Different coffee consumption patterns affect HbA1c via propionic acid-producing gut microbiota Authors: Zhaoming Cao, Yu An, Yage Du, Guangyi Xu, Jingya Wang, Yanhui Lu Published: 2025 ABSTRACT AbstractThis study investigates the causal relationship between coffee consumption patterns, propionic acid-producing gut microbiota, and glycated haemoglobin (HbA1c) levels using Mendelian randomisation (MR). Genome-wide association study datasets from the UK Biobank and MiBioGen consortium were analysed, identifying a significant mediating pathway: unsweetened filter coffee consumption promotes Veillonella abundance, which in turn reduces HbA1c levels. The inverse variance weighted method served as the primary analytical approach, supported by sensitivity analyses to ensure robustness. These findings highlight the potential of unsweetened filter coffee as a dietary intervention to improve glycaemic control by modulating gut microbiota. This research provides novel insights into the coffee-intestinal microbiota-metabolism axis and offers practical guidance for diabetes management, emphasising the importance of coffee preparation methods and dietary habits in harnessing its therapeutic potential.Subject terms: Endocrine system and metabolic diseases, Nutrition, Quality of life Introduction Diabetes is one of the most prevalent and severe metabolic diseases globally,type 2 diabetes accounts for about 90% of all diabetics. According to the International Diabetes Federation, the number of diabetes patients worldwide exceeded 500 million by 2021, with projections reaching 643 million by 2030 and 783 million by 20451. This trend poses a considerable challenge to global healthcare systems, as diabetes not only precipitates multiple complications, such as cardiovascular disease, chronic kidney disease, and retinopathy but also profoundly impacts patients’ quality of life. Therefore, achieving effective blood glucose control in diabetic patients is of paramount importance. Glycated hemoglobin (HbA1c) has been widely recognized as a pivotal monitoring biomarker for diabetes management for a considerable period2–4. HbA1c is generated through non-enzymatic glycosylation reactions between hemoglobin in red blood cells and blood sugars (primarily glucose). This non-enzymatic reaction is continuous, gradual, and irreversible, resulting in HbA1c levels that reflect past, rather than current, blood glucose concentrations, regardless of factors such as fasting status, insulin administration, or hypoglycemic therapy. HbA1c concentration effectively reflects average blood glucose levels over the past 8–12 weeks5. HbA1c is comprised of HbA1a, HbA1b, and HbA1c isoforms, with HbA1c constituting approximately 70% of the total6. Owing to its stable structure properties, HbA1c is routinely utilized in clinical practice for monitoring diabetes management, aiding in the assessment of treatment efficacy and the prediction of complication risks. Therefore, maintaining HbA1c levels within reasonable ranges is crucial for the long-term health of diabetic patients. In research on blood glucose management, short-chain fatty acids (SCFAs), specifically propionate, have garnered growing attention7–10. Recent clinical randomized controlled trials and basic experimental studies have found that propionate exerts significant effects on regulating mitigating systemic inflammation, and managing blood glucose levels. A study published in Nature Metabolism in 201911 clinically validated that propionate intake can significantly decrease fasting blood glucose levels and positively impact insulin sensitivity. Additionally, basic research has indicated that beneficial gut microbiota such as Veillonella12, Parabacteroides13, Akkermansia14, and Bifidobacterium15,16 can elicit anti-inflammatory and hypoglycemic effects through propionate production. However, whether the propionate-producing probiotics has a direct impact on on HbA1c remains uncertain, leaving ample space for further research. Coffee stands as one of the most popular beverages globally, with its consumption increasing at an average annual rate of 1.05% between 2017 and 2022, ultimately reaching 10.0767 million tons in 202217. Meanwhile, numerous studies have confirmed coffee’s potential benefits in lowering blood glucose. For instance, a systematic review by Van Dam and Hu, published in JAMA18 indicated an inverse relationship between coffee consumption and the risk of type 2 diabetes, particularly among individuals with high coffee intake. Additionally, a systematic review and meta-analysis by Carlstrom and Larsson, featured in Nutrition Reviews19, echoed this notion, revealing a dose-response relationship between coffee consumption and a decreased risk of type 2 diabetes. However, research findings concerning the benefits of coffee in lowering blood glucose have been inconsistent. For example, a study by Keijzers et al., published in Diabetes Care20 showed that caffeine may impair insulin sensitivity, thereby potentially nullifying its potential benefits for blood glucose regulation. Furthermore, a randomized controlled trial by Alperet et al., appearing in the American Journal of Clinical Nutrition21, similarly failed to detect significant improvements in insulin sensitivity following coffee consumption. Moreover, studies have shown that certain coffee components, such as chlorogenic acid, may indirectly affect metabolic health by influencing gut microbiota composition and function22. On the basis of this information, we hypothesize that the variability in coffee’s impact on blood glucose levels could be attributed to the variety of coffee consumption patterns (such as decaffeinated coffee, filtered coffee, milk-added instant coffee, etc.) and their interactions with beneficial gut microbiota. Nevertheless, comprehensive studies exploring the relationships between various coffee consumption patterns and specific gut probiotics, as well as their impact on blood glucose, remain scarce. To overcome the constraints inherent in observational studies and randomized controlled trials, Mendelian Randomization (MR) has been introduced, employing genetic factors that are robustly associated with exposure as instrumental variables to assess causal relationships between the exposure and outcomes23. As an analytical approach that integrates genetic data, MR exhibits distinct advantages in elucidating causal relationships. Using genetic variants as instrumental variables, MR provides an effective method for exploring causal associations between exposure factors and outcomes. In contrast to traditional observational studies, which grapple with issues of reverse causation and confounding factors, MR studies can infer causality with relative robustness. MR research has become increasingly favored by researchers, especially in situations with constrained infrastructure or funding, owing to its straightforward methodology and compelling persuasiveness. This study aims to use MR to explore the associations between different coffee consumption patterns and HbA1c, assess the regulatory role of propionate-producing probiotics, and quantify the magnitude of their direct and indirect effects. This endeavor not only provides foundational data for future investigations but also provides guidance for promoting healthy lifestyle choices among the public. Results Screening of instrumental variables A total of 15 SNP loci significantly associated with Intake of sugar added to coffee, 20 SNP loci significantly linked to filtered coffee intake were identified after removing Linkage disequilibrium (LD), (R2 < 0.001, kb=10,000) SNP loci significantly associated with Intake of artificial sweetener added to coffee were identified, 14 SNP loci significantly associated with Intake of artificial sweetener added to coffee were identified, 12 SNP loci significantly associated with Intake of artificial sweetener added to coffee were identified, 18 SNP loci significantly associated with Added milk to instant coffee were identified and 18 SNP loci significantly associated with Added milk to instant coffee were identified. Added milk to instant coffee, identified 19 SNP loci significantly associated with Added milk to filtered coffee, identified 14 SNP loci significantly associated with decaffeinated coffee, 120 SNP loci significantly associated with Bifidobacterium significantly associated with Bifidobacterium, 17 SNP loci significantly associated with Veillonella, 11 SNP loci significantly associated with Parabacteroides and 27 SNP loci significantly associated with Akkermansia (P < 5×10-6). All SNP loci had F values > 10, indicating that all SNP loci were strong instrumental variables; A total of 39 SNP loci were excluded from the GWAS for HbA1c and its typing due to their non-retrieved. These excluded loci included 2 SNP that were associated with the intake of sugar added to coffee. 1 SNP locus that was associated with the intake of filtered coffee,, 2 SNP loci that were associated with the intake of artificial sweetener added to coffee, 3 SNP loci that were associated with the addition of milk to instant coffee, 3 SNP loci that were associated with the addition of milk to filtered coffee, 1 SNP locus that was associated with the intake of decaffeinated coffee, 3 SNP loci that were associated with Akkermansia, 15 SNP loci that were associated with Bifidobacterium, 4 SNP loci that were associated with Veillonella, and 5 SNP loci that were associated with Parabacteroides. In addition, 20 palindromic SNP sites were excluded during the subsequent MR analysis. Effect of propionic acid-producing probiotic gut microbial abundance on HbA1c levels In this study, we investigated the impact of four propionic acid-producing probiotics, Akkermansia, Bifidobacterium, Parabacteroides and Veillonella, on HbA1c levels. The causal relationships between these probiotics and HbA1c levels were evaluated using several MR methods, namely IVW, MR Egger regression, weighted median, and weighted mode methods. The IVW analysis revealed that Akkermansia, Bifidobacterium, Parabacteroides, and Veillonella all exhibited a protective effect on HbA1c levels (Table 1,Fig. 1). No inverse causality of HbA1c on four gut microbial. Table 1MR analysis of the abundance of propionic acid-producing probiotic bacteria on glycosylated hemoglobin (mmol/mol)Propionic acid-producing bacteriaMethods of analysisOR (95%CI)SEP 值AkkermansiaMR Egger1.00 (-0.16 ~ 0.16)0.081.00Weighted median0.96 (-0.09 ~ 0.02)0.030.17Inverse variance weighted0.95 (-0.09 ~ -0.01)0.020.02Simple mode0.96 (-0.15 ~ 0.07)0.050.49Weighted mode0.97 (-0.13 ~ 0.07)0.050.52BifidobacteriumMR Egger0.89 (-0.27 ~ 0.04)0.080.15Weighted median0.95 (-0.08 ~ -0.03)0.010.00Inverse variance weighted0.95 (-0.07 ~ -0.04)0.01<0.01Simple mode0.94 (-0.11 ~ -0.01)0.030.02Weighted mode0.94 (-0.11 ~ -0.01)0.030.02ParabacteroidesMR Egger0.92 (0.66 ~ 1.28)0.170.65Weighted median0.87 (0.76 ~ 0.99)0.070.03Inverse variance weighted0.87 (0.79 ~ 0.95)0.05<0.01Simple mode0.81 (0.68 ~ 0.96)0.090.06Weighted mode0.81 (0.68 ~ 0.97)0.090.08VeillonellaMR Egger1.00 (0.22 ~ 4.51)0.771.00Weighted median0.97 (0.91 ~ 1.03)0.030.35Inverse variance weighted0.95 (0.90 ~ 0.99)0.030.04Simple mode0.98 (0.88 ~ 1.09)0.050.66Weighted mode0.98 (0.87 ~ 1.09)0.060.68 MR analysis of the abundance of propionic acid-producing probiotic bacteria on glycosylated hemoglobin (mmol/mol) Forest plot for MR analysis results of impact of Akkermansia, Bifidobacterium, Parabacteroides and Veillonella on HbA1c. The results of the heterogeneity and pleiotropy tests demonstrated consistent effects among the selected instrumental variables, with no significant evidence of pleiotropy observed (Table 2). Table 2Results of the test for heterogeneity and multiplicity of the results of MR analysis of the effects of Akkermansia, Bifidobacterium, Parabacteroides and Veillonella on HbA1cExposureOutcomeCochran’s QMR EggerMR PRESSOQPinterceptPRSSobsPAkkermansiaHbA1c10.800.99-0.0030.5412.280.99BifidobacteriumHbA1c11.7610.0060.4416.231ParabacteroidesHbA1c3.320.65-0.0060.738.180.96VeillonellaHbA1c5.960.92-0.0040.948.180.96 Results of the test for heterogeneity and multiplicity of the results of MR analysis of the effects of Akkermansia, Bifidobacterium, Parabacteroides and Veillonella on HbA1c Effect of different coffee consumption preferences on HbA1c levels Among the six coffee consumption patterns examined, only filtered coffee demonstrated a statistically significant association with HbA1c levels. The inverse variance weighted (IVW) analysis revealed that filtered coffee consumption was significantly associated with decreased HbA1c levels (OR = 0.97, 95% CI: 0.94–0.99, P = 0.04). For other consumption patterns, including sugar-added coffee (OR = 1.01, 95% CI: 0.96–1.07, P = 0.64), artificial sweetener-added coffee (OR = 1.03, 95% CI: 0.96–1.11, P = 0.41), milk-added instant coffee (OR = 1.00, 95% CI: 0.95–1.05, P = 0.97), milk-added filtered coffee (OR = 0.98, 95% CI: 0.93–1.04, P = 0.47), and decaffeinated coffee (OR = 0.99, 95% CI: 0.92–1.07, P = 0.79), no statistically significant associations with HbA1c levels were observed (Table 3). No inverse causality of HbA1c levels on coffee consumption preferences. Table 3MR analysis of different coffee consumption preferences on glycosylated hemoglobin levels (mmol/mol)Different coffee consumption preferencesMethods of analysisOR (95%CI)SEPIntake of sugar added to coffeeMR Egger1.10 (0.95 ~ 1.27)0.070.22Weighted median1.01 (0.94 ~ 1.09)0.040.78Inverse variance weighted1.01 (0.96 ~ 1.07)0.030.64Simple mode1.01 (0.90 ~ 1.12)0.060.91Weighted mode1.01 (0.90 ~ 1.13)0.060.92Filtered coffee intakeMR Egger1.08(0.98 ~ 1.19)0.050.13Weighted median0.99(0.95 ~ 1.03)0.020.56Inverse variance weighted0.97(0.94 ~ 0.99)0.020.04*Simple mode1.01(0.94 ~ 1.09)0.040.76Weighted mode1.01(0.94 ~ 1.08)0.040.79Intake of artificial sweetener added to coffeeMR Egger1.12 (0.97 ~ 1.29)0.070.16Weighted median1.04 (0.95 ~ 1.14)0.050.43Inverse variance weighted1.03 (0.96 ~ 1.11)0.040.41Simple mode1.07 (0.93 ~ 1.23)0.070.39Weighted mode1.05 (0.93 ~ 1.19)0.060.41Added milk to instant coffeeMR Egger0.90 (0.79 ~ 1.03)0.070.14Weighted median1.02 (0.96 ~ 1.08)0.030.61Inverse variance weighted1.00 (0.95 ~ 1.05)0.030.97Simple mode1.04 (0.93 ~ 1.17)0.060.46Weighted mode1.05 (0.93 ~ 1.17)0.060.46Added milk to filtered coffeeMR Egger0.98(0.88 ~ 1.10)0.060.77Weighted median0.95(0.88 ~ 1.02)0.040.18Inverse variance weighted0.98(0.93 ~ 1.04)0.030.47Simple mode0.90(0.77 ~ 1.04)0.080.18Weighted mode0.90(0.77 ~ 1.04)0.080.18Decaffeinated coffeeMR Egger0.97 (0.83 ~ 1.13)0.080.69Weighted median0.98 (0.89 ~ 1.07)0.050.60Inverse variance weighted0.99 (0.92 ~ 1.07)0.040.79Simple mode0.92 (0.77 ~ 1.09)0.090.33Weighted mode0.93 (0.77 ~ 1.12)0.090.46 MR analysis of different coffee consumption preferences on glycosylated hemoglobin levels (mmol/mol) To assess the robustness of these findings, multiple sensitivity analyses were performed. Cochran’s Q test revealed potential heterogeneity in filtered coffee consumption (Q = 18.39, P = 0.24). The MR Egger intercept test showed no significant horizontal pleiotropy across all exposure variables (all P > 0.05), supporting the validity of our instrumental variables. The MR PRESSO global test identified potential pleiotropy in milk-added filtered coffee (RSSobs=1695.65, P = 0.02), suggesting that these results should be interpreted with caution. These comprehensive sensitivity analyses strengthen the reliability of our findings while highlighting the complexity of the relationship between coffee consumption patterns and glycemic control (Table 4). Table 4Heterogeneity and test of multiple validity of the results of MR analysis of the effect of different coffee consumption preferences on HbA1cExposureOutcomeCochran’s QMR EggerMR PRESSOQPInterceptPRSSobsPIntake of sugar added to coffeeHbA1c7.470.82<-0.010.258.600.89Filtered coffee intakeHbA1c18.390.240.080.13105.960.37Intake of artificial sweetener added to coffeeHbA1c14.970.18<-0.010.2418.590.26Added milk to instant coffeeHbA1c20.410.085<0.010.1130.610.05Added milk to filtered coffeeHbA1c12.960.61<0.010.951695.650.02Decaffeinated coffeeHbA1c15.650.21<0.010.7519.590.20 Heterogeneity and test of multiple validity of the results of MR analysis of the effect of different coffee consumption preferences on HbA1c Effect of different coffee consumption preferences on the abundance of propionic acid-producing probiotic gut microorganisms According to the results of MR analysis, there were some differences in the effects of different coffee consumption preferences on four probiotic gut microorganisms capable of producing propionic acid, namely Akkermansia, Bifidobacterium, Parabacteroides, and Veillonella. Coffee with added sugar showed a significant positive correlation with Bifidobacterium (IVW method: OR = 1.56, 95%CI: 1.05-2.33, P = 0.03) and a borderline significant negative correlation with Veillonella (IVW method: OR = 0.65, 95%CI: 0.423–0.99, P = 0.05). Filtered coffee showed a significant positive correlation with Veillonella (IVW method: OR = 1.29, 95%CI: 1.03–1.63, P = 0.03). Instant coffee with added milk showed a significant negative association with Bifidobacterium (IVW method: OR = 0.71, 95%CI: 0.523–0.97, P = 0.03). For other coffee consumption patterns, including coffee with added artificial sweeteners, filter coffee with added milk, and decaffeinated coffee, no statistically significant associations were observed with the four propionic acid-producing bacterial genera that produce propionic acid: Akkermansia, Bifidobacterium, Parabacteroides, and Veillonella (Table 5). Effect of different coffee consumption preferences on gut microbiota abundance of propionic acid-producing probiotic bacteria No reverse causality for different coffee consumption preferences. Table 5MR analysis of four intestinal probiotics capable of producing propionic acid, Akkermansia, Bifidobacterium, Parabacteroides, and Veillonella, by different coffee consumption preferencesDifferent coffee consumption preferencesPropionic acid-producing bacteriaMethods of analysisOR (95%CI)SEP 值Intake of sugar added to coffeeAkkermansiaMR Egger3.11(0.26 ~ 37.38)1.270.39Weighted median0.69(0.41 ~ 1.14)0.260.15Inverse variance weighted0.84(0.58 ~ 1.23)0.190.38Simple mode0.64(0.28 ~ 1.48)0.430.32Weighted mode0.64(0.29 ~ 1.45)0.410.31BifidobacteriumMR Egger7.85(0.74 ~ 83.17)1.200.14Weighted median1.45(0.86 ~ 2.44)0.270.17Inverse variance weighted1.56(1.05 ~ 2.33)0.200.03Simple mode1.11(0.47 ~ 2.65)0.440.82Weighted mode1.11(0.47 ~ 2.61)0.430.81ParabacteroidesMR Egger2.65(0.34 ~ 20.66)1.050.38Weighted median1.14(0.72 ~ 1.78)0.230.58Inverse variance weighted1.11(0.81 ~ 1.52)0.160.52Simple mode1.19(0.59 ~ 2.40)0.360.63Weighted mode1.18(0.58 ~ 2.38)0.360.66VeillonellaMR Egger0.45(0.03 ~ 7.30)1.420.59Weighted median0.65(0.37 ~ 1.13)0.290.13Inverse variance weighted0.65(0.42 ~ 0.99)0.220.05Simple mode0.57(0.22 ~ 1.51)0.500.29Weighted mode0.57(0.24 ~ 1.32)0.430.22Filtered coffee intakeAkkermansiaMR Egger0.30(0.09 ~ 1.01)0.610.07Weighted median0.86(0.66 ~ 1.11)0.130.24Inverse variance weighted0.94(0.77 ~ 1.14)0.100.52Simple mode0.80(0.51 ~ 1.24)0.230.33Weighted mode0.78(0.52 ~ 1.19)0.210.27BifidobacteriumMR Egger0.47(0.14 ~ 1.63)0.630.26Weighted median1.00(0.78 ~ 1.29)0.130.98Inverse variance weighted0.93(0.76 ~ 1.13)0.100.47Simple mode1.19(0.74 ~ 1.93)0.250.48Weighted mode1.11(0.73 ~ 1.68)0.210.64ParabacteroidesMR Egger0.69(0.20 ~ 2.38)0.630.57Weighted median1.02(0.81 ~ 1.28)0.120.87Inverse variance weighted0.94(0.77 ~ 1.14)0.100.52Simple mode1.12(0.73 ~ 1.71)0.220.61Weighted mode1.10(0.75 ~ 1.61)0.200.64VeillonellaMR Egger0.79(0.20 ~ 3.17)0.710.74Weighted median1.22(0.88 ~ 1.68)0.170.24Inverse variance weighted1.29(1.03 ~ 1.63)0.120.03Simple mode1.18(0.66 ~ 2.10)0.300.59Weighted mode1.12(0.65 ~ 1.94)0.280.69Intake of artificial sweetener added to coffeeAkkermansiaMR Egger0.53(0.06 ~ 4.79)1.120.61Weighted median1.33(0.63 ~ 2.80)0.380.45Inverse variance weighted1.41(0.78 ~ 2.56)0.300.25Simple mode1.33(0.52 ~ 3.40)0.480.58Weighted mode1.33(0.59 ~ 3.03)0.420.53BifidobacteriumMR Egger1.13(0.13 ~ 10.00)1.110.92Weighted median0.90(0.45 ~ 1.79)0.350.76Inverse variance weighted1.03(0.61 ~ 1.77)0.270.90Simple mode0.84(0.32 ~ 2.15)0.480.73Weighted mode0.87(0.43 ~ 1.79)0.370.73ParabacteroidesMR Egger2.25(0.36 ~ 14.00)0.930.45Weighted median1.07(0.58 ~ 1.96)0.310.84Inverse variance weighted0.92(0.56 ~ 1.50)0.250.73Simple mode1.13(0.51 ~ 2.52)0.410.78Weighted mode1.11(0.55 ~ 2.24)0.360.79VeillonellaMR Egger1.60(0.03 ~ 99.27)2.110.84Weighted median1.50(0.62 ~ 3.64)0.450.37Inverse variance weighted0.94(0.35 ~ 2.49)0.500.90Simple mode1.73(0.45 ~ 6.66)0.690.47Weighted mode1.63(0.62 ~ 4.28)0.490.38Added milk to instant coffeeAkkermansiaMR Egger1.46(0.31 ~ 6.79)0.790.64Weighted median0.92(0.63 ~ 1.35)0.190.67Inverse variance weighted1.00(0.74 ~ 1.36)0.160.99Simple mode0.92(0.51 ~ 1.64)0.300.77Weighted mode0.92(0.52 ~ 1.63)0.290.78BifidobacteriumMR Egger0.56(0.17 ~ 1.85)0.610.37Weighted median0.76(0.51 ~ 1.13)0.210.18Inverse variance weighted0.71(0.52 ~ 0.97)0.160.03Simple mode0.80(0.43 ~ 1.47)0.310.49Weighted mode0.77(0.44 ~ 1.35)0.290.38ParabacteroidesMR Egger0.78(0.29 ~ 2.08)0.500.63Weighted median0.95(0.69 ~ 1.31)0.160.75Inverse variance weighted0.99(0.78 ~ 1.27)0.130.95Simple mode0.91(0.54 ~ 1.54)0.270.73Weighted mode0.90(0.56 ~ 1.45)0.240.67VeillonellaMR Egger1.09(0.19 ~ 6.37)0.900.92Weighted median0.98(0.61 ~ 1.56)0.240.92Inverse variance weighted1.05(0.74 ~ 1.49)0.180.78Simple mode0.93(0.43 ~ 2.01)0.390.86Weighted mode0.96(0.47 ~ 1.97)0.370.91Added milk to filtered coffeeAkkermansiaMR Egger2.56(0.36 ~ 18.33)1.000.39Weighted median0.92(0.42 ~ 2.04)0.400.84Inverse variance weighted1.36(0.78 ~ 2.37)0.290.28Simple mode0.73(0.20 ~ 2.58)0.650.64Weighted mode0.74(0.20 ~ 2.69)0.660.66BifidobacteriumMR Egger0.38(0.08 ~ 1.75)0.780.26Weighted median1.19(0.62 ~ 2.29)0.330.60Inverse variance weighted1.34(0.83 ~ 2.18)0.250.23Simple mode1.19(0.41 ~ 3.40)0.540.76Weighted mode1.19(0.39 ~ 3.57)0.560.77ParabacteroidesMR Egger2.74(0.65 ~ 11.54)0.730.22Weighted median0.97(0.52 ~ 1.82)0.320.93Inverse variance weighted0.83(0.50 ~ 1.37)0.250.47Simple mode1.19(0.44 ~ 3.24)0.510.75Weighted mode1.19(0.45 ~ 3.11)0.490.74VeillonellaMR Egger2.24(0.26 ~ 19.58)1.110.50Weighted median1.07(0.48 ~ 2.37)0.410.87Inverse variance weighted1.03(0.55 ~ 1.94)0.320.93Simple mode1.08(0.31 ~ 3.80)0.640.91Weighted mode0.99(0.31 ~ 3.17)0.590.98Decaffeinated coffeeAkkermansiaMR Egger1.33(0.05 ~ 34.95)1.670.87Weighted median1.30(0.60 ~ 2.78)0.390.50Inverse variance weighted1.11 (0.59 ~ 2.08)0.320.76Simple mode1.72(0.56 ~ 5.28)0.570.39Weighted mode1.73(0.53 ~ 5.63)0.600.41BifidobacteriumMR Egger0.21(0.01 ~ 4.35)1.540.37Weighted median1.00(0.47 ~ 2.12)0.390.99Inverse variance weighted1.07(0.60 ~ 1.89)0.290.82Simple mode0.97(0.35 ~ 2.68)0.520.96Weighted mode1.03(0.37 ~ 2.89)0.530.96ParabacteroidesMR Egger1.74(0.07 ~ 46.41)1.680.76Weighted median0.66(0.33 ~ 1.31)0.350.23Inverse variance weighted0.82(0.47 ~ 1.44)0.290.48Simple mode0.62(0.24 ~ 1.64)0.500.38Weighted mode0.61(0.23 ~ 1.59)0.490.36VeillonellaMR Egger0.15(0.00 ~ 8.88)2.100.41Weighted median0.62(0.24 ~ 1.60)0.480.32Inverse variance weighted0.64(0.31 ~ 1.32)0.370.22Simple mode0.72(0.16 ~ 3.17)0.760.68Weighted mode0.77(0.22 ~ 2.65)0.630.70 MR analysis of four intestinal probiotics capable of producing propionic acid, Akkermansia, Bifidobacterium, Parabacteroides, and Veillonella, by different coffee consumption preferences The results of the heterogeneity and pleiotropy tests (Table 6) showed that the Cochran’s Q test did not show significant heterogeneity in any of the analyses (all P > 0.05). The MR-Egger intercept test failed to identify any significant horizontal pleiotropy (all P > 0.05), and the MR-PRESSO test did not detect any significant outliers (all P > 0.05), indicating the robustness of our analysis. Table 6Results of heterogeneity and test of multiple validity of MR analysis of the effect of different coffee consumption preferences on propionic acid-producing probiotic gut microorganismsDifferent coffee consumption preferencesGut microbiotaCochran’s QMR EggerMR PRESSOQPInterceptPRSSobsPIntake of sugar added to coffeeAkkermansia7.020.72-0.030.323.830.76Bifidobacterium4.190.760.030.2218.220.18Parabacteroides5.060.89-0.020.426.930.88Veillonella3.550.970.010.804.470.98Filtered coffee intakeAkkermansia8.770.850.040.099.900.88Bifidobacterium18.530.180.030.3021.600.24Parabacteroides20.300.120.010.6322.850.16Veillonella12.540.480.020.4919.080.35Intake of artificial sweetener added to coffeeAkkermansia1.910.750.020.439.900.90Bifidobacterium3.660.450.000.935.210.63Parabacteroides2.100.72-0.020.393.150.83Veillonella8.380.08-0.010.8119.070.32Added milk to instant coffeeAkkermansia3.170.96-0.010.644.100.99Bifidobacterium4.620.800.010.7112.060.67Parabacteroides3.820.980.010.636.820.94Veillonella5.550.780.000.9710.610.61Added milk to filtered coffeeAkkermansia5.710.46-0.020.549.900.89Bifidobacterium4.620.800.020.717.980.65Parabacteroides8.630.28-0.030.1411.520.37Veillonella3.910.69-0.020.505.090.80Decaffeinated coffeeAkkermansia2.610.760.000.923.840.78Bifidobacterium3.520.620.030.355.050.64Parabacteroides5.670.34-0.010.678.500.34Veillonella5.000.420.030.516.810.49 Results of heterogeneity and test of multiple validity of MR analysis of the effect of different coffee consumption preferences on propionic acid-producing probiotic gut microorganisms Mediating role of propionic acid-producing probiotic gut microbes between different coffee consumption preferences and glycosylated hemoglobin levels Based on the mediation analysis, we identified a significant mediating role of Veillonella in the relationship between filtered coffee consumption and HbA1c levels. The analysis revealed that filtered coffee consumption was positively associated with the abundance of Veillonella (βa = 0.26, 95% CI: 0.03~0.49, P = 0.03), while Veillonella demonstrated a significant negative association with HbA1c levels (βb = -0.05, 95% CI: -0.10~0.00, P < 0.01). The indirect effect, calculated as the product of paths a and b, was statistically significant (βc’ = -0.013*, 95% CI: -0.02~-0.01, P < 0.01). Importantly, the direct effect of filtered coffee on HbA1c levels remained significant (βc = -0.03, OR = 0.97, 95% CI: 0.943–0.99, P = 0.04), indicating a partial mediation effect where the indirect effect through Veillonella accounts for 43.33% of the total effect (Fig. 2). Fully mediated effect of filtered coffee intake affecting HbA1c levels by influencing Veillonella. *:Three decimals are retained for more precise calculation of the mediating effect share. Discussion The present study identified an important mediating pathway through a two-step, two-sample Mendelian randomization analysis: filter coffee reduces HbA1c levels by promoting Veillonella abundance. Specifically, filter coffee intake was significantly and positively associated with Veillonella abundance (βa = 0.26, 95% CI: 0.92-0.98, P < 0.01), whereas Veillonella was significantly and negatively associated with HbA1c levels (βb = -0.05, 95% CI: 0.923–0.98, P < 0.01). The indirect effect calculated by the product of paths a and b was statistically significant (βc’=-0.013, 95% CI: -0.023–0.01, P < 0.01). Notably, the direct effect of filtered coffee on HbA1c levels remained significant (βc = -0.03, OR = 0.97, 95% CI: 0.943–0.99, P = 0.04), suggesting that this was a partially mediated effect, with the indirect effect via Veillonella accounting for 43.33% of the total effect. This finding provides a new perspective for understanding the regulatory effects of coffee on glucose metabolism. Firstly, we found that all four propionic acid-producing probiotics (Veillonella, Parabacteroides, Akkermansia and Bifidobacterium) included in the present study had a direct modulatory effect on HbA1c levels and all were negatively correlated. This is in line with previous studies, where a 12-month prospective cohort study conducted by Depommier et al.24 showed that diabetic patients with higher abundance of Akkermansia muciniphila at baseline had a more significant reduction in HbA1c levels (HR = 0.78, 95%CI: 0.65-0.94). A study by Plovier et al.25 in published in Nature Medicine further demonstrated that purified Akkermansia muciniphila membrane proteins significantly improved metabolic status in obese and diabetic mice. In addition recent studies have shown that Akkermansia improves insulin resistance by enhancing intestinal barrier function, reducing endotoxin translocation and systemic inflammatory response14. For Bifidobacterium, several clinical trials have confirmed its hypoglycemic effect. a randomized controlled trial by Kobyliak et al.26 showed that oral Bifidobacterium supplementation for 3 months significantly reduced HbA1c levels in diabetic patients by an average of 0.54% (p < 0.05). In addition Bifidobacterium16 can produce short-chain fatty acids through fermentation of dietary fiber, increase GLP-1 secretion and inhibit the mechanism of action of pro-inflammatory cytokine production. Pan et al.13 found that Parabacteroides distasonis can significantly enhance insulin sensitivity by increasing the production of succinic acid and secondary bile acids Zhang et al.27 showed that Parabacteroides could also improve metabolic status by regulating glucocorticoid-induced disorders of glucose and lipid metabolism.The hypoglycemic mechanism of Veillonella has been more thoroughly studied than the other three gut microorganisms, and in 2014, it was found that Veillonella could promote GLP-1 secretion through activating G protein-coupled receptors GPR41 and GPR43, improving the mechanism of pancreatic β-cell function28,29. One of the more notable findings of this study was that filter coffee was the only coffee consumption pattern that significantly reduced HbA1c levels (OR = 0.97, 95% CI: 0.943–0.99, P = 0.04). This finding is consistent with the results of a systematic review by Carlstrom and Larsson published in Nutrition Reviews19, who found that filtered coffee was associated with a dose-response relationship with a reduced risk of type 2 diabetes. A recent study published in Nature Microbiology30 further demonstrated significant alterations in the composition of the intestinal flora in those who regularly consumed filter coffee (≥3 cups/day), with a significant increase in the abundance of propionic acid-producing bacteria. This unique effect may be closely related to the preparation method and compositional characteristics of filter coffee. Filtered coffee retains higher levels of chlorogenic acid and other polyphenols compared to other coffee preparations, while the filtration process removes coffee lipids (diterpenes found in coffee beans), and these compositional differences may be key to the different effects. A study by Hong et al.31 found that filtered coffee contained significantly higher levels of chlorogenic acid, which has been shown to increase the intestinal flora via the AMPK signaling pathway to increase skeletal muscle glucose uptake. It has also been found that polyphenols in filter coffee can regulate the intestinal flora through multiple pathways: first, they can enhance the antioxidant capacity of intestinal epithelial cells by activating the Nrf2 signaling pathway, which provides a favorable environment for the colonization of beneficial bacteria32; second, these compounds can selectively promote the growth and metabolic activity of propionic acid-producing bacteria27; and third, they can optimize the growth and metabolic activity of propionic acid-producing bacteria by adjusting intestinal pH and the redox environment. redox environment, optimizing the survival conditions of propionic acid-producing bacteria. As for the finding that other coffee consumption patterns did not have a significant effect on HbA1c levels, this is broadly in line with the results of previous studies. van Dam et al.22, in a NEJM review, suggested that the addition of ingredients such as sugar and creamer may counteract the potential benefits of coffee by altering the intestinal microenvironment. Recent studies have shed further light on this complex interaction. A large number of existing studies point out that the addition of dairy products significantly alters the bioavailability of coffee polyphenols, and that interactions between phenolic compounds in coffee and milk proteins may be the main reason for the negative impact of milk on coffee functionality33, but this is less discussed in the literature at present, and in particular the role of the milk fat globules and their membranes has not been fully explored. The present study did not find a significant effect of sugar-containing coffee on hemoglobin, but this does not mean that sugar-containing coffee cannot influence glycolipid metabolic processes; a prospective cohort study conducted by researchers at Harvard University on which substance in coffee is the key factor in the promotion of weight loss demonstrated that unsweetened coffee, with or without caffeine, did not affect its weight loss effect, whereas the addition of sugar reduced this weight loss effect34. The coffee-Veillonella-HbA1c regulatory axis identified in this study has important clinical translational value. Recent chemical analyses have shown that filtered coffee retains significantly higher levels of bioactive substances, including chlorogenic acid at 3.23–4.5 mg/mL and caffeoylquinic acid at 1.83–2.6 mg/mL31. These polyphenolic compounds regulate intestinal microecology through multiple mechanisms. Toydemir et al.35 confirmed that coffee polyphenols can enhance the antioxidant capacity of intestinal epithelial cells through the nuclear factor E2-related factor 2 (Nrf2) signaling pathway, upregulate MUC2 gene expression, promote mucus secretion, and create a suitable microecological environment for Veillonella colonization through a human intestinal epithelial cell model. The lactate metabolism pathway, which is unique to Veillonella, converts lactate to propionic acid36, which improves β-cell function through activation of G protein-coupled receptors GPR41 and GPR43, increasing intracellular calcium ion concentration, and promoting the cytosolic action of insulin-secreting granules32. Based on the results of this study, we offer several important recommendations for clinical practice. First, moderate consumption of unsweetened filter coffee is recommended because studies have shown that the bioactive substances retained in it, especially chlorogenic acid and other polyphenolic compounds, can effectively promote the growth of beneficial intestinal flora. This type of consumption not only preserves the beneficial components of coffee, but also maximizes its modulatory effect on the intestinal flora. Second, the findings emphasize the importance of avoiding adding sugar or high-fat milk to coffee. These additives not only increase calorie intake, but more importantly may inhibit the metabolic activity of propionic acid-producing bacteria, thereby diminishing the health benefits of coffee. For those who are accustomed to adding sugar or milk to their coffee, it is recommended to gradually reduce the amount added and eventually transition to drinking plain coffee. At the same time, joint supplementation with prebiotics or dietary fiber can be considered, as these nutrients can further promote the growth of propionic acid-producing bacteria and enhance the metabolic regulatory effects of coffee. Finally, for patients with low diversity of intestinal flora or metabolic disorders, personalized dietary interventions under the guidance of a physician are recommended. This intervention should take into account the patient’s specific situation, including the degree of coffee tolerance, existing dietary habits, and metabolic status, to develop an individually appropriate coffee consumption program. At the same time, regular monitoring of blood glucose levels and intestinal flora is recommended for timely adjustment of the intervention strategy to achieve the best therapeutic effect. The present study elucidates the potential molecular mechanisms by which coffee consumption affects glucose metabolism, but several methodological limitations remain. The first limitation stems from the nature of the source of the study data: analyses based on the existing GWAS database mainly reflect the genetic characteristics of populations of European descent, and this limitation of population representativeness not only affects the extrapolation of the results, but may also mask race-specific metabolic modulation patterns resulting from the interaction of genetic polymorphisms, epigenetic modifications, and environmental factors. Second, this study focused only on the mediating effects of four specific propionic acid-producing genera, failing to comprehensively assess the dynamic balance of the intestinal microecosystem and its synergistic network of metabolomes. In addition, the study failed to establish an accurate dose-effect relationship model, and the modulatory effects of different coffee preparation processes and intake doses on the composition of intestinal flora and their metabolic activities still need to be further quantified. The inherent limitations of the cross-sectional study design also constrained the in-depth analysis of the temporal evolution of microbial communities and their metabolic interactions with the host. Based on the above limitations, future research should focus on the following aspects: firstly, there is an urgent need to carry out multi-center and large-scale population cohort studies to explore the regulatory effects of coffee intake on intestinal flora and glucose metabolism in different racial backgrounds through stratified analysis. Second, it is recommended to integrate multi-omics technology platforms, including but not limited to macro-genomics, transcriptomics, metabolomics and proteomics, to build a systematic molecular network to comprehensively analyze the regulatory mechanisms of coffee bioactive components on the intestinal microecosystem. Particular attention needs to be paid to changes in short-chain fatty acid profiles, remodeling of bile acid metabolism, and signal transduction pathways in the microbial-host metabolic axis. In addition, a standardized dose-effect assessment system should be established to systematically study the effects of different coffee preparation processes and intake patterns on the ecological niche and metabolic phenotype of intestinal flora, so as to provide a scientific basis for the development of precise nutritional intervention strategies. In conclusion, this study provides a new theoretical framework for elucidating the molecular mechanism of coffee regulation of glucose metabolism, but more in-depth mechanistic studies are still needed to improve the existing knowledge. Future studies should adopt a multidimensional and systematic research strategy to deeply analyze the regulatory network of the coffee-intestinal flora-host metabolism axis, so as to provide a more comprehensive theoretical basis and intervention targets for the prevention and treatment of metabolic diseases. Methods Study design This study was completed in November, 2024 In MR analysis, instrumental variables must adhere to the three fundamental assumptions: association, independence, and exclusivity. Association implies a strong link between the instrumental variable and the exposure factor. Independence signifies that the instrumental variable is unrelated to other confounding factors influencing the relationship between the exposure and the outcome. Exclusivity denotes that the instrumental variable can solely impact the outcome via the exposure factor, without any direct or indirect association with the outcome (Fig. 3). Diagram of the MR assumption. A Core Assumption 1 Selected genetic variants (instrumental variables) must be statistically significantly correlated with exposure. B Independence: the genetic variant (instrumental variable) is not associated with any possible confounders. C Exclusivity: the genetic variant (instrumental variable) can only contribute to the outcome through exposure factors. 1). In the present study, TwoSampleMR analyses were conducted using four beneficial propionic acid-producing bacterial genera(Bifidobacterium, Veillonella, Parabacteroides, and Akkermansia), which have been demonstrated to be able to produce propionic acid independently, as mediators in order to investigate the mediating role of propionic acid-producing bacteria in the association between different coffee consumption preferences and hemoglobin levels. Data sources Genome-Wide Association Studies (GWAS) data on different preferences for coffee consumption and HbA1c levels were obtained from the UK Biobank (UKB; official website: https://www.ukbiobank.ac.uk/), which is currently the world’s most globally renowned and accessible biobank, and has collected 500,000 blood and saliva samples from participants across the UK since its establishment in 2006, as well as well-established demographic, socioeconomic and health information. the basic information of which is shown in Table 7. Table 7General information of GWAS for traits included in the studyTypeVariableSample sizeConsortiumData download addressExposuresDecaffeinated coffee64,949UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ukb-b-12558/ (accessed on 5 August 2024)Filtered coffee intake64,949UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ukb-b-748/(accessed on 5 August 2024)Added milk to filtered coffee64,949UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ukb-b-5332/(accessed on 5 August 2024)Added milk to instant coffee64,949UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ukb-b-4051/(accessed on 5 August 2024)Intake of artificial sweetener added to coffee64,949UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ukb-b-1338/(accessed on 5 August 2024)Intake of sugar added to coffee64,949UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ukb-b-243/(accessed on 5 August 2024)Mediatorsgenus.Bifidobacterium18,340MiBioGenhttps://mibiogen.gcc.rug.nl/(accessed on 5 August 2024)genus.Veillonella18,340MiBioGenhttps://mibiogen.gcc.rug.nl/(accessed on 5 August 2024)genus.Parabacteroides18,340MiBioGenhttps://mibiogen.gcc.rug.nl/(accessed on 5 August 2024)genus.Akkermansia18,340MiBioGenhttps://mibiogen.gcc.rug.nl/(accessed on 5 August 2024)OutcomeGlycated haemoglobin HbA1c levels389,889UK Biobankhttps://gwas.mrcieu.ac.uk/datasets/ebi-a-GCST90014006/(accessed on 5 August 2024) General information of GWAS for traits included in the study The GWAS for gut microbes was derived from the largest 16S rRNA gene sequencing profile and genotyping data for gut microbiota to date published in 2021 by the MiBioGen consortium (https://mibiogen.gcc.rug.nl/), which comprises 24 population-based cohorts with a total of 18,340 participants, and the team identified 122,110 variant loci in 211 taxa (from genus to phylum level) to analyze variation in transgenic taxa in different populations. Four microbial taxa, Bifidobacterium, Veillonella, Parabacteroides and Akkermansia, were selected for analysis in this study (Table 7 provides access to these data). Data Organization Single nucleotide polymorphism (SNP) loci associated with HbA1c and six coffee consumption preferences, retrieved from the IEU Open GWAS project (mrcieu.ac.uk), were utilized for the targeted GWAS. To maintain an adequate number of instrumental variables (IVs) in mendelian randomization analysis, we selected P < 5 × 10-6 as our significance threshold. Independent SNPs (R2 < 0.001, distance >10000 kb) were identified and employed as instrumental variables. The European Thousand Genomes Project’s gene-wide information was referenced to assess linkage disequilibrium (LD) among these SNP loci. To ensure that the included SNP loci were associated with outcomes only through exposure factors, each SNP locus was screened using the PhenoScanner database, and the association hypothesis was further verified by calculating the F-value to assess whether there was a weak instrumental variable bias in the selected instrumental variables. The F-value calculation formula was as follows: where N is the exposure sample size, R2 is the proportion of variance in the exposure factor explained by the instrumental variables, β is the effect value of the allele, and EAF is the effect allele frequency; the F-value of a single SNP point was calculated as follows: * SE is a standard error. All effect estimates for HbA1c are reported in mmol/mol units with 95% confidence intervals. Effect estimates for gut microbiota abundances are reported as odds ratios representing standard deviation changes per unit increase in exposure. Statistical methods The inverse variance weighted (IVW) method was used as the main analytical method, while supplementary methods included MR Egger regression37, the weighted median, weighted mode38, and robust adjusted curve score (MR-RAPS)39, all aimed at evaluating the causal link between exposure factors and outcomes. The IVW method is characterized by its regressions models, which omit the intercept term and utilize the inverse of the outcome variance as weights. This method yields the most precise results when all included SNP loci are all valid instrumental variables40. Sensitivity analyses were performed by heterogeneity tests and multiple validity assessments. For SNP loci meeting the three hypotheses, the Cochran’s Q test was employed to assess heterogeneity among individual genetic variants. The MR Egger intercept method was utilized to detect violations of MR hypotheses due to horizontal pleiotropy. Sensitivity was tested using the leave-one-out approach to identify potential outliers among the final SNP loci. The stability of funnel plots was examined for asymmetries, and MR PRESSO was applied to detect outliers and assess their impact on the results. The mediated share (or mediation proportion) represents the percentage of the total effect that is explained through the mediator pathway. The mediated share is calculated as:Mediated Share = (Indirect Effect ÷ Total Effect) × 100% The above statistical analyses were performed based on the R 4.3.3 software packages “Mendelian Randomization”, “TwoSampleMR” and “MRPRESSO”. “ software packages were realized. The difference was considered statistically significant at two-sided P < 0.05. This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization (STROBE-MR) guidelines. STROBE-MR checklist in supplementary material. Supplementary information Supplementary Data 1 Acknowledgements This research was funded by the National Key Research and Development Program of China (grant number 2022YFA1103602 awarded to Yanhui Lu), the National Natural Science Foundation of China (grant number 82473618 awarded to Yanhui Lu), the National Natural Science Foundation of China (grant number 92357305 awarded to Yanhui Lu). Author contributions Conceptualisation, Z.C. and Y.A.; methodology, Z.C.; software, Z.C. and Y.A.; validation, G.X. and J.W.; formal analysis, Z.C. and Y.A.; investigation, Y.D.; resources, Y.A. and Y.L.; data curation, Z.C. and Y.A.; writing—original draft preparation, Z.C.; writing—review and editing, Y.L.; supervision, Y.L.; project administration, Y.D. All authors have read and agreed to the published version of the manuscript. Data availability The datasets generated and/or analysed during the current study are available in the IEU OpenGWAS repository, https://opengwas.io/datasets/. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Zhaoming Cao, Yu An. Change history 5/7/2026 A Correction to this paper has been published: 10.1038/s41538-026-00834-3 Supplementary information The online version contains supplementary material available at 10.1038/s41538-025-00655-w. References Associated Data Supplementary Materials Supplementary Data 1 Data Availability Statement The datasets generated and/or analysed during the current study are available in the IEU OpenGWAS repository, https://opengwas.io/datasets/.