The effect of vitamin D supplementation on the gut microbiome in older Australians – Results from analyses of the D-Health Trial

ABSTRACT

Observational studies suggest a link between vitamin D and the composition of the gut microbiome, but there is little evidence from randomized controlled trials of vitamin D supplementation. We analyzed data from the D-Health Trial, a randomized, double-blind, placebo-controlled trial. We recruited 21,315 Australians aged 60–84 y and randomized them to 60,000 IU of vitamin D₃ or placebo monthly for 5 y. Stool samples were collected from a sample of 835 participants (417 in the placebo and 418 in the vitamin D group) approximately 5 y after randomization. We characterized the gut microbiome using 16S rRNA gene sequencing. We used linear regression to compare alpha diversity indices (i.e. Shannon index (primary outcome), richness, inverse Simpson index), and the ratio of Firmicutes to Bacteroidetes between the two groups. We analyzed between-sample (beta) diversity (i.e. Bray Curtis distance and UniFrac index) using principal coordinate analysis and used PERMANOVA to test for significant clustering according to randomization group. We also assessed the difference in the abundance of the 20 most abundant genera between the two groups using negative binomial regression model with adjustment for multiple testing. Approximately half the participants included in this analysis were women (mean age 69.4 y). Vitamin D supplementation did not alter the Shannon diversity index (mean 3.51 versus 3.52 in the placebo and vitamin D groups, respectively, p = 0.50). Similarly, there was little difference between the groups for other alpha diversity indices, the abundance of different genera, and the Firmicutes-to-Bacteroidetes ratio. We did not observe clustering of bacterial communities according to randomization group. In conlusion, monthly doses of 60,000 IU of vitamin D supplementation for 5 y did not alter the composition of the gut microbiome in older Australians.

KEYWORDS:

• Gut microbiome
• vitamin D supplementation
• randomized controlled trial

Introduction

Microbiota are communities of microorganisms that co-exist with the host ecosystem in a specific environment. The term microbiome refers to the microbial genome. Current metagenomic sequencing has identified more than 10 million microbial genes in the human gut microbiome.¹

The gut microbiome plays an important role in health and disease; indeed, it is sometimes referred to as an organ in its own right. Studies have suggested that the gut microbiome influences the interaction between the central and enteric nervous systems (the gut-brain axis) (reviewed by Carabotti et al.²), with effects on health outcomes such as depression. Characteristics of an unhealthy gut microbiome (e.g., dysbiosis) have also been linked to several diseases, such as irritable bowel syndrome, inflammatory bowel disease, diabetes, and cardiovascular disease.³ Recent studies suggest a role of the gut microbiome in modulating the functions of the immune system (reviewed by Shreiner et al. and Yamamoto and Jorgensen),^4,5 indicating a possible effect on risk or severity of infection. For example, some components or metabolites of gut bacteria (e.g. polysaccharide A and short-chain fatty acids) promote anti-inflammatory responses.⁵

The composition of the gut microbiome is diverse, both within and between individuals, and it is challenging to characterize a healthy gut microbiome. However, it is generally accepted that gut microbial communities with greater diversity, richness, and stability, and a higher relative abundance of species associated with production of short-chain fatty acids are associated with better health outcomes.⁶

There has been increasing interest in the potential effect of vitamin D on the composition of the gut microbiome. The active form of vitamin D (calcitriol, 1,25(OH)₂D) binds to macrophages and induces the production of antimicrobial peptides, resulting in bacterial killing.⁷ Rodent studies have shown that vitamin D deficiency or a lack of vitamin D receptor is associated with an increase in both Bacteroidetes and Proteobacteria phyla, which are generally considered less healthy (reviewed by Yamamoto and Jorgensen).⁵ Vitamin D also helps maintain the physical and functional integrity of the gut mucosal barrier by reducing permeability of epithelial cells and modulating tight-junction proteins,⁵ preventing invasion of pathogenic bacterial species.

A systematic review of studies in humans found a link between vitamin D and the composition of the gut microbiome as measured by diversity and the abundance of different bacteria.⁸ Subsequent to this review, a non-randomized pre-post interventional study (n = 80) in healthy women found significant increases in gut microbial diversity and the abundance of health-promoting probiotic taxa, Akkermansia and Bifidobacterium, after 12 weeks of supplementation with 50,000 international units (IU) per week of vitamin D.⁹ We identified only three small randomized controlled trials (RCTs) assessing the effect of vitamin D supplementation on the gut microbiome,^10–12 two of which were published subsequent to the systematic review. All had a sample size ≤26 and intervention duration ≤16 weeks; one study was conducted in adults with cystic fibrosis,¹¹ one among overweight or obese adults,¹² and the third in healthy adults.¹⁰ These studies found some positive effects of vitamin D supplementation or of high versus low dose supplementation on some beneficial bacteria; however, effects on the diversity indices were mixed.

In light of the paucity of data from RCTs, we aimed to investigate the effect on the gut microbiome of supplementing older adults with 60,000 IU of vitamin D per month for five y, using a subsample (n = 835) of participants recruited from the large population-based D-Health Trial.

Results

Participant characteristics and composition of the gut microbiome

We invited 1130 D-Health participants to participate in the microbiome sub-study; 982 (87%) expressed interest and 880 (90% of those who agreed to participate) were eligible. We received 835 stool samples [417 (95%) of those eligible in the placebo group and 418 (95%) of those eligible in the vitamin D group] (Figure 1). The mean age of the participants included in this analysis was 69.4 y and approximately 50% of the participants were women. There was some imbalance in self-reported ancestry (p = 0.03), with 96.1% of the vitamin D group reporting British or European ancestry, compared with 91.7% of the placebo group (Table 1). All other baseline characteristics were well balanced between the two groups (Table 1). In blood samples collected from D-Health Trial participants (i.e., not necessarily those included in this analysis), the mean 25(OH)D concentration in the placebo and vitamin D groups was 77 (SD 25) and 115 (SD 30) nmol/L, respectively.

Figure 1. Participant flow (CONSORT flow diagram).

Table 1. Baseline characteristics according to the randomization group among included participants (n = 835).

Characteristic	N (%)		P value^a
	Placebo (N = 417)	Vitamin D (N = 418)
Age (years)			0.97
60–64	107 (25.7)	109 (26.1)
65–69	109 (26.1)	112 (26.8)
70–74	129 (30.9)	123 (29.4)
≥75	72 (17.3)	74 (17.7)
Sex			0.97
Men	207 (49.6)	208 (49.8)
Women	210 (50.4)	210 (50.2)
Body mass index (kg/m^2)			0.96
<25	132 (31.7)	133 (31.8)
≥25	285 (68.3)	285 (68.2)
Self-reported ancestry			0.03
British/European	375 (91.7)	393 (96.1)
Australian/New Zealander	20 (4.9)	10 (2.4)
Asian/Indigenous/Mixed/Other	14 (3.4)	6 (1.5)
Missing	8	9
Smoking history			0.62
Never	249 (60.4)	248 (59.6)
Ex-smoker	150 (36.4)	159 (38.2)
Current	13 (3.2)	9 (2.2)
Missing	5	2
Alcohol consumption (drinks/week)			0.23
<1	99 (24.4)	92 (22.4)
1 to 7	176 (43.5)	195 (47.6)
>7 to 14	73 (18.0)	82 (20.0)
>14	57 (14.1)	41 (10.0)
Missing	12	8
Self-rated overall health			0.96
Excellent or very good	254 (61.4)	255 (61.9)
Good	138 (33.3)	134 (32.5)
Fair or poor	22 (5.3)	23 (5.6)
Missing	3	6
Self-rated quality of life			0.96
Excellent or very good	301 (73.4)	303 (73.9)
Good	93 (22.7)	90 (22.0)
Fair or poor	16 (3.9)	17 (4.1)
Missing	7	8
Predicted 25(OH)D concentration (nmol/L)			0.69
<50	91 (21.8)	96 (23.0)
≥50	326 (78.2)	322 (77.0)
Antidiabetic therapy user*^b			0.39
No	355 (93.4)	371 (94.9)
Yes	25 (6.6)	20 (5.1)
Missing	37	27
Cardiovascular drug user*^c			0.51
No	162 (42.6)	176 (45.0)
Yes	218 (57.4)	215 (55.0)
Missing	37	27
Antidepressant user*^d			0.10
No	331 (87.1)	355 (90.8)
Yes	49 (12.9)	36 (9.2)
Missing	37	27
Number GP visits first quarter*^e			0.11
≤1	214 (52.6)	234 (58.2)
>1	193 (47.4)	168 (41.8)
Missing	10	16

^aP value from chi-squared test.

*Variables were derived using Pharmaceutical Benefits Scheme (PBS) and Medicare Benefits Schedule (MBS) data within three months after randomization as indicators of baseline use.

^bPBS items with ATC code starting with “A10”.

^cPBS items with ATC code starting with “C01-C10”.

^dPBS items with ATC code starting with “N06A” or “N06CA”.

^eMBS records with Item Category = A1 General Practitioner.

Among the 835 participants, we defined a total of 2617 OTUs. Firmicutes and Bacteroidetes phyla represented more than 80% of the total sequencing reads; the 20 most abundant genera are shown in Figure 2. In our observational analysis, men and those with BMI <25 kg/m² had significantly higher mean alpha diversity indices than women and people who were overweight or obese (Table S1).

Figure 2. Relative abundance of the 20 most abundant genera, presented according to randomization group.

Note. Genera sorted from the most common to the least common

Effects of vitamin D supplementation on composition of the gut microbiome

Vitamin D supplementation for 5 y did not alter measures of alpha diversity or the Firmicutes-to-Bacteroidetes ratio (Table 2, Figure 3). There was negligible change in results for the Shannon diversity index in sensitivity analyses, restricted to participants who reported the following: (1) British/European ancestry; or (2) British/European/Australian/New Zealand ancestry (Table S2). We did not find an effect of vitamin D supplementation on the beta diversity of the gut microbiome as measured by Bray-Curtis and UniFrac distances (p values 0.10 and 0.61, respectively) (Figure 4), or on the abundance of different genera (Table 3). There was a lower abundance of genus Bacteroides in the vitamin D group, although this was not statistically significant after adjustment for multiple testing (fold difference 0.91; 95% CI 0.84–0.98, adjusted p value = 0.15).

Figure 3. Boxplots of alpha diversity indices (Shannon, richness, and inverse Simpson) and Firmicutes to Bacteroidetes ratio according to randomization group.

Note: Dots are for outliers

Figure 4. The effect of vitamin D supplementation on beta diversity of the gut microbiome.

*PERMANOVA used to test for significant clustering. The models included age at randomization, sex, and state of residence at randomization.

Table 2. The effect of vitamin D supplementation on gut microbiome indices.

	Mean (95% CI)		Mean difference^a (95% CI)	P-value^a
	Placebo (N = 417)	Vitamin D (N = 418)
Shannon diversity	3.51 (3.48, 3.54)	3.52 (3.49, 3.56)	0.02 (−0.03, 0.06)	0.50
Richness	358.2 (354.2, 362.3)	361.6 (357.6, 365.6)	3.34 (−2.23, 8.92)	0.24
Inverse Simpson’s index	17.3 (16.6, 18.0)	17.7 (17.0, 18.4)	0.42 (−0.53, 1.37)	0.38
Firmicutes/Bacteroidetes	1.5 (1.4, 1.6)	1.6 (1.5, 1.7)	0.11 (−0.02, 0.25)	0.10

^aEstimated using linear regression; models included age, sex, and state of residence at randomization.

Abbreviation: CI = confidence interval.

Table 3. The effect of vitamin D supplementation on the abundance of different genera.

	Mean relative abundance (%) (95% CI)		Fold difference (95% CI)*	Adjusted p-value^#
Genus	Placebo (N = 417)	Vitamin D (N = 418)
Akkermansia^1	1.9 (1.6, 2.2)	1.8 (1.5, 2.1)	0.97 (0.78, 1.21)	0.92
Alkaliphilus^2	1.3 (1.2, 1.4)	1.3 (1.2, 1.4)	0.93 (0.83, 1.04)	0.51
Bacteroides^3	25.5 (24.5, 26.5)	23.7 (22.7, 24.7)	0.91 (0.84, 0.98)	0.15
Bifidobacterium^4	1.1 (0.9, 1.3)	0.9 (0.7, 1.1)	0.87 (0.72, 1.04)	0.39
Blautia^2	9.6 (9.2, 10.0)	10.0 (9.6, 10.4)	1.02 (0.95, 1.08)	0.82
Caloramator^2	0.8 (0.7, 0.9)	0.9 (0.7, 1.0)	1.09 (0.91, 1.31)	0.65
Clostridium^2	2.5 (2.4, 2.6)	2.6 (2.5, 2.8)	1.04 (0.98, 1.11)	0.46
Coprococcus^2	2.6 (2.4, 2.7)	2.5 (2.4, 2.7)	0.97 (0.89, 1.06)	0.78
Escherichia^5	2.6 (2.1, 3.1)	2.9 (2.4, 3.4)	1.11 (0.86, 1.43)	0.68
Faecalibacterium^2	8.7 (8.3, 9.1)	8.7 (8.3, 9.1)	0.99 (0.91, 1.08)	0.94
Flavobacterium^3	2.6 (2.4, 2.8)	2.5 (2.2, 2.7)	0.91 (0.81, 1.01)	0.32
Lachnospira^2	2.1 (2.0, 2.3)	2.4 (2.2, 2.5)	1.07 (0.97, 1.18)	0.46
Oscillospira^2	3.2 (3.0, 3.4)	3.2 (3.0, 3.4)	0.98 (0.92, 1.05)	0.83
Parabacteroides^3	3.0 (2.8, 3.2)	3.1 (2.9, 3.3)	1.03 (0.93, 1.14)	0.80
Phascolarctobacterium^2	1.1 (1.0, 1.1)	1.1 (1.0, 1.2)	0.98 (0.83, 1.15)	0.91
Prevotella^3	3.5 (2.8, 4.2)	4.3 (3.6, 5.0)	1.23 (0.96, 1.59)	0.36
Pseudomonas^5	1.6 (1.1, 2.1)	1.9 (1.5, 2.4)	1.24 (0.89, 1.71)	0.49
Roseburia^2	2.6 (2.4, 2.8)	2.5 (2.3, 2.7)	0.94 (0.82, 1.07)	0.62
Ruminococcus^2	5.9 (5.7, 6.2)	5.9 (5.6, 6.2)	0.97 (0.91, 1.04)	0.65
Sutterella^5	1.2 (1.1, 1.3)	1.1 (1.0, 1.2)	0.91 (0.80, 1.03)	0.40

Note: *Fold difference compared vitamin D to the placebo group.

*Estimated using DESeq2 R package; p-values were adjusted for multiple comparisons using the Benjamini and Hochberg false discovery rate method.

1 Phylum Verrucomicrobia 2 Phylum Firmicutes 3 Phylum Bacteroidetes 4 Phylum Actinobacteria 5 Phylum Proteobacteria.

Abbreviation: CI = confidence interval.

The effect of vitamin D supplementation on the Shannon diversity index was not modified by age, sex, BMI, or predicted baseline 25(OH)D concentration (Figure 5). In an exploratory analysis, the negative effect of vitamin D supplementation on genus Bacteroides was seen only in people with predicted baseline 25(OH)D concentration ≥50 nmol/L (fold difference 0.89; 95% CI 0.81–0.98, adjusted p value = 0.13, Table S3).

Figure 5. The effect of vitamin D supplementation on the Shannon diversity index, overall and in subgroups of participants.

Note. Estimates from linear regression; models included age at randomization, sex, and state of residence at randomization. The P value for interaction was from a likelihood ratio test that compared models with and without interaction between the randomization group and the subgroup variable.

Discussion

Analysis of data from the D-Health Trial did not find an effect of monthly doses of 60,000 IU vitamin D over 5 y on the composition of the gut microbiome of older Australians overall or in pre-specified subgroup analyses.

The gut microbiome profile of our participants is in line with findings from previous studies. Bacteroides were the most abundant genus in the gut microbiome of our participants; the Bacteroides enterotype is common in Western populations (e.g., Spain, Argentina, and the United States).¹³ The typical Firmicutes-to-Bacteroidetes ratio varies across samples from different countries (ranging from 0.19 to 3.87),¹³ and the mean ratio among our participants was 1.55. In our observational analysis, men and those with BMI <25 kg/m² had higher mean alpha diversity indices than women and those who were overweight or obese. We did not find any significant association with age. Current evidence regarding the associations between alpha diversity and age, sex, and BMI is inconclusive. A recent systematic review identified six studies that had reported the alpha diversity in the gut microbiome across age groups and found higher alpha diversity in the oldest age group compared with younger adults. There are limited data on sex differences, but previous studies reported non-significantly higher mean alpha diversity indices in men than in women.^14,15 A recent systematic review and meta-analysis did not find an overall significant difference in alpha diversity between obese and non-obese adults.¹⁶ Fourteen of the 22 identified studies reported lower alpha diversity (Shannon index) in obese adults, but the other eight found either a higher mean Shannon index or no difference between the obese and non-obese groups.¹⁶

We did not find an effect of vitamin D supplementation on either alpha or beta diversity of the gut microbiome. Previous studies in humans have yielded mixed results. Our results are similar to one RCT, which found that a 100,000 IU loading dose followed by 4000 IU of vitamin D daily for 16 weeks did not have an effect on either alpha or beta diversity in overweight or obese adults (N = 26).¹² A randomized trial among 20 adults with 25(OH)D concentration <75 nmol/L, in which vitamin D was administered at 600 IU, 4000 IU, or 10,000 IU daily for 8 weeks, found that alpha diversity did not differ between the study arms.¹⁰ In contrast, a RCT performed in adults with cystic fibrosis who had serum 25(OH)D concentration <75 nmol/L (N = 23) found that 12 weeks of supplementation with 50,000 IU/week of vitamin D resulted in a shift toward a healthier microbiome, with significant clustering according to the treatment group.¹¹ One non-randomized study found that supplementing women with serum 25(OH)D concentration <50 nmol/L (N = 80) with 50,000 IU of vitamin D weekly for 12 weeks increased the richness of the gut microbiome.⁹ Another non-randomized study also found increased richness in the mucosal microbiome of the upper gastrointestinal tract after 8 weeks of supplemental vitamin D equivalent to 980 IU/kg/week in 16 healthy adults.¹⁷ Overall, current evidence regarding the effect of vitamin D supplementation on the diversity of the gut microbiome is inconclusive, but there is limited evidence to support the benefit of vitamin D, particularly in light of the findings of the D-Health Trial.

We found little evidence for an effect of vitamin D supplementation on the abundance of different genera. There was some suggestion of a reduced abundance of genus Bacteroides in the vitamin D group, but this was not statistically significant after false discovery rate correction. Previous studies have suggested an effect of vitamin D supplementation on the bacterial composition of the gut microbiome, but the results were inconsistent in terms of which bacterial species was enhanced or inhibited by vitamin D. Two studies (one RCT and one non-randomized pre/post interventional study) found some indication of a positive link between vitamin D and the health-promoting probiotic taxa, Akkermansia, and that vitamin D supplementation decreased the Firmicutes-to-Bacteroidetes ratio.^9,10 The effect of vitamin D supplementation on the abundance of Bacteroidetes differed in two non-randomized pre/post interventional studies; one found an increase of Bacteroides in the upper gastrointestinal tract after 8 weeks of supplemental vitamin D equivalent to 980 IU/kg/week (N = 16)¹⁷ and the other found a reduction in stool samples after 9 weeks of 50,000 IU/week (N = 50).¹⁸ Two RCTs found some benefit of vitamin D supplementation in promoting other beneficial genera such as Lachnospira¹² and Lactococcus,¹¹ and there were reductions in the abundance of genera Blautia,¹² Veillonella, and Erysipelotrichaceae.¹¹

The D-Health Trial is the largest RCT to have investigated the effect of vitamin D supplementation on the composition of the gut microbiome, but some limitations should be noted. First, we restricted the study to participants who satisfied a number of eligibility criteria, potentially subverting the randomization and introducing bias. However, with the exception of slight differences in the distribution of self-reported ancestry, baseline characteristics were very well balanced between the two groups, suggesting that any bias was minimal. Second, we did not measure dietary patterns or gut microbiome profile at baseline so could not adjust for these factors in our study. Diet is known to influence the gut microbiome, and both may be associated with ancestry. However, accounting for self-reported ancestry made a negligible difference to results, and given the balance in other characteristics, we expect that the distribution of these factors would have been similarly well balanced. Third, the use of 16S rRNA sequencing that depends largely on primers, targeted variable regions, and PCR amplification can lead to variability in the results.¹⁹ We also did not measure gut metabolites, so were unable to assess the effect of vitamin D on these. Finally, we used the predicted instead of measured baseline 25(OH)D concentration. As the positive predictive value for 25(OH)D concentration <50 nmol/L was modest (23%), misclassification of participants’ baseline vitamin D status may have attenuated any effect in people predicted to have low vitamin D status at baseline.

The D-Health participants were largely vitamin D replete. The mean 25(OH)D concentration in the placebo group was a little higher than that reported in the 2011/2012 Australian Health survey (77 versus 69 nmol/L),²⁰ but this may be due to differences in the geographic distribution of participants or the timing of blood sampling. While our results are likely to be reasonably generalizable to the Australian population and other groups with a low prevalence of vitamin D deficiency, these findings cannot be used to infer the effect on the microbiome of treating vitamin D deficiency.

In conclusion, monthly doses of 60,000 IU vitamin D over 5 y did not alter the composition of the gut microbiome in a population that is largely vitamin D replete. Further investigation is needed to examine whether non-bolus doses of vitamin D would have an effect on the gut microbiome or whether vitamin D supplementation would be beneficial in populations with a higher prevalence of vitamin D deficiency.

Methods

Trial design, participants, and intervention

The D-Health Trial was a randomized, double-blind trial with two parallel arms. Details of the trial methods have been published previously.²¹ We recruited Australians aged between 60 and 79 y using a population register as the sampling frame. Volunteers aged 60–84 y were also included. People with current or previous diagnoses of hypercalcemia, hyperparathyroidism, kidney stones, osteomalacia, or sarcoidosis, or who were taking more than 500 IU of supplemental vitamin D per day were not eligible. Participants were randomly allocated (1:1 ratio) to monthly doses of either 60,000 IU of cholecalciferol (vitamin D₃) or matching placebo for up to 5 y. Vitamin D and placebo were manufactured by Lipa Pharmaceuticals Pty Ltd, and the same product was used throughout the intervention. We posted a 12-month supply of tablets to participants annually. Participants were asked to take tablets on the first day of each month, and we sent reminders via e-mail, mobile phone text message, or automated landline message on the last day of each month. The intervention phase was between January 2014 and February 2020. The primary outcome of the trial was all-cause mortality,²² with secondary outcomes being total cancer incidence and colorectal cancer incidence.²³

We used automated computer-generated permuted block randomization, stratified by age (60–64, 65–69, 70–74, 75+ y), sex, and state of residence (New South Wales, Queensland, South Australia, Tasmania, Victoria, Western Australia). Staff and investigators did not have access to the allocation list. The QIMR Berghofer Medical Research Institute Human Research Ethics Committee approved the trial and all participants gave written or online consent to participate.

The D-Health Trial is registered on the Australian New Zealand Clinical Trials Registry: ACTRN12613000743763. The QIMR Berghofer Medical Research Institute Human Research Ethics Committee approved the trial and all participants gave written or online consent to participate.

Baseline characteristics

We asked participants to complete a survey at baseline that asked about socio-demographic characteristics, lifestyle factors, and history of illness. We did not measure serum 25(OH)D concentrations at baseline. Rather, we predicted whether the deseasonalised baseline serum 25(OH)D concentration was ‘low’ [<50 nmol/L] using data collected from placebo group participants who provided a blood sample during the trial; the area under the receiver operating characteristic curve was 0.71 (95% confidence interval (CI) 0.63–0.78)].²⁴

Monitoring adherence and adverse events

In surveys administered annually, we asked participants to report the number of study capsules they had taken in the previous year and their use of off-study supplements containing vitamin D. In addition, we asked some participants, selected randomly each year, to provide a blood sample at the time they completed their annual survey. Participant selection for blood sampling was stratified by randomization group, age at randomization, sex, state of residence at randomization, and month of annual survey. We measured serum 25(OH)D concentrations to monitor the effect of the intervention on serum 25(OH)D concentration. Adverse events were reported via telephone or e-mail. We also captured diagnoses of hypercalcemia, kidney stones, and hyperparathyroidism in annual surveys.

Participants and eligibility criteria for the microbiome sub-study

Between 2018 and 2019, we invited a sample of D-Health participants to take part in the gut microbiome sub-study. The number of samples we were able to collect and analyze was limited due to budgetary constraints. We estimated that with a sample size of approximately 800 we would have 80% power to detect a difference of 0.05 in the Shannon diversity index at p < 0.05. To avoid reductions in power due to noncompliance and contamination, we randomly selected participants (within strata defined by randomization group, age, and sex) from among those who: (1) had not withdrawn prior to the final 4 months of their intervention period and (2) reported, on the annual survey prior to selection (the 4^th annual survey), taking 10–12 study capsules and ≤500 IU/day of off-study supplementary vitamin D. We sampled equal proportions of men and women. After selection, eligibility was further restricted to those who: (a) did not have inflammatory bowel disease; (b) had not taken antibiotics in the previous month; (c) were not taking any systemic immune suppressants; and (d) did not report increasing their supplementary vitamin D intake to >500 IU/day.

Sample collection and DNA sequencing

Participants who agreed to participate in the gut microbiome study were sent a sample collection swab (aCopan floq swab in an active drying tube). Stool samples were collected by participants and returned to the study center by mail, where they were stored at −20°C. The samples were sequenced using 16S rRNA gene sequencing (variable regions V3 and V4 were amplified with polymerase chain reaction) on the Illumina MiSeq platform. We used the QIIME pipeline for quality control and data processing.²⁵ The taxonomic classification was performed using the 16S rRNA gene database from Greengenes (http://greengenes.lbl.gov)²⁶ version 13_8. The operational taxonomic units (OTUs) were identified using a pre-trained classifier (trained at 97% OTU full-length sequences) against the Greengenes database.²⁶

Microbiome outcomes

We quantified alpha (within-sample) diversity using the Shannon index (primary outcome for this analysis), the inverse Simpson index, and richness. We used two distance metrics (Bray-Curtis and Unweighted UniFrac) to assess beta diversity (divergence in community composition between samples). We also evaluated the ratio of the Firmicutes to Bacteroidetes phyla, and the relative abundance of the 20 most abundant genera among all participants.

Blinding

In March 2020, at the end of the intervention period, participants were notified of their study group. Study staff, analysts, and investigators remained blinded until analyses of the primary outcome (mortality) were completed. Analyses for the microbiome study were performed blinded to the study group allocation following a pre-specified analysis plan. We developed statistical code to automatically generate all tables and figures using a dataset in which group allocation and participant identifiers had been removed and participants randomly assigned to two groups of equal size. We applied the code to the original data once it had been verified, and all investigators had approved the sample tables and figures. Any analyses performed thereafter are declared as exploratory.

Statistical methods

All analyses were performed in R version 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria) and SAS version 9.4 (SAS Institute, Inc., Cary, NC) and followed the intention-to-treat principle. To minimize inherent bias in amplicon sequencing, sequencing data were normalized using a random subsampling or rarefaction on OTU count. The samples were rarefied at a depth of 50,280 sequences.

Differences in baseline characteristics between the two groups were assessed using chi-squared tests. We compared alpha diversity indices and the Firmicutes/Bacteroidetes ratio between the two groups using linear regression and Wilcoxon signed rank tests. We analyzed beta diversity using principal coordinate analysis and used PERMANOVA to test for significant clustering according to randomization group. All estimates were adjusted for age, sex, and state of residence at randomization. For the analysis of the abundance of the 20 most abundant genera, we used the DESeq2 R package, which applied a negative binomial regression model with shrinkage estimation for dispersions and fold changes;²⁷ p-values were adjusted for multiple testing using the Benjamini and Hochberg false discovery rate method. We also assessed the associations between alpha diversity indices and selected baseline characteristics using linear regression; all models included age at randomization and sex.

Subgroup and exploratory analyses

We assessed whether the effect of vitamin D supplementation on our primary outcome (Shannon diversity index) varied across pre-specified subgroups of: (a) age at randomization (<70 versus ≥70 y); (b) sex (men versus women); (c) body mass index (BMI) at randomization (<25 versus ≥25 kg/m²); and (d) predicted baseline serum 25(OH)D concentration (<50 versus ≥50 nmol/L). We calculated the mean and standard deviation for each alpha diversity index within categories of selected baseline characteristics.

Informed by the results of a potential effect of vitamin D supplementation on genus Bacteroides, we performed an exploratory analysis that was not pre-specified, in which we further assessed whether the effect of vitamin D supplementation on the abundance of genus Bacteroides differed by the predicted baseline serum 25(OH)D concentration (<50 versus ≥50 nmol/L). We also performed two exploratory sensitivity analyses to account for the observed imbalance in self-reported ancestry, in which we assessed the effect of vitamin D supplementation on the primary outcome (i.e. Shannon diversity index) in data restricted to people who reported: (1) British/European ancestry; and (2) British/European or Australian/New Zealander ancestry.

Authors’ contributions

DW, PW, GH, DE, MK, RO, JV, AV, CB, BDR, PE, DM, BA, and RN designed the trial. CB, BDR, MW, DM, FH, and RN were involved in recruitment, data collection, and curation. HP, MW, FH, and RN carried out the investigations and formal analysis. HP wrote the first draft of the report with input from MW, FH, and RN. All authors approved the pre-specified analysis plan and were involved in reviewing and editing the final draft. MW, FH, and RN provided supervision.

Availability of data and materials

The data that support the findings of this study are not openly available. The data are available upon reasonable request and subject to approval.

List of abbreviations

BMI = body mass index

CI = confidence interval

IU = international units

OTU = operational taxonomic units

RCT = randomized controlled trial

Acknowledgments

We would like to acknowledge the D-Health Trial staff and members of the Data and Safety Monitoring Board (Patricia Valery, Ie-Wen Sim, and Kerrie Sanders), and Services Australia for supplying Pharmaceutical Benefits Scheme (PBS) and Medicare Benefits Schedule (MBS) data. We also extend our thanks to the D-Health Trial participants who committed to this research.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19490976.2023.2221429

Additional information

Funding

This work is supported by the National Health and Medical Research Council (NHMRC) project grants [Grant numbers GNT1046681, GNT1120682]. PR Ebeling, RE Neale, PM Webb, and DC Whiteman are/were supported by fellowships from the NHMRC [Grant numbers GNT1197958, GNT1060183, GNT1173346, GNT1155413]. DSA McLeod is supported by a Metro North Clinician Research Fellowship and a Queensland Advancing Clinical Research Fellowship. H Pham is supported by a University of Queensland PhD Scholarship. The vitamin D assays were performed at the University of Western Australia, supported by infrastructure funding from the Western Australian State Government in partnership with the Australian Federal Government, through Bioplatforms Australia and the National Collaborative Research Infrastructure Strategy (NCRIS).