2.1 Study design and population

South China Cohort (SCC)-deep was a multiomics subcohort conducted in Dongguan City, Guangdong Province, P.R. China.¹⁵ Ethical approval for this study was granted by the Ethics Committee of School of Public Health, Sun Yat-Sen University (2017-001), and all procedures were performed in accordance with the principles of the Declaration of Helsinki. Written informed consents were obtained from each individual. To exclude the complex effects of hormones on both gut microbiota and insulin sensitivity, only male participants were included in this study. Detailed inclusion and exclusion criteria of SCC-deep were provided in Supplementary Methods (Additional file 1).

Among the 647 male participants from SCC-deep, 87 individuals were further excluded based on the following reasons: (1) missing data of fasting glucose (n = 4); (2) missing information for sedentary time (n = 41); (3) ongoing hyperglycaemia treatment (n = 39); (4) use of drugs known to affect gut microbiota (n = 3). In total, 560 individuals were included in this analysis (Figure 1). Detailed procedures for data collection were provided in Supplementary Methods (Additional file 1).

2.2 Ascertainment of insulin resistance and sedentary time

Homeostatic Model Assessment for Insulin Resistance (HOMA-IR), calculated as [fasting insulin (pmol/L) × fasting glucose (mmol/L)]/135,¹⁶ was employed to determine insulin resistance. Individuals were categorised into high or low HOMA-IR groups based on a threshold of 2.5, as previously described.¹⁷ Sedentary time was assessed using the International Physical Activity Questionnaire Short Form (IPAQ-SF), a widely used tool for the quantification of long-term physical activity and sedentary time in large-scale population studies. The study participants reported their sedentary time by indicating the total time (in hours and minutes) spent on sitting on a typical day over the past 7 days.¹⁸ Sedentary time and HOMA-IR were primarily analysed as continuous variables. Given the lack of a universally accepted threshold of sedentary time, we categorised participants based on the median sedentary time of the study population (6.8 h per day) when group comparisons were conducted, as previously described.¹⁹

2.3 Fecal DNA isolation and sequencing

Fecal samples were collected and processed following a standardised protocol, with DNA extraction and metagenomic sequencing performed on the Illumina NovaSeq 6000 platform. Fecal samples were collected on the same day as blood drawing. More details about sample handling, quality control, and sequencing parameters are available in Supplementary Methods.

2.4 Plasma metabolomics profiling

After an overnight fast of at least 10 h, blood samples were collected concurrently with stool collection, and preserved at −80°C until analysis. An established protocol for large-scale profiling of broadly targeted metabolites was applied, as described in the previous study.²⁰ Additional information on metabolomics profiling and downstream analyses were provided in Supplementary Methods.

2.5 Human genomics and bidirectional Mendelian randomisation analysis

Host genomic DNA was isolated from the buffy coat using a commercial blood DNA extraction kit, according to the manufacturer's instruction. DNA samples with a concentration ≥80 ng/µL and total yield ≥1 µg were used for library construction. More details about genotyping and quality control were provided in Supplementary Methods. To explore host genetic variants linked to gut microbiota or microbial metabolites, a threshold of p < 5×10⁻⁵ was adopted to maximise explained genetic variance, as previously described.²¹ To investigate the causal associations between microbial species and metabolites on HOMA-IR variability, bidirectional Mendelian Randomisation (MR) analysis was performed using the MendelianRandomization R package (version 0.10.0). MR estimates were evaluated by three methods, including inverse-variance weighted (IVW), weighted median and MR-Egger regression, with the IVW method serving as the primary method. More details were provided in Supplementary Methods.

2.6 Statistical analysis

The normality of continuous variables was assessed using the Shapiro–Wilk test and Q–Q plots. Data with approximate normal distribution were summarised as mean and standard deviation (SD), while skewed variables were expressed as median with interquartile range (25^th–75^th percentile). The comparison of basic characteristics was performed using the Student's t-test or Wilcoxon rank-sum test, based on the distribution of the data. Categorical variables were expressed as counts and percentages, and analysed with Chi-squared test. Missing data were handled by multiple imputation via the mice R package. Spearman rank correlation coefficient was calculated using cor.test. To adjust for multiple comparisons, false discovery rate (FDR) correction was applied using 1000 permutations, with statistical significance defined as FDR p  <  .05. All analyses were conducted by R software (V4.4.1, R Foundation for Statistical Computing, Vienna, Austria).

3.1 Microbial species mediate the impact of sedentary time on insulin resistance

A total of 560 male participants (median age: 50 years, interquartile range: 44–56) without history of hyperglycaemia treatment or antibiotic use were analysed. Overall, individuals with longer sedentary periods were younger, more likely to have less physical activity, higher education attainment, worse lipid profiles, in addition to poorer glycaemic control (Table 1).

We then explored the potential role of gut microbiota in sedentary time and insulin resistance. At community level, the diversity of gut microbiota was found to be inversely correlated with both sedentary time (p < .05, Figure 2A) and insulin resistance (p < .05, Figure 2B), after adjustment for potential confounders that might affect gut microbiota, such as age, body mass index (BMI), smoking, drinking, diet diversity, educational attainment and household income. Moreover, a clear segregation in beta diversity was observed in individuals with higher or lower sedentary time (p < .05, PERMANOVA, Figure 2C). A similar, albeit marginal, segregation in beta diversity was observed between individuals with high or low HOMA-IR levels (p = .053, PERMANOVA, Figure 2D), suggesting a potential link between sedentary time, gut microbiota alterations, and insulin resistance. Of note, a total of 55 species, mainly from Bacillota (formerly Firmicutes), Actinomycetota and Bacteroidota phyla, showed significant associations with sedentary time (FDR p < .05, Figure 2E and Table S1). Of the 15 key microbial species closely associated with sedentary time, Fusobacterium varium, Bacteroides xylanisolvens, Parabacteroides distasonis, and Bacteroides vulgatus demonstrated a positive association with sedentary time, while the other 11 species showed a negative association (Figure 2F). The 15 species collectively accounted for 31.68% of the sedentary time-related variation in insulin resistance, after correcting for demographic, lifestyle and socioeconomic factors (p < .05, Figure 2G). Among the 15 microbial species associated with sedentary time, only 8 species, the majority of which were well-known SCFA producers, such as Intestinibacter bartlettii (I. bartlettii),²² Roseburia sp CAG:471, Firmicutes bacterium CAG:83, Fusicatenibacter saccharivorans (F. saccharivorans) and Veillonella atypica (V. atypica), were also significantly associated with HOMA-IR (Figure 2H). Consistently, all these 5 SCFA-producers were also found to be associated with better lipid profiles, lower fasting glucose, insulin levels and BMI (Figure 2I). On the contrary, Bacteroides xylanisolvens (B. xylanisolvens), Bacteroides vulgatus (B. vulgatus),²³ and Fusobacterium varium (F. varium),²⁴ which were previously reported to be enriched in insulin-resistant or obese individuals, demonstrated a positive association with longer sedentary time, higher HOMA-IR, elevated DBP and worse lipid profiles (Figure 2I). Overall, our findings indicated that microbial species may serve as intermediaries linking sedentary time and insulin resistance.

3.2 Causal association between sedentary-shaped species and the progression of insulin resistance

We conducted a bidirectional MR analysis to investigate the causal relationship between sedentary-associated microbial species and the development of insulin resistance. Only 5 species, including Roseburia sp CAG:471, I. bartlettii, F. saccharivorans, Firmicutes bacterium CAG:83 and B. xylanisolvens, were potentially causally linked to HOMA-IR (Figure 3A–I). Specifically, an IVW estimate indicated that Roseburia sp CAG:471 (Beta_IVW = –0.100, 95% CI = –0.125 to –0.076, p < .001), I. bartlettii (Beta_IVW = –0.064, 95% CI = –0.093 to –0.034, p < .001), F. saccharivorans (Beta_IVW = –0.060, 95% CI = –0.081 to –0.038, p < .001) and Firmicutes bacterium CAG:83 (Beta_IVW = –0.034, 95% CI = –0.050 to –0.018, p < .001, Figures 3B–E and S1A) enhanced insulin sensitivity. Additionally, these microbial species were negatively associated with longer sedentary time and higher insulin levels (Figure 2I). On the contrary, B. xylanisolvens (Beta_IVW = 0.060, 95% CI = 0.014 to 0.105, p = .010) exacerbated insulin resistance (Figures 3F and S1A), and was found to be positively associated with longer sedentary time, higher SBP, DBP, glucose and TG levels (Figure 2I). However, the IVW estimates did not support any causal associations of V. atypica, B. vulgatus and F. varium with HOMA-IR (Figure 3G–I). Importantly, the directionality of effect estimated from the IVW method were consistent with those derived from the weighted median and MR-Egger methods, and all instrumental variables (IVs) had F-statistics > 10, effectively eliminating the bias of weak IVs. Moreover, no significant heterogeneity was detected among these IVs based on Cochran's Q test from the IVW analysis. Similarly, the MR-Egger regression intercept analysis did not indicate any evidence of directional horizontal pleiotropy in the observed causal estimates (Figure S1A). Additionally, sensitivity analysis using a leave-one-out approach further demonstrated that the observed causal associations between Roseburia sp CAG:471, I. bartlettii, F. saccharivorans, Firmicutes bacterium CAG:83 and B. xylanisolvens and HOMA-IR were robust and not disproportionately influenced by any single SNP (Figure S2).

Moreover, reverse MR analysis revealed no significant causal effects of HOMA-IR on Roseburia sp CAG:471, Firmicutes bacterium CAG:83 and B. xylanisolvens in both IVW and weighted median analyses, while the effects on F. saccharivorans were significant, indicating a bidirectional effect (Figures 3J–R and S1B). Of note, although the causal effect of HOMA-IR on I. bartlettii was significant in IVW analysis, the direction was opposite to that in the MR-Egger method (Figures 3L and S1B), suggesting that this causal association was not reliable.²⁵ Collectively, the findings demonstrated that Roseburia sp CAG:471, I. bartlettii, Firmicutes bacterium CAG:83 were negatively correlated with sedentary time and causally linked to an amelioration of HOMA-IR, whereas B. xylanisolvens was positively associated with sedentary time and causally linked to a progression of insulin resistance.

3.3 Sedentary time-related microbial capacities contribute to the progression of insulin resistance

To further elucidate the mechanisms by which reduced Roseburia sp CAG:471, I. bartlettii, Firmicutes bacterium CAG:83, and increased B. xylanisolvens influenced host metabolism and promoted insulin resistance, microbial genes were functionally annotated using Kyoto Encyclopedia of Genes and Genomes (KEGG) database. After adjustment for demographics, lifestyle and socioeconomic factors, a total of 38 metabolic functions, spanning amino acid, fatty acid and glycan metabolism, were found to be significantly associated with at least three of the four pivotal sedentary-associated species (FDR p < .05, Figure 4A and Table S2). Collectively, these microbial functions explained 21.48% of the sedentary time-related variation in HOMA-IR (Figure 4B). Of note, almost all the microbial functions demonstrated a consistent direction of correlation with both sedentary time and HOMA-IR (Figure 4C). More specifically, microbial amino acid biosynthesis, such as arginine biosynthesis (ko00220), and cysteine and methionine metabolism (ko00270), and lipid metabolism, including both phosphatidylcholine biosynthesis (ko00564) and its upstream regulator glycerolipid metabolism (ko00561), were inversely associated with both sedentary time and HOMA-IR. Additionally, both phosphatidylcholine biosynthesis (ko00564) and glycerolipid metabolism (ko00561) were also negatively associated with adverse lipid profiles and obesity (Figure 4C). By contrast, sphingolipid metabolism (ko00600), particularly ceramide biosynthesis, glycosphingolipid biosynthesis (ko00603, ko00604), pathways involved in bacterial outer membrane components, such as lipopolysaccharide biosynthesis (ko00540), glycosaminoglycan degradation (ko00531) and O-antigen nucleotide sugar biosynthesis (ko00541) exhibited positive associations with sedentary time. In addition, glycosaminoglycan degradation (ko00531) was positively associated with high glucose levels (Figure 4C). Moreover, functions responsible for glucuronate supply and nucleotide sugar synthesis (ko00040, ko00520) demonstrated a positive association with both sedentary time and insulin resistance (Figure 4C). Taken together, these findings underscored the mediation of the sedentary lifestyle on insulin resistance, through alterations in microbial metabolism.

Given that metabolites often serve as effectors between gut microbiota and host metabolism,²⁶ integrated metagenomics-metabolomics analysis was employed to further explore how sedentary-related microbial metabolites modulated insulin resistance. After controlling for demographics, lifestyle and socioeconomic factors, 206 metabolites and 407 metabolites were significantly associated with sedentary time and HOMA-IR, respectively (FDR p < .05, Figures 4D and E and S3A and B). Among them, 92 metabolites spanning glycerophospholipids, nucleotide and its derivative, amino acids, organic acid and its derivatives, and heterocyclic compounds and bile acids, demonstrated significant associations with both sedentary time and HOMA-IR with consistent directionality (Figure 4F). Furthermore, 16 of these metabolites mediated the effect of Roseburia sp CAG:471, I. bartlettii, and Firmicutes bacterium CAG:83 on insulin resistance (FDR p_mediation < 0.05), while no obvious mediating link between B. xylanisolvens and HOMA-IR was observed, after controlling for demographics, lifestyle and socioeconomic factors (Figures 4G and S4A–S). Particularly, increased D-erythrose-4-phosphate accounted for 6.59% of the effect of decreased Roseburia sp. CAG:471 and 16.19% of the effect of decreased Firmicutes bacterium CAG:83, respectively, on worsening HOMA-IR (Figure S4E and H). Additionally, elevated L-norvaline and L-valine accounted for the largest proportion of the effect of reduced Roseburia sp. CAG:471 on insulin resistance, contributing 18.95% and 15.64%, respectively to the exacerbation of HOMA-IR (Figures 4G and S4A and D). Consistently, these two metabolites were also found to be positively correlated with worse lipid profiles, and higher waist circumference (Figure 4H). Conversely, reduced allantoin accounted for 12.39% of the effect of decreased Roseburia sp. CAG:471 on the worsening of insulin sensitivity (Figures 4G and S4B), and demonstrated an positive association with adverse metabolic profiles, such as obesity and elevated blood pressure (Figure 4H). In a similar fashion, increased levels of L-asparagine and lysophosphatidylserine (22:6/0:0) contributed to 21.76% and 16.07% of the effect of Firmicutes bacterium CAG:83 on improved insulin sensitivity, respectively (Figures 4G and S4O and M). Taken together, these findings highlighted the potential of microbial metabolites as molecular transducers linking prolonged sedentary time to insulin resistance.

3.4 L-citrulline and L-serine serve as key microbial effectors linking sedentary time to insulin resistance

Beyond numerical links observed above, metabolites with significant mediatory effects were further mapped to the KEGG database to explore their biological links. Intriguingly, argH and argF, key enzymes involved in arginine biosynthesis (ko00220), can be annotated in the genomes of Roseburia sp CAG:471 and Firmicutes bacterium CAG:83 through protein sequence BLAST (Figure 5A). In individuals with excessive sedentary time, the abundance of these two species was remarkably lower (Figure S5A and B), leading to a downregulation of arginine biosynthesis (ko00220) (Figure S5C). Such a downregulation was associated with a marked reduction in key intermediate metabolites along this pathway, such as L-arginosuccinate and L-ornithine, which in turn led to a suppressed production of L-citrulline, primarily due to a reduction in argF (Figure S5D–G), which can be encoded by both of these two species (Figure 5A). Notably, decreased L-citrulline accounted for 14.21% of the effect of decreased Roseburia sp. CAG:471 and 19.14% of the effect of decreased Firmicutes bacterium CAG:83, respectively on insulin resistance (Figure S4C and G). Consistently, L-citrulline was found to be substantially lower in individuals with longer sedentary time and demonstrated a negative association with insulin resistance, adverse lipid profiles, high BMI, and elevated blood pressure (Figures 4H and S5H). On the other hand, asd and thrC can also be annotated in the genome of Firmicutes bacterium CAG:83 (Figure 5B). Notably, L-aspartic acid served as a critical intermediate metabolite linking arginine biosynthesis (ko00220) and glycine, serine, and threonine metabolism (ko00260). In individuals with longer sedentary time, L-aspartic acid levels were found to be remarkably lower, which in turn contributed to a decreasing trend in ko00260 (Figures 5B and S5G–I). Consistent with the pathway-level inference, most of the key intermediate metabolites and enzymes within this pathway, including lysC, asd, hom, L-homoserine, L-threonine, ltaE, glycine and glyA exhibited a downregulated trend, and thrB was significantly lower in subjects with longer sedentary time (Figure S5K–S), which ultimately resulted in a markedly reduced biosynthesis of L-serine (Figures 5B and S5T). Remarkably, reduced L-serine accounted for 7.54% of the effect of Firmicutes bacterium CAG:83 on insulin resistance (Figure S4I), and was inversely associated with HOMA-IR, unfavourable lipid profiles, central obesity, and elevated blood pressure (Figure 4H).

Furthermore, bidirectional MR analysis demonstrated that both L-citrulline and L-serine were causally linked to the variation of HOMA-IR, whereas, no significant effects of HOMA-IR on these two metabolites were found (Figure 6A). Moreover, all IVs had F-statistics > 10, eliminating the risk of weak IV bias. Additionally, no significant heterogeneity was detected among these IVs based on Cochran's Q test from the IVW analysis. Similarly, the MR-Egger regression intercept analysis indicated no evidence of directional horizontal pleiotropy in these causal estimates (Figure S6A). Additionally, sensitivity analysis using a leave-one-out approach further demonstrated that the observed causal associations between L-citrulline and L-serine and HOMA-IR were robust and not disproportionately influenced by any single SNP (Figure S6B), suggesting that L-citrulline and L-serine were causally linked to an amelioration of insulin resistance. More importantly, the inverse associations between L-citrulline, L-serine, and HOMA-IR were further validated in an independent validation cohort²⁷ (Figure 6B), lending further support to the crucial role of L-citrulline and L-serine in the alleviation of insulin resistance induced by prolonged sedentary time. Furthermore, more than half of the downstream targets of L-citrulline and L-serine had been previously linked to the development of insulin resistance (Figure 6C). For instance, the nitric oxide synthase (NOS) family of enzymes, including NOS2 and NOS3,²⁸ were found to be potential downstream targets of L-citrulline by in silico analysis. Consistent with our finding, increased NOS2 expression in the skeletal muscle of obese mice had been reported to be associated with enhanced S-nitrosation of key insulin signalling components, such as the insulin receptor, insulin receptor substrate 1, and protein kinase B, which in turn disrupted insulin signalling and promoted insulin resistance.^{29, 30} In contrast, NOS3 deficiency had been found to be associated with reduced fatty acid oxidation,³¹ elevated circulating triglyceride levels, and fasting hyperinsulinemia.³² Similarly, nuclear factor kappa-B, a downstream target of both L-citrulline and L-serine, had been proved to be a key player in obesity-induced low-grade inflammation and insulin resistance.³³ Gene Ontology (GO) enrichment revealed that the downstream targets of L-citrulline and L-serine were primarily linked to metabolic processes and various signalling responses (Figure 6D). Notably, several GO terms associated with insulin receptor signalling or cellular response to insulin stimulus were identified (Figure 6E), supporting that reduced Roseburia sp. CAG:471 and Firmicutes bacterium CAG:83 due to excessive sedentary time may impair insulin signalling and promote insulin resistance. Overall, these findings indicated that decreased levels of L-citrulline and L-serine functioned as critical microbial effectors linking the detrimental effect of sedentary behaviour on insulin resistance.