Influence of the 403 gut bacterial taxa on longevity‑related traits

We first investigate the causal relationships between the gut microbiome and longevity by performing a two-sample MR analysis using GWAS summary data of 403 gut bacterial taxa (196 taxa sourced from MiBioGen consortium, and 207 taxa from DMP) and longevity-associated traits. We observed suggestive evidence for many bacterial taxa to be associated with longevity, and these causal relationships were statistically significant with a P-value of less than 0.05, at least when employing the IVW method (Table S2).

In terms of healthspan, our findings revealed that Intestinimonas (β = 0.033, P = 0.049), Olsenella (β = 0.044, P = 0.002), and Turicibacter (β = 0.045, P = 0.016) were positively correlated with healthspan, while Anaerostipes (β = − 0.051, P = 0.034), Tyzzerella3 (β = − 0.030, P = 0.026), Ruminiclostridium9 (β = − 0.059, P = 0.012), Ruminococcus obeum (β = − 0.045, P = 0.030), Bacteroides xylanisolvens (β = − 0.034, P = 0.020), Bacteroides vulgatus (β = -0.059, P = 0.009), and Bacteroides eggerthii (β = − 0.038, P = 0.004) were negatively correlated with healthspan. In the MR-Egger regression, there was no evidence of directional pleiotropic effects (intercept p-value > 0.05). There is a significant heterogeneity only for Ruminococcus obeum in the Cochran’s Q test (p = 0.031).

For lifespan, Defluviitaleaceae UCG-011(β = 0.038, P = 0.014), Erysipelotrichaceae UCG003 (β = 0.038, P = 0.016), Senegalimassilia (β = 0.052, P = 0.002), Tyzzerella3 (β = 0.036, P = 0.025), Odoribacter (β = 0.031, P = 0.030), Alistipes senegalensis (β = 0.038, P = 0.008), Bacteroides faecis (β = 0.013, P = 0.033), Holdemania unclassified (β = 0.019, P = 0.040), and Bilophila unclassified (β = 0.030, P = 0.045) were positively associated with lifespan. In comparison, Butyricimonas (β = − 0.034, P = 0.016), Lachnospira (β = − 0.073, P = 0.012), Lachnospiraceae UCG-001 (β = -0.037, P = 0.014), Streptococcus salivarius (Wald ratio, β =− 0.040, P = 0.046), and Collinsella aerofaciens (β = − 0.04, P = 0.005) were negatively associated with lifespan. Intercept of MR-Egger regression also showed no potential horizontal pleiotropy. There is a significant heterogeneity only for Tyzzerella3 (Cochran’s Q test, p = 0.009).

In relation to longevity, our findings revealed a positive correlation between Bilophila wadsworthia (β = 0.309, P = 5 × 10⁻4) and Adlercreutzia equolifaciens (β = 0.172, P = 0.022) with individuals who attained a lifespan exceeding the 90th percentile. Lachnospiraceae bacterium 3_1_46FAA was positively associated with longevity (> 90th percentile, β = 0.195, P = 0.014; > 99th percentile, β = 0.328, P = 0.008). On the contrary, Blautia (β = -0.246, P = 0.001), and Escherichia coli (β = -0.131, P = 0.016) were negatively correlated with longevity (> 90th percentile). Akkermansia muciniphila was negatively correlated with longevity (> 99th percentile, β = -0.219, P = 0.030). Bacteroides massiliensis was negatively associated with longevity (> 90th percentile, β = − 0.179, P = 0.006; > 99th percentile, β = − 0.216, P = 0.038). The forest plot and scatter plot about B. massiliensis were presented in Fig. 2 and Figure S1, while the outcomes of the leave-one-out analysis confirmed that the MR estimates were not driven by strong effect SNPs (Figure S2). In addition, Haemophilus parainfluenzae was also negatively associated with longevity (> 90th percentile, β = -0.154, P = 0.011; > 99th percentile, β = − 0.231, P = 0.016).

Fig. 2

The forest plot depicts the causal associations between Bacteroides massiliensis and traits associated with human longevity. The slope value equals the β-value calculated using the three methods (IVW, weighted median, and MR Egger), and it signifies the magnitude of the causal effect. A positive slope indicates that exposure is a contributory factor in promoting the outcome, whereas a negative slope suggests the opposite effect. Abbreviations: CI, confidence interval; OR, odds ratio; Nsnp, the number of single nucleotide polymorphisms (SNPs)

For parental longevity, genus Anaerofilum (β = 0.019, P = 0.023) was positively associated with father's age at death, while Collinsella (β = 0.035, P = 0.035), and Eubacterium rectale group (β = 0.046, P = 0.005) were positively associated with mother's age at death. Desulfovibrio (β = 0.031, P = 0.036), and Eubacterium xylanophilum group (β = 0.061, P = 0.001) were positively associated with combined parental age at death. Slackia (β = 0.033, P = 0.002), Enterorhabdus (β = 0.025, P = 0.031), and Lachnospiraceae bacterium 5_1_63FAA (β = 0.013, P = 0.004) increased parental longevity odds of father's attained age, while Erysipelatoclostridium (β = 0.017, P = 0.044), and Eubacterium rectale group (β = 0.041, P = 0.038) increased parental longevity odds of mother's attained age. Meanwhile, Akkermansia muciniphila was positively associated with mother's age at death (β = 0.021, P = 0.033), combined parental age at death (β = 0.025, P = 0.012), and both parents in top 10% (β = 0.012, P = 0.047). Bacteroides fragilis (β = 0.009, P = 0.034) and Coprobacter fastidiosus (β = 0.022, P = 0.006) were causally associated with mother's attained age. Moreover, Eubacterium eligens was positively associated with parental longevity (both parents in top 10%, β = 0.018, P = 0.007), while Eubacterium rectale was positively associated with combined parental age at death (β = 0.030, P = 0.045) in the IVW method. Regarding negative association, Bacteroides (β = − 0.060, P = 0.006), Butyricicoccus (β = -0.025, P = 0.036), Flavonifractor (β = − 0.041, P = 0.036), Lachnospiraceae UCG008 (β = -0.018, P = 0.020), Odoribacter (β = − 0.045, P = 0.043), Tyzzerella3 (β = − 0.015, P = 0.022), Bacteroides dorei (β = − 0.021, P = 0.021), and Eubacterium biforme (β = − 0.012, P = 0.011) were negatively linked to specific traits of parental longevity. Notably, Oxalobacter in the MiBioGen dataset was negatively associated with mother's age at death (β = − 0.019, P = 0.038), but Oxalobacter in the DMP dataset was positively associated with father's attained age (β = 0.011, P = 0.048).

Finally, frailty, which often accompanies aging, was analyzed as well. Genus Bifidobacterium (β = 0.042, P = 0.013), Clostridium innocuum group (β = 0.023, P = 0.036), Eubacterium coprostanoligenes group (β = 0.054, P = 0.003), Flavonifractor (β = 0.023, P = 0.046), and species Ruminococcus torques (β = 0.035, P = 0.032) were positively associated with the frailty index (FI). The positive associations between the Clostridium innocuum group, the Eubacterium coprostanoligenes group, and frailty are consistent with a previous study[38].

Influence of the 903 gut bacterial taxa on longevity‑related traits

The MR results of the significant associations between 903 gut bacterial taxa (430 taxa from 8956 German individuals, and 473 taxa from 5,959 Finland individuals in the FR02 cohort) and longevity-related phenotypes are summarized in Table S3.

For human healthspan, Porphyromonadaceae (β = 0.033, P = 0.001), Atopobiaceae (β = 0.065, P = 0.042), Ruminococcaceae (β = 0.021, P = 0.005), Caloranaerobacter (β = 0.140, P = 0.003), Sutterella (β = 0.035, P = 0.002), Oscillibacter prevalence (β = 0.019, P = 0.036), Bifidobacterium breve (β = 0.060, P = 0.018), Lawsonibacter sp002161175 (β = 0.124, P = 0.037), and Monoglobus pectinilyticus (β = 0.048, P = 0.017) were positively correlated with healthspan. On the contrary, Alphaproteobacteria (β = -0.038, P = 0.007), and Coprococcus (β = 0.124, P = 0.037) were negatively correlated with healthspan.

For lifespans, Clostridium XlVa (β = 0.074, P = 0.007), Oscillibacter abundance (β = 0.049, P = 0.005), Ruminococcus(β = 0.009, P = 0.020), and Faecalibacterium prausnitzii E (β = 0.063, P = 0.024) were positively correlated with lifespans, while Roseburia (β = − 0.042, P = 0.016), Helicobacter (β = − 0.121, P = 0.004), Alistipes (β = − 0.112, P = 0.038), Photobacterium (β =− 0.145, P = 0.024), Desulfovibrio (β = − 0.022, P = 0.001), Parasutterella prevalence (β = − 0.011, P = 0.049), Oscillibacter prevalence (β = − 0.016, P = 0.031), and Lactococcus lactis (β = -0.050, P = 0.036) were negatively correlated with lifespans. The significant causal relationship between Oscillibacter and longevity-related traits were summarized in Figure S3.

For longevity, Parabacteroides (β = 0.116, P = 0.035), Gordonibacter (β = 0.219, P = 0.030), Sutterella abundance (β = 0.103, P = 0.015), and Alistipes shahii (β = 0.166, P = 0.007) abundance in stool were linked to increased longevity (> 90th percentile). Prevotella sp900317685 (β = − 0.218, P = 0.013) and Blautia A sp900066355 (β = − 0.272, P = 0.026) were negatively correlated with longevity (> 90th percentile). Prevotella (β = − 0.065, P = 0.045), Bacteroides (β = − 0.068, P = 0.004), Escherichia flexneri (β = − 0.341, P = 0.002), and Coprobacillus cateniformis (β = − 0.376, P = 0.007), and Parabacteroides johnsonii (β = − 0.261, P = 0.047) were negatively correlated with longevity (> 99th percentile). Both the Parasutterella prevalence (β = − 0.085, P = 0.035) and Parasutterella abundance (β = − 0.078, P = 0.001) were negatively correlated with longevity (> 90th percentile).

Many of the 430 gut bacterial taxa from 8956 German individuals have been causally associated with parental longevity. Alistipes (β = 0.020, P = 0.009), and Subdoligranulum prevalence (β = 0.011, P = 0.025) were positively correlated with combined parental age at death. Alongside the previously mentioned results, more correlations between Alistipes and longevity were identified (Fig. 3A), especially the two Alistipes species of A. senegalensis and A. shahii (Fig. 3B). The results of the “leave-one-out” test showed that there was no abnormal IV in this analysis affecting the overall results (Figure S4). In addition, the Desulfovibrio abundance was negatively correlated with two parental longevity traits (both parents in top 10%, β = − 0.007, P = 0.038; combined parental age at death, β = − 0.015, P = 0.016), but was positively correlated with another two parental longevity traits (combined parental attained age, β = 0.016, P = 2 × 10–4; father's attained age, β = 0.013, P = 0.001). The Sutterella prevalence was negatively correlated with combined parental age at death (β = − 0.015, P = 0.045). Faecalibacterium abundance was negatively correlated with father’s age at death (β = − 0.021, P = 0.008). More results were summarized in Table S3.

Fig. 3

MR results reveal the causal associations between Alistipes and traits associated with human longevity. A Forest plot depicting the associations between Alistipes and longevity-correlated traits. B Scatter plot depicting the associations between Alistipes senegalensis and lifespan, and between Alistipes shahii and longevity (> 90th percentile)

In addition, many of the 473 taxa from 5,959 Finland individuals have been causally associated with parental longevity. For example, Leuconostoc mesenteroide was positively associated with combined parental attained age (β = 0.019, P = 0.046). Prevotella sp000436915 was positively associated with combined parental attained age (β = 0.013, P = 0.048), but was negatively correlated with father's age at death (β = − 0.022, P = 0.005). Bifidobacterium breve was negatively correlated with both parents in top 10% (β = − 0.031, P = 0.002), but was positively correlated with combined parental attained age (β = 0.034, P = 0.010). Bifidobacterium angulatum was negatively correlated with mother's age at death (β = − 0.019, P = 0.043). Dorea phocaeense was negatively correlated with combined parental age at death (β = − 0.050, P = 0.025), but was positively associated with mother's attained age (β = 0.034, P = 0.018) and combined parental attained age (β = 0.035, P = 0.045). Coprobacillus cateniformis abundance was positively associated with father's age at death (β = 0.021, P = 0.036), but was negatively associated with mother's age at death (β = -0.030, P = 0.004). Eubacterium callanderi abundance was positively correlated with parental longevity (father's age at death) (β = 0.050, P = 0.020), but was negatively associated with father's attained age (β = − 0.042, P = 0.010). The abundance of Lactococcus lactis was positively correlated with father's attained age (β = 0.034, P = 0.025), and on the contrary, Clostridium tertium was negatively associated with it (β = − 0.051, P = 0.010). The abundance of Enterococcus faecalis in stool was positively correlated with mother's attained age (β = 0.027, P = 0.045), while Blautia A sp900066355 was negatively linked to it (β = -0.036, P = 0.007). Enorma massiliensis was positively correlated with mother's attained age (β = 0.027, P = 0.013) but was negatively correlated with father's attained age (β = − 0.025, P = 0.022).

The MR results of the association between 903 gut bacterial taxa and frailty indicated that Sutterella prevalence (OTU99_116, β = 0.009, P = 0.009; TestASV_22, β = 0.014, P = 0.002) was positively correlated with FI, while the Sutterella abundance (β = − 0.040, P = 0.004) and Parasutterella abundance (β = -0.008, P = 0.021) were negatively correlated with FI. The abundance of Faecalibacterium (β = 0.033, P = 0.015), and Anaeromassilibacillus sp001305115 (β = 0.047, P = 0.003) were positively correlated with FI. Bacteroides sp002160055 (β = − 0.036, P = 0.017), Bacteroides stercoris (β = − 0.025, P = 0.032), Lawsonibacter sp002161175 (β = − 0.081, P = 0.037), Morganella (β = − 0.038, P = 0.042), and Subdoligranulum abundance (β = − 0.037, P = 0.009) in stool were negatively correlated with FI, at least in the IVW method. To sum up, Subdoligranulum was causally associated with multiple longevity-correlated traits (Fig. 4), and no abnormal IV in this analysis affecting the overall results (Figure S5). Moreover, all the notable associations between Parasutterella and longevity-related traits indicate Parasutterella has a negative impact on lifespan, longevity, and frailty (Figure S6).

Fig. 4

The causal associations between Subdoligranulum and traits associated with human longevity. The forest plot A and scatter plot B depicting the association between Subdoligranulum prevalence and Parental longevity (combined parental age at death), and between Subdoligranulum abundance and frailty index, respectively

Influence of the 205 gut functional pathways on longevity‑related traits

The gut microbial taxa potentially regulate longevity through their associated metabolic pathways, prompting us to also analyze the influence of gut functional pathways on longevity‑related traits. Only the gut microbiome from 7,738 participants of the DMP encompassed 205 functional pathways [32], whereas the other three GWAS datasets of gut microbiota exclusively comprised microbial taxa. A total of 65 pathways have been discovered to exhibit significant and robust associations with traits linked to longevity (Table S4). There was no evidence of directional pleiotropic effects (intercept p-value > 0.05), and no significant heterogeneity (Cochran’s Q test, p-value > 0.05).

For human healthspan, microbial pathways of PPGPPMET-PWY: ppGpp biosynthesis (β = 0.037, p = 0.028), and PWY-6507: 4-deoxy-L-threo-hex-4-enopyranuronate degradation (β = 0.017, p = 0.044) showed a positive association, in the IVW method. In contrast, pathways such as PWY-6284: superpathway of unsaturated fatty acids biosynthesis (E. coli) (β =  − 0.049, p = 0.005), HOMOSER-METSYN-PWY: L-methionine biosynthesis I (β =  − 0.051, p = 0.011), DENOVOPURINE2-PWY: superpathway of purine nucleotides de novo biosynthesis II (β =  − 0.042, p = 0.030), PWY-6163: chorismate biosynthesis from 3-dehydroquinate (β =  − 0.043, p = 0.035), and PWY-5913: TCA cycle VI obligate autotrophs (β =  − 0.047, p = 0.004), were negatively correlated with healthspan.

For lifespan, pathways of NONOXIPENT-PWY: pentose phosphate pathway (non-oxidative branch) (β = 0.055, P = 0.002), PWY-6630: superpathway of L-tyrosine biosynthesis (β = 0.023, P = 0.008), PWY-7196: superpathway of pyrimidine ribonucleosides salvage (β = 0.047, P = 0.011), PWY-7209: superpathway of pyrimidine ribonucleosides degradation (β = 0.022, P = 0.032), and PWY0-162: superpathway of pyrimidine ribonucleotides de novo biosynthesis (β = 0.030, P = 0.033) were positively associated. In contrast, UBISYN-PWY: superpathway of ubiquinol 8 biosynthesis (prokaryotic) (β = -0.040, P = 0.001), KDO-NAGLIPASYN-PWY: superpathway of (Kdo)2-lipid A biosynthesis (β = − 0.024, P = 0.006), PWY-7211: superpathway of pyrimidine deoxyribonucleotides de novo biosynthesis (β = -0.032, P = 0.017), PWY_REDCITCYC: TCA cycle VIII (helicobacter) (β = -0.030, P = 0.007), PWY-5918: superpathway of heme biosynthesis from glutamate (β = -0.028, P = 0.022), PWY0-1338: polymyxin resistance (β = − 0.020, P = 0.023), and ENTBACSYN-PWY: enterobactin biosynthesis (β = − 0.039, P = 0.028) demonstrated negative associations with lifespan.

Regarding longevity, the abundances of PWY-6731: starch degradation III was positively associated with age > 90th percentile (β = 0.084, P = 0.020), while PWY-5941: glycogen degradation II (eukaryotic) was positively associated with age > 99th percentile (β = 0.262, P = 0.030). Moreover, THRESYN-PWY: superpathway of L-threonine biosynthesis (β = − 0.338, P = 0.017), and PWY0-1261: anhydromuropeptides recycling (β = − 0.346, P = 0.020) were negatively correlated with longevity (age > 99th percentile).

Regarding parental longevity, 46 associations have been identified. For instance, PWY-7197: pyrimidine deoxyribonucleotide phosphorylation was positively associated with father's attained age in the IVW method (β = 0.019, P = 0.045). PWY-7254: TCA cycle VII (acetate producers) was positively associated with mother's age at death (β = 0.016, P = 0.046). PWY-7456: mannan degradation was positively associated with father’s attained age (β = 0.031, P = 0.001). PWY-6897: thiamin salvage II was positively associated with both parents in top 10% (β = 0.023, P = 0.007). PWY-5088: L-glutamate degradation VIII to propanoate were positively associated with father's attained age (β = 0.015, P = 0.009). GALACTARDEG-PWY: D-galactarate degradation I was positively associated with mother's attained age (β = 0.014, P = 0.021), and combined parental attained age (β = 0.014, P = 0.045). On the contrary, PWY-7323: superpathway of GDP-mannose derived O-antigen building blocks biosynthesis (β = -0.026, P = 0.014) was negatively associated with parental longevity (father’s age at death). PWY-5838: superpathway of menaquinol-8 biosynthesis I (β = − 0.022, P = 0.043), and HOMOSER-METSYN-PWY: L-methionine biosynthesis I (β = − 0.029, P = 0.043) were negatively correlated with mother's age at death. PWY-5667: CDP-diacylglycerol biosynthesis I was negatively correlated with father's attained age (β = − 0.019, P = 0.012). PWY0-1415: superpathway of heme biosynthesis from uroporphyrinogen III (β = − 0.016, P = 0.012), DAPLYSINESYN-PWY: L-lysine biosynthesis I (β = − 0.021, P = 0.017), and ANAEROFRUCAT-PWY: homolactic fermentation (β = − 0.037, P = 0.029) were negatively correlated with combined parental attained age. METHGLYUT-PWY: superpathway of methylglyoxal degradation (β = − 0.008, P = 0.034), and ORNDEG-PWY: superpathway of ornithine degradation (β = − 0.010, P = 0.027) were negatively correlated with both parents in top 10%. Notably, COA-PWY (coenzyme A biosynthesis I) exhibited a positive association with father's age at death (β = 0.035, P = 0.013) and combined parental age at death (β = 0.038, P = 0.023), but exhibited a negative association with combined parental attained age (β = − 0.030, P = 0.002), as illustrated in Fig. 5A. The pathway of P162-PWY: L-glutamate degradation V via hydroxyglutarate was negatively correlated with mother’s age at death (β = − 0.028, P = 0.012) and both parents in top 10% (β = -0.020, P = 0.014), but was positively correlated with mother’s attained age (β = 0.018, P = 0.044) and combined parental attained age (β = 0.019, P = 0.034). PWY-7013: L-1,2-propanediol degradation exhibited a negative correlation with father’s age at death (β = -0.012, P = 0.034) and combined parental age at death (β = − 0.016, P = 0.042), but exhibited a positive correlation with father’s attained age (β = 0.008, P = 0.043) and combined parental attained age (β = 0.009, P = 0.042).

Fig. 5

The forest plots reveal the impact of two microbial pathways, COA-PWY and PWY-5100, on traits associated with longevity. A The notable associations between COA-PWY (coenzyme A biosynthesis I) pathway and parental longevity. B The notable associations between PWY-5100 (pyruvate fermentation to acetate and lactate II) and traits associated with lifespan and parental longevity. Four MR methods including IVW, weighted median, simple median, and MR Egger were used. The causal relationships were presented using OR (odds ratio) and 95% CI (confidence interval)

In terms of frailty-correlated pathways, we found PWY-7456: mannan degradation (β = − 0.056, P = 0.008), TRNA-CHARGING-PWY: tRNA charging (β = − 0.032, P = 0.027), POLYAMSYN-PWY: superpathway of polyamine biosynthesis I (β = − 0.062, P = 0.003), and PWY-5101: L-isoleucine biosynthesis II (β = -0.032, P = 0.019) were negatively correlated with frailty index. In comparison, PWY-5920: superpathway of heme biosynthesis from glycine (β = 0.013, P = 0.044), PWY-6630: superpathway of L-tyrosine biosynthesis (β = 0.021, P = 0.049), GLUCONEO-PWY: gluconeogenesis I (β = 0.054, P = 0.009), and GLUCOSE1PMETAB-PWY: glucose and glucose 1-phosphate degradation (β = 0.036, P = 0.026) were positively associated with frailty index.

There are several multiple associations between specific one pathway and longevity traits among different datasets (Table S5). For example, the PWY-5100 (pyruvate fermentation to acetate and lactate II) pathway was positively associated with both lifespan (β = 0.028, P = 0.039) and parental longevity of combined parental age at death (β = 0.032, P = 0.009), but was negatively correlated with father's attained age (β = − 0.022, P = 0.006), which was depicted in Fig. 5B. The NONOXIPEN-PWY: pentose phosphate pathway (non-oxidative branch) exhibited a positive association with lifespan (β = 0.055, P = 0.002), parental longevity of both parents in top 10% (β = 0.026, P = 0.009), and combined parental age at death (β = 0.051, P = 0.006), but exhibited a negative association with mother’s attained age (β = − 0.026, P = 0.017), father’s attained age (β = − 0.024, P = 0.032), and combined parental attained age (β = -0.036, P = 0.001) (Fig. 6A). The PWY-7209: superpathway of pyrimidine ribonucleosides degradation exhibited a positive association with lifespan (β = 0.022, P = 0.032), but exhibited a opposite association with parental longevity of mother's attained age (β = − 0.016, P = 0.040) and combined parental attained age (β = -0.013, P = 0.046). Moreover, the pathway of REDCITCYC: TCA cycle VIII (helicobacter) was consistently negatively associated with lifespan, and three traits of parental longevity (both parents in top 10%, mother's age at death, and combined parental age at death) (Fig. 6B). The of HOMOSER-METSYN-PWY (L-methionine biosynthesis I) was consistently negatively associated with lifespan, and parental longevity of mother’s age at death. The PWY-6507 (4-deoxy-L-threo-hex-4-enopyranuronate degradation) pathway was consistently positively associated with lifespan and parental longevity (both parents in top 10%). In addition, the PWY-7456 (mannan degradation) pathway was positively associated with parental longevity of father's attained age, and negatively associated with FI.

Fig. 6

The forest plots reveal the impact of two significant microbial pathways, NONOXIPEN-PWY and REDCITCYC, on multiple traits associated with longevity. A The notable associations between NONOXIPEN-PWY (pentose phosphate pathway, non-oxidative branch) pathway and traits associated with longevity. B The notable associations between REDCITCYC: TCA cycle VIII (helicobacter) pathway and traits associated with longevity. Four MR methods including IVW, weighted median, simple median, and MR Egger were used. The causal relationships were presented using OR (odds ratio) and 95% CI (confidence interval)