INTRODUCTION
Intervertebral disc degeneration (IVDD) constitutes a multifactorial pathophysiological process, serving as the pathological foundation for various spinal degenerative disorders and playing a pivotal role in the onset of low back pain [1,2]. As the disease progresses, it may ultimately lead to a loss of labor and potentially impose a significant socioeconomic burden. Although the precise mechanisms driving IVDD development remain unclear, evidence indicates a significant involvement of gut microbiota (GM) and blood metabolites in its progression [3-7]. The human GM plays a crucial role in regulating aspects such as host metabolites and maintaining immunological homeostasis, among others. Dysbiosis in the GM has been linked to metabolic, immune, neurological, and musculoskeletal disorders [8-10]. Hence, we hypothesized that there might exist causal relationships between the GM, metabolites, and IVDD. Our aim was to elucidate these relationships and pinpoint potential metabolites that could serve as early diagnostic markers and clinical treatment targets.
Mendelian randomization (MR) is a powerful method utilized to uncover potential causal relationships between exposure and outcome [11]. It operates by leveraging genetic variations associated with the exposure as instrumental variables (IVs) to evaluate the connection between exposure and outcome [12]. Moreover, a growing body of evidence highlights the utility of utilizing human genetic data related to gut microbial characteristics in clinical investigations [13]. This approach allows us to apply MR as a tool for inferring causal relationships between GM and IVDD. In our study, we conducted bidirectional MR analyses and 2 mediation analyses using summary statistics derived from the most extensive and current genome-wide association studies (GWAS) of GM, blood metabolites, and IVDD. This enabled us to dissect the associations among these factors comprehensively.
MATERIALS AND METHODS
1. Study Design and Data Sources
Fig. 1 illustrates the study design, underscoring that the causal interpretations derived from MR estimates are contingent upon 3 fundamental assumptions. Specifically, the genetic variants, serving as IVs or single nucleotide polymorphisms (SNPs), are required to (1) strongly predict the exposures, (2) be associated with the outcome exclusively through these exposures, and (3) not be connected to any confounders that could affect the exposure-outcome relationship [12]. Ethical approval for each GWAS included in this study can be accessed through the respective original articles.
Genetic instruments for analyzing GM were sourced from the MiBioGen consortium (https://mibiogen.gcc.rug.nl), renowned for conducting the most comprehensive genome-wide meta-analysis that integrates human genome-wide genotypes with fecal 16S rRNA sequencing data [14]. This extensive analysis encompassed 18,340 individuals across 24 cohorts, mainly of European ancestry. At the lowest taxonomic level examined, genus, 131 genera were identified with a mean abundance greater than 1%, including 12 unidentified genera. Consequently, 119 known genus-level taxa were selected for inclusion in this analysis. Summary statistics for 1400 blood metabolites were obtained from the GWAS Catalog (https://www.ebi.ac.uk/gwas/), representing an unparalleled resource of genetic associations [15]. The outcome data were sourced from the FinnGen Consortium’s R10 release, an extensive genomic database with detailed phenotypic information. IVDD was identified using ICD-10 M51, ICD-9 722, and ICD-8 275 codes. Specifically, we focused on data labeled as “Other intervertebral disc disorders” within this dataset. Selecting this specific label was crucial to ensure an accurate and relevant definition of IVDD for our study [16].
2. Instrumental Variable
To select IVs, the following quality control procedures were implemented: First, SNPs with a p-value below the locus-wide significance threshold (1×10-5) were identified as IVs strongly linked to the GM and blood metabolites [13]. For reverse analysis, SNPs associated with IVDD meeting the conventional GWAS threshold (p<5×10-8) were selected. Second, to ensure IV independence, linkage disequilibrium among SNPs was assessed using clumping (R2<0.001, clumping distance = 10,000 kb). Third, screened SNPs were harmonized with the outcome’s GWAS summary statistics, excluding palindromic and ambiguous alleles. Fourth, the F-statistic of each SNP was calculated to assess its statistical strength, discarding those with F-statistics < 10 to mitigate weak IV bias [17]. Fifth, SNPs in the harmonized dataset strongly linked to the outcome (p<1×10-5) were manually reviewed and excluded. Sixth, the MR-PRESSO test was utilized to assess potential horizontal pleiotropy and identify outliers. The refined list of SNPs, post-outlier elimination, was then utilized for further MR analysis.
3. Two-Sample Bidirectional MR
We first conducted 2-sample bidirectional MR analyses to explore the causal relationship between the GM and IVDD. For effect estimates, we used the conventional MR approach: the inverse-variance-weighted (IVW) method. Results were reported as beta (β) values with standard errors for continuous outcomes, and odds ratios (ORs) with 95% confidence intervals (CIs) for binary outcomes; p-values below 0.05 were considered nominally significant. In addition, we performed 4 complementary methods: MR Egger, the weighted median method, the weighted mode method, and maximum likelihood.
4. Two-Step MR and MVMR
We further adopted 2 mediation approaches, 2-step MR (TSMR) and multivariable MR (MVMR) [18,19], to analyze the direct and indirect effects of the GM and 1400 blood metabolites on IVDD. TSMR assumes there is no interaction between the exposure and mediator. In addition to the basic effect estimates of GM on IVDD (β) obtained from univariate MR analyses, 2 additional estimates were calculated: (1) the causal effect of the mediator (1400 blood metabolites) on IVDD (β2), and (2) the causal effect of the exposure (the 5 bacterial genera identified in the primary MR analysis) on the mediator (β3).
Finally, we utilized MVMR as an additional method to validate the roles of metabolites identified in TSMR. MVMR estimates the controlled direct effect of the exposure on the outcome, which in our study includes the effect of metabolites on IVDD adjusting for bacteria (β2*), and the effect of bacteria on IVDD, adjusting for metabolites (β1*) [20]. The indirect effect, representing the causal effect of the GM on IVDD via mediators, can then be estimated using the product of coefficients method (β3×β2*). Consequently, the proportion mediated can be calculated as the “indirect effect divided by the total effect” ([β3×β2*]/β). Our analytic process was in accordance with the STROBE-MR guidelines [21].
5. Statistical Analysis
We conducted sensitivity analyses using the MR-Egger regression, leave-one-out, and MR-PRESSO methods. The MREgger regression and MR-PRESSO were utilized to test for and correct potential horizontal pleiotropy in our selected IVs. MR-Egger regression was employed to account for horizontal pleiotropy by relying on the InSIDE (Instrument Strength Independent of Direct Effect) assumption, which ensures the strength of genetic variants’ effect on the exposure is independent of their effect on the outcome [22]. MR-Egger estimates both the causal effect (slope) and pleiotropy (intercept), with a non-zero intercept indicating pleiotropic bias. The p-value from the MR-Egger regression intercept test was applied to evaluate horizontal pleiotropy, and the method provided bias-adjusted causal estimates and robustness assessments. Additionally, MR-PRESSO was used to detect and eliminate outliers among the IVs. MR-PRESSO includes 3 components: the Global Test to identify overall pleiotropy, the Outlier Test to flag and remove pleiotropic IVs, and the Distortion Test to assess the impact of outlier correction on causal estimates [23]. By systematically addressing pleiotropy, MR-PRESSO enhanced the reliability of our causal estimates. Furthermore, Cochrane Q statistic was applied to assess the variability of SNP estimates across each MR association, and MVMR analysis was implemented to adjust for confounding risk factors, effectively minimizing their influence on the causal inference. The statistical methods employed in our study have been thoroughly reviewed and rigorously validated by a statistician to ensure their accuracy and reliability.
To control for multiple comparisons in our analysis, we applied the false discovery rate (FDR) correction using the q-value approach. The q-value provides an estimate of the proportion of false positives among the set of significant results, allowing for a balance between statistical power and the control of false discoveries. This method was chosen due to its suitability for large-scale genomic data and its ability to provide interpretable significance thresholds [24]. All IVW results were adjusted for multiple testing using the FDR method, with a significance threshold set at q-value < 0.1. Results indicating a p-value < 0.05 but a q-value ≥ 0.1 were considered suggestive of an association. All MR analyses were conducted in R ver. 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria), utilizing the “TwoSampleMR” and “MendelianRandomization” packages. FDR q-values were estimated using the “p.adjust” function in R.
RESULTS
1. Selection of IVs
The number of SNPs used as IVs was 1,256 for 119 bacterial genera in the MiBioGen consortium, 28,860 for the 1400 blood metabolites in the GWAS Catalog, and 170 for IVDD in the FinnGen repository. All IVs exhibited F-values greater than 10, demonstrating their robustness against bias introduced by weak IVs. Details about the selected IVs and the manually excluded SNPs are provided in Supplementary Tables 1-4.
2. Causal Effects of GM on IVDD
When evaluating the causal effects of GM on IVDD, 4 bacterial genera were found to be negatively associated with IVDD, while 3 genera showed positive associations with IVDD using the IVW method (Fig. 2, Supplementary Table 5). Although the Eubacteriumbrachygroup and RuminococcaceaeUCG011 genera were linked causally to IVDD, the MR-Egger method indicated an opposite direction, suggesting that the causality might be invalid [22]. Among the other 5 genera, the Eubacteriumcoprostanoligenesgroup genus demonstrated the strongest effect on the risk of IVDD (OR, 0.842; 95% CI, 0.769–0.922; p<0.001). The IVW results for the Eubacteriumcoprostanoligenesgroup genus remained significant after corrections for multiple testing (FDR q-value=0.025). In the reverse MR analysis, we found no significant causal effect of IVDD on GM (Table 1). Furthermore, these results were deemed reliable, showing no evidence of pleiotropy, as confirmed by a sensitivity analysis (Supplementary Tables 6, 7).
3. Mediation Analyses of 1400 Blood Metabolites
In the TSMR analysis (Fig. 3), it was found that a total of 84 blood metabolites were causally associated with IVDD (Supplementary Table 8). In the sensitivity analysis, the MR-PRESSO test identified 5 SNPs with pleiotropy (rs6851940, rs4517148, rs147750540, rs2393791, rs3809095). After removing these SNPs, we found that the causal relationships between 2 blood metabolites and IVDD changed (p>0.05). In the leave-one-out analysis, we identified one pleiotropic IV (rs139446584). After removing this IV, the causal relationship between the corresponding blood metabolite and IVDD also changed (p>0.05). Additionally, although there were 14 blood metabolites causally associated with IVDD, the MR-Egger method showed opposite trends, suggesting that the causal relationship may not hold [22]. Among the remaining 67 blood metabolites, after multiple testing correction, the IVW result for 2-methylserine showed significance (FDR q=0.065). Although the Cochrane Q test indicated heterogeneity for 5 metabolites (p<0.05) (Supplementary Table 6), this heterogeneity did not affect the overall result. The presence of heterogeneity may be attributed to the data coming from different analytical platforms or experiments, etc.
Of the 5 bacterial genera causally associated with IVDD, 4 were significantly correlated with 9 of the 67 metabolites mentioned above (Table 2, Supplementary Table 9). However, for Eubacteriumfissicatenagroup with Phosphocholine and Lachnoclostridium with Phosphate/N-palmitoyl-sphingosine (d18:1 to 16:0), although they were causally related, the MR-Egger method indicated opposite directions, suggesting that the causality might be invalid [22]. The Lachnoclostridium genus, identified as a protective taxon against IVDD (OR, 0.914; 95% CI, 0.843–0.990; p=0.028), was associated with increased levels of histidine. The RuminococcaceaeUCG003 genus also demonstrated protective effects against IVDD (OR, 0.920; 95% CI, 0.856–0.989; p=0.024) by downregulating Pimeloylcarnitine/3-methyladipoylcarnitine (C7-DC) and X-24456. However, the Eubacteriumfissicatenagroup (OR, 1.058; 95% CI, 1.008–1.111; p=0.024) and Marvinbryantia (OR, 1.111; 95% CI, 1.007–1.226; p=0.035) genera were found to have detrimental effects on IVDD. Specifically, the Eubacteriumfissicatenagroup genus was associated with detrimental effects on IVDD through downregulation of X-24949 levels. Similarly, the Marvinbryantia genus exhibited detrimental effects on IVDD by upregulating 2-methylserine and downregulating Phosphocholine and N-delta-acetylornithine. Moreover, these results were deemed reliable, with no evidence of pleiotropy, as confirmed by a sensitivity analysis (Supplementary Tables 6, 7).
We conducted MVMR to validate the mediating effects of blood metabolites uncovered in TSMR (Table 3). We calculated the indirect effect and proportion mediated by these metabolites and found that the roles of X-24949, histidine, and C7-DC remained significant after adjusting for GM. Overall, we observed an indirect effect of X-24949 in the associations between the Eubacteriumfissicatenagroup genus and IVDD, with a mediated proportion of 8.1% (p=0.024); Histidine in the association between the Lachnoclostridium genus and IVDD with a mediated proportion of 18.1% (p=0.013); and C7-DC in the association between the RuminococcaceaeUCG003 genus and IVDD with a mediated proportion of -7.5% (p=0.041). The effects of X-24456, 2-methylserine, Phosphocholine, and N-delta-acetylornithine were insignificant after adjusting for the GM.
DISCUSSION
In this large-scale MR study, we found that 5 bacterial genera were causally associated with IVDD, while IVDD did not exhibit a significant causal effect on the GM. The IVW results for the Eubacteriumcoprostanoligenesgroup genus remained significant after corrections for multiple testing. Through a 2-step MR analysis, we identified several bacterial genera, including Eubacteriumfissicatenagroup, RuminococcaceaeUCG003, Lachnoclostridium, and Marvinbryantia, as well as metabolites such as X-24949, C7-DC, X-24456, histidine, 2-methylserine, Phosphocholine, and N-delta-acetylornithine, which were significantly associated with IVDD. Moreover, the IVW results for 2-methylserine remained significant after corrections for multiple testing. Additionally, our MVMR analysis revealed that the associations between Eubacteriumfissicatenagroup and IVDD were mediated by X-24949, between Lachnoclostridium and IVDD by histidine, and between RuminococcaceaeUCG003 and IVDD by C7-DC.
The Eubacteriumcoprostanoligenesgroup genus was observed to have a notable negative impact on IVDD in our study, exhibiting significant potency. These anaerobic gram-positive bacteria are capable of converting cholesterol into coprostanol, rendering it unabsorbable and thereby influencing cholesterol levels [25]. Previous research has highlighted cholesterol’s role in promoting IVDD through apoptosis and pyroptosis in NP cells, as well as extracellular matrix metabolism [26]. However, our study did not find any evidence of the Eubacteriumcoprostanoligenesgroup genus directly regulating cholesterol levels. The RuminococcaceaeUCG003 genus, a member of the Ruminococcaceae family, along with many bacteria within this family, are known to be general butyrate-producing bacteria. However, butyrate alleviates inflammatory response and nuclear factor-kappa B (NF-κB) activation in human degenerated intervertebral disc tissues [27]. Therefore, the influence of RuminococcaceaeUCG003 on IVDD might be mediated by butyrate. In this study, we found that the RuminococcaceaeUCG003 genus also demonstrated protective effects against IVDD by downregulating C7-DC and X-24456. C7-DC, an important acylcarnitine, may play a significant role in IVDD. Acylcarnitines are esterified forms of carnitine and are essential for transporting long-chain fatty acids (FAs) across the mitochondrial inner membrane for β-oxidation [28]. In metabolic disorders, the concentration of acylcarnitines is often altered, suggesting that C7-DC may influence the energy balance of intervertebral disc cells by regulating FA oxidation [29-31]. In the context of IVDD, the metabolic regulatory role of C7-DC may help slow down cellular energy depletion and degenerative changes. Overall, the role of C7-DC in IVDD warrants further investigation, particularly regarding its impact on FA metabolism and metabolic risks. Additionally, based on the existing results, the mediated proportion of C7-DC (-7.5%) suggests that it may play a certain role in the degenerative process of IVDD. We believe that although the clinical relevance of this effect needs further validation, it provides new directions for future research on C7-DC as a potential therapeutic target for IVDD. Lachnoclostridium genus is believed to be causally linked to a variety of diseases. In this study, it was also identified as a protective taxon against IVDD, mediated by increased levels of histidine. The metabolic product of histidine, histamine, is known to possess anti-inflammatory effects, primarily by regulating the NF-κB signaling pathway, which plays a central role in many inflammatory responses. In IVDD, proinflammatory cytokines such as interleukin (IL)-6, tumor necrosis factor-α and IL-1β activate NF-κB, triggering inflammation and degenerative changes in intervertebral disc tissue [32,33]. Studies have shown that histidine can inhibit NF-κB activation and reduce the degradation of IkBα, thereby alleviating the inflammatory response [34]. Therefore, histidine may help alleviate the degenerative process of IVDD by inhibiting the NF-κB pathway and reducing the production of inflammatory mediators such as IL-6. However, the precise underlying mechanism still requires further investigation.
The Eubacteriumfissicatenagroup genus is part of potentially disease-related bacteria that add to the risk of metabolic disorders and intestinal inflammation [35]. However, it is also recognized as a group of butyrate-producing bacteria and beneficial bacteria that suppress intestinal inflammation [36]. In this study, the Eubacteriumfissicatenagroup genus was found to be associated with detrimental effects on IVDD through the downregulation of X-24949 levels. Marvinbryantia represents a cellulose-degrading bacterial genus, comprising solely of one known species called Marvinbryantia formatexigens, which naturally resides within the human gut [37]. Marvinbryantia formatexigens is known as a gut acetogen that produces acetate in the human gut. Interestingly, a key pathway for producing butyrate in the human GM involves using externally sourced acetate to generate butyrate through butyryl-CoA: acetate Co-A transferase [38]. This suggests that the production of fecal butyrate may depend on acetate generation from Marvinbryantia formatexigens. However, butyrate alleviates inflammatory response and NF-κB activation in human degenerated intervertebral disc tissues [27]. Therefore, the influence of Marvinbryantia on IVDD might also be mediated by butyrate. Our study identifies Marvinbryantia genus as exerting detrimental effects on IVDD by upregulating 2-methylserine while concurrently downregulating Phosphocholine and N-delta-acetylornithine. Nevertheless, the precise mechanism underlying this influence necessitates further investigation.
To further enhance the clinical relevance of this work, these findings could be applied to the prevention and personalized treatment of IVDD by targeting the identified bacterial genera and metabolites. For instance, modulating the Eubacteriumcoprostanoligenesgroup genus, which negatively impacts IVDD, could provide a novel strategy for prevention by altering its metabolic pathways to reduce cholesterol accumulation. On the other hand, enhancing the protective effects of butyrate-producing bacteria, such as RuminococcaceaeUCG003, may alleviate inflammatory responses in the intervertebral discs by increasing butyrate production, thus promoting disc health. Additionally, the metabolites identified in this study, such as 2-methylserine and histidine, could serve as potential biomarkers for early diagnosis and personalized treatment plans for IVDD. Future studies should explore how these bacterial and metabolic alterations can be leveraged to develop more precise microbiome-based interventions, such as probiotics or metabolic modulators, to improve treatment outcomes and slow disease progression in IVDD patients.
Nevertheless, this study was subject to several limitations. To begin with, one limitation of our study is the assumption of linearity in the MR analysis, which may not fully capture the complexity of biological systems. Nonlinear relationships, threshold effects, and interactions between bacterial genera and host factors could play significant roles. Future research should incorporate nonlinear MR models and interaction analyses, and network-based approaches could provide further insights into these complex dynamics, enhancing our understanding of causal pathways. Furthermore, the MiBioGen consortium characterizes microbiome profiles using 16S ribosomal RNA gene sequencing, limited to genus-level taxonomic classification. However, metagenomic sequencing offers detail at the species level. Previous MR study of the GM found that higher taxonomic units sometimes yielded more significant p-values, indicating similar functionalities among species [39]. Additionally, disparities in genetic variant distribution among ethnic or racial groups may introduce population stratification, potentially skewing study outcomes [40]. Hence, caution should be exercised when extending findings to other demographic groups. Furthermore, we applied the FDR correction, which is particularly suitable for genomewide studies involving thousands of comparisons, as it strikes a balance between detecting true positives and controlling false discoveries. Unlike the overly conservative Bonferroni correction, which can reduce statistical power and miss subtle associations, FDR maintains sensitivity while providing interpretable significance thresholds [41]. Although some results with marginal q-values (e.g., q=0.095) fall within the FDR-controlled range, we recognize the importance of validating these findings in independent datasets. Future studies will focus on replicating these associations in diverse populations to enhance their reliability and generalizability. Additionally, future research should carefully consider the interaction effects between GM and IVDD in MR analyses to provide a more comprehensive understanding of these complex relationships.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement
Print
