Type 2 diabetes mediates the causal relationship between obesity and osteomyelitis: A Mendelian randomization study

Abstract

Mendelian randomization (MR) analysis was used to determine the causal relationship between Type 2 diabetes (T2D) and osteomyelitis (OM). We performed MR analysis using pooled data from different large-scale genome-wide association studies (GWAS). Instrumental variables were selected based on genome-wide significance, instrumental strength was assessed using F-values, and thresholds for the number of exposed phenotypes were further adjusted by Bonferroni correction. univariable and multivariable MR analyses were performed to assess causal effects and proportions mediated by T2D. IVW (inverse variance weighting) showed a significant genetic effect of osteomyelitis on the following: After correction by Bonferroni, univariable analyses showed that childhood body mass index (BMI) was not significantly associated with genetic susceptibility to OM [odds ratio (OR), 1.26; 95% confidence interval (CI), 1.02, 1.55; P = .030], not significantly associated with adulthood BMI (OR, 1.28; 95% CI, 1.02, 1.61; P = .034), significantly associated with waist circumference (OR, 1.84; 95% CI, 1.51, 2.24; P < .001), and significantly associated with hip circumference (OR, 1.52; 95% CI, 1.31, 1.76; P < .001). Meanwhile, multivariable analyses showed no significant effect of childhood BMI on OM (OR, 1.16; 95% CI, 0.84, 1.62; P = .370), no significant effect of adulthood BMI on OM (OR, 0.42; 95% CI, 0.21, 0.84; P = .015), a significant association between waist circumference and OM (OR, 4.30; 95% CI, 1.89, 9.82; P = .001), T2D mediated 10% (95% CI, 0.02, 0.14), and no significant association between hip circumference and OM (OR, 1.01; 95% CI, 0.54, 1.90; P = .968). Our study provides evidence for a genetically predicted causal relationship among obesity, T2D, and OM. We demonstrate that increased waist circumference is positively associated with an increased risk of OM and that T2D mediates this relationship. Clinicians should be more cautious in the perioperative management of osteomyelitis surgery in obese patients with T2D. In addition, waist circumference may be a more important criterion to emphasize and strictly control than other measures of obesity.

1. Introduction

Obesity has become a worldwide pandemic and is strongly associated with many health problems, significantly increasing the risk of various diseases such as Type 2 diabetes (T2D), myocardial infarction, stroke, dementia, and several types of cancers, leading to reduced quality of life and life expectancy.^[1] Childhood obesity tends to persist into adulthood, suggesting that childhood obesity is associated with adverse health outcomes throughout the life course.^[2] However, it is difficult to determine whether early obesity has an independent and long-lasting effect on disease risk, or whether its effects are mediated exclusively through late-life obesity. If the latter is the case, the potentially adverse consequences of childhood obesity may be avoided by attaining and maintaining a healthy weight in adulthood. Mendelian studies have suggested that the possible positive association between early obesity and risk of disease in adulthood can be attributed to individuals maintaining a larger body size in later life. However, smaller body sizes in childhood may still increase the risk of disease.^[3] In addition, body mass index (BMI) may be biased to represent obesity, so waist circumference and hip circumference are also used to represent obesity.^[4]

OM is a destructive bone tissue disease caused by infection by pathogenic microorganisms, characterized by chronic inflammation and infection in the bone.^[5,6] It is predominantly treated with antibiotics in combination with surgery, but its treatment is characterized by high difficulty, long duration, a high failure rate, and a high recurrence rate,^[7] making it one of the most challenging diseases faced by orthopedic surgeons. A study performed in the United States identified a prevalence of OM of 22 cases per 100,000 people, and found that the incidence of OM in patients with diabetes mellitus has increased substantially.^[8] Diabetic foot ulcers cause OM and are associated with amputation.^[9,10]

Mendelian randomization (MR) uses genetic variants that are closely associated with exposure to determine the causal relationship between exposure and outcome, reflecting the effective long-term effects of exposure on outcome. In this approach, given that genetic variants are randomly assigned at conception, outcome reliability is greatly increased, along with minimization of the influence of external confounders on outcome, avoidance of the various biases of observational studies in epidemiology and the confounding of reverse causality inference, and the provision of some immunity to small-sample limitations. In the MR analysis performed in this study, we aimed to determine whether obesity has a causal effect on OM and whether it is mediated by diabetes mellitus.^[11]

2. Methods

2.1. Data sources

In our analysis, we used pooled data obtained from published and publicly available genome-wide association studies (GWAS). We used childhood and adulthood BMI, waist circumference, and hip circumference to represent obesity, and obesity-related single nucleotide polymorphisms (SNPs) based on adulthood BMI from the GIANT Consortium, which analyzed a large sample of 236,781 individuals of European ancestry. Childhood BMI was obtained from the ECG Consortium, which analyzed a large sample of 39,620 European pedigrees, and waist and hip circumferences were obtained from the UK Biobank, which analyzed a large sample of 462,166 European pedigrees for waist circumference and a large sample of 462,117 European pedigrees for hip circumference. To examine the relationship between OM and genetics, we obtained data from the UK Biobank, which included 4836 cases of OM and 486,484 controls. GWAS summary statistics for T2D were obtained from the UK Biobank, the dataset of which used in our analyses consisted of data on 61,714 cases and 1178 controls.

2.2. SNPs in exposure and outcome selection

The genome-wide significance parameter for exposure instrumental variables (IVs) SNPs was set to P < 5e − 8. A linkage disequilibrium test was performed on IV-associated SNPs set to kb > 10 MB (R² < 0.0001) to ensure mutual independence of the selected genetic variants and to exclude palindromic SNPs with intermediate allele frequency. Significant associations with OM (P < 5e − 5) for the IVs were excluded. Steiger test showed no reverse causality for instrumental variables.^[12] F-value = (n-k-1/k)(R²/1-R²),^[13] where N = GWAS sample size and K = number variant instruments. In principle, an F-value >10 was chosen for the analysis, indicating that there is less likelihood of weak instrumental variable bias.^[14] R² = 2 × β2 × (1-EAF) × EAF,^[15] where EAF represents the effect allele frequency and β represents the estimated genetic effect per SNP. R² reflects the extent to which the instrumental variable explains exposure.

2.3. Statistical analysis

2.3.1. Univariable MR

In our study, inverse variance weighting (IVW) was the main univariable analysis method, and each valid SNP based on the Wald ratio method was calculated to provide consistent estimates.^[16] MR-Egger,^[17] weighted median,^[18] and weighted mode methods^[19] were used as complementary methods to ensure reliable and robust conclusions. The MR-Egger regression provided consistent estimates when 100% of genetic variation was for invalid IVs, and weighted median provided robust and consistent effect estimates when 50% of inheritance was for valid IVs. In terms of efficiency, weighted median estimates are generally almost as accurate as IVW estimates; both are substantially better than MR-Egger estimates, which are less precise and susceptible to external genetic variation. MR-Egger regression estimates are particularly imprecise when all IVs are associated with exposure to similar magnitudes.^[17,20]

Heterogeneity was tested using the Cochran Q heterogeneity test^[21] and the IVW and MR-Egger methods. When the Cochran Q heterogeneity test satisfied P < .05, it indicated that there was heterogeneity among SNPs; when P > .05, there was no heterogeneity among SNPs.^[22] Sensitivity analysis was performed using leave-one-out analysis to explore the effect of individual SNPs on causal associations. Horizontal pleiotropy analysis was performed using the MR_pleiotropy_test function, which indicated the presence of pleiotropy when P < .05. CAUSE modeled correlated and uncorrelated levels of pleiotropy estimated the posterior distributions of null, sharing, and causal effects of exposure, and compared the model fits by Expected Log Pointwise Posterior Density to avoid false positives that would occur due to levels of pleiotropy using other methods.^[23]

Bilateral P values < .05 were considered significant. We further adjusted the threshold for the number of exposure phenotypes by Bonferroni correction. Thus, the threshold for statistical significance for the 4 obesity exposures in IVW was set at P < .05/4 = 0.013. All statistical analyses were performed using the MR-PRESSO (version 1.0), 2-sample MR packages,^[24] and MR-CAUSE packages^[23] in R software (8.1.4).

2.3.2. Multivariable and mediation MR

In our study, we utilized multivariable MR to investigate the causal effect of obesity on the risk of OM. In Mendelian studies, there are often multiple correlated independent variables, and LASSO regression can reduce the effect of multicollinearity on the regression results by making the coefficients of the correlated independent variables zero through the correlation between the independent variables.^[25] Two-step MR mediation analysis was performed using a 2-step coefficient product method.^[11] Here, in the first step, the causal effect of obesity (exposure) on the risk of T2D (mediation) was determined, and in the second step, the causal effect of T2D (mediation) on OM (outcome) was determined. Then, the indirect effect through T2D as a mediation was obtained by taking the product of the effects of steps 1 and 2 (Fig. 1). Causal estimates for both steps were obtained using the IVW method.

Figure 1.

Visual representation of the relationship between obesity and osteomyelitis mediated by T2D. (A) Represents the regression coefficient of the relationship between obesity and T2D, (B) represents the regression coefficient of the relationship between T2D and osteomyelitis, (C) represents the total effect of the relationship between obesity and osteomyelitis, and (C’) represents the direct effect of obesity on osteomyelitis, taking into account the adjustments of T2D on osteomyelitis. T2D = Type 2 diabetes.

3. Results

3.1. Causal effect of obesity on OM (univariable)

In univariable MR, all F-values in this study IVs had a statistical range of more than 10, indicating no weak instrumental bias. Childhood BMI, adulthood BMI, waist circumference, and hip circumference were used as SNPs for inclusion in the analysis. After correction by Bonferroni, childhood BMI was not significantly associated with genetic susceptibility to OM [odds ratio (OR), 1.26; 95% confidence interval (CI), 1.02, 1.55; P = .030], and was not significantly associated with adulthood BMI (OR, 1.28; 95% CI, 1.02, 1.61; P = .034), significantly associated with waist circumference (OR, 1.84; 95% CI, 1.51, 2.24; P < .001), and significantly associated with hip circumference (OR, 1.52; 95% CI, 1.31, 1.76; P < .001) (Fig. 2). Summary results of the MR Method (Supplementary Table 1, http://links.lww.com/MD/M560). MR Visualization Supplement(Supplementary Data sheet 1, http://links.lww.com/MD/M564, http://links.lww.com/MD/M565, http://links.lww.com/MD/M566, http://links.lww.com/MD/M567, http://links.lww.com/MD/M568, http://links.lww.com/MD/M569, http://links.lww.com/MD/M570, http://links.lww.com/MD/M571, http://links.lww.com/MD/M572, http://links.lww.com/MD/M573, http://links.lww.com/MD/M574, http://links.lww.com/MD/M575, http://links.lww.com/MD/M576, http://links.lww.com/MD/M577, http://links.lww.com/MD/M578, http://links.lww.com/MD/M579). CAUSE analysis (Supplementary Table 2, http://links.lww.com/MD/M561) and found that waist circumference remains an independent cause of OM and that the causal waist circumference model OM is superior to the shared waist circumference model OM. IVW is the most common and robust method of analysis, and despite our Bonferroni correction, the P value for IVW is still <.013 after correction, which can be considered as a positive result.

Figure 2.

Effect of exposure on the likelihood of developing osteomyelitis in univariable and multivariable models.

3.2. Mediation effect of T2D on OM via obesity (multivariable)

After adjustment for single risk factors and in a fully adjusted model that considered all risk factors in a multivariable IVW analysis, which ultimately included 159 SNPs. After correction by Bonferroni BMI had no significant effect on OM in children (OR, 1.16; 95% CI, 0.84, 1.62; P = .370) or on OM in adults (OR, 0.42; 95% CI, 0.21, 0.84; P = .015), waist circumference was significantly associated with OM (OR, 4.30; 95% CI, 1.89, 9.82; P = .001), the proportion mediated by T2D was 10% (95% CI, 0.02, 0.14) (Supplementary Table 3, http://links.lww.com/MD/M562), and hip circumference was not significantly associated with OM (OR, 1.01; 95% CI, 0.54, 1.90; P = .968) (Fig. 2). In the LASSO regression analysis of this study (Supplementary Table 4, http://links.lww.com/MD/M563), the coefficients of the assessment beta do not have the value of 0, that is, there is no collinearity.

4. Discussion

In this study, we investigated the genetic predictive causal relationship between obesity and OM using large-scale GWAS summary data. In univariable analysis, GWAS summary data on childhood BMI and adulthood BMI representing obesity had no significant causal relationship with OM, while GWAS summary data on waist circumference and hip circumference had significant causal relationships with OM. Specifically, greater waist and hip circumferences were associated with a higher likelihood of suffering from OM. Meanwhile, in multivariable analyses, there was no significant correlation between childhood BMI, adulthood BMI, and hip circumference SNPs with OM. In addition, waist circumference had a significant genetic predictive relationship. Specifically, a greater waist circumference may increase the likelihood of OM, and it was suggested that the risk effect of obesity on OM is mediated by T2D mellitus.

The mechanisms by which obesity and diabetes mediate the increased risk of OM remain unclear, but may be related to the following. First, obesity is associated with altered immune responses^[26] and chronic inflammation,^[27] which may increase the risk of infection. A large prospective cohort study with a 7-year long-term follow-up period demonstrated that obesity increased the risk of hospitalization for abdominal infections, reproductive and urinary tract infections, skin and soft-tissue infections, OM, and necrotizing fasciitis. OM is one of the diseases with the greatest increase in risk in association with obesity. OM is the most common form of obesity.^[28] The pathogen most commonly causative of OM is Staphylococcus aureus, and obese individuals and those with T2D have been reported to have lower serum staphylococcal IgG antibody titers. This reduced level of anti-S aureus antibodies is a potential mechanism leading to increased susceptibility to S aureus infection in obese and T2D patients.^[26] Moreover, experiments in mice have demonstrated increased severity of S aureus infections in obese and T2D groups. This was associated with increased abscess formation, increased fibrin deposition, and significant adaptive upregulation of surface fibrin (pro)-binding proteins (especially coagulation factor A) by S aureus.^[29] Second, Derek et al found that increased BMI was the factor conferring the greatest risk of OM, resulting from the inclusion of 90 open fractures. They thus concluded that obesity was associated with impaired angiogenic histiocyte response to peripheral injury and the need for higher doses of antimicrobials.^[30,31] In obese individuals, impaired angiogenic progenitor cells produce fewer cells and exhibit lower adhesion. This reduces vascular regeneration and compromises wound healing.^[32] Third, mechanical stress in obese individuals also increases the burden on soft tissue. Obese individuals experience increased loads, especially on their bones and joints, potentially conferring a risk of bone damage. At sites with such damage, there is a tendency for an inflammatory response to be triggered, which may in turn lead to OM. Although BMI has advantages in assessing generalized obesity, it does not take into account the heterogeneity of sites where body fat is distributed and has limited accuracy in diagnosing obesity.^[33] Studies have shown that the correlations of waist circumference and BMI with the risk of developing cancer strongly overlap across all cancers, confirming that waist circumference can also be used to estimate disease-related risk.^[34] In fact, a large number of studies have shown that, in some specific diseases, waist circumference is a better fit than BMI for adverse events.^[4,35] Regarding the reasons for this, greater waist circumference is associated with more visceral fat deposition, which tends to cause adipocyte development, and adipocyte enlargement and dysfunction. Meanwhile, adipose tissue and peripheral stromal vascular cells can secrete adipokines (e.g., adiponectin, leptin, and resistin) and inflammatory factors (e.g., interleukin-6 and tumor necrosis factor-α), as well as fibrinogen activator imprinted agent-1, which cause a series of harms.^[36–38]

Despite the various strengths of this work, this study has several limitations. First, the results of other MR methods (MR-Egger, weighted median, and weighted mode) were not fully consistent with those of the IVW method. However, since MR analysis is primarily an IVW method, IVW results were preferred in the absence of pleiotropy. Second, the selection of SNPs in GWAS data may increase the sample overlap between exposure and outcome, thus biasing the outcome; however, the selection of GWAS data from as many different samples as possible and an F-value much >10 can minimize sample overlap.^[39] Third, 1 MR analysis on OM is already reported in the literature, but that work did not consider setting major depression as a confounder because the investigator exposure was set to a P1 value of < 5e − 7, a more lenient standard than the 1 set in the present study.^[40] Fourth, owing to the limitations of the GWAS database, the main study was conducted in a European population, so the findings may not be generalizable to other populations. Studies on a greater range of ethnicities are needed to validate the obtained results.

5. Conclusion

Our study provides evidence for a genetically predicted causal relationship among obesity, T2D, and OM. We demonstrate that increased waist circumference is positively associated with an increased risk of OM, with T2D playing a mediating role in this relationship. As such, clinicians should be more cautious in the perioperative management of OM surgery in obese patients with T2D. In addition, waist circumference may be a more important criterion to emphasize and strictly control than other measures of obesity.

Author contributions

Conceptualization: Heng-zhi Liu, Jie Liang, Ai-Xin Hu.

Data curation: Heng-zhi Liu.

Software: Heng-zhi Liu.

Supervision: Jie Liang, Ai-Xin Hu.

Visualization: Heng-zhi Liu.

Writing – original draft: Heng-zhi Liu, Jie Liang.

Writing – review & editing: Heng-zhi Liu, Ai-Xin Hu.

Supplementary Material

medi-103-e38214-s001.doc ^{(41KB, doc)}

medi-103-e38214-s018.doc ^{(30.5KB, doc)}

medi-103-e38214-s019.doc ^{(16.5KB, doc)}

medi-103-e38214-s020.doc ^{(17.5KB, doc)}