0.44; RA→dementia: ORIVW 0.99; 95% CI 0.98–1.01; pIVW = 0.73).
MR (Circulating Inflammatory Biomarker and AD)
Given the well-established link between inflammation and AD,25 we further performed bidirectional MR analyses for circulating inflammatory biomarker levels and the diseases of interest to identify inflammatory biomarkers that are a common cause or the consequence of inflammatory disease and AD. Figure 3 presents the results from the Wald ratio or IVW, and eTables 12 and 13 (links.lww.com/WNL/C486) present the sensitivity methods. After accounting for multiple comparisons, each doubling in odds of MS was associated with a lower circulating level of MCP1 (coding gene: CCL2; MS→MCP1: NSNP = 67; βIVW = −0.01; 95% CI −0.009 to −0.005; pIVW = 2.0 × 10−10). Each doubling in odds of RA was associated with a higher circulating level of MIP1a (CCL3; NSNP = 13; βIVW = 0.03; 95% CI 0.013–0.038; pIVW = 5.1 × 10−5) and MIP1b (CCL4; NSNP = 12; βIVW = 0.03; 95% CI 0.018–0.045; pIVW = 5.7 × 10−6). Each doubling in odds of psoriasis was associated with a higher circulating level of TRAIL (TNFSF10; NSNP = 31; βIVW = 0.01; 95% CI 0.004–0.011; pIVW = 6.0 × 10−5). MR estimates from the weighted median and MR-Egger regression showed consistent directions (Figure 4), and MR-Egger did not suggest horizontal pleiotropy. Genetically predicted AD was not associated with circulating levels of any biomarkers tested.
MR estimates were obtained from the Wald ratio method (NSNP = 1), from the inverse variance weighted (IVW) fixed-effects model (IVW-FE, NSNP = 2 or 3), or from the IVW random-effects model (IVW-RE, NSNP > 3). *p < 0.05 and **p < 0.05/533. AD = Alzheimer disease; Crohn = Crohn disease; IBD = inflammatory bowel disease; MR = Mendelian randomization; MS = multiple sclerosis; RA = rheumatoid arthritis; UC = ulcerative colitis.
MR estimates were obtained from the Wald ratio method (NSNP = 1), from the inverse variance weighted (IVW) fixed-effects model (IVW-FE, NSNP = 2 or 3), or from the IVW random-effects model (IVW-RE, NSNP > 3). Red symbol indicates p < 0.05/695; larger blue symbol indicates 95% CI and does not include the null value. AD = Alzheimer disease; Crohn = Crohn disease; IBD = inflammatory bowel disease; IVW = inverse variance weighting; MR = Mendelian randomization; MS = multiple sclerosis; RA = rheumatoid arthritis; UC = ulcerative colitis; WM = weighted median.
Among the 46 biomarkers investigated, 35 had at least 1 genetic instrument for MR analysis. After accounting for multiple comparisons, each SD higher circulating MIP1b (CCL4) level was associated with a higher risk of IBD (NSNP = 4; ORIVW 1.14; 95% CI 1.07–1.21; pIVW = 2.4 × 10−5), but lower risk of RA (NSNP = 4; ORIVW 0.84; 95% CI 0.79–0.90; pIVW = 9.4 × 10−8). Each SD higher circulating IL1RA (IL1RN) level was associated with a lower risk of RA (NSNP = 4; ORIVW 0.80; 95% CI 0.72–0.90; pIVW = 3.3 × 10−5). MR estimates from the weighted median and MR-Egger regression showed consistent direction (Figure 4), and MR-Egger did not suggest horizontal pleiotropy. We identified 7 more significant causal associations between circulating inflammatory biomarkers and inflammatory diseases after Bonferroni correction; however, less than 3 genetic instruments were available for these biomarkers. Thus, sensitivity methods were not performed. Particularly, genetically predicted circulating levels of interleukin (IL) 6 (IL6) and IP10 (CXCL10) were associated with more than 1 inflammatory disease. Each SD higher circulating IL6 (IL6) level was associated with higher risk of MS (NSNP = 2; ORIVW 2.45; 95% CI 2.05–2.93; pIVW = 1.3 × 10−22), but lower risk of RA (NSNP = 1; ORWald 0.54; 95% CI 0.43–0.67; pWald = 1.3 × 10−8). Each SD higher circulating levels of IP10 (NSNP = 1) was associated with higher risk of psoriasis and IBD (IP10→psoriasis: ORWald 2.97; 95% CI 1.74–5.05; pWald = 6.3 × 10−5; IP10→IBD: ORWald 1.67; 95% CI 1.42–1.97; pWald = 1.3 × 10−9), but lower risk of Crohn disease and UC (IP10→Crohn: ORWald 0.50; 95% CI 0.38–0.66; pWald = 5.7 × 10−7; IP10→UC: ORWald 0.54; 95% CI 0.41–0.70; pWald = 5.4 × 10−6). We also found a reverse association between the circulating level of IL2ra (IL2RA) and the risk of MS (NSNP = 1; ORWald 0.82; 95% CI 0.76–0.89; pWald = 3.2 × 10−6). The observed effect sizes for all these associations were larger than the minimum effect detectable at 85% power (eTable 14, links.lww.com/WNL/C486). However, most of the null findings in our analyses had an observed effect size smaller than the minimum detectable effect (eTable 14).
From the above MR analyses for circulating inflammatory biomarkers and inflammatory diseases, we found a bidirectional relationship between circulating MIP1b (CCL4) level and risk of RA. MR-Steiger suggested that both directions were valid; however, the RA→MIP1b association was less susceptible to measurement error (eTable 15, links.lww.com/WNL/C486). eFigure 6 (links.lww.com/WNL/C487) demonstrates that the MIP1b→RA association was driven by a single SNP (prs11574435-RA = 0.004), and the association did not sustain after exclusion of this SNP. The RA→MIP1b association remained nominally significant after excluding 2 SNPs nominally associated with MIP1b (pIVW = 0.004).
Genetically predicted circulating inflammatory biomarker levels were not associated with the risk of AD after accounting for multiple comparisons. Nevertheless, the top signal identified for AD (circulating MIG [CXCL9] level [NSNP = 2; ORIVW 1.23; 95% CI 1.06–1.42; pIVW = 0.007]) also was a suggestive causal biomarker for Crohn disease (NSNP = 2; ORIVW 0.73; 95% CI −0.62 to 0.86; pIVW = 1.3 × 10−4). The observed effect sizes for these associations were larger than the minimum effect detectable at 85% power (eTable 14, links.lww.com/WNL/C486). However, colocalization did not suggest a common causal SNP for AD and Crohn disease within the genomic region (±50 kb from the MIG coding gene CXCL9) (posterior probability [PP.H4] = 0.001; eFigure 7, links.lww.com/WNL/C487). A common causal SNP was also unlikely for MIG and AD (PP.H4 = 0.03), although possible for MIG and Crohn disease (PP.H4 = 0.74; eFigure 8). Colocalization suggested neither a common causal SNP for IP10 (CXCL10) and AD (PP.H4 = 0.09), nor for IP10 (CXCL10) and Crohn disease (PP.H4 = 0.09; eFigure 9).
Discussion
In this study, we performed real-world observational analysis and 2-sample MR to investigate the relationship between inflammatory diseases and AD. Consistent with previous observational studies,4,5,26 our real-world observational analysis based on CPRD data showed that inflammatory diseases were associated with a higher risk of AD. However, these associations were not supported by 2-sample MR analysis, suggesting that confounding factors might be driving the observed association between inflammatory diseases and AD. In MR analysis, we found that several biomarkers are likely to be affected by inflammatory diseases. Moreover, we found evidence for a potential causal role of MIG in Crohn disease. However, consistent with the results of the observational study, we did not find an inflammatory biomarker that could explain the association of inflammatory diseases and AD.
The role of inflammation in AD progression is supported by the identification of innate immune genes in genetic studies on AD, such as CD33 and TREM2,19,27 and the emerging understanding that microglia, which contributes to the regulation of immune response in the brain, likely is central to the pathogenesis of AD.28 Chronic inflammation may drive neuroinflammatory changes and chronic activation of microglia, leading to oxidative stress and enhanced deposition of toxic proteins in AD.1 Chronic inflammation also increases the risk of thromboembolic events and ischemic stroke, contributing to the development of vascular dementia, which leads to stepwise instead of gradual cognitive decline.29 However, it is also possible that the observational associations between inflammatory diseases and AD are due to common molecular mechanisms.
Our findings for the relationships between circulating inflammatory biomarkers and AD were not entirely consistent with other recent studies with similar designs.30,31 Both Yeung et al. and Pagoni et al. used genetic associations of 41 biomarkers from the previous Finnish GWAS, which had adjusted for BMI.30,–,32 Several studies have reported a causal effect of obesity or higher BMI on inflammatory biomarker levels,33,34 but BMI was not a causal factor of AD or dementia.35,36 Using BMI-adjusted genetic associations of biomarkers could result in collider bias in 2-sample MR because of unmeasured confounding.37 In our study, we updated the Finnish GWAS by including 5 additional biomarkers (activePAI, IL1a, sEselectin, sICAM, and sVCAM) and excluding BMI from the genetic model. We also took advantage of 2 additional GWASs to increase the sample size of 20 biomarkers (the maximum sample size was increased from 8,293 to 13,365).16
Our MR findings suggested a link between MIG and higher risk of AD, but lower risk of Crohn disease. MIG belongs to the CXC chemokine family and is also known as chemokine (C-X-C motif) ligand 9 (CXCL9). Chemokines have been found to involve in both neuroinflammatory and neurodegenerative processes and play a role in the development of Aβ plaques and neurofibrillary tangles, 2 pathologic hallmarks of AD.38 However, colocalization in our study only supported common causal SNPs for MIG and Crohn disease, indicating that the association between MIG and AD may be confounded by linkage disequilibrium. We previously found that circulating MIG causally affects the IP10 level.16 MIG (CXCL9), IP10 (CXCL10), and IP9 (CXCL11) share the same receptor, C-X-C chemokine receptor 3 (CXCR3),38 but colocalization in our study did not support common causal SNPs for IP10 and Crohn disease. This may suggest a more prominent role of MIG in Crohn disease. CXCL9 polymorphisms were found associated with Crohn disease, although only in pediatric patients.39,40 In addition, a recent study also identified a major role of MIG (CXCL9) in age-related chronic inflammation.41 Taken together, we did not find strong evidence supporting common inflammatory biomarkers affecting both risk of AD and inflammatory diseases. Our findings suggest that neuroinflammatory processes that are responsible for AD may be distinct from the mechanisms underlying inflammatory diseases. Nevertheless, the role of inflammatory biomarkers not being investigated in this study warrant further exploration.
There are limitations to our study. First, the observational analysis relies on the quality of data recording in the electronic health records. Some chronic autoimmune inflammatory diseases (e.g., RA, psoriasis) could be under-recorded, confounding estimates of their contributions to the risk of AD. The use of CPRD Medcodes also may provide a poor estimate of the proportion of patients with AD because nonspecific Medcodes are often used in primary care for suspected dementia and these can be misapplied.42 In addition, the diagnosis of AD was not confirmed by CSF biomarkers such as Aβ and tau protein. On the other hand, one systematic review assessed the validity of dementia recording in EHR and reported high false-positive rates of dementia recordings.43 However, recordings of AD diagnosis have improved in recent years, 44 and we used the linkage data set, which may provide additional patient records of AD diagnosis. Second, genetic factors explain a small proportion of the variance in the incidence of some inflammatory diseases.45,46 Thus, a lack of association between genetically predicted inflammatory diseases and risk of AD in our MR analysis may not rule out the contribution of inflammatory diseases to AD pathophysiology. Third, genetic associations for biomarkers were based on relatively small GWAS, which may raise the issue of lack of statistical power. Nevertheless, the observed effect sizes for our findings were larger than the minimum effect detectable at 85% power. Fourth, univariate MR analyses may not represent the direct effect of a specific biomarker on the disease outcomes because effects of one biomarker may be mediated by others in the complex biomarker network. Fifth, we primarily investigated peripheral inflammatory diseases; thus, our findings may not be able to inform on inflammation related to the CNS. In addition, circulating inflammatory biomarkers may not be relevant to neurologic diseases. However, it has been reported that some biomarkers can pass through blood-brain barrier, including IL1RA in our findings,47 and others can affect the integrity of blood-brain barrier, for example, MCP1.48 Monocytes can cross compromised blood-brain barrier and generate monocyte-derived macrophages.49 This may further suggest the synergistic effect of multiple biomarkers. Last but not least, all our analyses were based on individuals of European ancestry; generalization to other population requires further investigation.
In this study, we have integrated evidence from real-world observational analyses based on large sample size with a 2-sample MR analysis, which is less susceptible to confounding and reverse causation, to investigate the associations between chronic autoimmune inflammatory diseases and risk of AD and dementia. Our findings suggested that these associations were noncausal, but the factors confounding the observation have yet to be unraveled. Moreover, although some circulating inflammatory biomarkers were associated with the inflammatory diseases, evidence was weak that they affected both risk of AD and inflammatory diseases.
