FormalPara Key Summary Points

Why carry out this study?

Analyses of HbA1c levels are more consistent with the long-term effects of exposure on outcomes in MR studies. This study aimed to conduct a two-sample MR study to access the potential effect of genetically predicted levels of HbA1c on OA

What was learned from the study?

This study investigated the causal association between HbA1c and OA risk using a two-sample MR analysis based on a large GWAS summary statistics database. The current study supported a causal effect of genetically increased HbA1c on knee OA risk. Our study suggested that clinicians should timely manage the HbA1c levels in DM patients to avoid the negative effects of long-term high HbA1c levels on their joints

Introduction

Osteoarthritis (OA) is one of the most important chronic diseases worldwide, which is the leading cause of disability and negatively affects the quality of life in patients [1]. According to a systematic analysis of the Global Burden of Disease Study, more than 303 million people reported limited mobility due to OA, of which 263 million had knee OA [2]. The risk of OA is related to multiple factors. Older age, female gender, obesity, previous joint injury (e.g., ankle fracture or knee ligament tear), and abnormal joint anatomy (e.g., shallow acetabulum) are all associated with an increased risk of OA [1]. Furthermore, previous studies have reported that some chronic diseases such as diabetes mellitus (DM) are also linked to a higher risk of OA [3,4,5].

The negative effects of DM on OA may be related to the increase of oxidative stress and proinflammatory cytokines, as well as exposure to chronic high glucose concentrations [6]. Several studies have investigated the relationship between DM or hyperglycemia and OA, but no consistent conclusions have been reached [7,8,9,10]. A meta-analysis demonstrated that type 2 DM may be significantly associated with the development of OA [11]. However, an updated meta-analysis showed that DM was not an independent risk factor for OA, which may be related to body mass index (BMI) being the most important confounder [12]. Inconsistent results may be influenced by many factors, such as different study design types (cohort or cross-sectional), adjustment for confounders, location of OA (knee, hip, or hand), and diagnostic indicators (fasting blood glucose, postprandial 2 h blood glucose, or glycated hemoglobin). Existing observational studies cannot address confounding by various biases and reverse causality inference, and are limited by small sample sizes.

Mendelian randomization (MR) uses genetic variants strongly associated with exposure to determine the causal relationship between exposure and outcome, and it avoids reverse causality inference, reflecting the long-term effect of exposure on outcome [13]. A previous MR study between DM and OA did not find a causal relationship between genetically increased fasting and 2-h postprandial blood glucose levels with hip and knee OA risk [14]. However, their study did not analyze glycated hemoglobin (HbA1c) as exposure to OA. HbA1c can better reflect blood glucose levels over longer periods compared with fasting blood glucose and 2-h postprandial blood glucose [15]. Therefore, analyses of HbA1c levels are more consistent with the long-term effects of exposure on outcomes in MR studies. This study aimed to conduct a two-sample MR study to access the potential effect of genetically predicted levels of HbA1c on OA.

Methods

Study Design and Data Source

The study design is shown in Fig. 1. This study utilized a two-sample MR method to explore the causal effect of HbA1C on OA. Three assumptions are to be tested in this MR design: (1) the selected genetic variants are related to HbA1C levels; (2) these genetic variants are not associated with confounders (BMI [16], bone mineral density (BMD) [17], and tobacco smoking [18]); (3) these genetic variants are related to OA only via HbA1C. Summary single-nucleotide polymorphism (SNP)-phenotype association data were obtained from genome-wide association studies (GWAS) in the Integrative Epidemiology Unit (IEU) OpenGWAS database (https://gwas.mrcieu.ac.uk/datasets/). The OpenGWAS database is an open-access cloud-based data infrastructure that currently contains 126 billion genetic associations from 14,582 complete GWAS datasets, representing a diverse range of human phenotypes in diverse populations and disease outcomes [19]. HbA1c was measured from fasting or nonfasting whole blood using National Glycohemoglobin Standardization Program (CGSP)-certified methods [20]. The identification of OA patients was based on the patients who have been diagnosed with OA through the hospital as recorded in the database. [21]. Each study was approved by the respective institutional ethics review board and informed consent of the participants was obtained. Therefore, this study was exempt from institutional review board approval of the Wuhan No.1 Hospital because this study was based on a secondary analysis of the original study. All study conduct adhered to the tenets of the Declaration of Helsinki of 1964 and its later amendments.

Fig. 1
figure 1

Mendelian randomization model of glycated hemoglobin (HbA1c) and risk of osteoarthritis (OA)

Full size image

Instrument Variables

SNPs associated with exposure (HbA1c), outcome (OA), and confounders (BMI, BMD, and tobacco smoking) are summarized in Table 1. SNPs related to HbA1C are obtained from a larger meta-analysis of GWAS including a total of 46,368 participants [20]. OA was divided into overall OA, knee OA, and hip OA, and the SNPs associated with them were derived from the same GWAS study based on UK Biobank data [21]. Confounders in this study including BMI, heel BMD, forearm BMD, and tobacco smoking, and the SNPs related to them were obtained from the IEU OpenGWAS database. A series of quality control steps were performed to select instrumental variables. First, SNPs with genome-wide significant (P  < 0.00000005) were selected. Then, SNPs without linkage disequilibrium (defined by r2 < 0.001 and clump distance > 10,000 kb) were used as instrument variables. Finally, SNPs with minor allele frequency (MAF) < 0.01 and palindromic SNPs were excluded.

Table 1 Characteristics of studies and number of genetic instruments used for variant outcomes estimates in the in Mendelian randomization (MR) analysis
Full size table

Pleiotropy Analysis

For the instrument variables selected, pleiotropy needs to be removed and the exclusion restriction assumptions (assumptions 2 and 3). Genetic variants should be strongly associated with exposure, independent of confounders associated with exposure and outcome, and influence outcome only through exposure. SNPs that were directly related to the OA [P < 0.017 = 0.05/3 (one exposure and three outcomes)] and not through HbA1c were excluded (assumption 3). SNPs associated with confounders (BMI, heel BMD, forearm BMD, and tobacco smoking) were excluded (assumption 2). The intercept and P-value of MR–Egger regression were calculated to test for pleiotropic bias. MR–PRESSO analysis was utilized to conform to the horizontal pleiotropic outliers [22].

Statistical Analysis

Two-sample MR analysis was performed to estimate the effect of HbA1c on OA using summarized data of SNP-HbA1c and SNP-OA associations. The inverse-variance-weighted (IVW) method based on the fixed-effects model is the main analytical method used in this study to generate a causal effect estimate. The IVW method is based on the Wald ratio method to obtain an unconfounded estimate of the exposure of genetic prediction outcomes [23]. The weighted-median method provided robust and consistent estimates of outcomes, even though nearly 50% of the genetic variants were invalid instruments [24]. The MR–Egger method used weighted linear regression to detect broad horizontal pleiotropy. However, MR–Egger results often exhibit low precision and may be affected by abnormal genetic variation. The leave-one-out analysis was performed to assess the robust association of SNPs with HbA1c levels, and the symmetry in the resulting figure represents no pleiotropy. For testing the results, Cochran’s Q-test was conducted to evaluate the statistical heterogeneity between SNPs in the IVW method, with P < 0.05 considered significantly heterogeneous. Furthermore, we calculated the F-statistics and variance explained (R2) for each exposure was calculated to assess the instrument variable strength, and F-statistics > 10 were considered to imply adequate instrument strength [25]. An online tool, mRnd, was used to calculate the power to detect causal effect. The equation of the mRnd tool used an approximate linear model applied to the binary outcomes, and the parameters calculated by the equation included the sample size, the effect size of exposure on outcome at the epidemiological level, the standard deviation of exposure and outcome, and the proportion of phenotypic variation explained by instrument variable SNPs [26]. Power calculation demonstrated that the sample provided sufficient statistical power (≥ 88%) for causal analysis of HbA1c on knee and overall OA (Supplementary Table 1). All statistical analyses were completed in R 4.1.1 software (TwoSampleMR package), and the observed two-sided P < 0.017 was considered statistically significant.

Results

Instrument Variables Selection and Pleiotropy

A total of 244 SNPs associated with HbA1c were selected with genome-wide significance (P < 5 × 10–8) and no linkage disequilibrium. Table 2 presents the results of the pleiotropy analysis of the SNPs. A total of 221 SNPs overlapped between SNP-HbA1c and SNP-OA. After excluding palindromic SNPs and SNPs with MAF, 175 SNPs were retained. After removing SNPs that directly affect OA but not through HbA1c (assumption 3), the numbers of SNPs associated with knee OA, hip OA, and overall OA were 169, 168, and 166, respectively. SNPs associated with confounders were excluded (assumption 2), and the numbers of SNPs associated with knee OA, hip OA, and overall OA through HbA1c remained 148, 147, and 146, respectively. The MR–Egger analysis demonstrated no pleiotropic bias in the remaining SNPs.

Table 2 Selection of SNPs and pleiotropy analysis
Full size table

Two-sample MR Analysis for Causal Link of HbA1c with OA

Figure 2 presents the MR estimates for different methods to assess the causal effect of HbA1c on OA. The inverse-variance-weighted indicated that genetically predicted HbA1c was related to an increased risk of knee OA [odds ratio (OR) = 1.561; 95% confidence interval (CI), 1.110–2.197; P = 0.011]. The MR–Egger analysis demonstrated that genetically predicted HbA1c was suggestively associated with an increased risk of knee OA (OR = 3.175; 95% CI 1.080–9.315; P = 0.037). However, the associations of HbA1c with the hip OA (OR = 1.073; 95% CI 0.674–1.710; P = 0.766) and overall OA (OR = 1.141; 95% CI 0.904–1.441; P = 0.804) were not found.

Fig. 2
figure 2

The Mendelian randomization (MR) estimates for different methods to assess the causal effect of glycated hemoglobin (HbA1c) on osteoarthritis (OA)

Full size image

Table 3 presents the results of the instrumental variable strength and heterogeneity test. There was a strong association between SNPs and HbA1c (F = 21.138), and no evidence of heterogeneity on the inverse-variance-weighted analysis (Q = 150.625, P = 0.402). Furthermore, the leave-one-out analysis showed that there were no potential SNPs affecting the causal link and the results were stable (Supplement Fig. 1).

Table 3 The results of the instrumental variable (IV) strength and heterogeneity test
Full size table

Discussion

The association between DM and OA is controversial in the studies that have been reported. This study used summary statistics from GWASs to explore the causal association between HbA1c and OA. The results found that HbA1c causally increased knee OA, whereas the estimated effect of HbA1c on hip OA and overall OA were not significant, suggesting that HbA1c had different causal effects on OA at different sites.

Previous studies have demonstrated that type 2 DM was associated with an increased risk of OA [3, 4]. Dubey et al. found that type 1 and type 2 DM were closely related to knee OA [8]. Eymard et al. suggested that type 2 DM was significantly linked to reduced joint space in men with knee OA after adjusting for confounders [9]. A cohort study showed that high levels of HbA1c and fasting serum glucose were positively related to a higher risk of knee OA [5]. However, the conclusions of two meta-analyses on the relationship between DM and OA were inconsistent [11, 12]. The meta-analysis of Williams et al. found that DM was associated with the development of OA [11], whereas Khor et al. indicated that DM was not a risk factor for OA [12]. Previous observational and interventional studies on the association of HbA1c with OA risk may be limited to drawing a causal relationship. The current study explored the causal relationship between HbA1c and OA risk using MR analysis. Our results suggested a causal relationship between high HbA1c levels and the risk of knee OA. However, a previous MR study found no significant causal effects of genetically increased type 2 DM risk, fasting glucose, and 2-h postprandial glucose on hip OA and knee OA [14]. A possible reason for the inconsistency between our results and the previous study was that the study by Cui et al. focused on fasting glucose and 2-h postprandial glucose, while our study focused on HbA1c. Compared with fasting glucose and 2-h postprandial glucose, HbA1c can better reflect blood glucose levels in a relatively long period of time [15]. Furthermore, some confounders have been reported that may affect the accuracy of the results. Hartley et al. found a causal effect between high BMD levels and an increased risk of OA [17]. A large MR study based on the UK Biobank demonstrated that BMI had a major causal effect on OA risk at weight-bearing joints [16]. Our study excluded some SNPs related to confounders such as BMI, BMD, and tobacco smoking, which ensured the reliability of the results.

OA is related to a local and systemic low-grade inflammation state [6]. Some studies have reported the possible mechanisms between blood glucose and OA. The effect of type 2 DM on OA may be through the following pathways: high blood glucose exposure shows local toxicity to joint tissue, increasing local oxidative stress, cytokine, proteolytic enzyme production, and advanced glycation end products (AGEs) accumulation [6, 27, 28]. High intra-articular glucose levels promote cartilage degradation by increasing the expression of cyclooxygenase-2 (COX-2), and matrix metalloproteinase 13 (MMP-13), while reducing collagen type II and peroxisome proliferator-activated receptor gamma (PPARγ) synthesis [4]. High-glucose environment inhibited autophagy in chondrocytes and the activity of hemooxygenase-1, which has important antioxidant properties [29]. This results in increased oxidative stress and cartilage damage, and promotes the secretion of potent proinflammatory mediators such as prostaglandin E2 (PGE2), interleukin-1β (IL-1β) IL-6, IL-10, and tumor necrosis factor-α (TNF-α) [30]. Furthermore, the accumulation of AGEs can adversely affect cartilage turnover and mechanical properties [31].

In this study, we selected instrument variables to examine the causal relationship between HbA1c and OA through a series of quality control steps, such as selecting SNPs with the genome-wide association and independent inheritance without any LD, and excluding SNPs related to confounders. To ensure that the results were robust and reliable, sensitive analyses were performed and F-statistics were used to assess the strength of the association between the SNPs and exposure. However, some limitations of our study need to be considered. First, although some of the reported confounders such as BMI, BMD, and tobacco smoking were excluded from our study, there may be some potential confounders that could have influenced the results. Second, we were unable to estimate the degree of overlap of participants in the studies, since the exposure and outcome studies used in two-sample MR analysis should not involve overlapping participants. However, sample overlap bias can be minimized by strong instrumental variables. Third, due to database limitations, this study did not analyze the relationship between HbA1c and OA, and the relationship between HbA1c levels and non-weight-bearing joint OA needs to be explored.

Conclusions

This study investigated the causal association between HbA1c and OA risk using a two-sample MR analysis based on a large GWAS summary statistics database. The current study supported a causal effect of genetically increased HbA1c on knee OA risk. Our study suggested that clinicians should timely manage the HbA1c levels in DM patients to avoid the negative effects of long-term high HbA1c levels on their joints.