figure b

Introduction

In our previous work, we observed a borderline yet detectable effect of gluten-free diet (GFD) on residual beta cell capacity, metabolic control and the length of the partial remission period in children shortly after type 1 diabetes onset [1]. It is presently not known whether such effects might be mediated through the composition or function of the human gut microbiome (which changes upon introduction of this diet [2]), or through the gut permeability (affected by both gluten [3] and microbiota [4]).

The aim of the present study was to investigate samples and data obtained in our previous interventional trial [1] and to test whether GFD is associated with changes in gut bacteriome composition, in its function and in intestinal permeability, and whether such changes mediate the previously observed effects of GFD on the pace of beta cell loss.

Methods

Forty-five children (aged 10.2 ± 3.3 years) were recruited into a self-selected intervention trial primarily focused on determining the role of GFD on beta cell preservation (ClinicalTrials.gov NCT02867436). The detailed study population characteristics have been described previously [1]. This was a non-randomised dietary interventional trial using GFD in children recently diagnosed with type 1 diabetes. The study was completed per protocol by 22 GFD-compliant participants and 19 control participants (electronic supplementary material [ESM] Methods 1.1 and ESM Table 1). The primary study outcome was the AUC of the C-peptide levels in the standard mixed-meal tolerance test (MMTT) [5], assessed at 6 and 12 months and expressed as the percentual decline relative to the baseline test. In an adjusted longitudinal analysis of all three time points, the C-peptide AUC declined more slowly in the GFD group than in the control participants.

To explore the faecal bacteriome as a mediator of the above-mentioned effect of the GFD, we first: (1) explored associations of GFD with changes in the faecal bacteriome community composition and with the faecal metabolome; then (2) tested whether the observed changes are associated with the pace of beta cell loss (the hypothetical taxa or metabolites being a vertex of the mediator effect analysis in ESM Fig. 1a); and then (3) if such outcome-associated taxa or metabolites had been found, formal statistical testing was planned of the mediated (i.e. indirect, intermediate) effect of the bacteriome on the pathway between GFD and beta cell loss (ESM Fig. 1b). The collection of samples, extraction of DNA, PCR amplification of the V4 region of the 16S rDNA gene, amplicon sequencing, bioinformatic and statistical analyses were performed according to published protocols [6, 7], and are detailed in ESM Methods 1.2. The stool sample metabolome was explored by NMR spectroscopy: 1H NMR spectra were recorded and processed as previously described [8] and detailed in ESM Methods 1.3.

The intestinal permeability was assessed by two different methods: (1) urine lactulose to mannitol (L/M) excretion ratio was measured at the 12-month visit as described by Sequeira et al [9], using high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC/PAD) on a CarboPac MA1 column (Dionex, USA); and (2) as a presumed marker of human intestinal permeability [10], the human protein zonulin was tested in serum samples collected at visits at 1, 6 and 12 months using Human Zonulin ELISA kit (PromoCell, Heidelberg, Germany).

Nutrient intake and intervention adherence were assessed using 3-day nutritional questionnaires collected at each visit. The daily nutrient and energy intake were assessed by NutriServis PROFI software (Forsapi, Prague, Czechia), and compared between intervention and control participants using two-sample Welch t tests at 6 and 12 months to ensure that the GFD intervention was not accompanied by other, unwanted changes in the diet. Full dietary data were available for 32 compliant participants (15 GFD group, 17 control group) at 6 months and 30 (13 GFD group, 17 control group) at 12 months.

Every participant was tested for protocol adherence at three or more time points: the presence of faecal gluten immunogenic peptides (GIP) was assessed using the iVylisa GIP-S kit (Biomedal, Sevilla, Spain) [11]. Participants were not specifically reminded about this testing, although it had been covered in the written informed consent. The testing was performed retrospectively, and GIP positivity in any sample taken while on the GFD was considered non-compliance and resulted in exclusion of the child from the statistical analysis.

The study followed the STROBE guidelines [12] and was approved by the institutional Ethics Committee. The study protocol was registered at ClinicalTrials.gov (NCT 02867436). Written informed consent was granted by all participants’ parents/caregivers. The parents/caregivers of the participants in the intervention group were financially reimbursed for the added costs of the GFD.

Results

Participants and their samples for bacteriome and metabolome

Of the 45 children participating in the study, four in the GFD intervention group were positive for GIP in stool, and were consequently excluded from the per-protocol analysis owing to non-adherence. In the control group, detectable amounts of GIP were present in all of the tested stool samples. The remaining 41 study participants were compliant with the intervention protocol (22 from the GFD group, 19 from the control group; detailed in ESM Results 2.1). Here we report analyses of 240 stool samples collected throughout the study.

GFD intervention and the gut bacteriome

The alpha (within-sample) diversity of bacteriomes did not appreciably change upon applying GFD (ESM Fig. 2). Analyses of the overall bacteriome community patterns did not reveal appreciable changes upon GFD intervention, as assessed by transformation-based redundancy analysis, non-metric distance-based ordination or permutational multivariate ANOVA (ESM Results 2.2).

Individual taxonomic components of the bacteriome were tested in two tiers: first, we used sensitive DESeq2 modelling, then candidate taxa identified therein were confirmed in models for longitudinal data (details in ESM Results 2.3). Bifidobacterium genus decreased 3.3-fold while on GFD (Fig. 1a,b; adjusted p=1.4 × 10−4 in a DESeq2 model, p=0.026 in generalised estimating equations; this association projected upwards to phylum Actinobacteria), whereas Roseburia increased 2.4-fold on GFD (Fig. 1c,d; adjusted p=0.02 in DESeq2 model, p=0.002 in generalised estimating equations).

Fig. 1
figure 1

Bacterial genera with significant changes in their relative abundance during the intervention with GFD. (a, b) Changes in the relative abundance of genus Bifidobacterium in the intervention group (a) and control group (b). (c, d) Changes in the relative abundance of genus Roseburia in the intervention group (c) and control group (d)

Full size image

The gut bacteriome and beta cell residual function

If the previously observed effects of GFD on beta cell preservation were mediated by the bacteriome, then not only would the bacteriomes change upon the introduction of GFD, but these bacteriome changes would also be associated with the preservation of beta cell function. This was not the case: neither Bifidobacterium (p=0.86) nor Roseburia (p=0.83) were associated with the pace of decline of the beta cell function. Nor was there any convincing association with the residual beta cell function among the remaining taxa: nominally significant association signals were noted for the Bacteroidaceae family (nominal uncorrected p=0.013) whose positive association with better beta cell function was likely due to such associations of the genus Bacteroides (p=0.013, with the strongest contribution from amplicon sequence variants belonging to B. faecis or B. thetaiotaomicron). Further suggestive direct association was noted for Odoribacter (p=0.011) and its family Marinifilaceae (p=0.0063). The quantity of these taxa is plotted against the percentual preservation of beta cell function in Fig. 2a,b. The beta cell preservation was not associated with alpha diversity of the bacteriome (ESM Fig. 3) nor with overall bacteriome community patterns (ESM Fig. 4 and ESM Results 2.4). The above-mentioned body of evidence together indicates that the composition of the bacteriome did not act as an intermediate in the previously detected link between GFD and beta cell preservation.

Fig. 2
figure 2

Bacterial genera with the strongest direct association with beta cell residual function. (a) Mean relative abundance of genus Bacteroides against residual beta cell capacity in the intervention or control group. (b) Mean relative abundance of genus Odoribacter against residual beta cell capacity in the intervention or control group

Full size image

The gut metabolome

The NMR gut metabolome analysis revealed the presence of 31 compounds, whereas some of the other metabolites normally found in stool were overlapping with glycerol, which was added as cryoprotectant. There was a highly significant correlation between the distances of sample metabolomes, and the respective distances of bacteriomes (r=0.30, p<0.001 in the Mantel test; co-inertia plots in ESM Fig. 5).

The profile of the measured metabolites did not significantly change upon the introduction of GFD (predictors fitted onto principal component analysis of Euclidean distances among the scaled metabolite levels, p=0.11 for the GFD term, adjusted for the category of time since type 1 diabetes onset). The beta cell preservation categories were not associated with the overall metabolome profiles (p=0.57), with individual metabolites, or with the potential functional capacity of the bacteriome inferred by the PICRUSt program (ESM Results 2.5).

Intestinal permeability

The L/M excretion ratio at study month 12 did not differ significantly between the study groups. Similarly, no difference was noted for serum zonulin levels in any of the analysed time points (ESM Results 2.6 and ESM Fig. 6). There was no significant correlation between clinical variables or beta cell residual function and L/M excretion ratio (ESM Table 2) or zonulin levels (ESM Table 3).

Nutrient intake analysis

No significant differences were observed at 6 or 12 months between the intervention and control group in energy or macronutrient intake (carbohydrates, simple carbohydrates, polysaccharides, fat, saturated fat, mono- or polyunsaturated fatty acids, cholesterol, protein and fibre) (ESM Table 4).

Discussion

The administration of the GFD in children shortly after the onset of type 1 diabetes was accompanied by subtle yet significant changes in the gut microbiome. These changes were not paralleled by alterations of the stool metabolome and were not associated with beta cell function preservation.

The reaction of the bacteriome to GFD was modest—there was no statistically significant change in the overall community composition of the bacteriome, nor were there changes in stool metabolites that largely mirror the bacterial metabolic production. Two important bacterial genera were convincingly associated with GFD intervention: Bifidobacterium decreased, whereas Roseburia increased. Bifidobacterium reaction to GFD has been noted in previous studies [13,14,15], and probably reflects its frequent ability to utilise gluten as the sole nitrogen source [16]. Roseburia, one of the main producers of beneficial butyrate, may increase on GFD because some gluten-free products contain high amounts of fibre and resistant starch [17], which are among the main substrates for this species [18].

The subtle microbiome changes after introducing GFD are unlikely to mediate the previously observed association between GFD and beta cell function. First, neither Bifidobacterium nor Roseburia showed association with the preservation of residual beta cell function. Moreover, no other definable taxon, metabolite or other bacteriome trait could be pinpointed as associating with the pace of loss of the residual beta cell function. Other candidate pathways should therefore be explored, be it the change in the immune landscape of the gut or the direct toxic effect of gluten peptides.

The strengths of our study lie in its longitudinal nature, as well as the length of the study period which is—to date—the longest involving individuals without coeliac disease [2, 13, 19]. Furthermore, the frequent sampling of the study participants allowed us to discard transient changes likely linked to the period around the onset of type 1 diabetes. The study also benefited greatly from an objective method of GFD adherence testing of all stool samples which is not, to our knowledge, a commonplace feature of such studies [2, 13], and rigid non-compliance criteria which led to the exclusion of four non-compliant participants. Finally, the two-step statistical analysis helped eliminate falsely positive results that might have arisen from the overly sensitive DESeq2 [20].

Among the limitations are the small study size and the rather modest association of GFD with improvements in residual beta cell function—these make any causative inference difficult. The use of the 16S rDNA marker gene approach provides less data than bacterial metagenomic or metatranscriptomic sequencing. Another limitation is the relatively low metabolite coverage of the metabolomics approach, owing to glycerol (added as part of the biobank storage protocol) overlapping with proton signals of some of the metabolites normally found in stool. Intestinal permeability marker L/M excretion ratio was obtained only at the end of the study period, thereby precluding the observation of likely transient changes, and zonulin measurements alone may not be reliable in detecting the actual protein [21], nor correlate with the gold standard sugar-absorption tests [22].

In conclusion, the bacteriome composition is unlikely to mediate the previously observed effects of GFD on beta cell residual function. This may have implications for future studies of therapeutic modalities, and underlines the need to focus on other possible mechanisms.