Causal relationships between the gut microbiota, immune cells, and diabetic kidney disease: a Mendelian randomization and mediation analysis Authors: Han Yan, Haipeng Pang, Lingxiang Xie, Jin Ding, Yiman Peng, Yuanqin Yang, Zhiguang Zhou, Xi Li, Jingyi Hu Published: 2026 ABSTRACT AbstractObservational studies have indicated that the gut microbiota is associated with diabetic kidney disease (DKD). However, it remains unclear whether disturbances in the gut microbiota causally contribute to DKD. This study aimed to investigate the causal relationships between the gut microbiota and DKD and explore peripheral immune cell traits as potential mediators. Summary statistics were extracted from genome-wide association studies of the Dutch Microbiome Project (N = 7,738), SUMMIT Consortium (N = 10,875), and GWAS summary statistics for immune cell traits (N = 3,757). We used a two-sample two-step mediation Mendelian randomization analysis to investigate genetic causality between immune cell traits and the phenotypes of DKD, causal effect of gut microbiota/microbiota metabolism pathways on DKD, and mediation effects of immune cell traits. In total, seven immune cell traits and five taxa were causally associated with the DKD phenotypes after multiple testing corrections. We performed mediation analysis on gut microbiota and immune cell traits that passed the Bonferroni correction, and identified five mediation relationships. For instance, mediation analysis indicated that Pseudoflavonifractor capillosus had adverse effects on chronic kidney disease in type 2 diabetes by downregulating CD25 expression in CD4+ regulatory T cells (proportion mediated = 19.8%, p = 0.035). These results indicate that the gut microbiota has causal effects on DKD and that immune cell traits exert a mediation effect to some extent. Introduction Diabetic kidney disease (DKD), as a complication of diabetes, is the most common cause of chronic kidney disease (CKD) and an important contributor to end-stage renal disease (ESRD). Approximately half of patients with type 2 diabetes (T2D) and one-third with type 1 diabetes (T1D) will develop DKD [1–3]. Among the multiple chronic complications of diabetes, DKD imposes a substantial financial and health burden on diabetic patients because conventional treatments such as renin-angiotensin-aldosterone system blockade and blood glucose control only delay the disease progression [4]. The precise pathogenesis of DKD remains unclear, and a lack of comprehensive awareness of the progression of DKD restricts its prevention and management. Recently, disturbance of gut microbiota has been reported to contribute to various diseases, including DKD [5–7]. The gut microbiota is composed of complex microbial community with more than 100 trillion microorganisms and has 150 times more genes than the human genome [8]. Human gut microbiota plays an important role in physiological activities including protecting the gastrointestinal functions, absorbing the nutrients, and maintaining the immune-inflammatory homeostasis. The gut microbiota-derived metabolites such as short-chain fatty acids (SCFAs) maintain the immune homeostasis through regulating the immune cells via their respective receptors [9]. For instance, it has been indicated that SCFAs could promote the naïve T cells toward regulatory T cells (Tregs) [10]. In the context of DKD, the increased permeability of the intestinal barrier leads to transfer of toxins into the blood circulation, and impaired renal function further leads to endotoxin accumulation, causing persistent systemic inflammation [5]. It has been indicated that gut microbiota composition is associated with DKD occurrence [11]. Decreased diversity and biased community structure of gut microbiota are closely associated with kidney function failure [12]. Existing evidence supports the associative relationship between gut microbiota and DKD, but their causal relationship remains unknown. Mendelian randomization (MR), using genetic variants associated with proposed risk factors as instrumental variables, is a widely accepted methodology to estimate the causal associations between exposures and outcomes of interest [13]. The MR approach is generally considered to avoid confounding factors due to the random distribution of alleles [14]. In addition, the effect estimated by MR method is not affected by reverse causality bias since genetic variation is assigned at conception. Currently, a vast number of single nucleotide polymorphisms (SNPs) associated with DKD and gut microbiota have been identified by well-powered genome-wide association studies (GWASs), which enables us to estimate their causal association by means of MR method. Here, we used two-sample bidirectional MR and mediation MR methods to explore the causal relationship between gut microbiota and DKD and assess the potential effect of peripheral immune cell traits as mediators. This study aims to fill the gap by employing MR to rigorously establish causality. The primary novelty of our work lies in its two focuses: to definitively establish the causal role of the gut microbiota in DKD, moving beyond associative findings, and to investigate for the first time the role of peripheral immune cell traits as potential mediators in this causal pathway. By constructing an immune-mediated regulatory network from gut microbiota to DKD, we provide a new mechanistic perspective that could inform future therapeutic strategies. Materials and methods Study design The genetic summary data of gut microbiota, microbiota metabolism pathways, DKD, and immune cell signatures were extracted from the GWAS catalog (https://www.ebi.ac.uk/gwas/). The SNPs were functionally annotated using dbSNP155 database. We utilized SNPs as instrumental variables (IVs) to proxy for exposures. Firstly, to investigate the causal relationships between peripheral immune cell traits, gut microbiota and DKD, we conducted two-sample MR analysis by utilizing immune cell traits and gut microbiota as exposures and multiple phenotypes of DKD as outcomes, and screened candidate immunophenotypes and gut microbiota with significant causal relationships with DKD phenotypes. We further used the Bonferroni method for multiple testing to select core immune cell traits and gut microbiota whose P-value reached the significance threshold of 0.05/n (where n represents the effective number of independent exposures). Next, to construct the immune regulatory network from gut microbiota, immune cell traits to DKD, we analyzed the causal relationships between candidate gut microbiota and core immune cells, as well as between core gut microbiota and candidate immune cells. We established the causal pathways from gut microbiota to DKD through the mediating immunophenotypes by conducting mediation analysis based on two-step MR. Data sources and instrument variables selection For the exposure gut microbiota, the genetic variation was obtained from a GWAS comprising 7,738 European-ancestry individuals conducted by the Dutch Microbiome Project (DMP) [15]. Currently, this is the largest gut microbiota database with species-level data, and the participants are relatively homogeneous. The 206 taxa (5 phyla, 10 classes, 13 orders, 26 families, 48 genera and 104 species) and 205 bacterial metabolism pathways were identified by the shotgun metagenomic sequencing. For the outcome DKD, the summary genetic statistics were extracted from the SUMMIT Consortium [16]. There were 16 reported phenotypes of DKD. These phenotypes comprise: (1) diabetic kidney disease in diabetes (end-stage renal disease [ESRD] vs. no ESRD); (2) chronic kidney disease in type 2 diabetes (CKD in T2D); (3) early diabetic kidney disease in type 2 diabetes (early DKD in T2D); and other related traits. Throughout this study, we refer to these collectively as ‘DKD phenotypes’ for brevity, with the understanding that ‘CKD in T2D’ specifically denotes chronic kidney disease occurring in the context of type 2 diabetes, which is a manifestation of DKD. A detailed list of all 16 phenotypes is provided in Supplementary Table 1. For the mediators immunophenotypes, the summary statistics were available from the public GWAS Catalog, which profiled 3,757 Sardinians by flow cytometry [17]. A total of 731 immune cell traits including absolute cell counts (n = 118), relative cell counts (n = 192), median fluorescence intensities reflecting surface antigen levels (n = 389), morphological parameters (n = 32) were identified. SNPs associated with gut microbiota, immunophenotypes and DKD with p < 1 × 10−5 and effect allele frequency (EAF) > 0.01 were selected as potential IVs [14]. All identified SNPs were clumped at a linkage disequilibrium (LD) threshold of r2 < 0.001, window size = 10000 kb. For harmonization of exposure and outcome summary statistics, we used the harmonise_data function in TwoSampleMR (version 0.5.6). All palindromic SNPs with ambiguous strand orientation were dropped to avoid potential strand-flipping errors. We used the R packages ieugwasr and plinkr to conduct the local cluster analysis and set the 1000 Genomes European reference panel as the reference file (data_maf0.01_rs_ref. tgz). The F value of each exposure was also calculated, only the exposure with F > 10 was selected for the subsequent analysis. Statistical analysis The method selected for MR analysis depends on the number of IVs available. When more than one IV is present, the Inverse Variance Weighted (IVW) method is implemented. Conversely, in instances where only a single IV is available, the Wald ratio model is utilized. We also performed several other analyses including MR Egger, Weighted median, Simple mode, and Weighted mode to validate the robustness of results from IVW method. P-value less than 0.05 was considered to have a suggestively significant association. To assess heterogeneity and pleiotropy, we used Cochran’s Q test and the MR Egger intercept, respectively. If evidence of heterogeneity or pleiotropy (p < 0.05) was detected, the result would be excluded from subsequent analyses. In addition, Steiger test was also conducted to avoid reverse causality. In mediation analysis, two-step MR was used to decompose direct and indirect effects. The proportion of the effect mediated by immune cells was assessed by dividing the indirect effect by the total effect (βem×βmo/βeo). The βem was the effect of gut microbiota on the immunophenotype; βmo was the effect of immunophenotype on DKD; and βeo was the effect of gut microbiota on DKD. Firstly, we defined significant taxa, pathways, and immunophenotypes as those with raw p < 0.05. We then defined candidate taxa and immunophenotypes as those that have passed multiple testing correction. Next, we performed MR analysis by setting either the significant taxa or pathways as exposure and the candidate immunophenotypes as outcome, or alternatively, setting the candidate taxa as exposure and the significant immunophenotypes as outcome. We then assigned the P and β from the MR analysis of taxa or pathways to DKD phenotypes as Peo and βeo, respectively, assigned the P and β from the MR analysis of immunophenotypes to DKD phenotypes as Pmo and βmo, respectively, and assigned the P and β from the MR analysis of taxa or pathways to immunophenotypes as Pem and βem, respectively. We constructed mediation relationships based on these MR analysis results combined with the previous three parts of the MR analysis results. Only those factors with Peo, Pem, and Pmo all less than 0.05 and βem *βmo having the same sign as βeo were included in the mediation analysis. The MR analyses were performed on the R platform (4.2.3) and the TwoSampleMR package. Because the P-value of βem and βmo were both less than 0.05. Figure 1 displays the flow chart of this study. To exclude the possibility of reverse causality, we conducted reverse MR analyses by using genetic variants associated with each DKD phenotype (p < 1 × 10−5, clumping r2 < 0.001, window size = 10,000 kb) as the exposure and the gut microbiota or immune cell traits as the outcome. The same IV selection criteria (F > 10) were applied. IVW was used as the primary method, with weighted median and MR Egger as sensitivity analyses. Detailed information on significant instrumental SNPs used for each exposure and mediator, including effect allele, β-value, standard error, P-value, sample size, is provided in Supplementary Table 2. Flowchart of bidirectional two-sample Mendelian randomization and mediation analyses. Specifically, for each identified mediation relationship (exposure: gut microbiota or pathway; mediator: immune cell trait; outcome: DKD phenotype), we performed 10,000 bootstrap resamples of the original summary-statistics-based two-step MR estimates. In each bootstrap iteration, we resampled the SNP-exposure and SNP-outcome effect estimates with replacement, accounting for their standard errors, and recalculated the indirect effect as the product of the two path coefficients. From the 10,000 bootstrap estimates, we derived the 95% confidence interval (CI) using the percentile method. The indirect effect was considered statistically significant if the P is less than 0.05. To assess the direct causal effect of each gut microbiota taxon or pathway on DKD phenotypes while adjusting for potential confounding or mediating immune cell traits, we performed multivariable MR using the ‘MVMR’ package in R. Conditional F-statistics were calculated to evaluate instrument strength in the multivariable setting. The MVMR-IVW method was used as the primary analysis, and the direct effect estimates (β-direct) along with their 95% CI are reported. The Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) test was applied to detect and correct for horizontal pleiotropy. The global test was used to evaluate overall pleiotropy, and if significant (p < 0.05), the outlier-corrected causal estimates were reported after removing the identified outlier SNPs. MR-PRESSO was implemented using the ‘MRPRESSO’ R package with 5,000 permutations. For each primary association, we sequentially removed each instrumental SNP and recalculated the IVW causal estimate. The stability of the result was examined through leave-one-out plots, and no single SNP was identified as disproportionately driving the causal estimate. This analysis was performed using the ‘TwoSampleMR’ package in R. Results Causal associations of immune cell traits and DKD Among 731 immune cell signatures, 726 successfully acquired IVs and we detected causal effects of 388 immunophenotypes on DKD-related phenotypes with IVW p < 0.05 (Supplementary Table 3 and Supplementary File 1). After ruling out contradictory results in different DKD phenotypes, 80 immunophenotypes were negatively associated with DKD while 93 immunophenotypes were positively associated with DKD. The majority of immune cell traits significantly associated with DKD are T cell phenotypes, with 43/80 presenting a negative correlation and 37/93 showing a positive correlation. In addition, the proportion of CD25-positive traits in the negative association immunophenotypes was significantly higher than their proportion in all immunophenotypes (14/80 vs. 55/726, p = 0.003). After adjustment for multiple testing via the Bonferroni procedure (p < 0.05/731), 7 immunophenotypes were still significantly associated with DKD phenotypes (Figure 2A). The most significant result for DKD (DKD in diabetes, ESRD vs. no ESRD) is Natural Killer/lymphocyte (p = 7.28 × 10−5, OR = 0.762, 95% CI = 0.667–0.872). Two immunophenotypes have been found as risk factors for CKD and DKD in diabetes, including CD127-CD8 + T cell/CD8 + T cell (p = 1.48 × 10−6, OR = 1.503, 95% CI = 1.337–1.669) and CD45RA + CD8 + T cell Absolute Count (p = 4.92 × 10−5, OR = 1.057, 95% CI = 1.030–1.084). One immunophenotype, CD8 on CD28 + CD45RA-CD8+ T cell (p = 1.63 × 10−6, OR = 0.712, 95% CI = 0.557–0.866), was identified as a protective factor for CKD and DKD in diabetes. Moreover, Natural Killer T/T cell (p = 6.34 × 10−6, OR = 1.293, 95% CI = 1.181–1.404) and CD19 on B cell (p = 1.56 × 10−5, OR = 1.487, 95% CI = 1.307–1.667) were found to be risk factors for CKD in T2D and ESRD in diabetes, respectively. While CD25 on secreting CD4 regulatory T cell was found to be negatively associated with CKD in T2D (OR = 0.513, 95% CI = 0.189–0.838, p = 5.54 × 10−5). Reverse causal association was detected for CD45RA + CD8 + T cell absolute count and CKD and DKD in diabetes, therefore, we didn’t include it in the subsequent mediation analysis. Gut microbiota and immune cell associated with diabetic kidney disease (DKD). (A) Immune cell traits that were significantly associated with phenotypes of DKD after adjustment for multiple testing via the Bonferroni procedure. Each dot represents an immune cell trait. (B) Gut microbiota that was significantly associated with phenotypes of DKD after adjustment for multiple testing via the Bonferroni procedure. Each dot represents a microbial taxon. (C) Candidate microbiota metabolism pathways identified by Mendelian randomization. Each dot represents a microbiota metabolism pathway. We assessed instrument strength by calculating variance explained (R2) for each genetic variant, and performed Steiger directionality tests to verify the assumed causal direction (Supplementary Table 4). All instruments demonstrated strong explanatory power (exposure R2 > outcome R2) and highly significant Steiger P-values, confirming adequate instrument strength and the correct causal direction without evidence of reverse causation. Causal associations of gut microbiota and DKD All 206 taxa successfully acquired IVs and we detected causal effects of 195 taxa on DKD-related phenotypes with IVW or Wald ratio p < 0.05 (Supplementary Table 5 and Supplementary File 1). A total of 34 taxa showed significant causal effects on DKD. Among them, 15 taxa exhibited protective effects against DKD, while the others showed risky effects. Among the protective taxa, those belonging to the phylum Firmicutes have the highest proportion (9/15), while among the risk taxa, those belonging to the phylum Bacteroidetes have the highest proportion (7/19). The most significant result was for the phylum Firmicutes, which is a protective factor for DKD in T2D with an OR of 0.543 and a 95% CI of 0.412–0.717 (p = 1.56 × 10−5). Interestingly, the class Clostridia and order Clostridiales also showed a protective effect like Firmicutes on DKD in T2D. These two taxa have the same P value and OR (p = 3.16 × 10−5, OR = 0.553, 95% CI = 0.418–0.731). Species Bifidobacterium catenulatum was identified as a risk factor for DKD in diabetes (ESRD vs. no ESRD) with a P value of 1.74 × 10−4 (OR = 1.796, 95% CI = 1.490–2.102). All of these four taxa have passed the multiple testing correction (p < 0.05/206) (Figure 2B). For the other phenotypes, only class Clostridia and order Clostridiales passed multiple testing correction. They were found to be negatively associated with early DKD in T2D with a P value of 5.37 × 10−5 (OR = 0.412, 95% CI = 0.271–0.636) (Figure 2B). No reverse causal association has been observed for the taxa that passed the multiple testing correction. To assess whether our IVs had sufficient statistical power to detect causal associations, we performed power calculations using the mRnd online tool (https://shiny.cnsgenomics.com/mRnd/). For binary outcomes (DKD phenotypes), the total sample size was set to 40,340 (the total sample size of the SUMMIT consortium), with the proportion of cases set at 0.4. Assuming ORs of 0.9 (protective effect) and 1.1 (risk effect), which correspond to β-values of approximately 0.105 and 0.095, respectively, and an r2 of 0.09, the calculated statistical power was 0.87 for the protective effect and 0.81 for the risk effect. For the vast majority of exposure-outcome associations that reached statistical significance, the power exceeded 0.80, indicating that our analyses were adequately powered to detect true causal effects. Causal associations of microbiota metabolism pathways and DKD As with the taxa, all 205 microbiota metabolism pathways acquired IVs and we detected causal effects of 180 pathways on DKD-related phenotypes with IVW or Wald ratio p < 0.05 (Supplementary Table 6). After ruling out contradictory results in different DKD phenotypes, 18 pathways were identified as protective factors for DKD while 17 pathways were found to be risk factors of DKD. However, none of the pathways passed the multiple testing correction (p < 0.05/205) (Figure 2C). Top three pathways causally associated with DKD were ornithine degradation pathway (p = 0.004, OR = 0.651, 95% CI = 0.485–0.874), glucose and xylose degradation pathway (p = 0.005, OR = 0.608, 95% CI = 0.428-0.863) and L-glutamate degradation VIII to propanoate pathway (p = 0.006, OR = 0.771, 95% CI = 0.640–0.928), all of which were protective factors for DKD. The increase of the abundance of them would decrease the risk of DKD. Reverse Mendelian randomization analysis In reverse MR analyses using DKD-related genetic variants as the exposure, and immune cell traits and gut microbiota as the outcome, the IVW method showed that most immune cell phenotypes and gut microbiota were not significantly associated with DKD (p > 0.05), arguing against a systematic reverse causal effect (Supplementary Table 7). A few traits exhibited nominally significant associations (p < 0.05), but the effect directions were inconsistent across traits and sensitivity analyses (weighted median, MR Egger) did not uniformly support these findings. Overall, these results do not provide robust evidence for reverse causation from DKD to the examined immune cell traits and gut microbiota. Mediation analysis of microbiota, immunophenotype, and DKD In total, 5 mediation relationships have been identified (Table 1). Phylum Firmicutes and class Clostridia were causally associated with decreased risk of DKD. Mediation analysis showed that CD127 on CD28+ CD45RA- CD8+ T cell mediated the causal effects of phylum Firmicutes (proportion mediated = 5.98%) and class Clostridia (proportion mediated = 6.16%) on DKD. In addition, for other phenotypes of DKD, we also performed mediation analysis. For instance, species Pseudoflavonifractor capillosus was causally associated with the increased risk of DKD. Besides, we found that the species Pseudoflavonifractor capillosus causally decreased CD25 on CD4+ Tregs and this decrease was subsequently associated with an increased risk of chronic kidney disease in T2D, with a mediated proportion of 19.8% (Figure 3). Genus Eubacterium causally increased CD19 on B cell and this increase was subsequently associated with an increased risk of end-stage renal disease in diabetes, with a mediated proportion of 24.06% (Figure 3). The immune cell traits mediated the causal effect of gut microbiota on diabetic kidney disease. The β-values are the MR estimates. Table 1.Mediation analysis results.ExposureMediatorOutcomeORemPemOReoPeoORmoPmoProportion (%) p Species Pseudoflavonifractor capillosusCD25+ CD4+ regulatory T cellChronic kidney disease in T2D0.865(0.757–0.989)0.0341.632(1.117–2.385)0.0110.513(0.371–0.710)5.54E-0519.750.035Genus EubacteriumCD19 on B cellEnd-stage renal disease in diabetes1.341(1.006–1.788)0.0451.623(1.011–2.606)0.0451.487(1.242–1.781)1.56E-0524.060.048Phylum FirmicutesCD127 on CD28+ CD45RA- CD8+ T cellDiabetic kidney disease0.733(0.549–0.980)0.0360.543(0.412–0.717)1.56E-051.125(1.042–1.214)0.0025.980.038Class ClostridiaCD127 on CD28+ CD45RA- CD8+ T cellDiabetic kidney disease0.733(0.549–0.980)0.0360.553(0.419–0.731)3.16E-051.125(1.042–1.214)0.0026.160.038Coenzyme A biosynthesis I (COA-PWY)CD25+ CD4+ regulatory T cellChronic kidney disease in T2D0.779(0.610–0.995)0.0452.409(1.237–4.690)0.0100.513(0.371–0.710)5.54E-0518.930.044eo, exposure to outcome; em, exposure to mediator; mo, mediator to outcome. p Values were derived from 10,000 bootstrap resamples (percentile method). Mediation analysis results. eo, exposure to outcome; em, exposure to mediator; mo, mediator to outcome. p Values were derived from 10,000 bootstrap resamples (percentile method). Multivariable Mendelian randomization results Multivariable MR analyses adjusting for immune cell traits revealed that coenzyme A biosynthesis pathway I had a significant direct positive effect on chronic kidney disease in type 2 diabetes (β = 0.671, p = 0.033), while CD25 on CD4+ regulatory T cells showed a direct protective effect (β=-0.638, p = 0.002); however, Pseudoflavonifractor capillosus did not retain a significant direct effect after adjustment for the same immune trait (β = 0.322, p = 0.150). For DKD in diabetes (ESRD vs. no ESRD), class Clostridia consistently demonstrated direct protective effects independent of several immune mediators, whereas the immune cell traits themselves were not significantly direct. For early DKD in T2D, Holdemania filiformis showed a significant direct protective effect (β=–0.558, p = 0.002). No significant direct effects were observed for the other tested exposures or outcomes. Complete MVMR estimates are provided in Supplementary Table 8. Sensitivity analyses To assess the robustness of our MR findings and to detect potential horizontal pleiotropy, we performed MR-PRESSO global tests and leave-one-out analyses for all exposure-outcome pairs. The MR-PRESSO global test showed no evidence of significant horizontal pleiotropy for the vast majority of analyses (global p > 0.05) (Supplementary Table 9). For the few instances where the global test indicated pleiotropy (p < 0.05), we applied the MR-PRESSO outlier-correction method. The corrected causal estimates remained directionally consistent and statistically significant for all core exposure-mediator-outcome relationships that were subsequently included in the mediation analysis. Leave-one-out analyses further confirmed that no single SNP disproportionately drove the causal estimates, as the effect sizes and significance levels remained stable after sequentially removing each instrumental variable (Supplementary Table 10 and Supplementary File 2). Taken together, these sensitivity analyses support the robustness of our primary MR results and indicate that horizontal pleiotropy does not materially bias our conclusions. Discussion In this large-scale and comprehensive MR study, we identified 5 gut taxa and 7 peripheral immune cell traits that significantly affected DKD phenotypes. We also established 8 causal pathways from gut microbiota to DKD and its related phenotypes through mediation analysis. To our knowledge, this is the first study to explore the causal associations between gut microbiota and DKD, and the role of peripheral immune cell traits as potential mediators. In the past few years, multiple observational studies have revealed the contributory role of gut microbiota in the pathogenesis of DKD by using 16S rRNA sequencing. Disturbance of the gut environment and composition of the gut microflora lead to disorders of host-microbiota interaction and a systemic inflammatory response, thus influencing kidney function [18]. It has been revealed that the degree of diversity of bacterial groups and richness of gut microbiota was significantly different in DKD compared with diabetes [11]. Although numerous case-control or cross-sectional studies have characterized the gut microbiota in DKD, poor reproducibility of inconsistent results describing the microbial differences in DKD across different studies has been reported. For instance, the changes of the relative abundance of genus Lachnospira, Rothia and Lachnoclostridium were opposite in different studies [19,20]. The MR study design provides a more reliable method to detect causal relationship between gut microbiota and DKD by minimizing the influence of confounding factors [14]. Among the identified protective and risk microbial taxa, those belonging to the phylum Firmicutes and phylum Bacteroidetes have the highest proportion respectively. The phylum Firmicutes and phylum Bacteroidetes represent > 90% of the total community of human gut microbiota [8]. Perturbed Bacteroidetes/Firmicutes eubiosis has been described in obese subjects, and the Bacteroidetes/Firmicutes ratio has been considered as a potential hallmark for obesity [21]. Previous studies have also indicated the alterations of the phyla Firmicutes and Bacteroidetes in DKD. For instance, a meta-analysis showed that patients with diabetic microvascular complications (including DKD) were depleted in Firmicutes [22]. In addition, compositional changes of phyla Firmicutes and Bacteroidetes were found in DKD compared with diabetes, and Bacteroidetes/Firmicutes ratio was significantly lower in the early DKD group than in the diabetes group [7]. We have identified multiple immune cell traits causally associated with DKD related phenotypes, and T cell phenotypes have the highest proportion. In traditional perceptions, nonimmune factors such as metabolism and hemodynamics are considered the most critical causes of renal injury in DKD [23,24]. However, current studies have also highlighted that DKD is an inflammatory disease [25]. The infiltration of immune cells is involved in hyperglycemia-induced renal damage [26]. As early as 2012, the aberrant intrarenal infiltration and recruitment of T cells in DKD have been discovered [27]. Our results have reemphasized the importance of T cell in the pathogenesis of DKD. Our mediation analysis has identified 5 mediation relationships from gut microbiota to DKD related phenotypes mediated by immune cells. We found species Pseudoflavonifractor capillosus was causally associated with increased risk of chronic kidney disease in T2D. In the mediation analysis, species Pseudoflavonifractor capillosus causally decreased the CD25+ CD4+ regulatory T cell and subsequently associated with increased risk of DKD, with a mediated proportion of 19.8%. Species Pseudoflavonifractor capillosus belongs to SCFA-producing bacteria [28]. Previous study has indicated that Cyclocarya paliurus polysaccharides could alleviate T2D symptoms by modulating gut microbiota including species Pseudoflavonifractor capillosus [29]. Another study indicated the genus Pseudoflavonifractor might contribute to the perturbation of metabolites and influence progression of obesity-induced T2D [30]. Our study found that species Pseudoflavonifractor capillosus might affect DKD through regulating Tregs to some extent. Tregs were a unique lineage of T cells with the ability to inhibit autoimmune responses. Unlike other Th cells, Treg might provide protection against DKD [31]. It has been indicated that the Urine Albumin/Creatinine Ratio was negatively correlated with the proportions of Tregs in DKD patients [32]. Besides, a study has indicated the potential role of Tregs in reversing obesity-linked insulin resistance and DKD [31]. This provides a potential mechanistic link: a reduction in Pseudoflavonifractor capillosus could lead to decreased production of SCFAs. SCFAs are known to promote the differentiation and expansion of Tregs. Therefore, the depletion of Pseudoflavonifractor capillosus might reduce SCFA levels, impairing Treg generation and function. The loss of Treg-mediated immunosuppression would then permit a pro-inflammatory state, contributing to the pathogenesis and progression of DKD. Genus Eubacterium was causally associated with increased risk of end-stage renal disease in diabetes and CD19 on B cell served as a mediator with a mediated proportion of 24.06%. Previous study has indicated that genus Eubacterium_nodatum_group was involved in metabolic disorders in the progression of DKD, which was in agreement with our finding [6]. In addition, a MR study identified that genus Eubacterium_rectale_group was associated with the risk of diabetic retinopathy [33] and this flora has been identified to induce systemic inflammation [34]. Our results indicated that genus Eubacterium might contribute to DKD through affecting CD19 on B cell. Although the exact role of B cell in DKD is not extensively revealed, multiple studies have suggested that the B cell was closely associated with worsening DKD [35,36]. The mechanisms of B cell involved in DKD mainly include antigen presentation, antibody production, immune complex formation, and cytokine generation [37]. The antibody produced by B cell leads to formation of immune complexes that can deposit in kidney and induce inflammation and glomerulonephritis. In conclusion, our mediation analysis indicated that the gut microbiota could contribute to DKD through influencing immune cell traits. However, the precise mechanisms by which the gut microbiome affects immune cells need further investigation. Our study indicated that interventions targeting gut microbiota might be a promising strategy for preventing DKD progression in clinical practice. Unlike the host genome, the genome of microorganisms can be modified. The administration of prebiotics, probiotics, and synbiotics can alter the composition of the intestinal microbiota. Dietary fiber, which can profitably reshape the intestinal microbial ecosystem, has been reported to protect against DKD through modulation of gut microbiota and increased SCFA production [38]. A meta-analysis indicated that probiotics might ameliorate oxidative stress and inflammatory status in DKD [39]. However, further research to understand the biological functions of microbiota taxa in DKD and how they interact with other factors such as host genetics and lifestyle is needed before the clinical application. Our mediation analysis showed that approximately 20% of the causal effect of gut microbiota on DKD was mediated by immune cell traits. This finding highlights that immune regulation serves as a meaningful pathway linking gut dysbiosis to DKD. Although additional mechanisms also contribute, the identified immune mediators represent important targets for adjunctive therapy. Therefore, a combined approach is warranted: microbiome-directed interventions can address the primary microbial drivers, while immune modulation can serve as a valuable adjunct to reinforce protective effects. Such integrated strategies may offer additive or synergistic benefits and should be explored in future translational research for DKD prevention and treatment. There are some limitations in our study. Firstly, whether our findings apply to other populations needs to be further investigated given that our analysis was primarily performed in a population of European ancestry. The unequal distribution of genetic variants among different ethnic groups may bias MR results [40]. Therefore, the generalization of our findings to other ethnic groups should be made with caution, considering the presence of ethnicity-specific associations between host genomes and gut microbiota. Secondly, MR results reflect lifelong genetic exposures and the effect of short-term microbiome change may be ignored. The exact role of significant microbiota identified by MR analysis should be further investigated by animal experiments and clinical trials. Thirdly, in this study, we performed causal inference using summary statistics from three different GWAS consortia within a two-sample MR framework. It should be noted that, despite the different origins of these cohorts, a certain degree of sample overlap may still exist. Such sample overlap could potentially bias causal estimates and increase the risk of false-positive findings. To mitigate this bias, we preferentially selected GWAS datasets with minimal participant overlap and applied sensitivity methods that are robust to sample overlap, such as MR-PRESSO, to validate our primary results. In addition, it is important to acknowledge that the heritability of gut microbial taxa is generally low to moderate, as the composition of the gut microbiota is influenced by numerous environmental factors in addition to host genetics. This inherent characteristic may limit the strength of genetic instruments and could, in principle, affect the statistical power and validity of MR analyses. To mitigate this concern, we adopted a previously validated P-value threshold (p < 1 × 10−5) for instrument selection, which has been widely used in microbiome MR studies to ensure an adequate number of instrumental variables while maintaining reasonable instrument strength. Furthermore, we calculated the F-statistic for each exposure and only included instruments with an F-statistic > 10, indicating that weak instrument bias is unlikely to substantially influence our causal estimates. The power calculations we performed further demonstrated that for the associations that passed multiple testing correction, the statistical power exceeded 0.80. Nevertheless, we acknowledge that the modest heritability of certain microbial taxa remains an inherent limitation of microbiome MR studies, and future large-scale genome-wide association studies of the gut microbiota are needed to improve the precision of genetic instruments. Finally, MR estimates assume a linear relationship between exposures and outcomes, and cannot discern non-linear correlations. Therefore, potential non-linear and interaction effects between gut microbiota and DKD should be considered. In conclusion, this MR study provides novel evidence for a causal role of the gut microbiota in DKD. We identified specific microbial taxa, such as phylum Firmicutes and species Pseudoflavonifractor capillosus, that are causally linked to DKD. Importantly, we further demonstrated that these effects are partially mediated by peripheral immune cell traits, including regulatory T cells and B cells. This not only establishes a causal relationship but also reveals a potential immune-mediated mechanism, highlighting the gut microbiota-immune-kidney axis as a critical pathway in DKD pathogenesis. These findings offer promising targets for therapeutic intervention and lay a foundation for future research into microbiome-based strategies for DKD prevention and treatment. Supplementary Material Supplemental Material Supplemental Material Supplemental Material