High dietary B₁₂ is associated with reduced gut microbial B₁₂ biosynthesis capacity and lower fecal short-chain fatty acids in healthy United States adults

ABSTRACT

Vitamin B₁₂ is acquired through the consumption of animal-source foods and supplements. In animal models, interventions with B₁₂ and/or methionine influence fecal short-chain fatty acid (SCFA) concentration. Yet the relevance of dietary B₁₂ to microbially produced SCFAs in humans is unknown. This study determined associations between dietary B₁₂ and the gut microbiome in a deeply phenotyped cohort of healthy U.S. adults. Habitual diet and fecal shotgun metagenomes were integrated alongside measurements of fecal SCFAs, plasma SCFAs, and plasma B₁₂ (n = 277). Vitamin B₁₂ intake ranged from 2.4 to 1062 µg/day, and nearly all participants were B₁₂ replete. Stratification of participants into adequate (2.4–8.51 µg/day) and high B₁₂ intake (>8.51 µg/day) groups revealed the association of high intake with a reduction in bacteria capable of anaerobic B₁₂ biosynthesis. High B₁₂ intake was also associated with lower fecal SCFA concentrations even after controlling for fiber and methionine intake. Differences in microbial taxa between dietary groups were limited. However, machine learning models demonstrated the ability to predict fecal propionate and butyrate from microbial pathways in the adequate or no supplement groups, but not in the high intake or supplement groups. Our results indicate that dietary B₁₂ greater than 8.51 µg/day may be associated with reduced microbial synthesis of B₁₂ and lower fecal SCFA production.

Graphical abstract

Introduction

Animal source foods are the only natural source of vitamin B₁₂ in the human diet. Fortified foods and supplements offer a practical way to increase B₁₂ intake when intake of animal source foods is limited, whether due to food environment barriers, financial barriers, or dietary choices that restrict animal source foods. ^{1 , 2} In the United States, where animal-source food intake is generally sufficient to meet the daily recommended intake of B₁₂, many still fall short of meeting the estimated average requirements for several essential vitamins and minerals through diet alone. ³ Consequently, nearly 58% of Americans take vitamin supplements, with usage highest among women and older adults. ⁴

B₁₂ is initially absorbed in the small intestine, and its bioavailability is approximately 50% from dietary sources, with decreases in absorption efficiency detected at higher pharmacologic doses. ⁵ Dietary B₁₂ evades host digestion through a multi-step process that culminates in receptor-mediated uptake in the ileum. ⁶ Intrinsic factor-mediated host absorption is limited to a threshold of 1.5–2.5 µg per meal due to the delay in recycling of the endocytosed receptor back to the ileal enterocytes. ⁷ Thus, with higher B₁₂ intake, a greater fraction of dietary B₁₂ bypasses small intestinal absorption and passes to the large intestine for modification and utilization by the gut microbiota. ^{8 , 9} In the colon, B₁₂ and related cobamide structures are ecologically important to the gut microbiota as co-factors to microbial enzymes involved in carbon, nitrogen, and nucleotide metabolism, as well as fermentation of diet- or mucus-derived monosaccharides. ^9-12 B₁₂ also functions as a co-factor in microbial short-chain fatty acid (SCFA) production. Specifically, B₁₂ is a co-factor in the propanediol and succinate pathways of propionate synthesis as well as acetate synthesis via the Wood–Ljungdahl pathway. ^13-16

Recent attention has been given to the effects of dietary methionine restriction on, for example, metabolic health ^{17 , 18} and improved lifespan. ¹⁹ Moreover, previous research in animal models has indicated that excess dietary methionine may inhibit microbial SCFA production, ²⁰ resulting in a decrease in these health-promoting metabolites. ^{21 , 22} Yet the dietary sources and eating patterns that provide methionine are often rich in B₁₂ as well. In vitro work investigating the relationship between B₁₂ and SCFAs found mixed results that suggested an association dependent on gut microbial composition. ²³ In the present work, we analyzed the relationship between dietary B₁₂ and fecal SCFAs in healthy humans, while controlling for dietary methionine and other person-specific factors that likely influence fecal SCFA abundance.

Despite the known functions of B₁₂, it remains unclear if high dietary B₁₂ intake impacts the taxonomic structure and functional capacity of the gut microbiome in humans. A systematic review linking dietary B₁₂ to the gut microbiome highlighted the lack of human studies using metagenomic approaches. ²⁴ Our study addresses that gap by integrating dietary data with fecal shotgun metagenomes and fecal SCFA measurements from healthy U.S. adults in an observational cohort. Specifically, we tested the hypothesis that higher B₁₂ intake is associated with a reduction of microbially produced fecal SCFAs and a reduction in anaerobic B₁₂ biosynthesis capacity in the gut microbiome. After hypothesis testing, we used feature engineering and machine learning to examine associations between gut microbial taxa and metabolic pathways with fecal SCFA concentrations.

Materials and methods

Study design and participants

This work analyzes the USDA Nutritional Phenotyping Study, an observational, cross-sectional study that enrolled 393 healthy adult participants balanced by age (18–65 y), sex, and BMI (18.5–39.9 kg/m²) over the course of 4 y. ²⁵ Exclusion criteria included minor surgery, hospitalization, or antibiotic therapy in the previous 4 weeks, major surgery within the past 16 weeks, egg allergy, pregnancy or lactation, and diagnosed chronic disease treated with medication. The study protocol was approved by the Institutional Review Board at the University of California, Davis, and was prospectively registered on Clinicaltrials.gov (NCT02367287). All participants for whom the shotgun fecal metagenomes were sequenced and both plasma B₁₂ and fecal and/or plasma SCFAs were quantified were included in this analysis. Plasma B₁₂ was the most limiting analyte. In addition, we excluded participants who consumed less than the daily recommended intake of B₁₂ (2.4 µg/day) based on responses from dietary assessments (n = 10). The stepwise reduction in participants from the parent study to the final analytic sample of n = 277 is shown in Supplemental Figure 1. The subset of participants in our analyses retained balanced recruitment of age and sex, while recruitment from the highest BMI category (≥30  kg/m²) was lower than that from the other two BMI categories (p = 0.01; Supplemental Table 1).

Dietary assessment

Given that our outcomes of interest focused on microbial structure and function, we utilized habitual dietary estimates rather than recent intake, as habitual intake better reflects long-term B₁₂ consumption and was therefore more likely to reveal stable microbial patterns associated with diet. Habitual intake of B₁₂ from foods and supplements was estimated using the Block 2014 Food Frequency Questionnaire (FFQ; NutritionQuest, Berkeley, CA, USA). The FFQ estimates the habitual intake of 127 foods and beverages over the previous year. The questionnaire was filled out electronically under the supervision of trained personnel. Measuring cups and spoons were used as visual aids for determining portion sizes.

For our analyses, we stratified participants into four dietary groups: individuals whose B₁₂ intake was in the upper 50th percentile (>8.51 µg per day, herein ‘high intake group’), lower 50th percentile (2.40–8.51 µg per day, herein ‘adequate intake group’), individuals who consumed non-zero B₁₂ in the form of a supplement (herein ‘supplement use group’, n = 121), and those who did not consume B₁₂ in the form of a supplement (‘no supplement use group’, n = 156). Other dietary variables from the FFQ included in our analyses were soluble dietary fiber and dietary methionine (both in grams/day). Baseline participant characteristics remained similar between the vitamin B₁₂ intake groups and the supplement use groups (Supplemental Tables 2 and 3).

Plasma sample collection

Fasting blood specimens were collected using Vacutainer® heparin tubes (BD, Franklin Lakes, NJ, USA). After 30 min on wet ice, plasma was obtained by centrifugation at 1300 × g at 4 °C for 10 min. The plasma was immediately aliquoted and stored in Cryo-Store® vials at −80 °C until analysis.

Quantification of plasma B₁₂ concentration

B₁₂ concentrations in the plasma were analyzed using the cobas® e411 immunoassay analyzer employing the Elecsys® Vitamin B₁₂ II kit (Roche Diagnostics, Indianapolis, IN, USA) according to the manufacturer’s protocol. To ensure precision and accuracy in each analytical batch, we included the Roche PreciControl Varia measured at levels 0, 1, and 2, as well as an internal UTAK control (UTAK Laboratories Inc., Valencia, CA, USA).

Quantification of plasma SCFAs

Plasma SCFAs were measured in EDTA-treated plasma samples at the Columbia University Medical Center Biomarkers Core Laboratory using ultra-performance liquid chromatography‒tandem mass spectrometry as previously described. ²⁶

Stool collection

Sanitary collection supplies were provided, and participants received an email reminder 3–4 d prior to the scheduled site visit. Each participant was asked to provide a single stool sample and transport the sample on cold packs to the research center as soon as possible for same-day processing. The specimen was homogenized using a Stomacher® paddle blender, frozen, and stored in Cryo-Store® vials at −80 °C until analysis.

Quantification of fecal SCFAs

Fecal SCFAs were measured via gas chromatography‒mass spectrometry as previously described from approximately 150 mg of wet stool. ²⁶

Fecal DNA extraction and metagenomic sequencing

DNA from fecal samples was extracted using the ZymoBiomics DNA Miniprep kit (Zymo Research, Irvine, CA, USA), as previously described. ²⁷ Library preparation and sequencing were performed by the DNA Technologies & Expression Analysis Core Laboratory at the University of California, Davis Genome Center. Library insert sizes were 250–280 bp, and the exact size selection of DNA fragments was performed with a PippinHT instrument (Sage Science, Beverly, MA, USA). Libraries were combined into a single pool for each sequencing batch and sequenced in a 2 × 150-bp format on the NovaSeq6000 platform (Illumina, San Diego, CA, USA), with an S4 flow cell resulting in an approximately 20-bp overlap in the reads. Initially, 290 samples were sequenced and analyzed. ²⁷ Later an additional 40 samples were sequened and analyzed using identical methods. ²⁶ A combination of these previously described fecal shotgun metagenomes were analyzed in this publication (Supplemental Figure 1).

Metagenomic sequencing pre-processing

Metagenomic reads were processed as previously described. ²⁷ Briefly, BMTagger was used to remove reads aligning to the human genome version GRCh38.p13 from all samples. ^{28 , 29} Afterward, Trimmomatic version 0.33 was used to remove TruSeq paired-end adapters and reads were trimmed with a sliding window of 4 bp, a minimum average quality of 15, and a minimum length of 99 bp. ³⁰ Duplicated reads were then removed using FastUniq version 1.1 with default settings. ³¹ The resulting paired-end reads were assembled using FLASH version 1.2.11 with the overlapping length set between 10 bp to 100 bp and the mismatch ratio set at 0.1. ³² Code to reproduce the pre-processing of metagenomic sequences from this work can be found at https://github.com/dglemay/ARG_metagenome.

Read-based taxonomic and functional profiling of the microbiome

Merged reads output from FLASH were used as input to MetaPhlAn v4.0.6 against the vOct22 database, using default parameters and the --add_viruses flag. ³³ Prior to running the differential abundance analysis and machine learning (ML) models using these taxonomic data, the default MetaPhlAn taxonomy table was converted to Genome Taxonomy Database (GTDB) format using the utility script (sgb_to_gtdb_profile.py) provided by the MetaPhlAn developers.

Merged reads output from FLASH were also used as input to HUMAnN version 3.7 using the default parameters and the flag --search-mode uniref90. ³⁴ The ChocoPhlAn database used was v201901_v31, and the uniref90 database was v201901b_full. Pathway abundance files were merged and normalized to copies per million using the humann_renorm_table command.

Differential abundance analysis

The relative abundances of microbial taxa were used to construct multivariable associations with linear models using the MaAsLin3 package in R. ³⁵ We aimed to identify taxa that were differentially abundant between our dietary groups based on total B₁₂ intake and the use of B₁₂ supplements. To account for differences in the host-indigestible nutrients available for utilization by the gut microbiome, we added dietary soluble fiber intake, as estimated by the FFQ, to all model formulas. In addition, the arguments min_prevalence and min_abundance were set to 0.25 and 0.001, respectively. All other arguments were set to default.

Assembly-based functional profiling of the microbiome

Reads were individually assembled using MEGAHIT v1.2.9. ³⁶ Bowtie2 v2.5.2 was used to recruit metagenomic reads to contigs, and samtools v1.19.2 was used to convert SAM files to BAM files. ^{37 , 38} MetaBAT 2 v2.18 was used to generate read depth files from the sorted BAM files, and to subsequently bin contigs. ³⁹ Bin reconstruction quality was evaluated using CheckM2 v1.1.0. ⁴⁰ For our assembly-based analyses, we used medium-quality draft metagenome-assembled genomes (MAGs), defined as bins that were estimated as being greater than or equal to 50% complete and less than 10% contaminated. ⁴¹

Metabolic pathway reconstruction and functional profiling were performed using anvi’o v8. ⁴² MAG fasta files were imported and reformatted for use with anvi’o, then open reading frames were predicted using anvi-gen-contigs-database. Functional annotations were performed with anvi-run-kegg-kofams to assign genes to metabolic pathways using the v2023-09-22 KEGG snapshot. Annotated MAGs were then input to anvi-estimate-metabolism to predict the presence of metabolic pathways and summarize potential metabolic functions using default parameters. ⁴³

The proportion of microbial taxa with the genomic capacity for anaerobic B₁₂ biosynthesis was quantified using the presence of two composite KEGG modules, M00924 and M00122. A list of queried enzymes can be found in Supplemental Table 4, which were used to create a custom metabolic module for use with anvi’o. ^{44 , 45} A pathway was considered complete if ≥90% of the pathway was present in a MAG, given the following pathway definition: (K02302,(K02303,K13542,K00589,K02496,K13543) (K02304,K24866)) (K02190,K03795,K22011) (K03394,K13540) (K05934,K13541,K21479) K05936 (K02189,K13541) K02188 K05895 (K02191,K03399,K00595) K06042 K02224 (K00798,K19221) K02232 (K02225,K02227) K02231 K00768 (K02226,K22316) K02233. To examine the associations between dietary B₁₂ intake and microbial B₁₂ biosynthetic potential, the results were aggregated by subject and were visualized to represent the proportion of taxa with a complete pathway relative to the total number of recovered MAGs (per subject) between our dietary groups based on total B₁₂ intake and the use of B₁₂ supplements.

Machine learning

Random forest regression models were used to assess the predictive capacity of gut microbial composition and pathway abundances on the concentration of fecal acetate, propionate, and butyrate. Each of our models included the covariate features of age, sex, and BMI, alongside a summary dietary feature from the FFQ that estimated soluble dietary fiber (in grams) consumed per day. To explore how B₁₂ intake may shape microbial metabolism, we stratified participants into four dietary groups: individuals whose B₁₂ intake was in the upper 50th percentile (>8.51  µg per day, herein referred to as the ‘high intake group’), lower 50th percentile (<8.51 µg per day, herein referred to as the ‘adequate intake group’), individuals who consumed non-zero B₁₂ in the form of a supplement (herein referred to as the ‘supplement use group’, n = 101), and those who did not consume B₁₂ in the form of a supplement (‘no supplement use group’, n = 146). From each of these dietary groups, predictive ML models of each fecal SCFA were built using either the merged MetaPhlAn table or the merged and normalized HUMAnN table. Across the four dietary groups, this approach yielded 24 models (Supplemental Table 5).

The implementation of random forest regression models was done using DietML for microbiome pathway abundance models or TaxaHFE-ML for microbiome composition models (https://github.com/aoliver44/taxaHFE). ⁴⁶ DietML and TaxaHFE-ML have many optional arguments; as such, tuning was performed by using a random search grid (n = 500 combinations) to maximize R² (of the DietML or TaxaHFE-ML model), defining the following parameters: corr_level, n_perm, train_split, folds, metric, tune_length, tune_time, tune_stop, prevalence, abundance, and superfilter. Metabolite-specific tuning was done using full data sets, regardless of dietary intake. Once we identified the program arguments that yielded the top-performing models, we split the data into the four categories of B₁₂ intake and re-ran those top models 100 times using randomly generated seeds to account for variability in initial test-train splits. The test data scores were averaged. We calculated feature importance for the highest-performing models based on R²; additionally, these models had lower MAE compared to the null model (the null model is one that predicts the response using just the mean of the response calculated with the training data). Shapley Additive exPlanations (SHAP) values were calculated and used to explain feature importance.

Statistical analysis

All the statistical analyses were conducted in R version 4.4.3 with a significance threshold of α = 0.05. ⁴⁷ Visualizations were done using the ggplot2 and patchwork packages. ^{48 , 49} Chi-square goodness-of-fit tests were used to evaluate whether the categories of age, sex, and BMI were evenly distributed in study participants. Comparisons of age and BMI between categories of dietary intake and supplement use were done using independent t-tests, and chi-square goodness-of-fit tests were used for sex comparisons. Tests were implemented using the gtsummary package in R. ⁵⁰

Plasma B₁₂ concentration, dietary B₁₂, and dietary methionine were assessed for residual normality and transformed accordingly using the bestNormalize package. ⁵¹ Group differences in plasma B₁₂ concentrations, dietary B₁₂, and dietary methionine between the categories of dietary intake and supplement use were evaluated with Mann–Whitney U tests using the rstatix package. ⁵² Associations between plasma B₁₂ and predictors (age, sex, BMI, dietary B₁₂) were assessed using multivariable linear regression using the car package in R. ⁵³ Similarly, associations between dietary methionine and predictors (age, sex, BMI, soluble fiber, and dietary B₁₂) were assessed using multivariable linear regression using the car package.

Microbial alpha diversity was computed using the vegan package in R and modeled against the categories of dietary intake and supplement use as predictors using linear regression. ⁵⁴ Beta diversity analyses were done using the adonis2 function in the vegan package using 999 permutations.

Partial Spearman correlations were computed using the ppcor package in R, which rank-transformed variables (dietary B₁₂, methionine, fecal SCFAs, age, sex, BMI, and soluble fiber intake) prior to estimating partial correlations. ⁵⁵ Associations between dietary B₁₂ and fecal SCFAs were modeled separately using Mann–Whitney U tests in the rstatix package and generalized linear models (Gamma distribution, log-link), which adjusted for age, sex, BMI, soluble fiber intake, and dietary methionine. Model fit was assessed with the DHARMa package, ⁵⁶ and estimated marginal means were back-transformed to the original scale using the emmeans package. ⁵⁷ Pairwise contrasts were computed using the contrast() function within the emmeans package, and associated p-values were FDR-adjusted.

Marginal slopes of dietary B₁₂ were estimated using the emtrends() function from the emmeans package, providing covariate-adjusted estimates of the linear trend on the log scale. Two-sided p-values for slopes were computed using Wald z-tests. For visualization, model-adjusted fecal SCFA concentrations were generated across the observed range of normalized B₁₂ intake using estimated marginal means on the response scale. The predicted lines of best fit and 95% confidence intervals were plotted alongside raw data to illustrate covariate-adjusted relationships between B₁₂ intake and fecal SCFA concentrations.

As a preliminary test of group differences, Mann–Whitney U tests were conducted for each plasma SCFA against the categories of dietary intake and supplement use. Partial Spearman correlations were computed using the ppcor package in R, which rank-transformed variables (dietary B₁₂, plasma SCFAs, age, sex, BMI, dietary methionine, and soluble fiber intake) prior to estimating partial correlations. ⁵⁵ Inspection of the lower tail of plasma butyrate and propionate distributions revealed evidence of left-censoring. Accordingly, Tobit regressions were implemented using the VGAM package in R for plasma propionate and butyrate with the lower censoring threshold set to the observed minimum for each analyte while adjusting for age, sex, BMI, dietary fiber, and methionine in the models. ⁵⁸ Linear regressions adjusting for age, sex, BMI, dietary fiber, and methionine that used a normalized response variable generated using the bestNormalize package were implemented to model associations between plasma acetate and the categories of dietary intake and supplement use. The fit of all regression models was assessed with the DHARMa package. ⁵⁶

Machine learning performance was evaluated using mean absolute error (MAE) and R². For each model, the percent change in performance relative to a null model was calculated. Model performance was summarized as the mean ± standard error, calculated using the plotrix package. ⁵⁹ Model performance comparisons were conducted using a global Kruskal–Wallis test and Dunn’s post-hoc test using the FSA package. ⁶⁰ A Benjamini–Hochberg correction for multiple testing was applied to p-values, and statistical differences between ML models were visualized using compact letter displays with the multcompView package. ⁶¹ Visualizations and statistical analyses reported in this paper are provided in a public GitHub repository at https://github.com/margarethilliard/VB12-analysis.

Results

Higher dietary B₁₂ is associated with supplemented B₁₂

We estimated dietary intake of B₁₂ (µg/day) based on foods and supplements recorded in the food frequency questionnaire from 277 healthy individuals (Supplemental Table 1 and Figure 1). Dietary B₁₂ intake was positively skewed (Figure 1A) for our subset of participants who habitually consumed greater than the daily recommended intake of B₁₂ (2.4 µg/day). We split participants into two groups based on their intake of B₁₂ from foods and supplements: individuals whose B₁₂ intake was in the upper 50th percentile (herein the ‘high intake group’) versus those in the lower 50th percentile (herein the ‘adequate intake group’). We also split participants into two groups based on their consumption of vitamin B₁₂ supplements. We assessed B₁₂ supplement use using the food frequency questionnaire and found that nearly 44% of participants consumed non-zero B₁₂ in the form of a vitamin supplement (herein ‘supplement use group’, n = 121; ‘no supplement use group’, n = 156). We found that dietary B₁₂ was higher on average in the high intake group compared to the adequate intake group (p < 0.001, Figure 1C), and in the supplement use group compared to the no supplement use group (p < 0.001, Figure 1D). The number of individuals who did not supplement B₁₂ overlapped with the adequate intake group (n = 118, 67%; Figure 1B). Additionally, the number of individuals who use B₁₂ supplements overlapped with high habitual B₁₂ consumers (n = 102, 64%; Figure 1B). Thus, a majority of the high intake group used B₁₂ supplements, and a majority of the adequate intake group did not use B₁₂ supplements.

Figure 1.

An overview of vitamin B₁₂ intake in a healthy U.S. adult cohort. (A) The discrete distribution plot shows dietary B₁₂, rounded to the nearest whole number, across 277 individuals. The vertical red dashed line indicates the median split (8.51 µg/day), which we used to define our adequate (2.4–8.51 µg/day) and high (>8.51 µg/day) intake groups. (B) Venn diagrams show the number of participants and the percentage overlap between dietary grouping variables based on total B₁₂ intake and B₁₂ supplement use. Boxplots comparing vitamin B₁₂ intake based on (C) participant intake of vitamin B₁₂: adequate intake: 2.40–8.51 µg/day, high intake: > 8.51 µg/day) or (D) use of vitamin B₁₂ supplements (‘B₁₂ supplement use’, n = 121; ‘no B₁₂ supplement use’, n = 156). The upper and lower lines of the boxplots represent the upper and lower quartiles, while the middle line represents the median. The black dashed lines represent the recommended daily intake of vitamin B₁₂ (2.4 µg/day), and the black diamonds represent the mean intake in each group. Group differences were determined using a Wilcoxon rank sum test, and significance is denoted by asterisks (*p < 0.05, **p < 0.01, ***p < 0.001; ns = not significant).

Plasma B₁₂ concentrations are associated with B₁₂ intake from foods and supplements

Plasma B₁₂ concentrations were positively skewed and showed nearly all participants (>96%) were replete for plasma B₁₂ (148–300 pmol/L; Figure 2A, B). We found that plasma B₁₂ concentrations were higher on average in the high intake group compared to the adequate intake group (p < 0.001, Figure 2A), and in the supplement use group compared to the no supplement use group (p < 0.001; Figure 2B).

Figure 2.

An overview of plasma B₁₂ status in a healthy U.S. adult cohort. Boxplots comparing plasma vitamin B₁₂ concentrations based on (A) participant intake of vitamin B₁₂ (adequate intake: 2.40–8.51 µg/day, high intake: > 8.51 µg/day) or (B) use of vitamin B₁₂ supplements (‘B₁₂ supplement use’, n = 121; ‘no B₁₂ supplement use’, n = 156). The upper and lower lines of the boxplots represent the upper and lower quartiles, while the middle line represents the median. The black dashed lines represent the replete range for plasma B₁₂ (148–300 pmol/L), and the black diamonds represent the mean concentration in each group. Group differences were determined using a Wilcoxon rank sum test, and significance is denoted by asterisks (*p < 0.05, **p < 0.01, ***p < 0.001; ns = not significant). (C) Partial regression between the plasma B₁₂ concentration and B₁₂ intake, adjusting for age, sex, and BMI in the model. The black line represents the linear line of best fit, and the gray shading represents the 95% confidence intervals.

As expected, a partial regression adjusted for age, sex, and BMI revealed a strong positive relationship between plasma B₁₂ concentration and B₁₂ intake from diet and supplements (adjusted R² = 0.18, β = 0.41, p < 0.001; Figure 2C). Age had a weak negative relationship with the plasma B₁₂ concentration (partial regression β = −0.01, p = 0.01), while sex and BMI had no relationship (Supplemental Figure 2).

B₁₂ intake is weakly associated with changes in the proportion of bacterial taxa with genomic capacity for anaerobic B₁₂ biosynthesis

We next tested the hypothesis that greater B₁₂ intake would be associated with a community-wide reduction in the capacity for anaerobic B₁₂ biosynthesis. Specifically, we asked if the proportion of microbial taxa encoding the complete anaerobic B₁₂ biosynthesis pathway was different based on the habitual dietary habits of the host. Available read-based functional profiling tools (HUMAnN) quantify B₁₂ salvage pathways, but lack comprehensive annotation for the anaerobic B₁₂ biosynthesis pathway. Thus, we employed an assembly-based approach to detect the genomic capacity for anaerobic B₁₂ biosynthesis that enabled the species-level detection of our function of interest. We found a reduction in the proportion of taxa with the genomic capacity for B₁₂ biosynthesis among the high intake group (Figure 3A; p = 0.049). We found no difference in the mean proportion of total MAGs with complete B₁₂ biosynthesis pathways based on B₁₂ supplement use (Figure 3B).

Figure 3.

Differences in the proportion of microbial taxa with the genomic capacity for anaerobic vitamin B₁₂ biosynthesis between vitamin B₁₂ dietary groups. Anaerobic vitamin B₁₂ biosynthesis was quantified from medium-quality draft metagenome-assembled genomes (MAGs) using the presence of two composite KEGG modules, M00924 and M00122. A pathway was considered complete ≥ 90% of the total KEGG modules were present in a MAG. The proportion of complete B₁₂ biosynthesis pathways was calculated per subject based on the total number of MAGs. The bars represent the mean proportion of complete pathways across individuals in each group, and the error bars represent the standard error of the mean. (A) Proportion of complete B₁₂ pathways based on participant B₁₂ intake group (adequate intake: 2.40–8.51 µg/day, high intake: >8.51 µg/day). (B) Proportion of complete B₁₂ pathways based on participant use of vitamin B₁₂ supplements. Group differences were determined using a Wilcoxon rank sum test, and significance is denoted (ns = not significant).

Greater B₁₂ intake from food and supplements is associated with lower fecal SCFA concentrations

In animal models, diets high in the amino acid methionine have been implicated as being inhibitory to microbial fecal SCFA production. As animal source foods are high in both methionine and B₁₂, which one is responsible for reduced SCFA production, and whether this relationship exists in humans, is unknown. In the current cohort, we found that dietary methionine and B₁₂ were correlated with each other (partial Spearman’s rho = 0.22, p < 0.001). Interestingly, the relationship between dietary B₁₂ and methionine was strengthened when analyzing only participants in the adequate intake group and no supplement use group, but disappeared when looking at the high intake group and the supplement use group (Supplemental Figure 3A, 3C). Similarly, we found that dietary methionine intake was higher on average in the high intake group (p < 0.001), but was not different based on B₁₂ supplement use (p = 0.7, Supplemental Figure 3B, 3D). Thus, the high intake group and supplement users offer the ability to disentangle B₁₂ and methionine with respect to associations with fecal SCFAs because these participants consumed high amounts of B₁₂ without sustaining the positive relationship with dietary methionine intake that would otherwise be expected when consuming animal source foods. To our surprise, we found that dietary methionine was not correlated with fecal acetate (partial Spearman’s rho = 0.06, p = 0.4), propionate (partial Spearman’s rho = 0.08, p = 0.2), or butyrate (partial Spearman’s rho = 0.04, p = 0.5) after controlling for age, sex, BMI, soluble dietary fiber, and dietary B₁₂. In contrast, we found that dietary B₁₂ was negatively correlated with fecal acetate (partial Spearman’s rho = −0.2, p = 0.002) and propionate (partial Spearman’s rho = −0.15, p = 0.02) after controlling for age, sex, BMI, soluble dietary fiber, and dietary methionine. We did not find that dietary B₁₂ was correlated with fecal butyrate (partial Spearman’s rho = −0.1, p = 0.08).

We found that fecal SCFAs were significantly different between B₁₂ intake groups and based on supplement use (Supplemental Table 6). Thus, we extended the analyses of B₁₂ and fecal SCFAs to the categories of dietary intake and supplement use by using generalized linear models that included the covariates age, sex, BMI, soluble fiber intake, and dietary methionine. Pairwise contrasts of estimated marginal means (EMMs) revealed that model-adjusted SCFA concentrations were lower in the high B₁₂ intake group (Figure 4A) and in the supplement use group (Figure 4B). The participants in the adequate intake group had significantly higher adjusted mean concentrations of fecal acetate (EMM = 31.7 nmol/mg, 95% CI [30.1, 33.3]), propionate (EMM = 9.2 nmol/mg, 95% CI [8.6, 9.9]), and butyrate (EMM = 10.3 nmol/mg, 95% CI [9.3, 11.4]) compared to the high intake group (acetate: EMM = 26.8 nmol/mg, 95% CI [25.5, 28.2]; propionate: EMM = 7.8 nmol/mg, 95% CI [7.3, 8.4]; butyrate: EMM = 8.5 nmol/mg, 95% CI [7.7, 9.4]). All three pairwise contrasts were statistically significant (acetate: ratio = 1.2, 95% CI [1.1, 1.3], FDR-adj. p < 0.001; propionate: ratio = 1.2, 95% CI [1.1, 1.3], FDR-adj. p = 0.007; butyrate: ratio = 1.2, 95% CI [1.03, 1.4], FDR-adj. p = 0.02). A consistent pattern was observed for B₁₂ supplement use. Participants that did not use supplements had higher adjusted mean concentrations of acetate (EMM = 30.4 nmol/mg, 95% CI [29.0, 31.9]), propionate (EMM = 9.0 nmol/mg, 95% CI [8.4, 9.6]), and butyrate (EMM = 10.1 nmol/mg, 95% CI [9.2, 11.1]) compared to supplement users (acetate: EMM = 27.6 nmol/mg, 95% CI [26.1, 29.1]; propionate: EMM = 7.9 nmol/mg, 95% CI [7.3, 8.5]; butyrate: EMM = 8.4 nmol/mg, 95% CI [7.5, 9.3]), with all contrasts reaching statistical significance (acetate: ratio = 1.1, 95% CI [1.03, 1.2], FDR-adj. p = 0.01; propionate: ratio = 1.1, 95% CI [1.02, 1.3], FDR-adj. p = 0.02; butyrate: ratio = 1.2, 95% CI [1.04, 1.4], FDR-adj. p = 0.02).

Figure 4.

Fecal SCFA concentrations and their model-adjusted means between vitamin B₁₂ dietary groups. Raw concentrations of fecal acetate, butyrate, and propionate are shown alongside the estimated marginal means (EMMs) predicted from generalized linear models adjusted for age, sex, BMI, dietary soluble fiber, and dietary methionine. (A) Fecal SCFA concentrations based on participant dietary intake of vitamin B₁₂ (adequate intake group: 2.40–8.51 µg/day, high intake: >8.51 µg/day). (B) Fecal SCFA concentrations based on participant use of vitamin B₁₂ supplements. The upper and lower lines of the boxplots represent the upper and lower quartiles, while the middle line represents the median. The black diamonds and vertical black lines represent the EMMs and the 95% confidence intervals, respectively. Significance of pairwise contrasts is denoted by asterisks (*p < 0.05, **p < 0.01, ***p < 0.001; ns = not significant) after false discovery rate correction.

Considering these findings, we extended our hypothesis-driven inquiry to test if there was a negative dose response from B₁₂ from foods and supplements on fecal SCFA concentrations. Additional generalized linear models were constructed that adjusted for age, sex, BMI, soluble fiber intake, and dietary methionine. Estimated marginal slopes of fecal SCFA concentrations across normalized observations of dietary B₁₂ levels were calculated. Consistent with our previous results, the average marginal effect of dietary B₁₂ on fecal acetate was negative (β = −0.07, SE = 0.02, CI [−0.1, −0.03], p < 0.001; Figure 5A). Similarly, the average marginal effect of dietary B₁₂ on fecal propionate was negative (β = −0.07, SE = 0.03, CI [−0.1, −0.02], p = 0.008; Figure 5C). However, the average marginal effect of dietary B₁₂ on fecal butyrate was not significantly different than zero (β = −0.06, SE = 0.04, CI [−0.1, 0.02], p = 0.1; Figure 5B).

Figure 5.

Estimated marginal slopes of predicted fecal SCFA concentrations across the observed range of normalized B₁₂ intake. Model-adjusted estimates of fecal (A) acetate, (B) butyrate, and (C) propionate concentrations predicted from generalized linear models adjusted for age, sex, BMI, dietary soluble fiber, and dietary methionine. The lines represent adjusted mean values, the shaded areas indicate the 95% confidence intervals, and the displayed p-values indicate the significance of a two-tailed Wald z-test.

Dietary B₁₂ is not associated with plasma SCFA concentrations, microbial diversity, or differences in compositional abundance or prevalence

Given that microbially-derived SCFAs can enter systemic circulation, we investigated whether the differences we observed in fecal SCFAs based on B₁₂ intake from food and supplements would extend to associations with plasma SCFA concentrations. However, plasma acetate, propionate, and butyrate concentrations were not associated with B₁₂ intake, dietary intake group, or supplement use (Supplemental Table 7).

There were also no associations between dietary B₁₂ intake, or the four categories of B₁₂ intake, with alpha or beta diversity of the microbiome (Supplemental Table 8). Despite this, we conducted a differential abundance analysis to examine differences in taxonomic abundance and prevalence based on B₁₂ intake. We found no differences in compositional abundance or prevalence based on our dietary intake groups or by supplement use.

Fecal propionate concentrations were predicted by microbial and pathway abundance in participants with lower B₁₂ intake

Having established that B₁₂ intake is associated with fecal SCFA concentrations, we constructed random forest models within B₁₂ intake subgroups (adequate intake, high intake, supplement use, and no supplement use) to examine whether microbial pathway and taxonomic abundances could predict fecal SCFA concentrations in different dietary contexts, thereby exploring potential microbial associations with B₁₂ intake (Supplemental Table 5). In general, we found that propionate was frequently the most accurately predicted fecal SCFA, regardless of whether the models used pathway abundance or taxonomic composition features (Figure 6A, Supplemental Figure 4).

Figure 6.

Predicting fecal SCFA concentrations in vitamin B₁₂ dietary groups using machine learning. Bar charts depict the (A) mean percent change in the mean absolute error (MAE) between a trained model and a null model for predicting fecal SCFA concentrations from microbiome pathway abundance features. The solid vertical black line at x = 0 represents no difference in predictive accuracy between the trained and null models, and a lower mean percent change in MAE represents a more accurate model. The error bars represent the standard error of the mean (n = 100). Differences in means were determined using Dunn's test, where non-matching letters represent significant differences in means. (B) The Beeswarm plot shows the top pathway abundance features in a random forest regression model predicting fecal propionate in the subset of participants who do not use vitamin B₁₂ supplements. Points are colored by their respective feature value. Pathway features in bold are related to one-carbon metabolism.

Among the models using pathway abundance features, the prediction of fecal propionate was restricted to the adequate intake group and no supplement use groups, while the prediction of fecal butyrate was restricted to the no supplement use group. We further investigated the pathway features that drove the top models for predicting fecal propionate. In the no supplement use group, the top features included both person-specific and microbial predictors. Among the person-specific features, younger age and higher BMI were associated with predictions of higher fecal propionate concentrations. Among the microbial features, several top predictors were related to one-carbon metabolism. Namely, greater pathway abundances of two fotate transformation pathways linked to plants (pathway 3841) and E. coli (1CMET2) were associated with higher predictions of fecal propionate concentrations (Figure 6B). Similarly, the biosynthesis of the amino acid methionine surfaced as a top feature predicting greater fecal propionate concentrations in the no supplement use group. Conversely, a greater abundance of the tetrapyrrole biosynthesis II from the glycine pathway was associated with lower predictions of fecal propionate concentrations in this group.

Overall, in terms of predictive accuracy, we found that taxonomic-based models supported our pathway-based results. Among the microbial relative abundance feature set, accurate prediction of propionate and butyrate was widespread throughout all subsets of participants based on B₁₂ intake (Supplemental Figure 4A). Among the top predictors, higher abundances of Oscillospiraceae and Eggerthellaceae were consistently associated with lower fecal propionate concentrations across all subgroups (Supplemental Table 9). In addition, several recognized or putative butyrogenic taxa, including Agathobacter rectalis, Phocaeicola vulgatus, Bacteroides ovatus, Anaerostipes hadrus A, and Lachnospiraceae AM51-8, were all positively associated with fecal butyrate concentrations in a model predicting butyrate in the no supplement use group (Supplemental Table 9).

Discussion

This study examined the associations of B₁₂ intake from foods and supplements with plasma B₁₂ concentrations, gut microbiome composition, pathway abundances, and fecal and plasma SCFA concentrations in 240 healthy U.S. adults from the USDA Nutritional Phenotyping Study. As expected, plasma B₁₂ concentrations were positively correlated with the intake of B₁₂ from food and supplements. ⁶² All participants consumed adequate B₁₂ and nearly all were B₁₂ replete. Yet, there were notable differences in gut microbial functional capacity and fecal SCFAs levels between adequate and high consumers.

Assembly-based metagenomics revealed that approximately 20% of gut microbial genomes per subject encode a complete suite of B₁₂ biosynthetic genes, consistent with prior estimates. ⁶³ In humans, geography is associated with differences in B₁₂ biosynthetic pathway abundance, while dietary differences within a single country are not. ⁶⁴ We found that microbial B₁₂ biosynthetic potential in healthy U.S. adults was reduced in those with high B₁₂ intake. As this was a healthy cohort, the long-term implications of reduced microbial biosynthesis are unknown. Research in animals suggests that excessive dietary supplementation of B₁₂ is associated with less resilience to enteric challenge. ⁶⁵ Additionally, the participants in this study were almost entirely B₁₂ replete, and the relationship between diet and microbes would potentially differ in a markedly B₁₂-deficient state, ⁶⁶ underscoring the importance of evaluating both extremes of nutrient exposure.

We found that habitual consumption of greater than 8.51  µg per day of dietary B₁₂ and B₁₂ supplementation was associated with lower fecal SCFA concentrations. Our models controlled for soluble fiber and dietary methionine, as these factors would likely impact the available fermentable substrates for microbial production of SCFAs and are purported modulators of fecal SCFA production that are co-consumed with B₁₂ ²⁰ , respectively. The significance of the relationship between B₁₂ and fecal SCFAs despite adjustment for dietary pattern (soluble fiber, dietary methionine) suggests that isolated nutrients rather than broad dietary patterns may shape microbial metabolite production. This aligns with reports that dietary fiber interventions alone often fail to elicit sustained changes in fecal SCFA profiles. ^{67 , 68} Our results differ from murine work showing that B₁₂ supplementation had no effect on cecal SCFAs ^{65 , 69} and culture models showing no effect with B₁₂ ²³ or even positive effects with cobalamin supplementation on SCFA concentrations. ⁷⁰ However, our results are consistent with evidence that both high-methionine diets and animal-rich dietary patterns are linked to reduced SCFA production. ^{71 , 72} The inconsistent findings in animal studies could be due to the variable baseline gut microbes in those studies and the amount of fermentable substrates in their diet.

Reduced fecal SCFA concentrations are associated with states of dysbiosis in inflammatory bowel disease, ^{73 , 74} colorectal cancer, ⁷⁵ diabetes, ^76-78 kidney disease, ⁷⁹ and hypertension. ⁸⁰ SCFAs, especially butyrate, are vitally important to many aspects of human health. Given our finding that dietary B₁₂, in excess of 8.51  µg per day, is associated with lower fecal SCFAs in healthy adults, an intervention trial is sorely needed to determine if there should be an upper limit on B₁₂ consumption.

Unlike fecal SCFAs, there was no association of dietary or supplementary B₁₂ with plasma SCFAs. Others have noted that fecal SCFAs, but not plasma SCFAs, significantly associate with the gut microbiome. ^{26 , 81} This may be due to the metabolism of SCFAs by intestinal cells and the liver, resulting in low levels of SCFAs that appear in peripheral blood. ^{82 , 83} Mechanistic evidence for the effect of SCFAs on health has been mostly demonstrated by their local action in the gut, while the interplay with extraintestinal sites is less well-understood (reviewed in ⁸⁴ ). Therefore, even if higher B₁₂ intake does not have measurable effects on peripheral SCFAs, the effect of higher B₁₂ intake on gastrointestinal health and mucosal immunity via microbially-produced SCFAs is worth further pursuit.

Using machine learning models, we found fecal propionate concentrations were predictable using specific microbial metabolic pathways in participants whose B₁₂ intake ranged from 2.4—8.51 µg per day or who did not use B₁₂ supplements. Important features in a model specific to participants who did not use B₁₂ supplements were related to one-carbon metabolism that converged on methionine synthesis. We propose that the association between methionine biosynthesis pathway abundance and fecal propionate concentrations in this group arose because methionine synthase (MetH) requires B₁₂ as an enzymatic cofactor. ⁶³ In taxa that lack a B₁₂-independent alternative, luminal B₁₂ deficiency has the potential to prevent microbes from synthesizing methionine. ^{85 , 86} However, when B₁₂ is scarce, some microbes can instead synthesize methionine via a B₁₂-independent pathway (MetE) that uses a derivative of folate as a methyl donor. ⁸⁷ Thus, methionine biosynthesis and folate transformation pathway abundances may reflect the consumption of methionine from the luminal environment. The bacterial biosynthesis of methionine via folate-dependent mechanisms (MetE) may potentiate further microbial propionate production in these individuals. Further, less tetrapyrrole biosynthesis (from glycine) pathway abundance was predictive of higher fecal propionate. If this reflects a reduction in B₁₂ synthesis, it would be consistent with a shift toward folate-dependent methionine synthesis, as this pathway results in delta-aminolevulinic acid, the precursor to anaerobic B₁₂ biosynthesis. ⁸⁸ We propose that methionine biosynthesis and related pathways reflect fluxes in one-carbon metabolism, influencing the availability of intermediates that are channeled toward propionate synthesis in the microbiome.

Conversely, the machine learning models suggest that SCFA production is insensitive to microbial function when B₁₂ intake is excessive. Given that the taxonomic composition did not differ between the adequate and high intake groups, we speculate that excess dietary B₁₂ possibly suppresses intracellular microbially-produced B₁₂, which, in turn, could reduce propionate production. Bacteroides, a major gut commensal, has been shown to have sufficiently high affinity for B₁₂ that it can strip B₁₂ from human intrinsic factor, ⁸⁹ suggesting that dietary B₁₂ can be available to microbes in the colon. Secondly, bacterial riboswitches are known to regulate gene expression in response to availability of exogenous B₁₂ analogues ⁹⁰ and can suppress B₁₂ uptake ⁹¹ and synthesis. ⁹² Therefore, its possible that higher exogenous B₁₂ results in a scarcity of intracellular B₁₂, thereby reducing propionate production.

Our study has several notable strengths. First, through strategic participant recruitment, the parent observational cohort evenly covered all age groups, both sexes, and a broad range of BMIs. This balance was largely retained in the present analysis, with the exception of underrepresentation of participants in the highest BMI category (≥30 kg/m²), which we acknowledge as a limitation below. Nonetheless, the demographic breadth of these data minimized bias that may have otherwise been introduced to the study, given the influence of person-specific factors on fecal SCFA concentrations. ^{93 , 94} Second, we employed deep shotgun metagenomic sequencing, which enabled a direct assessment of microbial pathways rather than relying on inferred functions from amplicon-based sequencing. This approach provides higher taxonomic and pathway-level resolution, addressing a key limitation in the current literature. Third, we integrated metagenomic data with dietary intake data, allowing us to examine the relationships between host diet and the gut microbial community, with verification of B₁₂ status by direct measurement in plasma. Finally, we adopted a machine learning framework to explore microbial associations with fecal SCFA concentrations. Leveraging these emerging computational tools can facilitate hypothesis generation, benefitting future diet and microbiome research.

However, this study comes with notable limitations. Firstly, the observational study design precludes causal inference. We attempt to address this limitation by using multiple and generalized regression modeling to mitigate potential selection bias and confounding factors that can influence dependent outcomes. ⁹⁵ Secondly, as our sample size is limited (n = 277), caution should be exercised in generalizing our findings to other human cohorts. This work investigated a secondary objective of the parent study, which was not explicitly powered to explain the variation in SCFAs in response to dietary B₁₂. For the machine learning work, we attempt to mitigate this limitation through bootstrapping to estimate model performance variability associated with the initial test-train splits, and report the error associated with our model performance. ⁹⁶ Thirdly, in a cross-sectional study, we are limited to a single “snapshot” in time regarding our outcomes of interest. Also, while food frequency questionnaires are a cost- and time-effective way to estimate usual intake over long periods, they are subject to recall bias. ⁹⁷ Finally, the underrepresentation of participants in the highest BMI category (≥30 kg/m²) in our analysis introduces the possibility that our findings may not generalize to adults with obesity who are otherwise healthy.

In summary, our findings suggest that high B₁₂ intake, whether from foods or supplements, has fecal SCFA associations in humans, such as those observed with high methionine in animals. Consistent with evidence from animal models that demonstrate excess dietary B₁₂ can perturb the gut microbiome, our results indicate that while structural changes in microbial composition appear limited, higher B₁₂ intake may exert functional consequences related to fecal SCFA production. Future trials in humans are warranted to test the effects of a diet replete with fermentable fiber while high in B₁₂ (>8.51 µg/day) on microbial corrinoid metabolism and fecal SCFA production. As dietary guidance currently provides no upper limit on vitamin B₁₂ intake, our study suggests that such an investigation is urgently needed.

Supplementary Material

Supplemental tables and figures.docx

Supplemental tables and figures.docx

Funding Statement

This work was supported by the USDA Agricultural Research Service under Grants 2032-51000-022-00D, 2032-51530-026-000D, and 2032-10700-002-00D.

Disclosure of potential conflicts of interest

Margaret A. Hilliard has received compensation from AG1 for independent contractor services. This financial relationship is unrelated to the design, conduct, or reporting of the present research. The remaining authors report there are no competing interests to declare.

Data availability statement

Metagenomic reads are deposited in the NCBI Sequence Read Archive under two accession numbers: PRJNA795985 and PRJNA1090850. Requests for non-metagenomic data from the USDA Agricultural Research Service Western Human Nutrition Research Center (WHNRC) Nutritional Phenotyping Study used in this analysis should be made via an email to the senior WHNRC author on the publication of interest. Requests will be reviewed quarterly by a committee consisting of the study investigators. Code to reproduce our results is available in a public GitHub repository: https://github.com/margarethilliard/VB12-analysis.

Ethics approval statement

All participants provided informed consent. The study protocol was approved by the Institutional Review Board at the University of California, Davis, and was prospectively registered on Clinicaltrials.gov (NCT02367287).

Supplementary material

Supplemental data for this article can be accessed at https://doi.org/10.1080/19490976.2026.2718567.