2.1 Human subjects

We initially screened a total of 421 patients, out of whom 17 met inclusion criteria for unstable diabetes.^21-23 One participant withdrew due to overwhelming stress and two other patients dropped out due to poor protocol compliance. Eventually, 14 eligible participants completed the study (Table 1). Healthy donors from China Microbiota Transplantation System were screened based on published criteria.²⁴ Healthy Chinese Han females (body mass index < 25 kg/m², age 20–21 years) were selected. The study was approved by the Ethics Committee at Sir Run Run Hospital, Nanjing Medical University (ID: 2017-SR-001.S2) and conducted in accordance with the Declaration of Helsinki. The trial was registered in the Chinese Clinical Trial Registry (ChiCTR-ONN-17011279). All subjects gave written informed consent.

2.2 Experimental design

2.3 Calculation of GV indices

Our primary outcomes were daily insulin doses, glucose excursions during meal tests, and GV indices calculated using published methods: mean amplitude of glycemic excursion (MAGE),²⁶ SD of blood glucose (SDBG), mean blood glucose (MBG), coefficient of variation,²⁷ glucose percentage time in range (TIR: 3.9–10 mmol/L),²⁸ postprandial glucose excursions (PPGE), largest amplitude of glycemic excursions (LAGE), and hypoglycemic episodes (BG < 3.9 mmol/L).

2.4 16S rRNA gene sequencing

Our secondary outcomes were multiomics data and correlation analyzes in relation to GV phenotype. 16S rRNA gene sequencing was performed by Novogene (Beijing China) according to the workflow specified by the service provider.

2.5 Metagenomics sequencing and functional annotation

Metagenomics was performed by Novogene (Beijing China) according to the workflow specified by the service provider. The functional annotation to Kyoto Encyclopedia of Genes and Genomes (KEGG) database was performed through the DIAMOND software. Differential KEGG orthology (KOs) revealed by Metastats analysis were uploaded to KEGG Mapper tool (version 4.3) for enriched pathways. Network depicting connections of these pathways was constructed by Cytoscape (v 3.6.1).

2.6 Fecal and serum metabolomics profiling

Targeted fecal metabolomics and nontargeted serum metabolomics were performed by Metabo-Profile (Shanghai, China). Processing of fecal and serum samples were carried out as previously described.²⁹ Heatmaps with clustering showing differential metabolites were generated by Heatmapper website (http://www.heatmapper.ca/expression/).

2.7 Laboratory analysis

Plasma glucose, serum lipid panel including total cholesterol, low-density lipoprotein, high-density lipoprotein (HDL), and triglycerides were analyzed on a Cobas 8000 c701 Analyzer (Roche Diagnostics, Germany). HbA1c was measured on D-10 Hemoglobin Testing System (Bio-Rad, USA) using high-performance liquid chromatography. Serum C-peptide was detected on Cobas 8000 e602 Analyzer using chemiluminescence method. Plasma levels of enteroendocrine hormones, including glucagon-like peptide 1 (GLP-1; active; Millipore Cat# EGLP-35 K), GLP-2 (Millipore Cat # EZGLP2-37 K), and peptide YY (PYY; Millipore Cat# EZHPYYT-66 K) were analyzed by commercial ELISA kits. Plasma inflammatory cytokines were measured using the Meso scale discovery U-Plex assay system with customizedly configurated panels (Meso Scale diagnostics LLC, Rockville, MD). Transforming growth factor-β1 (TGF-β1) was measured by commercial ELISA kit (DAKEWE Cat # DKW12-1710).

2.8 Statistics

All statistical analyses were performed with GraphPad Prism, unless otherwise stated. Data are presented as either scatters, mean ± SEM, or box plot showing means with minimum and maximum values. Normal distribution of each dataset was accepted when passing both Shapiro–Wilk and Kolmogorov–Smirnov tests. For comparisons between two groups with matched datapoints and normal distribution, two-tailed paired t tests were used. For comparisons between two nonmatched groups, two-tailed unpaired t test or Mann–Whitney test was chosen. For comparisons among multiple groups with matched datapoints, repeated-measures one-way or two-way analysis of variance (ANOVA) were adopted for data with normal distribution, whereas Friedman test or Kruskal–Wallis test were adopted for data with nonnormal distribution. All multiple comparisons were adjusted by the original false discovery rate (FDR) method of Benjamini and Hochberg method. FDR or p values < .05 were considered statistically significant. Spearman's correlations between metabolites and clinical parameters were calculated based on a published method,³⁰ using pooled variables collected over all the follow-up time points. The post hoc power analysis was performed via G*Power software, using one-sided paired t test (α = 0.05, effect size = mean differences/SD; N = 14) as instructed.³¹

3.1 Overview of experimental design

Fourteen eligible patients with unstable diabetes (Table 1) received allogenic (healthy donors) WMT and were followed up at three time points: 1 week (T1W), 1 month (T1M), and 3 months (T3M). Their baseline parameters prior to WMT were documented (T0) and served as basis for later comparisons (Figure 1A). All patients were required to wear real-time CGM devices from T0 to T1W.

3.2 WMT reduces GV in patients with unstable diabetes

Daily insulin dose dramatically reduced at T1W and T1M, but rebounded by T3M (Figure 1B). HbA1c was unaffected by the end of the study (Table 1). Notably, glucose excursions during the SBMT significantly decreased at T1W (Figure 1C). Analysis with CGM data showed that the mean amplitude of glycemic excursion, SDBG, percentage of time spent with BG > 11.1 mmol/L, LAGE, and MBG all significantly decreased (Figure 1D–H), whereas glucose percentage TIR of 3.9–10 mmol/L significantly increased (Figure S1A). Post hoc power analysis of the aforementioned CGM GV indices with a sample size of 14 revealed an averaged statistical power of 85% (Table S1), which is typically considered as an acceptable power level in multiple WMT clinical studies.^{14, 32} In agreement, analysis with self-monitoring of blood glucose (SMBG) data also revealed TIR to be significantly greater at T1W and T1M (Figure 1I), suggesting the therapeutic effects of WMT could be maintained up to 1 month. Of note, hypoglycemic episodes reduced profoundly at T1W and T1M (Figure 1J). Representative 24 h CGM profiles obtained at T0 and T1W of two recipients are shown in Figure 1K. To our attention, the clinical effects of WMT on the aforementioned GV indices were not different between participants with T1DM and those with T2DM at any follow-up time points (Figure S2). Interestingly, WMT significantly increased the level of serum HDL (Table 1), consistent with an earlier report showing that gut microbiome is associated with postprandial lipemia.^{18, 33} Collectively, we demonstrated a strong ameliorating effect of WMT on GV in patients with unstable diabetes. No severe WMT-related adverse events were observed.

3.3 WMT alters gut microbiota composition in recipient patients

Next, we performed 16S rRNA gene sequencing to inspect WMT-induced changes of gut microbial community. First, we observed an increased alpha diversity, as reflected by the Observed Species, the Faith's phylogenetic diversity, and the ACE index (Figure 2A). Similar to GV indices, alpha diversities were not different between T1DM and T2DM (Figure S3). Second, WMT changed beta diversity as represented by the principal coordinates analysis (PCoA) that is calculated using Unweighted UniFrac distances to visualize dissimilarities among all samples (Figure 2B). Scatterplot and associated analysis of molecular variance (all adjusted p < .05) confirmed significantly shifted clustering of gut bacterial communities by WMT. To identify signature microbial taxa driving these compositional differences, we performed LEfSe analysis (Figure 2C) and found T1W was characterized by increased abundance of genus Prevotella and family Prevotelleceae. At T1M, gut microbiota was enriched with family Coriobacteriaceae, order Coriobacteriales, class Coriobacteriia and was depleted with the pathogenic lipopolysaccharide-containing species Escherichia coli, genus Escherichia-Shigella, family Enterobacteriaceae, order Enterobacteriales, class Gammaproteobacteria, and phylum Proteobacteria. Both T1W and T1M were characterized by lower abundance of genus Bacteroides and family Bacteroidaceae. Overall, WMT-induced changes in gut microbiota composition were reflected in specific hierarchical lineages and maintained down through constituent subtaxa. A bar graph showing differential phyla and a representative heatmap showing differential genera are displayed in Figure S4. No differences were observed between T1DM and T2DM in the relative abundance of the phylum Firmicutes or Bacteroidetes (Figure S4B,C).

Then, we built cooccurrence networks with identified genera to assess their interactions. Interestingly, the number and relative abundance of constituent genera (node size shown in Figure 2D), the number of significant connections (Table S2), and network complexity (network diameter, graph density, and average degree shown in Figure 2E) all increased after WMT, indicating a more intertwined ecological network.

3.4 WMT promotes divergent functional shifts in gut microbiome of recipients

To further understand WMT-induced functional changes of gut microbiome, we performed fecal shotgun metagenomics sequencing that generated a total of 1 344 921 nonredundant microbial genes, which were then annotated to KEGG KOs. WMT caused a clear and persistent shifting of microbial functional variations from T0 to become no different from those of donors (Figure 3A). Then, we focused on and mapped the 328 KOs that were initially different between donors and recipients pre-WMT (permutation test, p < .05), but reverted significantly toward the levels of donors at T1W and T1M (Figure 3B; Table S3), into KEGG database for pathway enrichment analysis (Figure 3C; Table S4). Metabolic connections between as-enriched pathways and associated biochemical reactions are depicted in Figure 3D.

The most enriched pathway was methane metabolism, an observation echoing our taxonomic results showing increased abundance of the phylum Euryarchaeota (Figure S4A), to which most methanogens belong.³⁴ Second, changes in pyruvate metabolism were represented by a reduced production from methylglyoxal via lactaldehyde dehydrogenase (K07248) and by an augmented oxidation to acetyl-CoA, selectively via 2-oxoacid ferredoxin oxidoreductase (K00174). To support, the pathway of pantothenate and CoA biosynthesis was highly enriched to promote the formation of coenzyme A, assisting pyruvate oxidation to acetyl-CoA. With respect to the citrate cycle, WMT promoted conversion of 2-oxoglutarate to succinyl-CoA by upregulating 2-oxoacid ferredoxin oxidoreductase (K00174, K00176, K00177). Another critical finding is that metabolism of amino acids was altered, particularly for those of aromatic amino acids (AAAs) and branched-chain amino acids (BCAAs). For example, WMT induced higher abundance of tyrosine decarboxylase (K18933) to promote tyrosine degradation. BCAA degradation was augmented by upregulation of 2-oxoisovalerate ferredoxin oxidoreductase (K00186, K00187) that unidirectionally convert the cognate α-ketoacids of each BCAA to their corresponding CoA thiol ester derivatives. In addition, butyrate metabolism was enriched, wherein succinate-semialdehyde dehydrogenase (K18119) that catalyzes formation of the precursor succinate semialdehyde from succinyl-CoA was upregulated. Altogether, our results demonstrated WMT reshaped gut microbiome to affect metabolic pathways including methane production, pyruvate oxidation, citrate cycle metabolism, amino acid degradation, and butyrate formation.

3.5 WMT alters profiles of fecal metabolites

Next, we performed targeted fecal metabolomics to explore whether WMT-induced compositional changes and functional shifts of gut microbiome led to alterations of metabolites (Figure 4A). Compared to T0, many metabolites derived from carbohydrate and amino acid metabolism changed in patterns generally consistent with those of genes encoding metabolic enzymes. In accordance with an altered citrate cycle pathway, isocitric acid increased at both T1W and T1M, and oxoglutaric acid decreased at T3M. For amino acid metabolism, L-valine, L-proline, L-tyrosine, and L-alanine were all downregulated at T3M, whereas hydroxyphenyllactic acid, a metabolite derived from L-tyrosine degradation, was persistently elevated at all three follow-up time points. In addition, 3-3-hydroxyphenyl-3-hydroxypropanoic acid and ortho-hydroxyphenylacetic acid, two organic acids derived from L-phenylalanine degradation, were persistently decreased. Other downregulated metabolites included L-alpha-aminobutyric acid at T3M, which is derived from cysteine and methionine metabolism, and aminoadipic acid at T1M, which is derived from L-lysine metabolism. Moreover, pyroglutamic acid was increased at T1M. Butyric acid was upregulated at both T1W and T1M.

Then, we performed Spearman's correlation analysis between fecal metabolites and clinical parameters (Figure 4B; Table S5). We found the significantly downregulated BCAA, L-valine, was positively correlated with daily insulin dose. Similar correlations were observed for other differential metabolites including 3-3-hydroxyphenyl-3-hydroxypropanoic acid, ortho-hydroxyphenylacetic acid, L-alpha-aminobutyric acid, pyroglutamic acid, and the citrate cycle intermediate oxoglutaric acid. Furthermore, downregulated aminoadipic acid correlated positively with PPGE.

3.6 WMT alters profiles of serum metabolites

To gain further mechanistic insights how WMT modulates systemic glucose metabolism, we performed nontargeted serum metabolomics (Figure 5A). In alignment with an altered pyruvate metabolism, serum pyruvic acid was significantly decreased at T1M. With respect to citrate cycle, oxoglutaric acid and maleic acid were decreased at T1M, whereas citric acid and isocitric acid were increased at T3M. Consistent with the enhanced capacity of amino acids degradation, L-methionine, L-tyrosine, L-phenylalanine, L-leucine, L-threonine, and L-alanine were all decreased at T1M and/or T3M. In harmony, abundance of ketoleucine, the first intermediate downstream of leucine catabolism, was significantly increased at T1W. Methylcysteine and methylmalonic acid were both decreased at T3M. Altogether, serum metabolomics revealed a differential profile of circulating metabolites that genuinely agreed with changes observed in metagenomics and fecal metabolomics.

Next, we performed Spearman's correlation analysis between serum metabolites and clinical parameters (Figure 5B, Table S6). Many downregulated amino acids, such as L-methionine, L-tyrosine, and L-alanine, were positively correlated with SBMT-area under the curve (AUC). In contrast, the upregulated methylcysteine, an L-cysteine derivative with antioxidant activity, correlated negatively with CGM-LAGE. The downregulated methylmalonic acid correlated negatively with SMBG-TIR. For citrate cycle intermediates, upregulated citric acid correlated negatively with CGM-MPG and BG > 11.1 mM, whereas downregulated maleic acid correlated positively with SBMT-AUC.

Finally, to identify serum metabolites that may have originated from the gut to exert systemic beneficial effects, we performed correlation analysis for assayed metabolites present in both feces and serum. A total of eight metabolites showed significant positive correlations (Figure 5C). Notably, L-tyrosine was the only metabolite that was downregulated either significantly or marginally in both compartments, suggesting its microbial metabolism may contribute to improved systemic glycemic control after WMT.

3.7 WMT does not affect immune status or enteroendocrine hormones

Intrigued by the well-established role of butyrate in suppressing inflammation, stimulating secretion of enteroendocrine hormones³⁵ and conserving beta cell functions in T1DM,³⁶ we measured serum inflammatory cytokines (interleukin [IL]-1β, IL-4, IL-6, IL-7, IL-10, IL-17A, IL-18, interferon gamma, tumor necrosis factor-α, and TGF-β1) (Figure S5), plasma enteroendocrine hormones (GLP-1, GLP-2, and PYY) during the SBMT (Figure S6), and serum C-peptide (Table 1). However, none of them was changed by WMT.