The double-blinded, randomized, placebo-controlled 4-month trial was performed from January 2018 to November 2018 in the Third Affiliated Hospital, Nantong University (renamed later as the Affiliated Hospital of Jiangnan University), Wuxi, Jiangsu, China. The trial was approved by the ethical review board of the Third Affiliated Hospital, Nantong University (ID: IEC201711001) and registered on the Chinese Clinical Trial Register as ChiCTR-IOR-17013656. Written informed consent was obtained from all participants before the intervention.

Ninety-six patients with T2D aged 35–75 years were enrolled in this study because they met the following criteria: 6.5%≤ HbA1c ≤13.0%. Besides sulfonylureas and insulins, most hypoglycemic drugs have been reported to be able to modulate gut microbiota (23). Therefore, only patients treated with sulfonylureas or insulin were included. Patients were excluded if they had type 1 diabetes, malignant hypertension, severe cardiac disease, renal failure (eGFR<15), kidney replacement, inflammatory bowel disease, gastrointestinal ulcer, an autoimmune disease, or cancer. Exclusion criteria also included patients receiving medication or surgery for losing weight within 3 months, receiving the administration of antibiotics within one month, receiving gastrointestinal surgery, or during or preparing for pregnancy and lactation. Subjects with poor compliance or protocol violation or unwillingness to continue the clinical trial were asked to withdraw from this study. In the end, ninety participants were eligible and decided to participate in this study.

All subjects were randomly assigned in a 1:2 ratio to the control group (Group C) and WCBE group (Group W). After enrollment, demographic data, including sex, age, disease history, medication and lifestyle, of all subjects were recorded. Anthropometric data, including body mass index (BMI), waist

circumference, hip circumference, blood pressure, and pulse rate, were collected. Sample size was calculated based on the primary outcome (HbA1c), considering a 0.4% difference in HbA1c and a common standard deviation of 0.6% were required to detect a significant improvement in primary outcomes with 95% level and 80% power (24). Considering 10% dropout, withdrawal and non-compliance, a total of 90 subjects were needed in Group C and Group W at a ratio of 1:2.

A simple randomization scheme generated by the computer was created by the trial statistician. The clinical trial was double-blinded because neither the subjects nor researchers knew which group every subject was assigned to, and the grouping of all subjects was unblinded by a statistician after they had completed their experiment. The progression of diabetic complications generally occur over a greater time period than 3 months. Additionally, HbA1c can be used to evaluate the mean level of blood glucose in the recent 2-3 months. In our study, at the end of Month 2, HbA1c was measured after the intense intervention. Parameters of glucose metabolism and diabetic complications were monitored in a part of participant to at Month 4 to determine whether the intervention could ameliorate diabetic complications and maintain the improvement of glucose metabolism. All of the participants received routine dietary guidance for diabetes. Subjects in Group W received 1.5 g WCBE rich in highly active α-amylase inhibitor (raw material preparation technology provided by Suzhou Langbang Nutrition Company and processed by Jiangsu Shouyuan Biotechnology Co. Ltd) half an hour before each meal. Subjects in Group C were given supplementation of 1.5 g maltodextrin as placebo After each participant was assessed for eligibility and assigned to one group, they experienced a 2-week wash-out phase in which their previous diet recipes were changed into dietary regimen for diabetic patient. Afterward, all participants received baseline biochemical and complication indicator measurements. This trial includes two phases. The first phase was a 2-month intense intervention. All participants received daily telephone follow-up, self-monitoring of fasting blood glucose and 2-hour postprandial blood glucose every 3 days, and a weekly face-to-face interview. All subjects only received biochemical examination at the end of this phase. As the progression of diabetic complications generally occur over a greater time period than 3 months, we then randomly selected 22 participants in Group C and 33 participants in Group W who were willing to continue to the next 2-month intervention phase. In this maintenance phase, participants received the intervention with the same dose of WCBE or maltodextrin as the first phase. During the second phase, they only received weekly telephone follow-up. At the end of Month 4, participants received biochemical examination for the third time and complication indicator measurement for the second time.

The change in the primary outcome, HbA1c, was detected during the intervention. Secondary outcomes included changes in the levels of fasting and postprandial plasma glucose, peripheral neuropathy nerve conduction, endothelial function and fecal microbiota. Outcomes representing lipid metabolism included changes in the levels of total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-c) and low-density lipoprotein cholesterol (LDL-c).

Feces and blood were collected from the subjects before the intervention and at the end of month 2 and 4. Venous blood was collected from the subjects after a 10-12 h fast, and the serum was separated for further laboratory measurements. Feces were immediately preserved in a -80°C freezer for further genome sequencing of the gut microbiota.

Clinical laboratory measurements were performed at Wuxi Third People’s Hospital. Blood cell measurements were conducted on an automated hematology analyzer (Sysmex K4500; Sysmex Corporation, Japan). The measurement of biochemical indicators, including fasting blood glucose (FBG), 0.5-hour postprandial blood glucose (0.5 h-PBG), 1-hour postprandial blood glucose (1 h-PBG), 2-hour postprandial blood glucose (2 h-PBG), 3-hour postprandial blood glucose (3 h-PBG), TC, Trig, HDL-c and LDL-c, was performed on an automatic biochemical analyzer (Beckman Coulter au6800; Beckman Corporation, America). Insulin levels were detected on an immunoassay system (Immulite 1000; Siemens Healthcare Diagnostics Inc., Germany). Urine samples collected for urine-microalbumin detection were measured on a biochemical analyzer (Beckman Coulter au5800; Beckman Corporation, America). Fecal samples collected at month 0, 2 and 4 were used to analyze the 16S rRNA gene profile for the gut microbiota.

The TCSS is based on clinical symptoms, lower limb reflexes and sensory tests, and it was used to evaluate peripheral neuropathy according to a previous study (25). Patients with TCSS scores ≤5 were considered to have no neuropathy, while those with scores ≥6 indicated that they might suffer from diabetic peripheral neuropathy (DPN).

The NCS was conducted using MedelecSynergy (VIASYS Health care, Philadelphia, PA, USA) based on standardized methodology for NCS (26). Four sensory nerves, including the median, ulnar, sural and superficial peroneal nerves, were tested bilaterally on all the participants. When conducting NCS, it was ensured that the skin temperature was maintained above 32°C over the upper limbs as well as the lower limbs. The parameters recorded in sensory NCS include sensory nerve conduction velocity potential (SNCV) and sensory nerve amplitude.

The measurements of ankle-brachial pressure index (ABI) indicate arterial sclerosis and the brachial-ankle pulse wave velocity (baPWV) indicates arterial stiffness and vascular damage. The measurement of these indices were conducted on a BP-203RPE III (Omron Health Co, Kyoto, Japan) according to the instructions for the instrument. The measurements were performed after at least 5 min of resting. Blood pressure of the upper and lower limbs was measured on oscillometric sensors. ABI was calculated bilaterally as the ratio of ankle SBP to arm SBP.

Fecal bacterial DNA extraction, NGS library preparation and sequencing were performed at Shanghai Honsun Biological Technology Co., Ltd. (Shanghai, China). Briefly, total DNA was extracted from the stool samples using the E.Z.N.A.^® Soil DNA Kit (Omega Biotek, Norcross, GA, U.S.) according to the manufacturer’s instructions. PCR for fecal DNA was performed to amplify the hypervariable V3-V4 region of the bacterial 16S rRNA gene using universal primers 338 F 5’-ACTCCTACGGGAGGCAGCAG-3’ and 806 R 5’-GGACTACHVGGGTWTCTAAT-3’. After purification and quantification, PCR-amplified fragments were used for NGS library construction. 16S rRNA gene sequence tags were generated using Illumina MiSeq PE300 (Illumina, San Diego, CA, USA) and were clustered into operational taxonomic units (OTUs) with a 97% sequence similarity cutoff using the UPARSE v.7.0 platform. The QIIME (version 1.9.1) pipeline was used for 16S rRNA data analysis. Singletons and chimeras were identified and removed before the cluster analysis. Each sequence was taxonomically classified and annotated using Ribosomal Database Project (RDP) Classifier v.2.2 (http://rdp.cme.msu.edu/) against the Silva (SSU138) 16S rRNA database with a confidence threshold of 70%.

The dataset was rarified to 31237 reads for the analysis of α-diversity using four metrics: Shannon, Simpson, Chao and Ace diversity. For beta diversity analysis, principal coordinates analysis (PCoA) was performed on the abundance matrix based on Brary-Curtis distance by R version 3.1.1 (https://cran.r-project.Org/bin/windows/base/old/3.1.1/). The linear discriminant analysis (LDA) effect size (LEfSe) method was applied to determine the OTUs most likely to explain differences between groups at the same time point and between the same group pre- and post-intervention. The pheatmap package (https://cran.r-project.org/src/contrib/Archive/pheatmap/) was used for ecological analysis and heatmap depiction.

Prism 8 (GraphPad Prism, San Diego, CA) statistical software was used to analyze the data. A two-way repeated-measures analysis of variance (ANOVA) with a mixed effects model was used for intra- and intergroup comparisons. Statistical tests were two-sided, and a P value < 0.05 was considered to be statistically significant.

Ninety-six patients (37 males and 59 females) aged 56-72 years old with T2D who met the inclusion criteria were recruited for this study. Four volunteers who met the exclusion criteria and two volunteers who declined further screening were asked to quit the trial. Thirty subjects were randomly assigned to Group C and sixty subjects to Group W. At baseline, all subjects received measurements of anthropometric, biochemical and complication indicators. At the end of the intense intervention phase, 26 subjects in Group C and 57 subjects in Group W finished the second biochemical examination. We then randomly selected 22 participants in Group C and 33 participants in Group W who were willing to continue the further maintenance phase of intervention. At the end of the second intervention phase, 21 participants in Group C and 31 participants in Group W finished the third biochemical examination and the second complication indicator examination. In this trial, there is no adverse events were reported.The trial profile is shown in Supplementary Figure S1 .

There was no significant difference in biochemical and diabetic complication indicators in subjects between the two groups at baseline, except for the levels of 0.5 hPBG, Trig, ABI on the left side, median nerve amplitude, and SCV-ulnar nerve ( Table 1 ). In addition, there was no change in the participants’ diet structure before and after the intervention ( Supplementary Figure S2 ).

The trial profile is shown in Figure 1 . The processed WCBE inhibited the activity of α-amylase to reduce the digestion and absorption of carbohydrates, leading to a decrease in the serum glucose level. HbA1c was the primary outcome. There was a greater reduction in HbA1c levels in Group W than in Group C at the end of the 2-month intense intervention (0.660 ± 0.468% vs. 0.222 ± 0.763%, p<0.05) and at the end of the second 2-month intervention (0.721 ± 0.742% vs. 1.059×10^-8 ± 0.942%, p<0.05) ( Figure 1A ). In OGTT glucose test, the FBG, 0.5 h PBG, 1 h PBG, 2 h PBG and 3 h PBG in Group W decreased after the 2-month and 4-month intervention, whereas almost all OGTT glucose parameters increased at the end of the 2nd month in Group C and continued to rise at the end of the 4th month. These results indicate an unsatisfactory glycemic control in Group C. Notably, the change in fasting and postprandial glucose levels were significantly different between the two groups at the end of 2nd and 4th month ( Figure 1B ). Accordingly, a significant reduction in the area under the curve (AUC) of glucose levels during the OGTT was observed in those administered WCBE, but this parameter was increased in those administered the placebo ( Figure 1C ). To evaluate the level of insulin resistance, HOMA-IR was calculated according to the fasting blood glucose and insulin levels at month 0, 2 and 4. After the 2-month intervention, HOMA-IR did not change significantly in either group. However, HOMA-IR was significantly increased after the 4-month intervention in Group C but remained steady compared with baseline in Group W. At the end of the 4th month, WCBE tended to lower HOMA-IR levels (P=0.078) ( Figure 1C ).

No significant differences were found between the two groups for changes in the lipid metabolism parameters of BMI, waist to hip ratio (WHR), TC, Trig and LDL-c after the 2- and 4-month intervention. However, there was a greater increase in HDL-c after the 2- and 4-month WCBE intervention compared with placebo (P<0.05). BMI slightly decreased in both groups at the end of the 2nd month. Although BMI showed a greater increase in both groups at the end of the 4th month compared with the 2nd month, BMI remained relatively steady in Group W. γ-Glutamyl transpeptidase (GGT) was slightly increased in Group C and decreased in Group W, and the change in GGT after 4 months was significantly different between the two groups. Notably, there was a greater enhancement of testosterone in male patients in Group W than in Group C ( Supplementary Figure S3 ).

TCSS, NCS and endothelial function analysis were used to assess diabetic vasculopathy and neuropathy. Considering that these diabetes complication indices could not be improved within a short time, we only conducted these measurements before and 4 months after the intervention. The decline in the Toronto score in Group W was greater than that in Group C (P<0.05) ( Figure 2A ). We further compared the proportion of patients with TCSS≥6 who could be diagnosed with diabetic peripheral neuropathy (DPN) pre- and post-intervention. We found that this proportion only significantly decreased in Group W after the intervention (P<0.01), but no change in this proportion was observed in Group C (P>0.05) ( Figure 2B ). Thus, the proportion of patients with TCSS≥6 was significantly lower in Group W than in Group C at the end of 4 months (P<0.05). ABI and baPWV are two sensitive markers of arterial stiffness. No difference in the change in ABI was observed between the two groups. As an index reflecting arterial stiffness in the arteries of the lower limbs, the baPWV level on the right side of the body showed greater reduction in Group W than in Group C after the 4-month intervention (P<0.05). However, there was no difference in the change of the baPWV level on the left side observed between the two groups ( Figure 2C ).

The NCS is considered the gold standard for the diagnosis of DPN. In this study, both the left and right sural sensory nerve conduction velocity (SNCV-left sural and SNCV-right sural) slightly decreased in Group C and slightly increased in Group W. There was a significant difference in the variation in the SNCV of the left sural nerve between the two groups after the intervention (P<0.05), and a slight but not significant difference in the variation in the SNCV of the right sural nerve was observed (P=0.0888) ( Figure 2D ). In addition, no difference in the variation in other motor or sensory NCVs were detected between the two groups ( Supplementary Figure 4 ).

At baseline, parameters representing community richness and diversity, including the Shannon, Simpson, Chao and Ace index, were not different between participants in Group W and Group C. After 2 months and 4 months of intervention, these α-diversity indices did not change significantly (P > 0.05, Supplementary Figure S5 ).

PCoA based on the Bray–Curtis distance was applied to evaluate the gut microbiota compositional discrimination between subjects in different groups at the same time point and in the same group at different time points. Before the intervention, there was no significant difference in gut microbiota structure between Group C and Group W (PERMANOVA, P>0.05). As expected, treatment with the WCBE resulted in a significant difference in the gut microbiota structure compared with Group C at the end of the 2nd month (PERMANOVA, P<0.01). At the end of the 4th month of intervention, the gut microbiota structure remained significantly different between Group C and Group W (PERMANOVA, P<0.05) ( Figure 3 ).

We further compared the gut microbiota structure at pre-intervention with that at 2 and 4 months post-intervention in each group. Significant variations were discovered in Group W after treatment with the α-amylase inhibitor compared with pre-intervention (PERMANOVA, P<0.01). However, no variation in the gut microbiota structure could be observed pre- and post-intervention in Group C (PERMANOVA, P>0.05) ( Figure 3A ). In the gut microbial community, a higher abundance of Faecalibacterium was observed in Group W than in Group C 2 months after the intervention. However, 4 months after the intervention, no difference in the abundance of this genus was discovered between the two groups. The level of Bifidobacterium was only increased in Group W 2 and 4 months after the intervention ( Figures 3A, B ).

LEfSe analysis was performed to identify bacterial taxa with differential abundance among the groups, and only those taxa with a log LDA score >2 were considered. At the end of the 2nd month of intervention, the abundances of Bifidobacterium, Faecalibacterium and Anaerostipes in participants in Group W were higher, whereas Weissella, Klebsiella, Cronobacter and Enterobacteriaceae_unclassified were less abundant than those in Group C ( Figure 4A ). After 4 months of intervention, Bifidobacterium remained more abundant in Group W than in Group C. The genus of Adlercreutzia also showed higher abundance in participants after the 2- and 4-month α-amylase inhibitor intervention than in those who received placebo treatment. After the 4 months of intervention, Citrobacter, Cronobacter and Enterobacteriaceae_unclassified were less abundant in Group W. Notably, these less abundant genera after the intervention in Group W all came from the family Enterobacteriaceae, which contains many opportunistic pathogens ( Figure 4B ).

Differential gut bacteria were also screened pre- and post-intervention. In Group W, the abundance of Lactobacillus at the end of 2 months of intervention and Bifidobacterium, Romboutsia and Faecalitalea at the end of 4 months of intervention were higher than those before the intervention. The concentrations of Fusobaterium, Roseburia, Citrobacter, Klebsiella and Enterobacteriaceae decreased after 4 months of WCBE intervention ( Figure 4C ).

From the heatmap of the abundance of the key OTUs responsible for the different variations in the gut microbiota structure between the two groups after the intervention, we found that WCBE could enrich SCFA-producing bacteria, such as Faecalibacterium and Bifidobacterium, at the end of the 2nd month after the intervention. However, after 4 months of the intervention, only Bifidobacterium was greatly enriched. The other bacterial genera were not as enriched as those at the end of the 2nd month ( Figure 5 ). Moreover, WCBE administration inhibited opportunistic pathogens, such as Klebsiella, after 2 and 4 months of intervention, but no significant change was observed in Group C.