Abstract
Observational studies link matrix metalloproteinases (MMPs) and tissue inhibitors of metalloproteinases (TIMPs) to ankylosing spondylitis (AS), but causality is uncertain. We performed a bidirectional, 2-sample Mendelian randomization (MR) analysis using genetic variants to assess causal relationships between genetically predicted circulating levels of 14 MMPs and 4 TIMPs and AS risk, and to evaluate reverse causation. Genetic instruments for circulating MMPs and TIMPs were derived from a large proteogenomic genome-wide association study of 5368 individuals. Summary statistics for AS were obtained from the FinnGen Release 12 GWAS (3838 cases; 353,224 controls). Causal estimates were generated using inverse-variance weighting (primary analysis), complemented by MR-Egger, weighted median, and weighted mode methods. Sensitivity analyses included Cochran’s Q test, MR-Egger intercept, MR-PRESSO, and leave-one-out analyses. Forward MR identified a nominal association between genetically predicted higher circulating MMP13 levels and an increased risk of AS (IVW odds ratio = 1.223, 95% confidence interval, 1.048–1.428; P = .011, FDR = 0.198). This association showed a consistent direction in the weighted median estimates and showed no evidence of heterogeneity or horizontal pleiotropy. No other MMPs or TIMPs showed convincing evidence of an association with AS. Reverse MR analyses provided no evidence that genetic liability to AS causally influences circulating MMP or TIMP levels. This bidirectional MR analysis provides genetic support for a potential causal relationship between MMP13 and the development of AS, while suggesting that other MMPs and TIMPs are unlikely to be causal drivers. Further experimental and clinical studies are warranted to clarify the biological relevance of MMP13 in AS.
Keywords: ankylosing spondylitis, causality, genome-wide association studies, Matrix metalloproteinase, Mendelian randomization
1. Introduction
Ankylosing spondylitis (AS) is a chronic, progressive, and systemic inflammatory disorder primarily affecting the axial skeleton. It is clinically characterized by inflammatory back pain, progressive stiffness, and restricted spinal mobility, while its pathological features include enthesitis, extracellular matrix remodeling, and aberrant new bone formation.[1] AS is associated with substantial comorbidity, including diabetes,[2] malignancy,[3] venous thromboembolism,[3] cardiovascular disease,[4] and inflammatory bowel disease,[5] contributing to a significant clinical and socioeconomic burden. Although genetic predisposition – particularly HLA-B27 – explains a considerable proportion of disease susceptibility,[6-8] the molecular pathways underlying inflammation and pathological structural changes remain only partly understood.[9,10] Identifying additional causal mediators is therefore essential for advancing future therapeutic strategies.[11]
Matrix metalloproteinases (MMPs) are a family of zinc-dependent endopeptidases involved in extracellular matrix degradation and tissue remodeling.[12] Their enzymatic activities are tightly regulated by tissue inhibitors of metalloproteinases (TIMPs), and imbalance in the MMP–TIMP system contributes to numerous inflammatory and degenerative musculoskeletal diseases.[13] Observational evidence suggests that several MMPs – including MMP1, MMP3, and MMP9 – are elevated in the serum, synovial fluid, or spinal tissues of patients with AS.[14-17] Moreover, MMP expression levels have been reported to correlate with disease activity and inflammatory imaging markers.[14,15,17] Despite these associations, observational studies cannot reliably distinguish causal drivers from downstream inflammatory biomarkers due to confounding and reverse causation.
Mendelian randomization (MR) is an analysis of genetic variables that follows Mendel’s law of inheritance, which exploits single nucleotide polymorphisms (SNPs) as instrumental variables (IVs) to infer the causality of an observed association between a modifiable exposure and a clinically relevant outcome.[18] Alleles are randomly segregated during meiosis, so MR can reduce the bias caused by confounding factors.[19] In addition, because genetic variation occurs before the disease and the order of the 2 cannot be reversed, MR can also avoid the interference of reverse causality.[19]
Given the uncertainty of whether circulating MMPs or TIMPs contribute causally to AS pathogenesis, MR provides an opportunity to clarify these relationships using genetic instruments derived from genome-wide association studies (GWAS). To address this evidence gap, we conducted a comprehensive bidirectional 2-sample MR analysis to evaluate the causal effects of genetically predicted circulating levels of 14 MMPs and 4 TIMPs on AS risk, and to assess whether genetic liability to AS influences circulating MMP/TIMP profiles.
2. Material and methods
This was an MR study based on publicly available GWAS data and was conducted under 3 main hypotheses. First, there must be a strong independent association between the selected IVs and the exposure. Second, the IVs should not be associated with potential confounders. Third, the effect of IVs on AS should operate exclusively through exposure. This study was reported in accordance with the STROBE‑MR checklist.[20,21] The flowchart of this study is shown in Figure S1, Supplemental Digital Content 1.
2.1. Data source
We obtained GWAS summary statistics for SNPs associated with circulating plasma levels of 14 MMPs (MMP1, MMP2, MMP3, MMP7, MMP8, MMP9, MMP10, MMP12, MMP13, MMP14, MMP16, MMP17, MMP19, MMP20) and 4 TIMPs (TIMP1, TIMP2, TIMP3, TIMP4) from a large-scale proteogenomic study in 5368 individuals.[22] Data are freely available from the public database GWAS Catalog: http://ftp.ebi.ac.uk/pub/databases/gwas/summary_statistics/GCST90088001-GCST90089000. The GWAS summary data for AS were obtained from the FinnGen Release 12 database (https://www.finngen.fi/en), which includes individuals of Finnish ancestry. The sample sizes of AS in this study were 3838/3,53,224 (case/control). Minimal sample overlap is expected between the UK Biobank-based exposure GWAS and the FinnGen outcome GWAS, reducing potential bias from this source. All data were obtained from public databases, and the underlying studies had received ethical approval. Therefore, additional ethical approval was not required for the present study.
2.2. Instrumental variable selection
IVs included in this study should meet the following criteria: SNPs associated with circulating MMPs and TIMPs were initially selected using the conventional genome-wide significance threshold of P < 5 × 10-8. However, application of this threshold resulted in fewer than 3 eligible IVs for some protein exposures, which was insufficient for the planned sensitivity analyses. Therefore, the threshold was relaxed to P < 5 × 10-6 for the forward MR analyses to obtain a sufficient number of IVs while retaining relatively stringent evidence of exposure association. For the reverse MR analysis, the conventional threshold of P < 5 × 10-8 was retained.[23] SNPs with minor allele frequency (MAF) > 0.01 were selected. According to R2 < 0.001 and window size = 10,000 kb, the linkage disequilibrium (LD) effect between SNPs was eliminated. When a selected IV was unavailable in the outcome summary data statistics, a proxy SNP with high LD (R2 > 0.8) was identified and used instead. The F-statistic for each SNP in IVs was calculated to evaluate instrument strength and minimize potential bias due to weak instruments. The calculation formula is as follows: F = R2 * (N − 2)/(1 − R2), where R2 is the proportion of variation in exposure that can be explained by SNP in IVs, and an F-statistic > 10 was considered indicative of sufficient instrument strength. Finally, to ensure consistency of effect alleles, we harmonized the exposure and outcome datasets by removing unclear SNPs with mismatched alleles and palindromic SNPs with intermediate minor allele frequencies.
2.3. MR analysis
We performed an MR analysis to investigate the causal association between MMPs/TIMPs and AS. Four commonly used MR methods were performed. Random effects inverse variance weighted (IVW) was used as the main analysis method,[24] while MR-Egger,[25] weighted median[26] and weighted mode[27] were used as supplementary analysis methods. The false discovery rate (FDR) was applied for multiple‑testing correction. All statistical analyses were performed using R (version 4.3.2) with the “TwoSampleMR” R package (version 0.6.22; MRC Integrative Epidemiology Unit [MRC IEU], University of Bristol). Forest plots, scatter plots, and funnel plots were used for visualization.
2.4. Sensitivity analysis
Cochran’s Q test was used to evaluate the heterogeneity among IVs. When P > .05, heterogeneity was considered low, indicating that variation among the IVs was unlikely to substantially affect the IVW estimates. MR-Egger regression was used to identify potential pleiotropy (P < .05 was considered as the presence of directional pleiotropy).[25] MR‐Pleiotropy Residual Sum and Outlier methods (MR-PRESSO) was used to assess possible outliers, and the causal association was reestimated after elimination, thus correcting for horizontal pleiotropy. The leave-one-out analysis was used to test the robustness and consistency of the results.
3. Results
3.1. SNP selection
In the forward analysis, we identified a total of 362 SNPs as IVs for 14 MMPs and 4 TIMPs (Table S1, Supplemental Digital Content 2). The number of IVs ranged from 9 (for MMP14) to 42 (for TIMP3). The mean F-statistics for all exposures were well above the threshold of 10, ranging from 22.97 to 90.82, suggesting the absence of weak instrument bias. Due to the exclusion of palindromic SNPs (Table S2, Supplemental Digital Content 3), unmatched outcomes (Table S1, Supplemental Digital Content 2), and outlier SNPs (Table S1, Supplemental Digital Content 2), 297 SNPs were finally included and analyzed (Table S3, Supplemental Digital Content 4).
In the reverse analysis, 19 SNPs were selected as IVs for AS (Table S4, Supplemental Digital Content 5). The instruments’ strength was robust (mean F-statistic = 215.09, all > 10). Following quality control, we excluded 2 palindromic SNPs (rs35730213, rs4713605) and 4 SNPs (rs188736974, rs114799031, rs181316459, rs113827118) that were absent from the outcome data. In addition, 2 outliers were eliminated when AS was exposed and TIMP1 was the outcome (rs2743940, rs652888). The details of the SNPs included in the analysis are shown in Table S5, Supplemental Digital Content 6.
3.2. Effects of MMPs/TIMPs on ankylosing spondylitis
The IVW suggested a nominally significant association between higher MMP13 levels and risk of AS (odds ratio [OR] = 1.223, 95% confidence interval [CI]: 1.048–1.428, P = .011; Table 1). The weighted median analysis showed a similar direction of effect (OR = 1.222, 95% CI: 1.001–1.491, P = .048; Table 1). However, neither the IVW nor weighted median association remained statistically significant after FDR correction (FDR = 0.198 and 0.432, respectively). The causal estimates from various MR methods for the association between MMP13 and AS are visually summarized in a forest plot (Fig. S2A, Supplemental Digital Content 7). A scatter plot of the SNP effects on MMP13 versus their effects on AS showed that most SNPs were aligned with the positive slope of the regression line, indicating a consistent causal direction (Fig. S2B, Supplemental Digital Content 7). In contrast, IVW analyses identified no statistically significant effects on AS risk for the remaining 13 MMPs and 4 TIMPs (all P > .05). Although the weighted median analysis suggested a potential association for MMP1 (OR = 1.095, P = .030), the corresponding IVW estimate was not statistically significant (P = .089), and the finding was therefore not considered robust (Table 1).
Table 1.
Mendelian randomization estimates for the causal effect of circulating MMPs and TIMPs on the risk of ankylosing spondylitis.
| Exposure | Outcome | nSNPs | Method | OR (95% CI) | P | FDR |
|---|---|---|---|---|---|---|
| MMP1 | Ankylosing spondylitis | 27 | Inverse variance weighted | 1.058 (0.991–1.129) | .089 | 0.801 |
| MMP1 | Ankylosing spondylitis | 27 | MR Egger | 1.111 (1.001–1.232) | .058 | 0.515 |
| MMP1 | Ankylosing spondylitis | 27 | Weighted median | 1.095 (1.009–1.188) | .03 | 0.432 |
| MMP1 | Ankylosing spondylitis | 27 | Weighted mode | 1.084 (0.999–1.175) | .062 | 0.911 |
| MMP10 | Ankylosing spondylitis | 13 | Inverse variance weighted | 0.950 (0.833–1.083) | .441 | 0.882 |
| MMP10 | Ankylosing spondylitis | 13 | MR Egger | 1.075 (0.837–1.380) | .583 | 0.757 |
| MMP10 | Ankylosing spondylitis | 13 | Weighted median | 0.951 (0.825–1.095) | .485 | 0.774 |
| MMP10 | Ankylosing spondylitis | 13 | Weighted mode | 0.954 (0.826–1.103) | .539 | 0.941 |
| MMP12 | Ankylosing spondylitis | 21 | Inverse variance weighted | 1.003 (0.926–1.085) | .948 | 0.948 |
| MMP12 | Ankylosing spondylitis | 21 | MR Egger | 1.051 (0.938–1.177) | .402 | 0.757 |
| MMP12 | Ankylosing spondylitis | 21 | Weighted median | 1.011 (0.928–1.102) | .795 | 0.871 |
| MMP12 | Ankylosing spondylitis | 21 | Weighted mode | 1.012 (0.930–1.102) | .781 | 0.942 |
| MMP13 | Ankylosing spondylitis | 11 | Inverse variance weighted | 1.223 (1.048–1.428) | .011 | 0.198 |
| MMP13 | Ankylosing spondylitis | 11 | MR Egger | 1.094 (0.772–1.551) | .624 | 0.757 |
| MMP13 | Ankylosing spondylitis | 11 | Weighted median | 1.222 (1.001–1.491) | .048 | 0.432 |
| MMP13 | Ankylosing spondylitis | 11 | Weighted mode | 1.233 (0.963–1.578) | .127 | 0.911 |
| MMP14 | Ankylosing spondylitis | 8 | Inverse variance weighted | 0.979 (0.848–1.130) | .77 | 0.924 |
| MMP14 | Ankylosing spondylitis | 8 | MR Egger | 0.683 (0.452–1.031) | .119 | 0.515 |
| MMP14 | Ankylosing spondylitis | 8 | Weighted median | 0.985 (0.819–1.185) | .871 | 0.871 |
| MMP14 | Ankylosing spondylitis | 8 | Weighted mode | 0.964 (0.749–1.240) | .782 | 0.942 |
| MMP16 | Ankylosing spondylitis | 9 | Inverse variance weighted | 0.906 (0.753–1.089) | .292 | 0.876 |
| MMP16 | Ankylosing spondylitis | 9 | MR Egger | 0.931 (0.643–1.347) | .715 | 0.757 |
| MMP16 | Ankylosing spondylitis | 9 | Weighted median | 0.930 (0.729–1.187) | .56 | 0.774 |
| MMP16 | Ankylosing spondylitis | 9 | Weighted mode | 1.033 (0.677–1.575) | .885 | 0.975 |
| MMP17 | Ankylosing spondylitis | 13 | Inverse variance weighted | 1.084 (0.900–1.304) | .397 | 0.882 |
| MMP17 | Ankylosing spondylitis | 13 | MR Egger | 0.969 (0.654–1.436) | .879 | 0.879 |
| MMP17 | Ankylosing spondylitis | 13 | Weighted median | 1.026 (0.798–1.319) | .839 | 0.871 |
| MMP17 | Ankylosing spondylitis | 13 | Weighted mode | 0.993 (0.750–1.316) | .964 | 0.975 |
| MMP19 | Ankylosing spondylitis | 19 | Inverse variance weighted | 1.024 (0.906–1.156) | .708 | 0.91 |
| MMP19 | Ankylosing spondylitis | 19 | MR Egger | 0.942 (0.711–1.249) | .685 | 0.757 |
| MMP19 | Ankylosing spondylitis | 19 | Weighted median | 0.946 (0.797–1.123) | .527 | 0.774 |
| MMP19 | Ankylosing spondylitis | 19 | Weighted mode | 0.945 (0.717–1.247) | .696 | 0.942 |
| MMP2 | Ankylosing spondylitis | 14 | Inverse variance weighted | 1.034 (0.896–1.192) | .65 | 0.91 |
| MMP2 | Ankylosing spondylitis | 14 | MR Egger | 1.422 (1.016–1.991) | .063 | 0.515 |
| MMP2 | Ankylosing spondylitis | 14 | Weighted median | 1.130 (0.929–1.374) | .221 | 0.774 |
| MMP2 | Ankylosing spondylitis | 14 | Weighted mode | 1.206 (0.887–1.641) | .253 | 0.911 |
| MMP20 | Ankylosing spondylitis | 14 | Inverse variance weighted | 1.072 (0.949–1.212) | .262 | 0.876 |
| MMP20 | Ankylosing spondylitis | 14 | MR Egger | 1.289 (0.973–1.706) | .102 | 0.515 |
| MMP20 | Ankylosing spondylitis | 14 | Weighted median | 1.069 (0.907–1.261) | .426 | 0.774 |
| MMP20 | Ankylosing spondylitis | 14 | Weighted mode | 1.080 (0.830–1.407) | .575 | 0.941 |
| MMP3 | Ankylosing spondylitis | 15 | Inverse variance weighted | 1.061 (0.925–1.218) | .396 | 0.882 |
| MMP3 | Ankylosing spondylitis | 15 | MR Egger | 1.224 (0.900–1.665) | .22 | 0.566 |
| MMP3 | Ankylosing spondylitis | 15 | Weighted median | 1.088 (0.924–1.279) | .311 | 0.774 |
| MMP3 | Ankylosing spondylitis | 15 | Weighted mode | 1.104 (0.947–1.286) | .226 | 0.911 |
| MMP7 | Ankylosing spondylitis | 17 | Inverse variance weighted | 1.070 (0.955–1.198) | .242 | 0.876 |
| MMP7 | Ankylosing spondylitis | 17 | MR Egger | 1.094 (0.863–1.387) | .47 | 0.757 |
| MMP7 | Ankylosing spondylitis | 17 | Weighted median | 1.050 (0.911–1.212) | .499 | 0.774 |
| MMP7 | Ankylosing spondylitis | 17 | Weighted mode | 1.061 (0.891–1.263) | .514 | 0.941 |
| MMP8 | Ankylosing spondylitis | 25 | Inverse variance weighted | 1.024 (0.937–1.119) | .604 | 0.91 |
| MMP8 | Ankylosing spondylitis | 25 | MR Egger | 0.965 (0.825–1.128) | .657 | 0.757 |
| MMP8 | Ankylosing spondylitis | 25 | Weighted median | 1.016 (0.890–1.159) | .815 | 0.871 |
| MMP8 | Ankylosing spondylitis | 25 | Weighted mode | 1.002 (0.863–1.165) | .975 | 0.975 |
| MMP9 | Ankylosing spondylitis | 17 | Inverse variance weighted | 0.972 (0.875–1.081) | .604 | 0.91 |
| MMP9 | Ankylosing spondylitis | 17 | MR Egger | 0.858 (0.707–1.042) | .143 | 0.515 |
| MMP9 | Ankylosing spondylitis | 17 | Weighted median | 0.940 (0.812–1.087) | .403 | 0.774 |
| MMP9 | Ankylosing spondylitis | 17 | Weighted mode | 0.937 (0.803–1.093) | .419 | 0.941 |
| TIMP1 | Ankylosing spondylitis | 12 | Inverse variance weighted | 1.038 (0.863–1.249) | .689 | 0.91 |
| TIMP1 | Ankylosing spondylitis | 12 | MR Egger | 0.877 (0.571–1.348) | .563 | 0.757 |
| TIMP1 | Ankylosing spondylitis | 12 | Weighted median | 1.069 (0.831–1.375) | .602 | 0.774 |
| TIMP1 | Ankylosing spondylitis | 12 | Weighted mode | 1.052 (0.738–1.498) | .785 | 0.942 |
| TIMP2 | Ankylosing spondylitis | 17 | Inverse variance weighted | 1.006 (0.885–1.145) | .925 | 0.948 |
| TIMP2 | Ankylosing spondylitis | 17 | MR Egger | 1.202 (0.924–1.562) | .19 | 0.566 |
| TIMP2 | Ankylosing spondylitis | 17 | Weighted median | 1.078 (0.913–1.273) | .373 | 0.774 |
| TIMP2 | Ankylosing spondylitis | 17 | Weighted mode | 1.117 (0.943–1.324) | .22 | 0.911 |
| TIMP3 | Ankylosing spondylitis | 34 | Inverse variance weighted | 1.007 (0.937–1.082) | .849 | 0.948 |
| TIMP3 | Ankylosing spondylitis | 34 | MR Egger | 1.048 (0.937–1.172) | .419 | 0.757 |
| TIMP3 | Ankylosing spondylitis | 34 | Weighted median | 1.032 (0.942–1.130) | .504 | 0.774 |
| TIMP3 | Ankylosing spondylitis | 34 | Weighted mode | 1.049 (0.957–1.150) | .312 | 0.936 |
| TIMP4 | Ankylosing spondylitis | 13 | Inverse variance weighted | 0.800 (0.543–1.178) | .258 | 0.876 |
| TIMP4 | Ankylosing spondylitis | 13 | MR Egger | 1.245 (0.550–2.821) | .61 | 0.757 |
| TIMP4 | Ankylosing spondylitis | 13 | Weighted median | 0.926 (0.796–1.076) | .315 | 0.774 |
| TIMP4 | Ankylosing spondylitis | 13 | Weighted mode | 0.947 (0.809–1.108) | .508 | 0.941 |
CI = confidence interval, FDR = false discovery rate, MMP = matrix metalloproteinase, nSNPs = number of single nucleotide polymorphisms, OR = odds ratio, TIMP = tissue inhibitors of metalloproteinase.
The robustness of the MMP13 association was further evaluated by sensitivity analyses. Cochran’s Q test indicated no significant heterogeneity among the IVs for MMP13 (Q statistic = 6.581, P = .764). Furthermore, the MR-Egger intercept test showed no evidence of directional pleiotropy (Intercept P = .502; Table 2), and the MR-PRESSO analysis did not detect any significant outliers (Global P = .817; Table 3). The symmetrical distribution of SNPs in the funnel plot for MMP13 further supported the absence of directional pleiotropy (Fig. S2C, Supplemental Digital Content 7). A leave-one-out sensitivity analysis demonstrated that the effect of MMP13 on AS was not driven by any single SNP (Fig. S2D, Supplemental Digital Content 7). For the other null associations, the sensitivity analyses did not reveal any significant directional pleiotropy.
Table 2.
Results of heterogeneity and horizontal pleiotropy tests for the causal effect of MMPs and TIMPs on ankylosing spondylitis.
| Exposure | Outcome | Heterogeneity | Pleiotropy | ||
|---|---|---|---|---|---|
| Q statistic (IVW) | P value | MR-Egger intercept | P value | ||
| MMP1 | MMP1 | 30.36 | .253 | −0.012 | .251 |
| MMP2 | Ankylosing spondylitis | 13.76 | .391 | −0.047 | .064 |
| MMP3 | Ankylosing spondylitis | 19.381 | .151 | −0.024 | .328 |
| MMP7 | Ankylosing spondylitis | 19.777 | .23 | −0.004 | .837 |
| MMP8 | Ankylosing spondylitis | 21.324 | .62 | 0.012 | .372 |
| MMP9 | Ankylosing spondylitis | 16.19 | .44 | 0.029 | .154 |
| MMP10 | Ankylosing spondylitis | 17.991 | .116 | −0.024 | .281 |
| MMP12 | Ankylosing spondylitis | 25.562 | .181 | −0.013 | .274 |
| MMP13 | Ankylosing spondylitis | 6.581 | .764 | 0.021 | .502 |
| MMP14 | Ankylosing spondylitis | 6.101 | .528 | 0.071 | .117 |
| MMP16 | Ankylosing spondylitis | 6.899 | .548 | −0.004 | .872 |
| MMP17 | Ankylosing spondylitis | 15.134 | .234 | 0.023 | .539 |
| MMP19 | Ankylosing spondylitis | 17.284 | .504 | 0.013 | .532 |
| MMP20 | Ankylosing spondylitis | 9.171 | .76 | −0.032 | .18 |
| TIMP1 | Ankylosing spondylitis | 9.269 | .597 | 0.018 | .413 |
| TIMP2 | Ankylosing spondylitis | 18.718 | .284 | −0.029 | .153 |
| TIMP3 | Ankylosing spondylitis | 45.92 | .067 | −0.009 | .37 |
| TIMP4 | Ankylosing spondylitis | 9.132 | .52 | −0.013 | .521 |
IVW = inverse variance weighted, MMPs = matrix metalloproteinases, TIMPs = tissue inhibitors of metalloproteinases.
Table 3.
MR-PRESSO analysis for outlier detection in the forward MR of MMPs and TIMPs on ankylosing spondylitis.
| Exposure | Outcome | Raw | Outlier corrected | Global P | Number of outliers | Distortion P | ||
|---|---|---|---|---|---|---|---|---|
| OR (95% CI) | P | OR (95% CI) | P | |||||
| MMP1 | Ankylosing spondylitis | 1.06 (0.99–1.13) | 0.101 | NA (NA-NA) | NA | .3 | NA | NA |
| MMP2 | Ankylosing spondylitis | 1.03 (0.90–1.19) | 0.657 | NA (NA-NA) | NA | .383 | NA | NA |
| MMP3 | Ankylosing spondylitis | 1.06 (0.92–1.22) | 0.411 | NA (NA-NA) | NA | .23 | NA | NA |
| MMP7 | Ankylosing spondylitis | 1.07 (0.96–1.20) | 0.259 | NA (NA-NA) | NA | .278 | NA | NA |
| MMP8 | Ankylosing spondylitis | 1.02 (0.94–1.11) | 0.587 | NA (NA-NA) | NA | .596 | NA | NA |
| MMP9 | Ankylosing spondylitis | 0.97 (0.88–1.08) | 0.611 | NA (NA-NA) | NA | .466 | NA | NA |
| MMP10 | Ankylosing spondylitis | 0.95 (0.83–1.08) | 0.456 | NA (NA-NA) | NA | .177 | NA | NA |
| MMP12 | Ankylosing spondylitis | 1.00 (0.93–1.09) | 0.948 | NA (NA-NA) | NA | .276 | NA | NA |
| MMP13 | Ankylosing spondylitis | 1.22 (1.08–1.39) | 0.01 | NA (NA-NA) | NA | .817 | NA | NA |
| MMP14 | Ankylosing spondylitis | 0.98 (0.86–1.12) | 0.763 | NA (NA-NA) | NA | .567 | NA | NA |
| MMP16 | Ankylosing spondylitis | 0.91 (0.76–1.07) | 0.289 | NA (NA-NA) | NA | .541 | NA | NA |
| MMP17 | Ankylosing spondylitis | 1.08 (0.90–1.30) | 0.413 | NA (NA-NA) | NA | .275 | NA | NA |
| MMP19 | Ankylosing spondylitis | 1.02 (0.91–1.15) | 0.707 | NA (NA-NA) | NA | .546 | NA | NA |
| MMP20 | Ankylosing spondylitis | 1.07 (0.97–1.19) | 0.205 | NA (NA-NA) | NA | .735 | NA | NA |
| TIMP1 | Ankylosing spondylitis | 1.04 (0.88–1.23) | 0.671 | NA (NA-NA) | NA | .624 | NA | NA |
| TIMP2 | Ankylosing spondylitis | 1.01 (0.88–1.14) | 0.926 | NA (NA-NA) | NA | .296 | NA | NA |
| TIMP3 | Ankylosing spondylitis | 1.01 (0.94–1.08) | 0.85 | NA (NA-NA) | NA | .079 | NA | NA |
| TIMP4 | Ankylosing spondylitis | 0.88 (0.78–1.00) | 0.072 | NA (NA-NA) | NA | .586 | NA | NA |
| rs9271708 | ||||||||
CI = confidence interval, MMP = matrix metalloproteinase, MR-PRESSO = MR‐Pleiotropy Residual Sum and Outlier methods, OR = odds ratio, TIMP = tissue inhibitors of metalloproteinase.
3.3. Effects of ankylosing spondylitis on MMPs and TIMPs (reverse MR)
The IVW analysis did not reveal any effects of AS on MMPs and TIMPs (all P > .05, Table S6, Supplemental Digital Content 8). Conversely, the weighted median method indicates a potential association between MMP8 and AS (OR = 1.050, 95% CI: 1.002–1.100, P = .039). Cochran’s Q test found no significant heterogeneity across any of the analyses (all P > .05; Table S7, Supplemental Digital Content 9). The MR-Egger intercept test revealed no evidence of directional pleiotropy (all P > .05; Table S7, Supplemental Digital Content 9). The MR-PRESSO analysis identified no significant outliers (all Global Test P > .05; Table S8, Supplemental Digital Content 10).
4. Discussion
This bidirectional 2-sample MR analysis provides suggestive genetic evidence supporting a potential causal role of circulating MMP13 in the development of AS. Among 14 MMPs and 4 TIMPs examined, MMP13 showed the most consistent positive association across MR methods. However, this association was nominally significant but did not remain significant after FDR correction. Reverse MR analyses further indicated that genetic liability to AS does not causally alter circulating MMP or TIMP levels, suggesting that the observed association was not supported by evidence of reverse causation.
MMP13, or collagenase-3, is a potent zinc-dependent endopeptidase with a well-established role in degrading extracellular matrix components, most critically type II collagen, the primary structural protein in articular cartilage. Its role in cartilage catabolism has been extensively documented in osteoarthritis, where MMP13 inhibition has been investigated as a potential therapeutic strategy.[28] Similarly, in rheumatoid arthritis, MMP13 is identified as a key inflammatory and degradative enzyme,[29] and its expression is often driven by pro-inflammatory cytokines like IL-17, a central player in spondyloarthropathies.[30] Together, these findings provide biological plausibility for a potential involvement of MMP13 in the inflammatory and tissue-remodeling processes underlying AS, although the precise role of MMP13 in AS remains to be established.
Emerging clinical and translational evidence further supports a possible relationship between MMP13 and AS-related pathological changes. Recent studies have reported correlations between AS-associated metabolic disturbances and MMP13 expression,[31] as well as interactions between complement-related proteins and MMP13 that may contribute to aberrant bone formation, a characteristic feature of AS.[32] MMP13 has also been implicated in endochondral ossification and osteoblast differentiation, suggesting a mechanistic link between cartilage degradation and pathological new bone formation.[33] Taken together, these observations, together with our genetic findings, suggest that MMP13 may represent a biologically relevant component at the intersection of inflammation, matrix remodeling, and abnormal bone formation in AS. However, because our analysis assessed genetically predicted circulating MMP13 levels rather than tissue-specific MMP13 activity, these findings should not be interpreted as evidence that circulating MMP13 directly drives pathological changes within the entheses or spine. Moreover, although MMP13 may warrant further investigation as a candidate therapeutic pathway, the present MR evidence alone is insufficient to establish the efficacy or clinical utility of MMP13-targeted therapies. Further tissue-specific and experimental studies are needed to clarify its mechanistic and therapeutic relevance in AS.
Conversely, our study did not support a causal role for the other 13 MMPs or the 4 TIMPs in the development of AS. This is particularly noteworthy for proteins like MMP1 and MMP3. In recent years, multiple studies have shown that changes in matrix metalloprotein levels are associated with AS. Among the MMPs, the correlation between MMP3 and AS has attracted considerable attention. In a cohort study of 30 patients with AS and 10 healthy volunteers, serum MMP3 levels were significantly elevated in patients with AS, and a higher level of MMP3 was correlated with the Bath AS Disease Activity Index.[34] In addition, some studies have observed that MMP1[35] and MMP7[36] are associated with an increased risk of AS. Overall, these clinical studies suggest that certain MMPs may be associated with the development of AS. However, other clinical studies have reported that the MMP3 genotype was not associated with susceptibility or severity of AS.[37] This discrepancy highlights the strength of the MR design in dissecting correlation from causation. Although these observational associations are informative, our null findings suggest that the elevated levels of these other MMPs in AS patients may be a downstream consequence of the inflammatory disease process or reflect local tissue activity, rather than being primary causal drivers of disease onset.
This study has several limitations. First, the analyses were based primarily on individuals of European ancestry, which may limit generalizability to diverse populations with different genetic backgrounds or environmental exposures. Second, MR estimates reflect the effects of lifelong genetically predicted circulating protein levels and may not fully capture the dynamic or tissue-specific expression of MMPs in key anatomical sites such as the entheses and spine. Third, the forward MR analyses used a relaxed SNP selection threshold of P < 5 × 10-6. Although this approach has been used in previous MR studies, the possibility of weak instrument bias or less specific instrument–exposure associations cannot be completely excluded, and the findings should therefore be interpreted cautiously. Integrative analyses combining MR with tissue-level proteogenomics, single-cell transcriptomics, and functional experiments may help clarify these mechanisms.
In conclusion, this comprehensive and bidirectional 2-sample MR study provides suggestive genetic evidence for a potential causal relationship between circulating MMP13 levels and AS risk. However, the association did not remain statistically significant after FDR correction and should therefore be interpreted cautiously. We found no evidence for causal effects of the other 13 MMPs or 4 TIMPs on AS, nor did we find evidence of reverse causation. Further tissue-specific, experimental, and clinical studies are needed to clarify the biological and therapeutic relevance of MMP13 in AS.
Author contributions
Conceptualization: Guoqing Li.
Data curation: Huaguo Zhao.
Formal analysis: Yang Wang.
Investigation: Huaguo Zhao.
Methodology: Huaguo Zhao.
Resources: Xiaoyang Sun.
Writing – original draft: Xiaoyang Sun, Guoqing Li, Huaguo Zhao, Yang Wang.
Writing – review & editing: Xiaoyang Sun, Guoqing Li, Huaguo Zhao, Yang Wang.
Abbreviations:
- AS
- ankylosing spondylitis
- CI
- confidence interval
- FDR
- false discovery rate
- GWAS
- genome-wide association study
- IV
- instrumental variable
- IVW
- inverse variance weighted
- MMP
- matrix metalloproteinase
- MR
- Mendelian randomization
- MR‐PRESSO
- Mendelian randomization-Pleiotropy Residual Sum and Outlier methods
- OR
- odds ratio
- SNP
- single nucleotide polymorphism
- TIMP
- Tissue inhibitors of metalloproteinase
This study was supported by the Liangshan Science and Technology Program (23ZDYF0055); Ningbo Clinical Research Center for Orthopedics, Sports Medicine & Rehabilitation (2024L004); Agricultural and Rural Social Science Technology Planning Project of Yinzhou District (2018YZQ02028); Zhejiang Provincial Medical and Health Science and Technology (2025KY1484).
This article is a Mendelian randomization study. The data for this study were obtained from publicly available databases and published literature data and does not require ethical approval and written informed consent.
The authors have no conflicts of interest to declare.
All data generated or analyzed during this study are included in this published article (and its supplementary information files).
Supplemental Digital Content is available in the online version of this article (http://dx.doi.org/10.1097/MD.0000000000051374).
How to cite this article: Sun X, Li G, Zhao H, Wang Y. Association of 14 matrix metalloproteinases and 4 tissue inhibitors of metalloproteinases on ankylosing spondylitis: A bidirectional Mendelian randomization study. Medicine 2026;105:44(e51374).
Contributor Information
Xiaoyang Sun, Email: sunxiaoyang2012@foxmail.com.
Huaguo Zhao, Email: zhaohuaguo@foxmail.com.
Yang Wang, Email: cxhszx11ba@163.com.
References
- [1].Chen CW, Wei JC, Gu J, Yu D. Editorial: advances in pathogenesis, etiology, and therapies for ankylosing spondylitis. Front Immunol. 2021;12:822582. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Lee YH, Song GG. Diabetes mellitus in ankylosing spondylitis. Ann Rheum Dis. 2021;80:e142. [DOI] [PubMed] [Google Scholar]
- [3].Ma X, Cui Y, Shi C, Han Z. The causal relationship between ankylosing spondylitis and the risk of ovarian cancer. Hum Immunol. 2024;85:110738. [DOI] [PubMed] [Google Scholar]
- [4].Navarini L, Caso F, Costa L, et al. Cardiovascular risk prediction in ankylosing spondylitis: from traditional scores to machine learning assessment. Rheumatol Ther. 2020;7:867–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Rosenbaum J, Chandran V. Management of comorbidities in ankylosing spondylitis. Am J Med Sci. 2012;343:364–6. [DOI] [PubMed] [Google Scholar]
- [6].Khan MA, Yong SB, Wei JC. Ankylosing spondylitis: History, epidemiology, and HLA-B27. Int J Rheum Dis. 2023;26:413–4. [DOI] [PubMed] [Google Scholar]
- [7].Moon KH, Kim YT. Medical treatment of ankylosing spondylitis. Hip Pelvis. 2014;26:129–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Reveille JD. The genetic basis of ankylosing spondylitis. Curr Opin Rheumatol. 2006;18:332–41. [DOI] [PubMed] [Google Scholar]
- [9].Yilmaz E. Ankylosing spondylitis occurring in a patient with Behçet disease after chronic brucellosis infection: a case report and review of the literature. Int J Rheum Dis. 2023;26:386–90. [DOI] [PubMed] [Google Scholar]
- [10].Schutz C, Baraliakos X. What do we know about co-stimulatory and co-inhibitory immune checkpoint signals in ankylosing spondylitis? Clin Exp Immunol. 2023;213:288–300. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Perrotta FM, Scriffignano S, Ciccia F, Lubrano E. Therapeutic targets for ankylosing spondylitis - recent insights and future prospects. Open Access Rheumatol. 2022;14:57–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Amar S, Smith L, Fields GB. Matrix metalloproteinase collagenolysis in health and disease. Biochim Biophys Acta Mol Cell Res. 2017;1864:1940–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Visse R, Nagase H. Matrix metalloproteinases and tissue inhibitors of metalloproteinases: structure, function, and biochemistry. Circ Res. 2003;92:827–39. [DOI] [PubMed] [Google Scholar]
- [14].Mattey DL, Packham JC, Nixon NB, et al. Association of cytokine and matrix metalloproteinase profiles with disease activity and function in ankylosing spondylitis. Arthritis Res Ther. 2012;14:R127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Holm Nielsen S, Sun S, Bay-Jensen AC, et al. Levels of extracellular matrix metabolites are associated with changes in Ankylosing spondylitis disease activity score and MRI inflammation scores in patients with axial spondyloarthritis during TNF inhibitor therapy. Arthritis Res Ther. 2022;24:279. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Skjot-Arkil H, Schett G, Zhang C, et al. Investigation of two novel biochemical markers of inflammation, matrix metalloproteinase and cathepsin generated fragments of C-reactive protein, in patients with ankylosing spondylitis. Clin Exp Rheumatol. 2012;30:371–9. [PubMed] [Google Scholar]
- [17].Zhu J, Yu DT. Matrix metalloproteinase expression in the spondyloarthropathies. Curr Opin Rheumatol. 2006;18:364–8. [DOI] [PubMed] [Google Scholar]
- [18].Holmes MV, Ala-Korpela M, Smith GD. Mendelian randomization in cardiometabolic disease: challenges in evaluating causality. Nat Rev Cardiol. 2017;14:577–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Zheng J, Baird D, Borges MC, et al. Recent developments in mendelian randomization studies. Curr Epidemiol Rep. 2017;4:330–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [20].Skrivankova VW, Richmond RC, Woolf BAR, et al. Strengthening the reporting of observational studies in epidemiology using mendelian randomization: the STROBE-MR statement. JAMA. 2021;326:1614–21. [DOI] [PubMed] [Google Scholar]
- [21].Skrivankova VW, Richmond RC, Woolf BAR, et al. Strengthening the reporting of observational studies in epidemiology using mendelian randomisation (STROBE-MR): explanation and elaboration. BMJ. 2021;375:n2233. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Gudjonsson A, Gudmundsdottir V, Axelsson GT, et al. A genome-wide association study of serum proteins reveals shared loci with common diseases. Nat Commun. 2022;13:480. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Sun S, Jiao M, Han C, et al. Causal effects of genetically determined metabolites on risk of polycystic ovary syndrome: a mendelian randomization study. Front Endocrinol (Lausanne). 2020;11:621. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol. 2013;37:658–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Burgess S, Thompson SG. Interpreting findings from Mendelian randomization using the MR-Egger method. Eur J Epidemiol. 2017;32:377–89. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol. 2016;40:304–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46:1985–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [28].Roy HS, Murugesan P, Kulkarni C, et al. On-demand release of a selective MMP-13 blocker from an enzyme-responsive injectable hydrogel protects cartilage from degenerative progression in osteoarthritis. J Mater Chem B. 2024;12:5325–38. [DOI] [PubMed] [Google Scholar]
- [29].Cai Y, Deng L, Yao J. Analysis and identification of ferroptosis-related diagnostic markers in rheumatoid arthritis. Ann Med. 2024;56:2397572. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [30].Chen X, Chen X, Xie J, Guo D. Promoter methylation of WIF1 is involved in IL-17-induced chondrocyte inflammatory injury and matrix degradation via promoting Wnt5a/MAPK-JNK signaling. Mol Biotechnol. 2026;68:496–506. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [31].Jeon C, Jang Y, Lee SH, et al. Abnormal kynurenine level contributes to the pathological bone features of ankylosing spondylitis. Int Immunopharmacol. 2023;118:110132. [DOI] [PubMed] [Google Scholar]
- [32].Lee JH, Lee SH, Jeon C, et al. The complement factor H-related protein-5 (CFHR5) exacerbates pathological bone formation in ankylosing spondylitis. J Mol Med (Berl). 2024;102:571–83. [DOI] [PubMed] [Google Scholar]
- [33].Arai Y, Lee SH. MMP13-overexpressing mesenchymal stem cells enhance bone tissue formation in the presence of collagen hydrogel. Tissue Eng Regen Med. 2023;20:461–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [34].Soliman E, Labib W, el-Tantawi G, Hamimy A, Alhadidy A, Aldawoudy A. Role of matrix metalloproteinase-3 (MMP-3) and magnetic resonance imaging of sacroiliitis in assessing disease activity in ankylosing spondylitis. Rheumatol Int. 2012;32:1711–20. [DOI] [PubMed] [Google Scholar]
- [35].Neidhart M, Baraliakos X, Seemayer C, et al. Expression of cathepsin K and matrix metalloproteinase 1 indicate persistent osteodestructive activity in long-standing ankylosing spondylitis. Ann Rheum Dis. 2009;68:1334–9. [DOI] [PubMed] [Google Scholar]
- [36].Hu Z, Lin D, Qi J, et al. Serum from patients with ankylosing spondylitis can increase PPARD, fra-1, MMP7, OPG and RANKL expression in MG63 cells. Clinics (Sao Paulo). 2015;70:738–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [37].Jin L, Weisman M, Zhang G, et al. Lack of association of matrix metalloproteinase 3 (MMP3) genotypes with ankylosing spondylitis susceptibility and severity. Rheumatology (Oxford). 2005;44:55–60. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.