Gut microbiome and frailty: insight from genetic correlation and mendelian randomization

ABSTRACT

Observational studies have shown that the gut microbiome is associated with frailty. However, whether these associations underlie causal effects remains unknown. Thus, this study aimed to assess the genetic correlation and causal relationships between the genetically predicted gut microbiome and frailty using linkage disequilibrium score regression (LDSC) and Mendelian Randomization (MR). Summary statistics for the gut microbiome were obtained from a genome-wide association study (GWAS) meta-analysis of the MiBioGen consortium (N = 18,340). Summary statistics for frailty were obtained from a GWAS meta-analysis, including the UK Biobank and TwinGene (N = 175,226). We used LDSC and MR analyses to estimate the genetic correlation and causality between the genetically predicted gut microbiome and frailty. Our findings indicate a suggestive genetic correlation between Christensenellaceae R-7 and frailty. Moreover, we found evidence for suggestive causal effects of twelve genus-level gut microbes on frailty using at least two MR methods. There was no evidence of horizontal pleiotropy or heterogeneity in the MR analysis. This study provides suggestive evidence for a potential genetic correlation and causal association between several genetically predicted gut microbes and frailty. More population-based observational studies and animal experiments are required to clarify this association and the underlying mechanisms.

Introduction

Frailty is a clinical syndrome related to aging that mainly manifests as a state of nonspecific vulnerability, reduced multisystem physiological reserves, and reduced resistance to stressors¹. Frailty is a major public health challenge worldwide. A recent systematic review and meta-analysis involving 62 countries and regions showed that the combined prevalence of physical frailty in older adults was 12%². The prevalence of frailty has created a serious burden on older adults, families, and society. Several meta-analyses have shown that frailty is associated with an increased risk of all-cause mortality, cause-specific mortality from cardiovascular disease (CVD), cancer, and respiratory illness^3–5. In addition, frailty also increases the cost of medical care for older people⁶, resulting in catastrophic health expenditure⁷. A recent simulation prediction study in Japan showed that by 2043, it is estimated that 97 billion US dollars will be spent on frail care⁸. However, there are no specific drugs to prevent and treat frailty, and non-drug interventions such as nutritional interventions are still one of the main means of preventing and treating frailty⁹.

Although the pathophysiological mechanisms of frailty have not been fully elucidated, current studies identify inflammation as one of the core mechanisms^10,11. In recent years, a new hypothesis regarding the origin of inflammation in the digestive tract, especially the imbalance of intestinal homeostasis, has attracted the attention of the academic community¹². Previous studies have shown that intestinal ecological imbalance causes the transformation of gut microbiota to pathogenic bacteria and the reduction of microbial diversity¹³, thus increasing the permeability of the mucosal barrier and allowing bacteria and their products to enter the human body through the intestine, leading to systemic inflammation^14–16. Thus, it can be inferred that the gut microbiota may be related to frailty. Some observational studies have preliminarily explored this relationship. A cross-sectional study of older adults in South Korea found that the frailty score was positively correlated with the Bacteroides and negatively correlated with Prevotella ¹⁷. A recent systematic review and meta-analysis based on observational studies showed that the species richness index and species diversity index of gut microbiota in frail older adults were significantly lower than those in non-frail¹⁸. A systematic review involving 10 case-control and one cohort study found that, compared with healthy older people, frail older adults exhibit decreased gut microbiota diversity and lower abundance of short chain fatty acids (SCFAs) producers¹⁹. Studies using mouse models also suggested that changes associated with aging-related disorders and frailty in gut microbiota involved a decrease in SCFAs producers such as Akkermansia ^20–23. SCFAs, as epigenetic modifiers of DNA and histone proteins, can participate in regulating age-related aging-associated chronic low-grade inflammation, muscle loss, glucose and lipid metabolism, and other pathophysiological process of frailty^24,25.

However, it should be noted that owing to the limitation that observational research mentioned above cannot infer the causal relationship, it is still unknown whether there is a causal relationship between gut microbiota and frailty. Therefore, the exploration of the causal relationship between the two needs to be deepened. In recent years, statistical methods based on genome-wide association study (GWAS) have been proposed to estimate the correlation and causality between traits. Linkage disequilibrium score regression (LDSC) can evaluate the genetic correlation from GWAS summary statistics and is not biased by sample overlap²⁶. Furthermore, Mendelian randomization (MR) has attracted wide attention in the medical field by inferring the causal relationship between variables by means of the instrumental variable of genetic variation. Because genotype precedes phenotype and alleles are randomly assigned at conception, using genetic variation as an instrumental variable to estimate causality can avoid measurement bias, confounding bias, and reverse causality interference²⁷. Therefore, this study assessed the genetic correlation and causal relationships between the genetically predicted gut microbiome and frailty using LDSC and MR.

Materials and methods

Study design

We conducted LDSC and two-sample MR to estimate the genetic correlation and causal relationships between the genetically predicted gut microbiome and frailty. An overview of the study design is presented in Figure 1 (by Figdraw).

Figure 1.

Overview of study design.

Data sources

Genetic variants associated with the gut microbiome were obtained from the largest GWAS meta-analysis published to date conducted by MiBioGen consortium²⁸. This study coordinated 16S rRNA gene sequencing profiles and genotyping data from 18,340 participants from 24 cohorts from the USA, Canada, Israel, South Korea, Germany, Denmark, the Netherlands, Belgium, Sweden, Finland, and the UK. Most of participants had European ancestry (N = 13,266). Among them, genus was the lowest taxonomic level, and 131 genera (including 12 unknown genera) with a mean abundance higher than 1% were identified²⁹.

We downloaded GWAS summary statistics for frailty via the GWAS catalog. This study included 175,226 individuals of European descent (164,610 UK Biobank individuals aged 60–70 years and 10,616 Swedish TwinGene individuals aged 41–87 years) and used the frailty index (FI) to measure frailty³⁰. FI combines dozens of parameters, including symptoms, signs, disease status, and disability, and reflects the accumulation of potential health deficits during the life course³¹.

Genetic IVs

We performed a series of selection criteria to filter eligible genetic IVs: (1) Since the number of eligible IVs less than the genome-wide significance threshold (p < 5 × 10⁻⁸) was extremely small, based on previous studies^28,29,32,33, a relatively less stringent threshold (p < 1 × 10⁻⁵) was selected to capture potential sets of variants likely to be enriched for association and obtain more comprehensive results³⁴. (2) We conducted a clumping procedure (R² <0.001, window size = 10,000 kb) to exclude variants in strong linkage disequilibrium (LD) and ensure the independence of each SNP. (3) SNPs with a minor allele frequency of < 0.01, ambiguous SNPs with non-concordant alleles, and palindromic SNPs were excluded. (4) We applied a PhenoScanner^35,36 search to identify all known phenotypes associated with genetic IVs (p < 5 × 10⁻⁸). If the genetic IV is associated with any other known phenotype, it would be excluded from subsequent MR analysis. We also refer to the largest GWAS published to date, which included 20 dietary habits such as raw vegetable intake, fresh fruit intake, and oily fish intake et al³⁷. We removed genetic IVs associated with the aforementioned 20 dietary habits.

Statistical analysis

Genetic correlation analysis

We estimated the genetic correlation(r_g) between gut microbiota and frailty using LDSC. GWAS summary statistics were filtered according to HapMap3 ref. Variants that were not SNPs (e.g., indels) and SNPs that were strand-ambiguous, repeated, and had a minor allele frequency (MAF) <0.01 were excluded. The LDSC examines the association between test statistics and linkage disequilibrium to quantify the contribution of inflation from a true polygenic signal or bias³⁸. This method can evaluate genetic correlation from GWAS summary statistics and is not biased by sample overlap²⁶. The z-scores of each variant from Trait 1 are multiplied by the z-scores of each variant from Trait 2. The genetic covariance was estimated by regressing this product against the LD score³⁹. The genetic covariance normalized by SNP-heritability represents the genetic correlation. p < 0.0004 (0.05/119, after strict Bonferroni correction) was considered statistically significant. 0.0004 < p < 0.05 was considered to be suggestive evidence for potential genetic correlation.

MR analysis

Before the MR analysis, we calculated the F-statistic of the microbiome IVs to determine whether there was a weak IV bias⁴⁰. An F-statistic <10 indicates a weak IV bias⁴¹. The formula for F-statistics is shown in Figure 2.

Figure 2.

The formula for F statistics.

Note: *p < 0.05

In this study, we explored the causal relationship between the gut microbiome and frailty using five methods: inverse variance weighted (IVW), MR-Egger, weighted median, weighted mode, and Robust Adjusted Profile Score (RAPS). The IVW method is considered the most accurate and powerful method for estimating causal effects when all selected SNPs are valid IVs⁴².

A consistent casual effect of the gut microbiome on frailty across several methods could be more reliable^33,43. In our study, the causal effect with an adjusted p < 0.0004 (0.05/119, after strict Bonferroni correction) in at least two analysis methods was considered significant. In at least two analysis methods, 0.0004 < p < 0.05, were considered to be suggestive evidence for potential causality. We applied the MR-Egger method to detect horizontal pleiotropy⁴⁴. If pleiotropy was present, the analysis yielded an intercept of p < 0.05. The Cochran Q test was used to assess heterogeneity. Outlier variants and potential horizontal pleiotropy were assessed using the MR-PRESSO method⁴⁵. We eliminated outliers based on this.

All statistical analyses were performed using the LDSC Version 1.0.1, “TwoSampleMR” package²⁷ and the “MR-PRESSO” package in R version 4.2.0.

Results

LDSC regression analysis

We performed LDSC regression analysis to evaluate the genetic correlation between 119 genus-level gut microbes and frailty. Owing to limitations such as low heritability and sample size, some genera cannot be used for the above analysis. Finally, we obtained the estimations of genetic correlation between the 61 genera and frailty. As shown in Table 1 and Figure 3, LDSC showed a suggestive correlation between Christensenellaceae R-7 and frailty (r_g= −0.212, p = 0.047). Detailed information regarding all genetic correlation results is listed in Table S1.

Table 1.

The genetic correlations between gut microbes and frailty.

Trait1	Trait2	rg	SE	P-Value
Christensenellaceae R-7	Frailty	−0.212	0.107	0.047

Figure 3.

Circular heat map of suggestive genetic correlation between gut microbes and frailty.

MR analysis

According to the criteria of screening, 1132 SNPs were selected as IVs for 119 genus-level gut microbes. The F statistics of all selected IVs were > 10, indicating a small possibility of weak instrument bias. Details of all the selected IVs are shown in Table S2. 12 genus-level gut microbes were examined to have suggestive associations with frailty in at least two MR methods (Table 2 and Figures 4, 5, 6). Scatter plots displayed the associations of the SNP effects on 12 genus-level gut microbes against the SNP effects on frailty. The results of the estimates of causal associations between 119 genera and frailty are presented in Table S3.

Table 2.

Significant MR results of causal association between gut microbes and frailty.

Exposure	Method	N.snp	beta	95%CI	p
Clostridium innocuum	Inverse variance weighted	9	0.023	(0.001,0.044)	0.036
Clostridium innocuum	RAPS	9	0.024	(0.001,0.046)	0.043
Eubacterium coprostanoligenes	Weighted median	13	0.068	(0.020,0.116)	0.006
Eubacterium coprostanoligenes	Inverse variance weighted	13	0.054	(0.019,0.090)	0.003
Eubacterium coprostanoligenes	RAPS	13	0.055	(0.016,0.094)	0.005
Eubacterium ruminantium	Inverse variance weighted	18	−0.027	(−0.051,-0.003)	0.028
Eubacterium ruminantium	RAPS	18	−0.029	(−0.050,-0.008)	0.007
Akkermansia	Inverse variance weighted	10	−0.042	(−0.074,-0.011)	0.009
Akkermansia	RAPS	10	−0.044	(−0.078,-0.010)	0.011
Bifidobacterium	Inverse variance weighted	11	0.044	(0.010,0.079)	0.012
Bifidobacterium	RAPS	11	0.045	(0.009,0.082)	0.015
Butyrivibrio	Inverse variance weighted	13	−0.019	(−0.035,-0.003)	0.020
Butyrivibrio	RAPS	13	−0.020	(−0.037,-0.003)	0.021
Catenibacterium	Inverse variance weighted	4	−0.031	(−0.058,-0.003)	0.030
Catenibacterium	RAPS	4	−0.031	(−0.061,-0.001)	0.046
Christensenellaceae R-7 group	Inverse variance weighted	7	−0.059	(−0.107,-0.011)	0.017
Christensenellaceae R-7 group	RAPS	7	−0.061	(−0.113,-0.008)	0.023
Defluviitaleaceae UCG-011	Weighted median	8	−0.042	(−0.082,-0.002)	0.039
Defluviitaleaceae UCG-011	Inverse variance weighted	8	−0.034	(−0.063,-0.005)	0.024
Defluviitaleaceae UCG-011	RAPS	8	−0.035	(−0.067,-0.004)	0.027
Howardella	Weighted median	9	0.027	(0.001,0.053)	0.047
Howardella	Inverse variance weighted	9	0.024	(0.004,0.045)	0.019
Howardella	RAPS	9	0.025	(0.003,0.046)	0.028
Ruminococcus 1	Weighted median	9	0.069	(0.016,0.122)	0.011
Ruminococcus 1	Inverse variance weighted	9	0.054	(0.013,0.096)	0.010
Ruminococcus 1	RAPS	9	0.057	(0.016,0.098)	0.007
Allisonella	Inverse variance weighted	8	0.032	(0.007,0.057)	0.012
Allisonella	RAPS	8	0.034	(0.013,0.055)	0.001

Figure 4.

Suggestive causal effects of 12 genus-level gut microbes on frailty.

Figure 5.

Scatter plots for causal effects of gut microbes on frailty.

Note: (a) Clostridium innocuum-frailty (b) Eubacterium coprostanoligenes-frailty (c) Eubacterium ruminantium-frailty (d) Akkermansia-frailty (e) Bifidobacterium-frailty(f) Butyrivibrio-frailty

Figure 6.

Scatter plots for causal effects of gut microbes on frailty.

Note: (a) Catenibacterium-frailty (b) Christensenellaceae R-7 group-frailty (c) Defluviitaleaceae UCG-011-frailty (d) Howardella-frailty (e) Ruminococcus 1-frailty(f) Allisonella-frailty

Sensitivity analysis

Through visual inspection of the scatter plot, there were potential outliers of the IVs of Eubacterium ruminantium, Akkermansia, Butyrivibrio, Defluviitaleaceae UCG-011, Ruminococcus 1, and Allisonella. However, according to the results of the radial MR-Egger intercept and MR-PRESSO global tests, there was no evidence of horizontal pleiotropy (Table S5). In addition, no significant heterogeneity was found in the Cochran’s Q test (Table S4).

Discussion

To the best of our knowledge, this is the first study to explore the genetic correlation and potential causality between the gut microbes and frailty by using GWAS summary statistics. Our findings indicate a suggestive genetic correlation between Christensenellaceae R-7 and frailty. Moreover, we found evidence for suggestive causal effects of twelve genus-level gut microbes on frailty in the MR analysis. These results will help us further explore the role of the gut microbes in aging and provide references for the development of future interventions and potential therapeutic targets.

Eubacterium coprostanoligenes are known to be associated with cholesterol metabolism⁴⁶. Early animal experiments found that feeding Eubacterium coprostanoligenes significantly decreased blood cholesterol concentrations in mice⁴⁷ and rabbits⁴⁸. A recent intervention experiment in mice also found that supplementation with a mixture of Opuntia ficus-indica, Theobroma cacao, and Acheta domesticus increased the abundance of Eubacterium coprostanoligenes in obese mice, thereby reducing serum cholesterol levels⁴⁹. Meanwhile, a large-scale study based on the UK Biobank found that lower levels of total, low-density lipoprotein (LDL), and high-density lipoprotein (HDL) cholesterol were associated with a higher prevalence of frailty⁵⁰. It can be deduced from this that the higher the abundance of Eubacterium coprostanoligenes, the lower cholesterol level may be, which may increase the risk of frailty. However, it is worth noting that there is not yet sufficient evidence to support the association between Eubacterium coprostanoligenes and lower blood cholesterol concentrations in humans. Considering the racial differences between humans and animals such as rabbits and mice, future studies need to further explore whether the positive effect of Eubacterium coprostanoligenes on frailty is mediated by cholesterol concentration in human subjects.

Allisonella was associated with decreased bone mineral density and bone metabolic indicators in postmenopausal women⁵¹. Additionally, it was more abundant in individuals with a high inflammatory index⁵². Interleukin 6 (IL-6) and tumor necrosis factor-alpha (TNF-α) have been recognized as the biomarkers of frailty^53,54. The proinflammatory properties of Allisonella seem to be a possible explanation for its suggestive association with frailty. In addition, this study found that a higher abundance of Bifidobacterium may increase the risk of frailty, similar to a previous observational study¹⁸. Bifidobacterium is considered to be a physiologically beneficial bacterium that can maintain intestinal homeostasis, regulate immune function, and reduce the growth of harmful bacteria⁵⁵. Surprisingly, an increased abundance of Bifidobacterium was found in both Parkinson’s patients⁵⁶ and frail older people¹⁸. A previous case-control study also found a significantly higher proportion of Bifidobacterium in patients with active inflammatory bowel than in healthy controls⁵⁷. Given that direct evidence is currently lacking, further exploration of the relationship between Bifidobacterium and frailty is warranted in the future.

Moreover, researchers analyzed the gut microbiota characterization of 29 subjects using Illumina MiSeq sequencing and found that the abundance of Howardella in the prediabetic group was significantly higher than that in healthy subjects⁵⁸. Recent studies have pointed out that diabetes and concomitant impaired glucose homeostasis and dysregulated nutrient-sensing participate in pathways linked to the metabolism of aging, which may weaken physiological reserves and lead to frailty⁵⁹. This may be one of the potential mechanisms underlying the suggestive association between Howardella and frailty. A study of 85 community-dwelling adults suggested that the module of co-occurring microbial genera composed of Ruminococcus, Eggerthela, and Coprobacillus was positively correlated with the frailty index, and the association remained robust after correction for body mass index (BMI), subject age, antibiotic use, and other confounding factors⁶⁰. Another study reported that Ruminococcus 1 was associated with lower adjusted body weight in older men⁶¹. This further implies that the cooperation between Ruminococcus 1 and other virulent symbionts may be involved in chronic inflammation⁶² and the subsequently acceleration of the aging process and frailty trajectory⁶⁰.

In addition to the above five gut microbiota, we identified six genera that may be associated with a reduced risk of frailty. The first was Eubacterium ruminantium. A previous study involving 27 hospitalized elderly patients showed that the abundance of Eubacterium ruminantium in participants with frailty was lower than that in the non-frailty group⁶³. The second was Akkermansia. The relative abundance of Akkermansia genus was higher in chronic kidney disease patients with sarcopenia⁶⁴ and community-dwelling older adults with frailty⁶⁵ but lower in cirrhotic patients with sarcopenia⁶⁶. Although the results of few population-based studies are inconsistent, MR results suggested that Akkermansia may have a potential negative effect on frailty. These further supports previous animal experiments^21,22. The third was Butyrivibrio. There is currently no observational study that has found an association between Butyrivibrio and frailty. However, a recent cohort study found that increased abundance of Butyrivibrio was associated with 1-year improvement in insulin status among elderly Mediterranean population at high cardiovascular risk.⁶⁷ In general, Eubacterium ruminantium ^{68, 69}Akkermansia ²¹ and Butyrivibrio ⁷⁰ are believed to be new generation of “possibly helpful microbe” with the ability to produce SCFAs. SCFAs may prevent or alleviate frailty in the following ways: First, they regulate the differentiation, recruitment, and activation of immune cells, and reduce the secretion of inflammatory cytokines to play an anti-inflammatory role^71,72. Second, they promote cognitive function by influencing the integrity of microglia and microglia-related activation involved in neuroinflammation, inducing the secretion of glucagon-like peptide 1 (GLP1) and peptide YY (PYY)⁷³. Third, they regulate the synthesis and degradation of muscle proteins to maintain muscle quality and function⁷⁴. In addition, they mediate the nuclear erythroid 2-related factor 2 (Nrf2)-related pathway, which can reduce oxidative and mitochondrial stress to delay aging⁷⁵.

We found that a higher abundance of Catenibacterium may be associated with a lower risk of frailty. A recent study on patients with chronic liver diseases found that the abundance of Catenibacterium was lower in the sarcopenia group⁷⁶. However, other studies have not obtained such significant results⁷⁷. As a gram-positive anaerobic bacteria⁷⁸, Catenibacterium typically has proinflammatory property and is highly abundant in obesity and infectious diseases^79,80. However, our study implied that frailty seems to be associated with a lower risk of frailty. This finding may need further validation in experimental studies, as only four instrumental variables may have implicit associations.

We also found a suggestive genetic correlation and causal association between a higher abundance of Christensenellacea R-7 and a lower risk of frailty, which further corroborates the findings of previous observational studies. A cross-sectional study including 35 Italian community dwellers over 70 years old reported that the abundance of Christensenellaceae in physical frailty and sarcopenia (PF&S) group was significantly lower than non-PF&S group⁸¹. It seems to be gradually being identified as a potential biomarker of longevity in studies on the characteristics of the gut microbiome from centenarians^82,83. Age-related adipose tissue dysfunction can lead to the infiltration of immune cells, secretion of proinflammatory cytokines and chemokines, and increased senescence-associated secretory phenotype⁸⁴. Chronic low-grade inflammation, insulin resistance, metabolic disturbances, and redistribution of adipose tissue caused by the above pathophysiological changes are considered core processes of frailty^85,86. As a butyrate-producing bacterium, Christensenellaceae R-7 is associated with lower insulin resistance⁸⁷, reduced visceral adipose tissue accumulation, immune regulation, and improved metabolic health⁸⁸. This appears to be the underlying mechanism of its suggestive causal effect on frailty.

Finally, a higher abundance of Defluviitaleaceae UCG-011 may be associated with a lower risk of frailty. A previous study found that, after dietary intervention in mice with cognitive impairment induced by a high-fat diet, the abundance of Defluviitaleaceae UCG-011 increased⁸⁹. After herbal interventions in mice with depression and cognitive decline induced by chronic mild stress, the abundance of Defluviitaleaceae UCG-011 also increased⁹⁰. This implies that Defluviitaleaceae UCG-011 may participate in the protection of cognitive function as a component of the gut-brain axis.

However, this study has some limitations. First, the gut microbiome may be influenced by demographic factors, diet, or drugs et al. Most of them have heterogeneity, inter-individual variability, and low heritability (representing the variance explained by genetics), which decreases the statistical efficacy and robustness of the results. Second, although most individuals in the GWAS meta-analysis of the gut microbiome were of European descent, there was still the possibility of interference by a small number of participants from other races, which may cause minimal bias and affect the universality of the results. Moreover, to obtain more comprehensive results and conduct horizontal pleiotropy detection and sensitivity analysis, selected genetic IVs did not reach the traditional GWAS significance threshold (p < 5 × 10⁻⁸), which may increase the possibility of false positives.

Conclusion

In summary, this study provides evidence for a suggestive genetic correlation between the genetically predicted Christensenellaceae R-7 and frailty. Furthermore, MR analysis indicated suggestive causal effects of genetically predicted 12 genus-level gut microbes on frailty.

Funding Statement

This study was supported by the National Natural Science Foundation of China (Nos.81803910 and 81973615), Capital’s Funds for Health, and the Qi-Huang Scholar Chief Scientist Program of National Administration of Traditional Chinese Medicine Leading Talents Support Program (2021).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Author contributions

Guanghui Cui, Shaojie Li, and Xuezhi Zhang contributed to study design. Guanghui Cui and Shaojie Li analyzed the data and drafted the manuscript. Yao Yang, Xiaofen Jia, Miaomiao Lin, Yue Feng, Zicheng Wang, Yingming Chu and Zongming Shi compiled the data and provided comments on the draft. Hui Ye and Xuezhi Zhang revised the manuscript. All authors have read and approved the final manuscript.

Data availability statement

The datasets analyzed in the current study can be downloaded from the website https://mibiogen.gcc.rug.nl/, https://www.ebi.ac.uk/gwas/downloads/summary-statistics.

Ethics approval

The datasets used in the current study were publicly available and ethical approval and informed consent were obtained prior to implementation. Therefore, our study did not require any additional informed consent or ethical approval.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19490976.2023.2282795.