Characteristics of the murine and human samples

In the mouse study, compared to WT mice, all three groups of diabetic mice (VG, MET, SGLT2i + MET) exhibited characteristic features of obesity, including higher body and liver weights (Table 1). However, the liver to body weight ratios were similar among all groups, suggesting proportional organ weight increases with body mass (Table 1).

Table 1 Characteristics of the murine samples

Diabetic mice also displayed hyperglycemia, as evidenced by elevated blood glucose and insulin levels. Dyslipidemia was observed with elevated cholesterol and triglyceride levels, and systemic inflammation was evident with elevated C-reactive protein levels (Table 1). Following two weeks of daily monotherapy with metformin or combination therapy in diabetic mice, significant reductions in blood glucose levels were observed. The MET mice showed a 29.1% reduction in blood glucose levels, while the SGLT2i + MET mice exhibited a substantial 70.0% reduction. In contrast, the VG and WT mice experienced modest reductions of 4.7% and 1.9%, respectively (Table 1). The SGLT2i + MET treatment led to lower body, insulin, and triglyceride levels compared to MET alone, but with higher cholesterol and C-reactive protein levels, suggesting a broader metabolic impact of combination therapy.

Cross-sectional validation of our findings from mouse studies was conducted in the KORA and QBB human studies, involving a total of 478 T2D patients and available metabolite profiles. The mt-T2D patients, compared to metformin-naïve T2D patients in both studies, were older and exhibited higher levels of HbA1C and fasting glucose, suggesting more advanced dysglycemia in the mt-T2D group (Table 2).

Table 2 Characteristics of the KORA F4 and QBB cross-sectional study samples (N = 478)

Longitudinal validation was conducted in the prospective KORA study, spanning from the baseline survey (S4) to the 7-year follow-up (F4). Characteristics of these two groups, 34 T2D patients and 628 controls, were assessed at the S4 and F4 surveys (Table 3). The mt-T2D patients were found to be older, more sedentary, and had a higher prevalence of obesity compared to the group of 628 metformin-naïve individuals. Additionally, mt-T2D patients displayed higher blood pressure, triglyceride levels, and glycemic parameters at both surveys (Table 3).

Table 3 Characteristics of the KORA S4 → F4 prospective study samples (N = 662)Metformin effects on the blood metabolites in mouse and human studies

In the mouse study, among the 351 metabolites analyzed, plasma levels of seven metabolites were significantly altered following two weeks of metformin treatment in db/db mice. These results were determined to be statistically significant after applying Bonferroni correction (Fig. 1a, Supplementary Table 2, Additional File 1). Out of the seven metformin-associated metabolites, three were identified as intermediates of the TCA cycle, namely fumarate, malate, and α-ketoglutarate (α-KG). Of these, six metabolites showed upregulation in response to metformin treatment (e.g., fumarate displayed a positive β-estimate in the linear regression analysis when comparing MET with VG mice as shown in Fig. 1b, and the relative concentration in MET mice was higher than that of VG mice as displayed with boxplots in Fig. 1c).

Fig. 1

Effects of metformin and leptin receptor mutation in murine plasma. a Volcano plot of linear regression analysis result (β-estimate and P value) for 351 plasma metabolites in pairwise comparison of MET with VG diabetic mice. The upper, middle and lower dashed lines represent Bonferroni-corrected, FDR and nominal (P = 0.05) significance levels, respectively. b Seven metformin-associated metabolites of β-estimates with confidence intervals are shown. c Boxplots of seven metabolites in three groups. WT wild type mice, VG vehicle gavaged diabetic mice, MET metformin-treated diabetic mice, FDR false discovery rate, 2-AB 2-aminobutyrate, 4-HB 4-hydroxybutyrate, α-KG α-ketoglutarate. See also Supplementary Table 2, Additional File 1

In the comparison of VG with WT mice, none of the seven metabolites exhibited a significant difference after applying Bonferroni correction. However, the values of four of these metabolites (fumarate, malate, 4-hydroxybutyrate [4-HB], and uracil) were nominally affected (P < 0.05) by the genetic background of the mice as well as HFD (Supplementary Table 2, Additional File 1). Interestingly, three metabolites (α-KG, citrulline, and 2-aminobutyrate [2-AB]) did not show a significant difference between VG and WT mice. This suggests that the changes induced by metformin in these three metabolites in db/db mice were independent of the physiological consequences of the leptin receptor mutation besides the HFD.

In the human studies, three of the seven metformin-associated metabolites identified in murine plasma (malate, citrulline, and 2-AB) were measured in serum samples obtained from the KORA participants at both the S4 and F4 surveys. The analysis included a comparison between 70 individuals with mt-T2D and 114 individuals with ndt-T2D patients. Using a Bonferroni cutoff for significance (P < 0.017) for the three analyzed metabolites, two of them, citrulline and malate, were found to be significantly different in the fully adjusted model (Fig. 2a, Supplementary Table 3, Additional File 1). These findings were independently replicated in the plasma samples of patients from the QBB study, comparing 146 mt-T2D with 148 ndt-T2D patients, despite approximately 72% of the patients in the QBB study being non-fasting (Fig. 2b, Supplementary Table 3, Additional File 1). However, no significant correlation was observed for 2-AB in any of the comparisons (Fig. 2a, b, Supplementary Table 3, Additional File 1).

Fig. 2

Cross-sectional and longitudinal analyses reveal specific pattern of metformin action in human serum and plasma. a, b β-estimates with confidence intervals of three metabolites in KORA and QBB cross-sectional human studies. c Mean relative residue of two metabolites in longitudinal KORA study. All analyses are based on the fully adjusted model (age, sex, BMI, physical activity, high alcohol intake, smoking status, systolic blood pressure, HbA1C, fasting glucose levels, high density lipoprotein cholesterol, triglycerides). 2-AB 2 aminobutyrate. See also Supplementary Tables 3, 4, Additional File 1

In the longitudinal KORA S4-F4 study, we specifically examined the impact of metformin on the serum levels of malate and citrulline (Fig. 2c). Our analysis confirmed that metformin use intiated after the baseline (S4) led to a significant increase in malate and a decrease in citrulline by the follow-up (F4). The changes were Bonferroni significant in basic model (malate: β = 0.39, P = 3.31 × 10–4), while nominal significant in the full model (malate: β = 0.25, P = 0.043), based on a comparison of 34 patients who started metformin treatment with 628 who did not (Supplementary Table 4a, Additional File 1).

Further validation through sensitivity analyses, which matched participants by age and sex across four case–control ratios, consistently showed Bonferroni significant results for both metabolites in all comparisons, except for malate in the ratio of 1:10 in the full model, which showed nominal significance, similar to the population-based study (Supplementary Tables 4b–e, Additional File 1). These sensitivity analyses further reinforce the robustness of the association between metformin use and alterations in malate and citrulline levels over time in different ratios of cases to controls.

Metformin’s effects on the hepatic and renal metabolites

Among the 391 analyzed metabolites in the liver, two metabolites, glutamate and taurine, showed Bonferroni-significant associations with metformin (Fig. 3a, Supplementary Table 2, Additional File 1). Metformin treatment resulted in an upregulation of glutamate and a downregulation of taurine compared to VG mice. Additionally, the values of fumarate and malate in the db/db liver were upregulated by metformin at nominal significance levels. In the comparison between VG and WT mice, glutamate, fumarate, and malate were downregulated at an FDR significant level, suggesting that these changes may be associated with the genetic background and the HFD. Notably, the observed up and down regulation of these metabolites (glutamate, fumarate, and malate) among the MET, VG and WT mice may indicate beneficial effects of metformin in the liver of the db/db mice.

Fig. 3

Hepatic and renal effects of metformin. Volcano plots of linear regression results for 391 hepatic (a) and 447 renal (b) metabolites for the comparison between MET and VG. The upper, middle and lower dashed lines represent Bonferroni-corrected, FDR and nominal (P = 0.05) significance levels, respectively. Boxplots of selected metformin-associated metabolites in WT, VG and MET mice are shown. MET metformin-treated db/db mice, VG vehicle-gavaged db/db mice, WT wild type mice, 2-AB 2-aminobutyrate, 2-HG 2-hydroxyglutarate. See also Supplementary Table 2, Additional File 1

Moving to the kidneys of db/db mice, among the 447 analyzed metabolites, three metabolites (2-hydroxyglutarate [2-HG], 2-AB, and 5,6-dihydrouracil) exhibited Bonferroni-significant upregulation due to metformin treatment (Fig. 3b, Supplementary Table 2, Additional File 1). Additionally, malate values in the db/db kidneys were altered by metformin at a nominal significant level. In the comparison between VG and WT mice, 2-HG had a Bonferroni-significant downregulation, malate showed an upregulation at a FDR significant level, while comparable levels of 2-AB and 5,6-dihydrouracil were observed between VG and WT mice.

Metabolic effects of adding SGLT2i to metformin

Of 716 analyzed metabolites in the three tissues, three (butyrylglycine, N-acetyl glycine, and indole lactate) in plasma and two (choline and X-10460) in the liver were found to have Bonferroni-significant associations with the combination therapy when comparing SGLT2i + MET with MET mice (Fig. 4a, b, Supplementary Table 5, Additional File 1). Except for X-10460, all four identified metabolites were upregulated in the combination therapy group. However, none of the 447 analyzed renal metabolites showed either Bonferroni or FDR significant differences in the pairwise comparison between SGLT2i + MET and MET mice (Fig. 4c).

Fig. 4

Metabolic effects of combination therapy in the three murine tissues. Volcano plots in plasma (a), liver (b) and kidney (c) when compare SGLT2i + MET with MET mice. The upper, middle and the lower dashed lines represent Bonferroni-corrected, FDR and nominal (P = 0.05) significance levels, respectively. Boxplots of selected metabolites in MET and SGLT2i + MET mice are shown. MET metformin-treated db/db mice, SGLT2i + MET SGLT2i and metformin treated db/db mice. See also Supplementary Table 5, Additional File 1

In addition to the Bonferroni-significant metabolites associated with the combination therapy, several metabolites related to the TCA cycle showed FDR or nominal significant alterations. In the pairwise comparison between SGLT2i + MET and MET mice, the levels of malate, α-KG, and pyruvate in plasma, malate and glutamate in the liver, were downregulated. On the other hand, taurine and citrulline in the liver were upregulated (Fig. 4a, b, Supplementary Table 5, Additional File 1).

Tissue- and drug-specific effects of TCA cycle metabolites and its anaplerosis

Collectively, we observed different tissue-dependent responses to metformin and its combination with SGLT2i for intermediates of TCA cycle and its anaplerosis. Specifically, circulating malate levels in all three db/db mice (VG, MET, and SGLT2i + MET) were higher than WT ones, whereas their hepatic values were comparable between WT and MET mice, but lowered in both VG and SGLT2i + MET mice (Fig. 5a). These observations may indicate that the TCA cycle activity in leukocytes (erythrocytes lack mitochondria and functional TCA cycle) and hepatocytes responds differently to metformin. Moreover, in the three group of db/db mice, similar patterns for circulating malate and α-KG, hepatic malate and glutamate, and renal 2-HG, were observed (e.g., highest levels were observed in MET mice). Whereas, taurine levels were lowest in the MET mice in the liver (Fig. 5a). These results suggested that add-on SGLT2i to metformin reversed abundances of these metabolites, thereby suggesting a bidirectional modulation of TCA cycle metabolites and anaplerosis.

Fig. 5

Alteraration of TCA cycle metabolites and related pathways. a Boxplots showing selected metabolites across four db/db mouse groups. b Schematic overview of metformin’s effects on TCA cycle related metabolites in liver, kidney, and plasma of db/db mice and serum/plasma in humans. c Schematic of combined SGLT2i and metformin therapy in the same mouse tissues. This figure provides a simplified overview intended to facilitate general understanding of the alterations in TCA cycle metabolites across different tissues and treatment modalities. It is not meant to depict precise metabolic flows or fully detailed pathway interactions. Each organ has unique metabolic functions; thus, interpretations should consider the specific metabolic context of each tissue. 2-HG 2-hydroxyglutarate, α-KG α-ketoglutarate, TCA citric acid, NO nitric oxide