https://orcid.org/0000-0002-0671-4392
Abstract
Gut dysbiosis has been linked to frailty, but its association with early mobility decline is unclear.
First, we determined the cross-sectional associations between walking speed and the gut microbiome in 740 older men (84 ± 4 years) from the MrOS cohort with available stool samples and 400 m walking speed measured in 2014–2016. Then, we analyzed the retrospective longitudinal associations between changes in 6 m walking speed (from 2005–2006 to 2014–2016, calculated by simple linear equation) and gut microbiome composition among participants with available data (702/740). We determined gut microbiome composition by 16S sequencing and examined diversity, taxa abundance, and performed network analysis to identify differences in the gut microbiome network of fast versus slow walkers.
Faster 400 m walking speed (m/s) was associated with greater microbiome α-diversity (R = 0.11; p = .004). The association between a slower decline in 6 m walking speed and higher α-diversity (R = 0.07; p = .054) approached borderline significance. Faster walking speed and less decline in walking speed were associated with a higher abundance of genus-level bacteria that produce short-chain fatty acids, and possess anti-inflammatory properties, including Paraprevotella, Fusicatenibacter, and Alistipes, after adjusting for potential covariates (p < .05). The gut microbiome networks of participants in the first versus last quartile of walking speed (≤0.9 vs ≥1.2 m/s) exhibited distinct characteristics, including different centrality measures (p < .05).
Our findings suggest a possible relationship between gut microbiome diversity and mobility function, as indicated by the associations between faster walking speed and less decline in walking speed over 10 years with higher gut microbiome diversity in older men.
Aging is associated with a decline in walking speed at a rate of approximately 1% per year between the ages of 65 and 69 which accelerates to approximately 4% per year in adults aged 80 years and older (1,2). The decline in walking speed and slowness (3,4) have been identified as a reliable predictor of several negative health outcomes in older adults, such as falls (5), disability, hospitalization (6), and mortality (7). Emerging research has proposed the existence of a complex bidirectional relationship between physical function and the gut microbiome (8,9). Studies have shown that gut dysbiosis, characterized by an imbalance in the gut microbiome composition, may contribute to mobility impairments (10,11). Conversely, enhancing physical function through exercise has been proposed to promote a healthy gut microbiome composition (12,13). However, much of our understanding of the relationship between gut dysbiosis and mobility function is derived from animal models of aging (14) or studies on athletes (15), and the precise nature of this relationship in older adults is not fully understood.
The gut microbiome composition undergoes changes with aging, including a reduction in the levels of beneficial bacteria (eg, Bifidobacterium (16) and Lactobacillus (17)), a reduction in short-chain fatty acid (SCFA) producing bacteria (18,19), and an increase in the abundance of Proteobacteria and Bacteroidetes (18). These changes along with age-related physiological alterations (eg, dentition (20), polypharmacy (21), reduced intestinal motility (22)), and lifestyle changes (eg, reduced physical activity (23)) contribute to age-related gut dysbiosis. Dysbiosis has been linked to several age-related diseases, including type 2 diabetes (24), hypertension (25), dementia (26), Alzheimer’s (27), and osteoporosis (28,29). Also, few cross-sectional observational studies (18,30–32) have shown that a dysbiotic shift in the gut microbiome composition, characterized by the overgrowth of Enterobacteriaceae and reductions in Lactobacillus (18,30) is associated with frailty, possibly through increased systemic inflammation (18). Yet, it remains to be determined if gut dysbiosis is associated with early stages of mobility decline in large-scale human studies. Moreover, there is a dearth of human studies investigating the longitudinal associations between changes in mobility function and gut microbiome composition. Examining this relationship offers valuable insights into the potential causal relationship between mobility impairment and gut microbiome and informs the development of targeted interventions (eg, dietary changes and probiotic supplementation) to preserve mobility function and improve overall health in older adults.
Here, our primary aim was to determine the cross-sectional associations between walking speed and gut microbiome composition among 740 older men (≥80 years) from the Osteoporotic Fractures in Men (MrOS) Study in 2014–2016. Our secondary aim was to determine retrospective longitudinal associations between changes in walking speed (from the 2005–2006 visit to the 2014–2016 visit) and gut microbiome composition among participants with available data (702/740). We hypothesized a positive association between walking speed and changes in walking speed with gut microbiome diversity, as well as potentially different abundances of certain types of bacteria in the gut.
Method
Study Population
The MrOS study is a multicenter longitudinal cohort of 5 994 community-dwelling older men (https://mrosonline.ucsf.edu) (33,34). The study was originally designed to determine risk factors of incident fractures in older men (≥65 years). Eligible participants had no history of bilateral hip replacement, were able to walk without the assistance of another person, and were enrolled from 6 clinical centers in the United States: Birmingham, AL; Minneapolis, MN; Monongahela Valley near Pittsburgh, PA; Palo Alto, CA; Portland, OR; and San Diego, CA. The study was approved by the Institutional Review Board at each clinical center and all participants provided written informed consent.
All the surviving participants were contacted in 2014–2016 (or Visit 4) and asked to provide stool samples for the Microbiome Ancillary study. Out of 1 328 participants who were invited, 982 men agreed to provide a stool sample (Supplementary Figure 1). Participants who refused to provide stool samples were similar to those who agreed in terms of race, education, marital status, smoking habits, and the prevalence of diabetes, hypertension, and cancer (Supplementary Table 1). However, they were slightly older and had poorer cognitive function and a marginally lower walking speed. In terms of dietary intake (total energy, carbohydrate, fat, and fiber consumption), they were comparable to those who provided samples, except for a slightly lower proportion of calorie intake from protein (15.6 ± 3% per day) compared to those who provided samples (16.3 ± 2.8% per day), as shown in Supplementary Table 1. To determine the cross-sectional associations between walking speed and gut microbiome, we included participants who had available stool samples and 400 m walking speed assessed at the 2014–2016 visit (n = 777). We further excluded participants with walking speed <0.3 m/s (n = 1) and those taking oral antibiotics over the past 2 weeks (n = 36). Our final analytical sample was N = 740 participants (Supplementary Figure 1). We also analyzed the retrospective longitudinal associations between changes in 6 m walking speed (between the 2005–2006 visit and the 2014–2016 visit) and gut microbiome composition among participants with available data (702/740) at both time points.
Assessment of 400 m Walking Speed (m/s) in 2014–2016
During the 2014–2016 visit, the participant’s usual walking pace without overexertion was recorded over a 400-meter distance, which consisted of 10 laps on a 20-meter course (35). The walking speed was measured in meters per second (m/s). During testing, participants were allowed to take a break of up to 60 seconds at any point, provided they did not lean on any surfaces or sit down. The use of assistive devices was not permitted, except for a single straight cane. Testing was stopped for distress (eg, labored breathing, confusion, and unresponsiveness), pain, resting >60 seconds, leaning on a surface twice during rest, requesting an assistive device other than a single straight cane, or requesting to stop.
Changes in 6 m Walking Speed (m/s) Between the 2005–2006 Visit and the 2014–2016 Follow-Up Visit
The 6 m walking test was conducted according to a standardized protocol (33). Participants were asked to walk 6 m along a straight obstacle-free hallway at their usual pace using ambulatory aids as needed. The test began when the participant’s foot crossed a line on the floor at the beginning of the walking course and ended when the participant’s foot crossed a second line at the end of the course. The time taken to complete the walk was recorded in seconds. The test was performed twice, and the faster time was used for analysis. The 6 m walking speed was then calculated in meters per second. This test was performed during both the 2005–2006 and 2014–2016 visits. The change in walking speed was calculated by subtracting the walking speed at the 2014–2016 visit from the walking speed at the 2005–2006 visit using a simple linear equation.
Stool Sample Collection and Gut Microbiome Profiling
At the 2014–2016 visit, participants used the OMNIgene•GUT stool collection DNA kit (OMR-200, DNA Genotek) to collect stool samples at home. The OMNIgene•GUT stool collection kit, is designed to stabilize the DNA in the sample immediately upon collection, ensuring its stability for up to 60 days at room temperature. Participants were asked to record the date of sample collection in a short questionnaire that was included in the collection kit. On average, there was a time span of 3.3 ± 7.8 days between collecting stool samples and participants’ clinic visits. Once collected, participants mailed samples to the MrOS Administrative Center at Oregon Health & Science University (Portland, OR) for storage. As soon as the samples were received, the study staff promptly opened them to verify the adequacy of the stool samples and to confirm that the paperwork, including the date of collection, was filled out properly (36). Following this, the samples were stored in a freezer at −80° C. In cases where a stool sample had not been received at the central lab within 2 weeks of the expected date, the corresponding study site of the participant was informed. Subsequently, a follow-up phone call was made to the participant to ensure that a sample had been collected (36). Stool samples were then shipped to the Baylor College of Medicine (Houston, TX) for taxonomic profiling of the gut microbiome by 16S rRNA gene sequencing (37).
In brief, bacterial DNA was extracted using the MO BIO PowerSoil DNA Isolation Kit (MO BIO Laboratories, Inc., Carlsbad, CA) and subjected to 16S (v4 region) rRNA amplification using PCR. The resulting sequences were analyzed on the Illumina MiSeq platform and bioinformatically processed using the DADA2 (version 1.7.0). After filtering and trimming the sequences, amplicon sequence variants were assigned taxonomic identities using the SILVA reference database (version 128). Finally, a phylogeny was constructed using the msa package and FastTree software (version 2.1.9). The gut microbiome processing in MrOS was carried out in 2 separate batches, with all steps of sample processing, including collection, storage, 16S amplification, and sequencing, being identical between the 2 batches. Although the characteristics of study participants were comparable between the 2 batches, we observed slight yet statistically significant differences in microbiome diversity measures with Batch 2 showing a 2.8% higher Shannon Index compared to Batch 1 (37). As a result, we adjusted our analysis for the batch number.
Covariates
Sociodemographiccharacteristics, including age, race/ethnicity, education, marital status, and smoking status were obtained by self-administered questionnaires (33).
Anthropometric measures and body composition (33) were assessed by measuring weight and height using a balance beam or digital scales and Harpenden stadiometers (Dyved UK), respectively, to calculate body mass index. Total lean mass and fat mass were assessed by dual-energy X-ray absorptiometry (Hologic, Inc., Waltham, MA). Upon availability, we also determined retrospective changes in body composition measures between the 2005–2006 and 2014–2016 visits.
Physicalactivityandstrength
The self-administered Physical Activity Scale for the Elderly (PASE) questionnaire (score 0–793) was used to establish the level of physical activity. Using a handheld Jamar dynamometer (Sammons Preston Rolyan), grip strength (in kilograms) was assessed by conducting 2 trials on each hand, and the maximum grip strength was recorded for analysis (38).
Dietaryintake
Total daily energy and nutrient intake (proportion of calorie intake from dietary protein, carbohydrate, and fat) and dietary fiber intake was estimated by the Block 98.2 MrOS brief Food Frequency questionnaire at the 2014–2016 visit (39).
Healthand mental assessment
The prevalence of comorbidities, including diabetes, hypertension, and cancer, was determined based on self-reported information, physician diagnosis, and/or medication use. Participants also self-rated their health status as excellent, good, fair, poor, and very poor. The total number of medications was determined by asking participants to bring in all medications they had taken in the past 30 days during the clinic visit. The Teng Modified Mini-Mental State Examination (3MS; 0–100) (40) and Geriatric Depression Scale (0–15) (41) were used to evaluate cognitive function and depressive symptoms, respectively.
Statistical Analysis
Participant characteristics are presented as either means ± standard deviations (SDs) or as numbers with percentages. We used Chi-square test for categorical variables, ANOVA for normally distributed, and Kruskal–Wallis for non-normally distributed variables to examine differences in participants’ characteristics across quartiles of 400 m walking speed (m/s) that was measured at the 2014–2016 visit. We performed the following statistical analyses to determine the cross-sectional associations between 400 m walking speed and gut microbiome composition as well as the retrospective longitudinal associations between changes in 6 m walking speed and gut microbiome composition:
First, we determined α-diversity measures (ie, Shannon and Inverse Simpson) to examine the correlations between 400 m walking speed (m/s) and the changes in 6 m walking speed (m/s), as continuous variables, with gut microbiome α-diversity. We also compared differences in microbiome α-diversity across quartiles of 400 m walking as well as quartiles of changes in 6 m walking speed using the Kruskal–Wallis test. Significant P values were adjusted by the Bonferroni correction for multiple comparisons.
Second, we performed β-diversity analysis to elucidate similarity (or dissimilarity) of microbiome composition across quartiles of 400 m walking speed using weighted and unweighted UniFrac Principal Co-ordinates Analysis (PCoA). To statistically test differences in β-diversity, we performed Permutation Multivariate ANOVA (PERMANOVA) analysis, adjusted for age, race, clinical center, education, marital status, health status, weight, height, physical activity, batch number, number of medications, total energy intake, and dietary fiber.
We repeated the β-diversity analysis to investigate similarities in microbiome composition across quartiles of changes in 6 m walking speed measured between the 2005–2006 visit and the 2014–2016 follow-up visit. To control for potential confounding factors, we adjusted our PERMANOVA analysis for all the covariates mentioned above as well as for the 400 m walking speed measured at the 2014–2016 visit, which measures endurance and physical fitness rather than the burst of mobility measured in the 6 m walking test. This allowed us to better isolate the effect of changes in 6 m walking speed on gut microbiome composition. We utilized the betadisper function in the Vegan R package to test the assumption of homogeneity of dispersion required for PERMANOVA. The results showed that for both unweighted and weighted UniFrac PCoA analyses of 400 m walking speed and changes in 6 m walking speed, the assumption of homogeneous dispersion was met, as indicated by obtaining nonsignificant p values. Both α- and β-diversity measures were determined after rarefying data using the minimum number of reads per sample (5 429 reads/sample). Of note, to avoid any potential influence of our rarefication on the evaluation of β-diversity, we utilized a normalization approach known as total sum scaling (TSS) to our microbiome data (42).
Third, we performed univariate differential abundance tests at the genus level to determine taxa that were differentially abundant according to the 400 m walking speed and changes in 6 m walking speed (as continuous variables) using Analysis of Compositions of Microbiomes with Bias Correction (ANCOM-BC) (43). Unlike the existing methods for differential abundance analysis, ANCOM-BC provides a statistically valid test with appropriate P values and standard errors (SEs) for a differential abundance of each taxon while controlling for the false discovery rate (ie, Hommel adjusted for multiple comparisons) and potential confounding variables (ie, age, race, clinical center, education, marital status, health status, weight, height, physical activity, batch number, number of medications, total energy intake, and dietary fiber).
Finally, we conducted a microbial association network analysis at the genus level using the NetCoMi R package (44) to uncover microbial community interactions based on 400 m walking speed levels. Specifically, we compared bacterial associations between slow walkers (ie, the first quartile [Q1] of 400 m walking speed) and fast walkers (ie, Q4), using the SPRING method as the association measure. The estimated partial correlations were transformed into dissimilarities using the “signed” distance metric, and the corresponding similarities were used as edge weights. We used eigenvector centrality to define hubs and scale node sizes. Additionally, we used the Jaccard index to quantitatively assess the differences between the sets of most central nodes in the networks of fast and slow walkers. To account for multiple test, we applied the Benjamini–Hochberg method for adjustment. Analyses were performed in RStudio statistical system version 2023.03.0 + 386 and IBM SPSS Statistics version 28.0 for Windows (SPSS Institute, Cary, NC).
Results
Out of 740 older men included in our study, the mean age was 83.9 ± 3.8 years, and most participants were White (88%). Table 1 shows participants’ characteristics across quartiles of 400 m walking speed at the 2014–2016 visit. Demographic characteristics varied by quartile with a greater proportion of participants in the higher quartiles of 400 m walking speed being married and reporting better health status compared to those in the lower quartiles. Individuals in the highest quartile (Q4, with a walking speed of ≥1.16 m/s) were predominantly from Portland and San Diego. Body composition also differed across quartiles, with participants in the higher quartiles having lower body weight attributed to lower fat mass and being taller, but their lean mass was similar to that of participants in the lower quartiles. Although changes in body composition were similar across quartiles between the 2005–2006 visit and the 2014–2016 visit, individuals in the higher quartiles experienced less lean mass loss than those in the lower quartiles (p = .018), resulting in a marginal decline in total body weight among faster walkers compared to slower walkers (p = .05). In terms of physical function, participants in the higher quartiles of 400 m walking speed were physically more active and had higher grip strength than those in the lower quartiles. Moreover, they experienced less decline in their 6 m walking speed and grip strength between the 2005–2006 visit and the 2014–2016 visit (Table 1).
Participant Characteristics by Quartiles of 400 m Walking Speed (m/s) at 2014–2016 Visit in the MrOS Study*
| Walking speed quartiles at Visit 4 | |||||
|---|---|---|---|---|---|
| Q1 | Q2 | Q3 | Q4 | p Values | |
| ≤0.899 m/s | 0.900–1.032 m/s | 1.033–1.158 m/s | ≥1.159 m/s | ||
| (n = 185) | (n = 185) | (n = 185) | (n = 185) | ||
| White, n (%) | 167 (90) | 162 (88) | 166 (90) | 158 (85) | .571† |
| ≤High school, n (%) | 43 (23) | 35 (19) | 32 (17) | 18 (10) | .063† |
| Clinic center, n (%) | <.001† | ||||
| Birmingham | 32 (17) | 25 (14) | 24 (13) | 17 (9) | |
| Minneapolis | 24 (13) | 36 (20) | 36 (20) | 28 (15) | |
| Palo Alto | 23 (12) | 32 (17) | 28 (15) | 25 (14) | |
| Pittsburgh | 39 (21) | 26 (14) | 19 (10) | 9 (5) | |
| Portland | 28 (15) | 21 (11) | 35 (19) | 52 (28) | |
| San Diego | 39 (21) | 45 (24) | 43 (23) | 54 (29) | |
| Married, n (%) | 109 (59) | 137 (74) | 137 (74) | 143 (77) | .002† |
| Never smoker, n (%) | 68 (37) | 67 (36) | 78 (42) | 92 (50) | .090† |
| Excellent health, n (%) | 35 (19) | 64 (35) | 79 (43) | 105 (57) | <.001† |
| Diabetes, n (%) | 26 (14) | 32 (17) | 31 (17) | 12 (7) | .009† |
| Hypertension, n (%) | 106 (57) | 94 (51) | 94 (51) | 86 (47) | .220† |
| Cancer, n (%) | 92 (50) | 87 (47) | 83 (45) | 92 (50) | .745† |
| Medication, n | 9.1 ± 3.9 | 9.0 ± 4.6 | 8.5 ± 4.3 | 8.4 ± 4.8 | .102 |
| Age, y | 85.5 ± 4.3 | 84.4 ± 3.5 | 83.3 ± 3.5 | 82.4 ± 2.9 | <.001 |
| Teng 3MS, 0–100 | 92.0 ± 6.6 | 91.9 ± 7.5 | 93.4 ± 5.7 | 94.2 ± 5.8 | <.001 |
| Depression score, 0–15 | 2.2 ± 2.0 | 1.6 ± 1.8 | 1.4 ± 1.4 | 1.0 ± 1.4 | <.001 |
| Height, cm | 172.0 ± 6.8 | 171.6 ± 6.8 | 172.1 ± 6.9 | 173.6 ± 6 | .012 |
| Weight, kg | 81.7 ± 13.6 | 78.6 ± 12.7 | 79.3 ± 12.1 | 77.5 ± 10.2 | .039 |
| Δ Weight, % | −4.2 ± 7.7 | −4.5 ± 7.1 | −3.6 ± 6.4 | −2.7 ± 5.9 | .050 |
| BMI, kg/m2 | 27.6 ± 4.1 | 26.7 ± 3.7 | 26.7 ± 3.5 | 25.7 ± 2.9 | <.001 |
| Δ BMI, kg/m2 | −0.42 ± 2.19 | −0.61 ± 1.91 | −0.48 ± 1.79 | −0.28 ± 1.57 | .353 |
| Fat mass, % | 29.1 ± 5.7 | 27.9 ± 6.2 | 27.1 ± 5.2 | 26.1 ± 5.6 | <.001 |
| Δ Fat mass, % | 6.8 ± 16.1 | 7.0 ± 15.6 | 8.5 ± 14.9 | 8.7 ± 14.2 | .633 |
| Lean mass, kg | 53.6 ± 7.0 | 52.6 ± 6.6 | 53.6 ± 6.5 | 53.3 ± 5.8 | .535 |
| Δ Lean mass, % | −6.6 ± 5.4 | −6.9 ± 4.8 | −6.3 ± 4.9 | −5.3 ± 4.4 | .018‡ |
| 400-m walking speed, m/s | 0.77 ± 0.11 | 0.97 ± 0.04 | 1.1 ± 0.04 | 1.27 ± 0.09 | <.001 |
| Δ walking speed, m/s | −0.22 ± 0.19 | −0.16 ± 0.18 | −0.09 ± 0.17 | −0.05 ± 0.22 | <.001 |
| Grip strength, kg | 30.0 ± 6.7 | 33.5 ± 7.0 | 34.3 ± 7.4 | 36.4 ± 6.9 | <.001 |
| Δ Grip strength, % | −19.7 ± 12.9 | −16.5 ± 14.0 | −15.9 ± 10.3 | −14.0 ± 10.5 | <.001 |
| PASE score, 0–793 | 108 ± 65 | 129 ± 60 | 134 ± 67 | 144 ± 63 | <.001 |
| Energy intake, kcal/d | 1 664 ± 780 | 1 502 ± 629 | 1 473 ± 597 | 1 439 ± 566 | .079 |
| Fat, % of kcal/d | 40.5 ± 6.6 | 41.1 ± 6.8 | 41.2 ± 6.8 | 40.2 ± 7.6 | .425‡ |
| Protein, % of kcal/d | 16.1 ± 2.8 | 16.3 ± 2.7 | 16.3 ± 3.0 | 16.5 ± 2.8 | .517 |
| Carbohydrate, % of kcal/d | 46.0 ± 7.0 | 45.4 ± 7.2 | 45.3 ± 7.3 | 46.5 ± 8.2 | .401 |
| Fiber, g/d | 16.3 ± 8.9 | 16.0 ± 7.9 | 15.6 ± 7.3 | 16.6 ± 8.3 | .790 |
| Walking speed quartiles at Visit 4 | |||||
|---|---|---|---|---|---|
| Q1 | Q2 | Q3 | Q4 | p Values | |
| ≤0.899 m/s | 0.900–1.032 m/s | 1.033–1.158 m/s | ≥1.159 m/s | ||
| (n = 185) | (n = 185) | (n = 185) | (n = 185) | ||
| White, n (%) | 167 (90) | 162 (88) | 166 (90) | 158 (85) | .571† |
| ≤High school, n (%) | 43 (23) | 35 (19) | 32 (17) | 18 (10) | .063† |
| Clinic center, n (%) | <.001† | ||||
| Birmingham | 32 (17) | 25 (14) | 24 (13) | 17 (9) | |
| Minneapolis | 24 (13) | 36 (20) | 36 (20) | 28 (15) | |
| Palo Alto | 23 (12) | 32 (17) | 28 (15) | 25 (14) | |
| Pittsburgh | 39 (21) | 26 (14) | 19 (10) | 9 (5) | |
| Portland | 28 (15) | 21 (11) | 35 (19) | 52 (28) | |
| San Diego | 39 (21) | 45 (24) | 43 (23) | 54 (29) | |
| Married, n (%) | 109 (59) | 137 (74) | 137 (74) | 143 (77) | .002† |
| Never smoker, n (%) | 68 (37) | 67 (36) | 78 (42) | 92 (50) | .090† |
| Excellent health, n (%) | 35 (19) | 64 (35) | 79 (43) | 105 (57) | <.001† |
| Diabetes, n (%) | 26 (14) | 32 (17) | 31 (17) | 12 (7) | .009† |
| Hypertension, n (%) | 106 (57) | 94 (51) | 94 (51) | 86 (47) | .220† |
| Cancer, n (%) | 92 (50) | 87 (47) | 83 (45) | 92 (50) | .745† |
| Medication, n | 9.1 ± 3.9 | 9.0 ± 4.6 | 8.5 ± 4.3 | 8.4 ± 4.8 | .102 |
| Age, y | 85.5 ± 4.3 | 84.4 ± 3.5 | 83.3 ± 3.5 | 82.4 ± 2.9 | <.001 |
| Teng 3MS, 0–100 | 92.0 ± 6.6 | 91.9 ± 7.5 | 93.4 ± 5.7 | 94.2 ± 5.8 | <.001 |
| Depression score, 0–15 | 2.2 ± 2.0 | 1.6 ± 1.8 | 1.4 ± 1.4 | 1.0 ± 1.4 | <.001 |
| Height, cm | 172.0 ± 6.8 | 171.6 ± 6.8 | 172.1 ± 6.9 | 173.6 ± 6 | .012 |
| Weight, kg | 81.7 ± 13.6 | 78.6 ± 12.7 | 79.3 ± 12.1 | 77.5 ± 10.2 | .039 |
| Δ Weight, % | −4.2 ± 7.7 | −4.5 ± 7.1 | −3.6 ± 6.4 | −2.7 ± 5.9 | .050 |
| BMI, kg/m2 | 27.6 ± 4.1 | 26.7 ± 3.7 | 26.7 ± 3.5 | 25.7 ± 2.9 | <.001 |
| Δ BMI, kg/m2 | −0.42 ± 2.19 | −0.61 ± 1.91 | −0.48 ± 1.79 | −0.28 ± 1.57 | .353 |
| Fat mass, % | 29.1 ± 5.7 | 27.9 ± 6.2 | 27.1 ± 5.2 | 26.1 ± 5.6 | <.001 |
| Δ Fat mass, % | 6.8 ± 16.1 | 7.0 ± 15.6 | 8.5 ± 14.9 | 8.7 ± 14.2 | .633 |
| Lean mass, kg | 53.6 ± 7.0 | 52.6 ± 6.6 | 53.6 ± 6.5 | 53.3 ± 5.8 | .535 |
| Δ Lean mass, % | −6.6 ± 5.4 | −6.9 ± 4.8 | −6.3 ± 4.9 | −5.3 ± 4.4 | .018‡ |
| 400-m walking speed, m/s | 0.77 ± 0.11 | 0.97 ± 0.04 | 1.1 ± 0.04 | 1.27 ± 0.09 | <.001 |
| Δ walking speed, m/s | −0.22 ± 0.19 | −0.16 ± 0.18 | −0.09 ± 0.17 | −0.05 ± 0.22 | <.001 |
| Grip strength, kg | 30.0 ± 6.7 | 33.5 ± 7.0 | 34.3 ± 7.4 | 36.4 ± 6.9 | <.001 |
| Δ Grip strength, % | −19.7 ± 12.9 | −16.5 ± 14.0 | −15.9 ± 10.3 | −14.0 ± 10.5 | <.001 |
| PASE score, 0–793 | 108 ± 65 | 129 ± 60 | 134 ± 67 | 144 ± 63 | <.001 |
| Energy intake, kcal/d | 1 664 ± 780 | 1 502 ± 629 | 1 473 ± 597 | 1 439 ± 566 | .079 |
| Fat, % of kcal/d | 40.5 ± 6.6 | 41.1 ± 6.8 | 41.2 ± 6.8 | 40.2 ± 7.6 | .425‡ |
| Protein, % of kcal/d | 16.1 ± 2.8 | 16.3 ± 2.7 | 16.3 ± 3.0 | 16.5 ± 2.8 | .517 |
| Carbohydrate, % of kcal/d | 46.0 ± 7.0 | 45.4 ± 7.2 | 45.3 ± 7.3 | 46.5 ± 8.2 | .401 |
| Fiber, g/d | 16.3 ± 8.9 | 16.0 ± 7.9 | 15.6 ± 7.3 | 16.6 ± 8.3 | .790 |
Notes: BMI = body mass index; MrOS = Osteoporotic Fractures in Men Study; PASE = Physical Activity Score for Elderly; Q = Quartile; Δ = Changes from 2005–2006 visit to 2014–2016 visit; Teng 3MS = The Teng Modified Mini-Mental State Examination.
P values were derived from the Kruskal–Wallis test unless otherwise indicated.
*Values are shown as means ± SD or n (%).
†From the Chi-square test.
‡From ANOVA.
Participant Characteristics by Quartiles of 400 m Walking Speed (m/s) at 2014–2016 Visit in the MrOS Study*
| Walking speed quartiles at Visit 4 | |||||
|---|---|---|---|---|---|
| Q1 | Q2 | Q3 | Q4 | p Values | |
| ≤0.899 m/s | 0.900–1.032 m/s | 1.033–1.158 m/s | ≥1.159 m/s | ||
| (n = 185) | (n = 185) | (n = 185) | (n = 185) | ||
| White, n (%) | 167 (90) | 162 (88) | 166 (90) | 158 (85) | .571† |
| ≤High school, n (%) | 43 (23) | 35 (19) | 32 (17) | 18 (10) | .063† |
| Clinic center, n (%) | <.001† | ||||
| Birmingham | 32 (17) | 25 (14) | 24 (13) | 17 (9) | |
| Minneapolis | 24 (13) | 36 (20) | 36 (20) | 28 (15) | |
| Palo Alto | 23 (12) | 32 (17) | 28 (15) | 25 (14) | |
| Pittsburgh | 39 (21) | 26 (14) | 19 (10) | 9 (5) | |
| Portland | 28 (15) | 21 (11) | 35 (19) | 52 (28) | |
| San Diego | 39 (21) | 45 (24) | 43 (23) | 54 (29) | |
| Married, n (%) | 109 (59) | 137 (74) | 137 (74) | 143 (77) | .002† |
| Never smoker, n (%) | 68 (37) | 67 (36) | 78 (42) | 92 (50) | .090† |
| Excellent health, n (%) | 35 (19) | 64 (35) | 79 (43) | 105 (57) | <.001† |
| Diabetes, n (%) | 26 (14) | 32 (17) | 31 (17) | 12 (7) | .009† |
| Hypertension, n (%) | 106 (57) | 94 (51) | 94 (51) | 86 (47) | .220† |
| Cancer, n (%) | 92 (50) | 87 (47) | 83 (45) | 92 (50) | .745† |
| Medication, n | 9.1 ± 3.9 | 9.0 ± 4.6 | 8.5 ± 4.3 | 8.4 ± 4.8 | .102 |
| Age, y | 85.5 ± 4.3 | 84.4 ± 3.5 | 83.3 ± 3.5 | 82.4 ± 2.9 | <.001 |
| Teng 3MS, 0–100 | 92.0 ± 6.6 | 91.9 ± 7.5 | 93.4 ± 5.7 | 94.2 ± 5.8 | <.001 |
| Depression score, 0–15 | 2.2 ± 2.0 | 1.6 ± 1.8 | 1.4 ± 1.4 | 1.0 ± 1.4 | <.001 |
| Height, cm | 172.0 ± 6.8 | 171.6 ± 6.8 | 172.1 ± 6.9 | 173.6 ± 6 | .012 |
| Weight, kg | 81.7 ± 13.6 | 78.6 ± 12.7 | 79.3 ± 12.1 | 77.5 ± 10.2 | .039 |
| Δ Weight, % | −4.2 ± 7.7 | −4.5 ± 7.1 | −3.6 ± 6.4 | −2.7 ± 5.9 | .050 |
| BMI, kg/m2 | 27.6 ± 4.1 | 26.7 ± 3.7 | 26.7 ± 3.5 | 25.7 ± 2.9 | <.001 |
| Δ BMI, kg/m2 | −0.42 ± 2.19 | −0.61 ± 1.91 | −0.48 ± 1.79 | −0.28 ± 1.57 | .353 |
| Fat mass, % | 29.1 ± 5.7 | 27.9 ± 6.2 | 27.1 ± 5.2 | 26.1 ± 5.6 | <.001 |
| Δ Fat mass, % | 6.8 ± 16.1 | 7.0 ± 15.6 | 8.5 ± 14.9 | 8.7 ± 14.2 | .633 |
| Lean mass, kg | 53.6 ± 7.0 | 52.6 ± 6.6 | 53.6 ± 6.5 | 53.3 ± 5.8 | .535 |
| Δ Lean mass, % | −6.6 ± 5.4 | −6.9 ± 4.8 | −6.3 ± 4.9 | −5.3 ± 4.4 | .018‡ |
| 400-m walking speed, m/s | 0.77 ± 0.11 | 0.97 ± 0.04 | 1.1 ± 0.04 | 1.27 ± 0.09 | <.001 |
| Δ walking speed, m/s | −0.22 ± 0.19 | −0.16 ± 0.18 | −0.09 ± 0.17 | −0.05 ± 0.22 | <.001 |
| Grip strength, kg | 30.0 ± 6.7 | 33.5 ± 7.0 | 34.3 ± 7.4 | 36.4 ± 6.9 | <.001 |
| Δ Grip strength, % | −19.7 ± 12.9 | −16.5 ± 14.0 | −15.9 ± 10.3 | −14.0 ± 10.5 | <.001 |
| PASE score, 0–793 | 108 ± 65 | 129 ± 60 | 134 ± 67 | 144 ± 63 | <.001 |
| Energy intake, kcal/d | 1 664 ± 780 | 1 502 ± 629 | 1 473 ± 597 | 1 439 ± 566 | .079 |
| Fat, % of kcal/d | 40.5 ± 6.6 | 41.1 ± 6.8 | 41.2 ± 6.8 | 40.2 ± 7.6 | .425‡ |
| Protein, % of kcal/d | 16.1 ± 2.8 | 16.3 ± 2.7 | 16.3 ± 3.0 | 16.5 ± 2.8 | .517 |
| Carbohydrate, % of kcal/d | 46.0 ± 7.0 | 45.4 ± 7.2 | 45.3 ± 7.3 | 46.5 ± 8.2 | .401 |
| Fiber, g/d | 16.3 ± 8.9 | 16.0 ± 7.9 | 15.6 ± 7.3 | 16.6 ± 8.3 | .790 |
| Walking speed quartiles at Visit 4 | |||||
|---|---|---|---|---|---|
| Q1 | Q2 | Q3 | Q4 | p Values | |
| ≤0.899 m/s | 0.900–1.032 m/s | 1.033–1.158 m/s | ≥1.159 m/s | ||
| (n = 185) | (n = 185) | (n = 185) | (n = 185) | ||
| White, n (%) | 167 (90) | 162 (88) | 166 (90) | 158 (85) | .571† |
| ≤High school, n (%) | 43 (23) | 35 (19) | 32 (17) | 18 (10) | .063† |
| Clinic center, n (%) | <.001† | ||||
| Birmingham | 32 (17) | 25 (14) | 24 (13) | 17 (9) | |
| Minneapolis | 24 (13) | 36 (20) | 36 (20) | 28 (15) | |
| Palo Alto | 23 (12) | 32 (17) | 28 (15) | 25 (14) | |
| Pittsburgh | 39 (21) | 26 (14) | 19 (10) | 9 (5) | |
| Portland | 28 (15) | 21 (11) | 35 (19) | 52 (28) | |
| San Diego | 39 (21) | 45 (24) | 43 (23) | 54 (29) | |
| Married, n (%) | 109 (59) | 137 (74) | 137 (74) | 143 (77) | .002† |
| Never smoker, n (%) | 68 (37) | 67 (36) | 78 (42) | 92 (50) | .090† |
| Excellent health, n (%) | 35 (19) | 64 (35) | 79 (43) | 105 (57) | <.001† |
| Diabetes, n (%) | 26 (14) | 32 (17) | 31 (17) | 12 (7) | .009† |
| Hypertension, n (%) | 106 (57) | 94 (51) | 94 (51) | 86 (47) | .220† |
| Cancer, n (%) | 92 (50) | 87 (47) | 83 (45) | 92 (50) | .745† |
| Medication, n | 9.1 ± 3.9 | 9.0 ± 4.6 | 8.5 ± 4.3 | 8.4 ± 4.8 | .102 |
| Age, y | 85.5 ± 4.3 | 84.4 ± 3.5 | 83.3 ± 3.5 | 82.4 ± 2.9 | <.001 |
| Teng 3MS, 0–100 | 92.0 ± 6.6 | 91.9 ± 7.5 | 93.4 ± 5.7 | 94.2 ± 5.8 | <.001 |
| Depression score, 0–15 | 2.2 ± 2.0 | 1.6 ± 1.8 | 1.4 ± 1.4 | 1.0 ± 1.4 | <.001 |
| Height, cm | 172.0 ± 6.8 | 171.6 ± 6.8 | 172.1 ± 6.9 | 173.6 ± 6 | .012 |
| Weight, kg | 81.7 ± 13.6 | 78.6 ± 12.7 | 79.3 ± 12.1 | 77.5 ± 10.2 | .039 |
| Δ Weight, % | −4.2 ± 7.7 | −4.5 ± 7.1 | −3.6 ± 6.4 | −2.7 ± 5.9 | .050 |
| BMI, kg/m2 | 27.6 ± 4.1 | 26.7 ± 3.7 | 26.7 ± 3.5 | 25.7 ± 2.9 | <.001 |
| Δ BMI, kg/m2 | −0.42 ± 2.19 | −0.61 ± 1.91 | −0.48 ± 1.79 | −0.28 ± 1.57 | .353 |
| Fat mass, % | 29.1 ± 5.7 | 27.9 ± 6.2 | 27.1 ± 5.2 | 26.1 ± 5.6 | <.001 |
| Δ Fat mass, % | 6.8 ± 16.1 | 7.0 ± 15.6 | 8.5 ± 14.9 | 8.7 ± 14.2 | .633 |
| Lean mass, kg | 53.6 ± 7.0 | 52.6 ± 6.6 | 53.6 ± 6.5 | 53.3 ± 5.8 | .535 |
| Δ Lean mass, % | −6.6 ± 5.4 | −6.9 ± 4.8 | −6.3 ± 4.9 | −5.3 ± 4.4 | .018‡ |
| 400-m walking speed, m/s | 0.77 ± 0.11 | 0.97 ± 0.04 | 1.1 ± 0.04 | 1.27 ± 0.09 | <.001 |
| Δ walking speed, m/s | −0.22 ± 0.19 | −0.16 ± 0.18 | −0.09 ± 0.17 | −0.05 ± 0.22 | <.001 |
| Grip strength, kg | 30.0 ± 6.7 | 33.5 ± 7.0 | 34.3 ± 7.4 | 36.4 ± 6.9 | <.001 |
| Δ Grip strength, % | −19.7 ± 12.9 | −16.5 ± 14.0 | −15.9 ± 10.3 | −14.0 ± 10.5 | <.001 |
| PASE score, 0–793 | 108 ± 65 | 129 ± 60 | 134 ± 67 | 144 ± 63 | <.001 |
| Energy intake, kcal/d | 1 664 ± 780 | 1 502 ± 629 | 1 473 ± 597 | 1 439 ± 566 | .079 |
| Fat, % of kcal/d | 40.5 ± 6.6 | 41.1 ± 6.8 | 41.2 ± 6.8 | 40.2 ± 7.6 | .425‡ |
| Protein, % of kcal/d | 16.1 ± 2.8 | 16.3 ± 2.7 | 16.3 ± 3.0 | 16.5 ± 2.8 | .517 |
| Carbohydrate, % of kcal/d | 46.0 ± 7.0 | 45.4 ± 7.2 | 45.3 ± 7.3 | 46.5 ± 8.2 | .401 |
| Fiber, g/d | 16.3 ± 8.9 | 16.0 ± 7.9 | 15.6 ± 7.3 | 16.6 ± 8.3 | .790 |
Notes: BMI = body mass index; MrOS = Osteoporotic Fractures in Men Study; PASE = Physical Activity Score for Elderly; Q = Quartile; Δ = Changes from 2005–2006 visit to 2014–2016 visit; Teng 3MS = The Teng Modified Mini-Mental State Examination.
P values were derived from the Kruskal–Wallis test unless otherwise indicated.
*Values are shown as means ± SD or n (%).
†From the Chi-square test.
‡From ANOVA.
Associations Between α-Diversity Measures and Walking Speed
In our cross-sectional analysis, we observed a modest positive correlation between 400 m walking speed and both Shannon (R = 0.105, p = .004) and Inverse Simpson (R = 0.108, p = .003) α-diversity measures (Figure 1, Panel A). To further investigate this relationship, participants were categorized into quartiles based on their 400 m walking speed. We tested for trends across quartiles and found a significant linear trend for both Shannon (p = .004) and Inverse Simpson (p < .001) α-diversity measures. Moreover, we observed a significant difference in α-diversity measures between those in the 4th quartile (with ≥1.16 m/s walking speed) versus those in the 1st (<0.90 m/s) and 2nd (0.9–1.03 m/s) quartiles (p < .05) (Figure 1, Panel B). Also, there were significant differences in Inverse Simpson α-diversity between the 3rd quartile (1.03–1.15 m/s) and the 1st quartile of 400 m walking speed.
Spearman correlations between α-diversity indices (ie, Shannon and Inverse Simpson) and 400 m walking speed measured at Visit 4 (2014–2016; Panel A) and changes in 6 m walking speed (from the 2005–2006 visit to the 2014–2016 visit; Panel C). Panel B and Panel D show comparison of α-diversity indices across quartiles of 400 m walking speed (Q1 ≤ 0.899, 0.900 ≤ Q2 ≤ 1.032, 1.033 ≤ Q3 ≤ 1.158, and Q4 ≥ 1.159 m/s) and quartiles of changes in 6 m walking speed (Q1 ≤ −0.246, −0.247 ≤ Q2 ≤ −0.133, −0.134 ≤ Q3 ≤ −0.002, and Q4 ≥ −0.001 m/s), respectively, by Kruskal–Wallis test and Bonferroni adjusted P values for multiple comparisons. Only p < .1 are shown in the figure. P trend across quartiles of quartiles of 400 m walking speed and Shannon was p = .004; Inverse Simpson was p < .001. P trend across quartiles of changes in 6 m walking speed and Shannon was p = .091; Inverse Simpson was p = .046.
In our retrospective longitudinal analysis, we observed that participants who experienced a greater decline in 6 m walking speed between the 2005–2006 and 2014–2016 visits had lower Shannon (R = 0.072, p = .054) and Inverse Simpson (R = 0.067, p = .071) α-diversity compared to those who experienced a lesser decline in walking speed. However, these associations were of borderline statistical significance (Figure 1, Panel C). When we categorized participants according to quartiles of changes in 6 m walking speed, with Q1 indicating a higher decline in 6 m walking speed (ie, <−0.246 m/s) and Q4 indicating a lower decline (ie, ≥−0.001 m/s), we observed similar associations between changes in 6 m walking speed and α-diversity measures (Figure 1, Panel D). However, these associations also did not reach statistical significance levels. Also, we tested for trends across quartiles and found a trend toward significance for Shannon (p = .091) and a significant trend for Inverse Simpson (p = .046). We also observed that participants in the lower quartiles of 6 m walking speed (ie, those who experienced more decline) had slower 400 m walking speeds, lower physical activity levels, and lower grip strength at the 2014–2016 visit (Supplementary Table 2). They were slightly older, with mean ages of 84.3 in Q1 versus 83.1 in Q4, had lower cognitive function as indicated by Teng 3MS scores, and reported more symptoms of depression, poorer self-reported health, and experienced a significant decline in grip strength from the 2005–2006 visit to the 2014–2016 visit compared to those in the upper quartiles. However, with regards to body weight, changes in body composition, and dietary intake, no differences were observed across the quartiles of changes in 6 m walking speed (Supplementary Table 2).
Associations Between β-Diversity Measures and Walking Speed
To determine differences (or dissimilarity) in microbiome composition in relation to walking speed and its changes over time, we performed distance-based analysis, that is, weighted and unweighted UniFrac multivariate differential abundance, and used PERMANOVA. Across quartiles of 400 m walking speed measured at the 2014–2016 visit, our analysis revealed that only unweighted UniFrac PCOA trended towards significance (p = .088, Supplementary Table 3), while weighted β-diversity did not reach statistical significance (Supplementary Table 4), adjusting our models for age, race, clinical center, education, marital status, batch number, self-reported health status, weight, height, physical activity, energy, and fiber intake, and number of medications.
In our retrospective longitudinal analysis, we did not observe any differences between quartiles of changes in 6 m walking speed (from 2005–2006 to 2014–2016 visits) and unweighted (Supplementary Table 5) and weighted (Supplementary Table 6) β-diversity measures. To account for differences in library size, we rarefied our microbiome data (using the minimum number of reads per sample) prior to β-diversity assessment. To verify that this approach did not affect our β-diversity assessment, we normalized our reads using the TSS method. Our findings using TSS normalization were consistent with our β-diversity assessment using rarefied reads.
Differential Abundance Analysis at the Genus Level
We used ANCOM-BC to determine genus-level taxa that were differentially abundant per 1 unit increase 400 m walking speed (m/s; Figure 2; Panel A) and per 1 unit change in 6 m walking speed (m/s; Figure 2; Panel B), both as continuous variables. Differences in the abundance of taxa are presented as log-fold changes with signs determining the direction of association. Therefore, a positive value indicates that higher walking speed is associated with an increased abundance of a particular taxon, while a negative value indicates that it is associated with a decreased abundance.
Genus-level taxa that are responsible for differences in microbiome composition per one-unit increase in 400 m walking speed (measured at Visit 4; Panel A) and per one-unit changes in 6 m walking speed (from 2005–2006 to 2014–2016; Panel B), using the ANCOM-BC differential abundance analysis R package. Taxa are categorized and displayed according to their phylum level, and data are presented by effect size (log-fold change) and SE bars. p Values for all the genus-level taxa shown in the figure were <.05. p Values were adjusted for age, race, clinical center, education, marital status, health status, weight, height, physical activity, batch number, number of medications, total energy intake, and dietary fiber and Hommel adjusted for multiple comparisons.
Overall, 86 genera were found to be differentially abundant per 1 unit increase in 400 m walking speed after adjusting for age, race, clinical center, education, marital status, health status, weight, height, physical activity, batch number, number of medications, total energy intake, and dietary fiber and Hommel adjusted for multiple comparisons (Figure 2; Panel A). Most of these genera (76%) belonged to the Firmicutes phylum, followed by Bacteroidetes and Proteobacteria. Fifty-five out of 86 genera, including Fusicatenibacter, Dorea, and Paraprevotella were more abundant as walking speed increased, while the abundance of 31 genera out of 86, including Escherichia/Shigella and Ruminococcaceae_UCG-014 were lower.
Moreover, we found 64 genera to be differentially abundant per 1 unit change in 6 m walking speed (Figure 2, Panel B), the majority of which (77%) belonged to the Firmicutes phylum. Out of the 64 genera, almost an equal number of genera increased (31/64) or decreased (33/64) in abundance.
Comparison of Bacterial Microbiome Networks Between Slow and Fast Walkers
We performed microbial association network analysis to explore microbial community dynamics and identify keystone taxa at the genus level based on 400 m walking speed levels. Our focus was to compare bacterial microbiome networks between slow walkers (representing the first quartile [Q1] of 400 m walking speed) and fast walkers (Q4; Figure 3). We observed almost a similar number of clusters in slow and fast walker networks (8 vs 7 clusters, respectively). However, both networks had different numbers of hub nodes and agreed on only 1 hub node. That is, the slow walker’s network had 3 hub nodes (ie, Bacteroides, Christensenellaceae_R-7_group, and Ruminococcaceae_UCG-005), while the fast walker network contained only 1 hub node (ie, Ruminococcaceae_UCG-005).
Comparison of bacterial associations at genus level between slow walkers (1st quartile of 400 m walking speed; Q1 ≤ 0.899; n = 185) and fast walkers (Q4 ≥ 1.159; n = 185), using the NetCoMi R package. The SPRING method was used as an association measure. The “signed” distance metric was used to transform estimated correlations into dissimilarities. Corresponding similarities were used as the edge weights. Eigenvector centrality was used to determine hubs and scaling node sizes. Clusters were determined using greedy modularity optimization and node colors was used to represent clusters. In both networks, clusters have the same color if they share at least 2 genera. Green color edges were used to show positive associations, while red edges show negative correlations. Nodes that were not connected were removed from both networks.
For a quantitative comparison of the similarity of sets of most central nodes and the hub nodes among slow and fast walkers, we used Jaccard indices (Table 2). Among our 4 centrality measures and hub taxa, degree (ie, the number of edges that are connected to a node in a network), closeness (ie, the reciprocal of the sum of the shortest paths between a node and all other nodes in the network), and Eigenvector centrality (ie, a measure of the importance of a node based on the connections it has to other important nodes in a network) were significantly different between slow and fast walker networks. That is, the number of connections between the nodes (eg, Prevotella_2, Ruminococcaceae_NK4A214_group, and Ruminococcaceae_UCG-002) was higher in the fast walker’s network compared to the slow walker’s network. This indicates a higher degree centrality in the fast walker’s network. Additionally, we observed that some nodes had higher Eigenvector centrality values in fast walkers (eg, Ruminococcaceae_UCG-005) while others had higher centrality values in slow walkers (eg, Lachnospiraceae_NK4A136_group). Eigenvector centrality can be high even when there are few connections to other nodes if the node is very well connected. Finally, compared to the slow walker networks, several nodes in the fast walker’s group had higher closeness centrality values, such as Prevotella_7 and Prevotella_9. Edge density was also higher in fast walkers than in the slow walkers group.
Quantitative Comparisons of Bacterial Networks for the Slow Walkers (1st Quartile of 400 m Walking Speed at Visit 4 (Q1 ≤ 0.899 m/s; n = 184) and Fast Walkers (Q4 ≥ 1.159 m/s; n = 185) by Jaccard index (j), Using the NetCoMi R Package
| Centrality measures | Jacc (j) | p (J ≤ j) | p (J ≥ j) |
|---|---|---|---|
| Degree | 0.700 | .997 | .020* |
| Betweenness centrality | 0.462 | .896 | .241 |
| Closeness centrality | 0.667 | .996 | .019* |
| Eigenvector centrality | 0.667 | .996 | .019* |
| Hub taxa | 0.500 | .889 | .556 |
| Centrality measures | Jacc (j) | p (J ≤ j) | p (J ≥ j) |
|---|---|---|---|
| Degree | 0.700 | .997 | .020* |
| Betweenness centrality | 0.462 | .896 | .241 |
| Closeness centrality | 0.667 | .996 | .019* |
| Eigenvector centrality | 0.667 | .996 | .019* |
| Hub taxa | 0.500 | .889 | .556 |
Note: (j) index shows similarity between sets of most central nodes and hubs between the 2 networks. P (J ≤ j) is the probability that j index takes a value less than or equal to the calculated index j for the present total number of genera in both sets. p (J ≥ j) is analogous, that is, j is greater than expected at random. To account for multiple testing, we applied the Benjamini–Hochberg method for adjustment. *Indicates significance at p < 0.05.
Quantitative Comparisons of Bacterial Networks for the Slow Walkers (1st Quartile of 400 m Walking Speed at Visit 4 (Q1 ≤ 0.899 m/s; n = 184) and Fast Walkers (Q4 ≥ 1.159 m/s; n = 185) by Jaccard index (j), Using the NetCoMi R Package
| Centrality measures | Jacc (j) | p (J ≤ j) | p (J ≥ j) |
|---|---|---|---|
| Degree | 0.700 | .997 | .020* |
| Betweenness centrality | 0.462 | .896 | .241 |
| Closeness centrality | 0.667 | .996 | .019* |
| Eigenvector centrality | 0.667 | .996 | .019* |
| Hub taxa | 0.500 | .889 | .556 |
| Centrality measures | Jacc (j) | p (J ≤ j) | p (J ≥ j) |
|---|---|---|---|
| Degree | 0.700 | .997 | .020* |
| Betweenness centrality | 0.462 | .896 | .241 |
| Closeness centrality | 0.667 | .996 | .019* |
| Eigenvector centrality | 0.667 | .996 | .019* |
| Hub taxa | 0.500 | .889 | .556 |
Note: (j) index shows similarity between sets of most central nodes and hubs between the 2 networks. P (J ≤ j) is the probability that j index takes a value less than or equal to the calculated index j for the present total number of genera in both sets. p (J ≥ j) is analogous, that is, j is greater than expected at random. To account for multiple testing, we applied the Benjamini–Hochberg method for adjustment. *Indicates significance at p < 0.05.
Discussion
Our study found that faster walking speed was associated with higher gut microbiome diversity in older men. Additionally, we observed a trend towards higher microbiome diversity in individuals with a lower decline in walking speed over a 10-year period preceding the microbiome measures. We also observed significant differences in the abundance of gut bacteria based on walking speed and the magnitude of changes in walking speed. Specifically, higher walking speed levels and lower decline in walking speed were associated with higher abundance of genera, such as Paraprevotella, Fusicatenibacter, and Alistipes, which are known to produce SCFAs and have anti-inflammatory properties. Moreover, the gut microbiome networks of fast versus slow walkers showed significant differences in their characteristics, including the presence of unique hubs and variations in centrality measures.
Walking Speed and Gut Microbiome Composition: Cross-Sectional and Retrospective Longitudinal Associations
Walking speed is a complex phenotype and is a widely accepted indicator of mobility disability (45), frailty (46), and mortality (7). Walking speed is influenced by various factors, including age, physical fitness, and underlying health conditions, such as musculoskeletal and neurological disorders (7). Most of our knowledge about the impact of gut microbiome on mobility function is from animal studies (14), as only a few human studies on older adults have investigated this association. Moreover, most human studies on the relationship between gut microbiome and physical function are cross-sectional and mainly focused on frailty (18,30–32), thereby overlooking the potential link between gut dysbiosis and the initial stages of mobility limitations. Here, our aim was to determine both cross-sectional and retrospective longitudinal associations between walking speed and gut microbiome composition. Our cross-sectional analysis revealed that faster walking speed was associated with higher gut microbiome diversity. We also observed that a greater decline in walking speed over a 10-year period was marginally associated with lower gut microbiome diversity in our retrospective longitudinal analysis. Our findings from the retrospective longitudinal analysis indicate that a sustained decline in mobility is associated with various health impairments, including diminished physical activity and muscle strength, reduced diversity of the gut microbiome, compromised overall well-being, and cognitive decline. These observations underscore the potential of gait speed as a longitudinal marker of health status.
Although gut-muscle (9) and gut-brain axes (47) have been proposed as potential mechanisms linking the gut microbiome to physical function, the exact mechanisms underlying the relationship between the gut microbiome and walking speed are still not fully understood due to the complex interactions between multiple factors that influence walking speed. Our findings suggest the existence of a gut-mobility axis, highlighting the relationship between walking speed and gut microbiome composition and the importance of targeting both factors for healthy aging.
The gut-mobility axis is a complex and bidirectional relationship. For example, the production of beneficial metabolites by gut bacteria, such as SCFAs, can have a positive impact on muscle metabolism and energy production (48). Additionally, age-associated dysfunction of the gut mucosal barrier function can cause fecal pro-inflammatory metabolites, like lipopolysaccharide to enter the bloodstream, which can lead to chronic inflammation (9). Chronic inflammation can contribute to muscle loss and other organ dysfunction by promoting muscle protein breakdown and impairing muscle protein synthesis. Furthermore, alterations in gut microbiome composition can also contribute to anabolic resistance and reduce the ability of muscle to respond to anabolic stimuli, such as protein intake and exercise, ultimately affecting muscle size and function (9,48). Conversely, muscle function can also influence the gut. For instance, exercise has been shown to have beneficial effects on the gut microbiome composition by promoting the growth of beneficial bacteria (12,13), reducing inflammation, and intestinal permeability (49).
The Abundance of Specific Bacteria in the Gut Is Linked to Walking Speed and Changes in Walking Speed
Our differential abundance analysis indicates a possible association between the abundance of specific bacterial genera, primarily belonging to the Firmicutes and Bacteroidetes phyla, and both walking speed and changes in walking speed. We observed a greater abundance of SCFA-producing bacteria that also have anti-inflammatory properties (eg, Dorea, Fusicatenibacter, Alistipes, and Oscillibacter), as well as a lower abundance of bacteria with pro-inflammatory properties (eg, Pseudomonas, Turicibacter, and Escherichia/Shigella) per 1 unit increase in 400 m walking speed and per 1 unit change in 6 m walking speed. Consistent with our result, in one of the earliest studies on the relationship between gut microbiome composition and muscle function in older adults, van Tongeren S. et al. (30), reported a lower abundance of SCFA-producing bacteria (eg, Faecalibacterium prausnitzii) in frail participants in a cross-sectional analysis of a small number of advanced agers (N = 23; 86 years). A lower abundance of SCFA-producing bacteria, including Faecalibacterium prausnitzii OUT was also observed in a cross-sectional analysis of 728 female twins (50). Further studies are needed to confirm the role of specific bacteria in the development of mobility impairments.
In this study, we observed that bacterial abundance and α-diversity measures were different based on walking speed and its changes, however, beta-diversity was the same. This suggests that there were compositional differences in the gut microbiome within individuals, as reflected in the differences in bacterial abundance and α-diversity measures. However, beta-diversity, which measures the diversity between individuals, was not significantly different, indicating that these compositional differences within individuals do not contribute significantly to the overall variation in gut microbiome diversity between individuals. For example, certain bacterial species may be more abundant in some individuals than in others, resulting in differences in α-diversity and bacterial abundance. However, these differences in abundance may be relatively small and do not have a significant impact on the overall variation in β-diversity between individuals. Also, to ensure that our rarefication method for adjusting for the library size did not affect our assessment of β-diversity, we applied normalization TSS to our microbiome data (42). This method accounts for differences in sequencing depth without rarefying the data. The results obtained using TSS normalization were consistent with our previous findings, indicating that our choice of normalization method did not influence the assessment of β-diversity.
Bacterial Network Analysis Reveals Keystone Taxa Associated With Walking Speed Levels
To capture complex interactions among different bacterial taxa in the gut microbiome, we conducted a bacterial network analysis based on walking speed. Unlike simple abundance analysis, this approach reveals the collaborative or competitive relationships between bacteria in the gut (44). Our aim was to identify key differences in the gut microbiome network of fast versus slow walkers that play a crucial role in maintaining the stability and functionality of the gut microbiome. Here, we observed that the gut microbiome networks of slow walkers had more hub nodes compared to fast walkers. There is currently no consensus on whether having more or fewer hub nodes may cause instability or dysbiosis in certain cases. Bacteroides and Christensenellaceae_R-7_group genera were among the nodes of slow walkers. A higher abundance of the genera Christensenellaceae R-7 group has been reported in less active older adults (step counts < 6 500) compared to active individuals (step counts ≥ 6 500) in a small cross-sectional analysis of 49 older adults (51). Furthermore, fast and slow walker networks agreed on only 1 hub node, ie, Ruminococcaceae UCG-005. Ruminococcaceae UCG-005 belongs to the family Ruminococcaceae, which is part of the phylum Firmicutes and has been reported to be butyrate-producing bacteria (52). Targeting hub nodes may represent a potential strategy for modulating the gut microbiome to improve health outcomes.
Moreover, we observed a higher number of connections between nodes and a denser and more interconnected network in the fast versus slow walker’s network. These findings may have implications for potential differences in mobility function and health outcomes between fast and slow walkers. However, additional research is needed to confirm these results and explore their underlying mechanisms.
Strength and Limitations
Based on our knowledge, our study is the first human study that investigated the relationship between gut microbiome composition and walking speed, particularly in the context of the initial stages of mobility decline in a large sample of older adults. Another strength of our study is looking at the relationship between walking speed and gut microbiome both cross-sectionally and longitudinally. Our results are conservative because we adjusted for multiple comparisons. Our longitudinal design can provide a more comprehensive understanding of the relationship between these variables, and it acknowledges the complex and bidirectional relationship between gut microbiome composition and muscle function. However, causality cannot be established because of the retrospective longitudinal design of this study, hence, further studies that prospectively follow participants can provide more robust evidence of causality. Furthermore, our study lacks the longitudinal assessment of the gut microbiome which can illuminate its dynamic changes over time and explore its correlation with mobility function.
Moreover, unlike previous small-scale studies, our study has a large sample size of older adults, which increases our statistical power and ensures that the results are more representative of the older adult population. While our study provides valuable insights into the composition of the gut microbiome in advanced-age adults, it is important to acknowledge that the generalizability of our findings may be limited by the fact that our participants were relatively healthy compared to the broader community of older adults. Also, the MrOS cohort is limited to men, the majority of whom were White; therefore, our findings may not be generalizable to women and other race/ethnic groups. In addition, we used a variety of robust analytical techniques, including differential abundance analysis (by ANCOM-BC package) and bacterial network analysis (by NetComi package), which provides a comprehensive analysis of the relationship between gut microbiome composition and walking speed.
Conclusion
Our findings provide evidence to support the existence of the gut-mobility axis and suggest that optimizing gut microbiome health could play a role in preventing or delaying mobility decline in older individuals. This highlights the potential for targeted exercise programs or dietary interventions that focus on improving gut microbiome health to improve overall health in older adults. However, more studies are needed to better understand the mechanisms underlying the relationship between gut microbiome and mobility function.
(Medical Sciences Section)
Funding
The Osteoporotic Fractures in Men (MrOS) Study is supported by National Institutes of Health funding. The following institutes provide support: the National Institute on Aging (NIA), the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS), and the National Center for Advancing Translational Sciences (NCATS) under the following grant numbers: U01 AG027810, U01 AG042124, U01 AG042139, U01 AG042140, U01 AG042143, U01 AG042145, U01 AG042168, U01 AR066160, R01 AG066671, and UL1 TR002369. S.F. is supported by a career development award from the NIA (K01 AG071855) and the Pittsburgh Older Americans Independence Center Scholar (P30AG024827). D.P.K.’s effort and a portion of gut microbiome genotyping in MrOS were supported by the NIAMS (R01 AR061445).
Conflict of Interest
D.P.K. serves on the Scientific Advisory Boards of Solarea Bio and Radius Health; has received grants for his institution from Amgen and Solarea Bio; and receives royalties from Wolters Kluwer for publications in UpToDate. E.S.O. serves as a consultant for Amgen, Bayer, BioCon, and Radius; receives research support from Mereo; and has received travel funding from the OI Foundation.All other authors report no conflicts of interest.
Data Availability
Data described in the manuscript, code book, and analytic code will be made publicly and freely available without restriction.
Author Contributions
The authors’ contributions were as follows—S.F.: designed, analyzed, and wrote the manuscript; J.A.C., P.M.C., L.L., E.S.O., D.M.K., D.P.K., and A.B.N.: were involved in the interpretation of data and manuscript critical review; S.F.: had primary responsibility for the final content; and all authors: read and approved the final manuscript.
