All 70-year-old women identified through the Swedish Gothenburg County registry (Brohall et al., 2006; Karlsson et al., 2013), were invited to participate in the previous research (Karlsson et al., 2013). The public metagenomic sequence datasets of the present study cohort were obtained from previously reported research with European women (Karlsson et al., 2013). The accession code was Sequence Read Archive ERP002469. The individuals with NGT or T2D were included in the present study. A restriction criterion was applied to investigate the information on gut microbiota producing LPS in patients with T2D and healthy women. Participants were invited to take screening examinations including oral glucose tolerance tests (OGTT). Women with known diabetes who were taking oral antidiabetic drugs or insulin did not have to undergo OGTT, whereas those with dietary treatment for diabetes and fasting blood glucose (FBG) < 7.5 mmol/L were examined by OGTT. A 75-g OGTT was conducted in the morning (before 11 a.m.), and glucose oxidase technique was performed to measured fasting- and 2-h postload capillary blood glucose values. Participants were fast overnight, avoid heavy physical activity the day before, and avoid smoking the morning prior to the test. According to World Health Organization (WHO) criteria (World Health Organization, 1999), T2D was defined as FBG ≥6.1 (≥110 mg/dl) and/or ≥11.1 mmol/L (≥200 mg/dl) 2 h after glucose load measure on two occasions. The use of drugs, especially metformin, has been implicated in alterations of gut microbiota and microbial products (Mueller et al., 2021). Hence, patients with T2D were excluded if they were under insulin or oral anti-diabetic treatment, such as metformin and sulphonylurea. Furthermore, subjects who reported a cancer diagnosis, chronic inflammatory disease, serious mental disorder, other serious illness, or drug addictions were excluded. Ultimately, the metagenomic sequence datasets for 72 individuals were downloaded with the accession codes, including individuals with NGT (n = 43), and untreated patients suffered T2D (T2D; n = 29).

The fecal samples were obtained from each participant, and transferred to the laboratory after the screening examination. Samples were stored at –80°C until DNA extraction. Metagenomic DNA was extracted from faecal samples using a standard procedure (Salonen et al., 2010). The concentration of DNA was measured with a Nanodrop instrument (Thermo Scientific, Waltham, Massachusetts, United States) and DNA quality was evaluated by agarose gel electrophoresis. All the isolated DNA samples subjected to sequencing in the Illumina HiSeq2000 instrument with up to 10 samples pooled in one lane. DNA libraries were prepared with a fragment length of 300 bp. With 100 bp in the forward and reverse directions, paired-end reads were generated. Bioinformatics analysis, including gene catalog construction, prediction of genes, taxonomic annotation, and relative abundance calculation, was performed. The genes and related proteins of the gut bacterial genomes were predicted from the assembled contigs through the MetaGeneMark prokaryotic hidden Markov model (v2.10) (Zhu et al., 2010). The Cluster Database at High Identity with Tolerance (CD-HIT, Version 4.5.8, La Jolla, California, United States) was applied to build a non-redundant gene library. The sequence identity cutoff was 0.95, and the minimum coverage cutoff was 0.9. The realignment of the reads to the gene catalog was carried out with SOAP2, and the parameters applied to determine the gene abundance was–m200 –×400 –s119. Genes with more than two mapped reads were included for further assessment. The abundance information of the genes was obtained via calculating the total number of reads and normalizing by the gene length. For evaluating the taxonomic annotation, DIAMOND (Version 0.7.9.58, default parameters except that–k 50-sensitive–e 0.00001, Tübingen, Germany) was used to blast the genes to the NR database. A significant match of genes was assessed using e-values ≤10 × e-value of the top hit, and the retained matches were used to differentiate taxonomy. The taxonomical level and abundance of each annotated gene were determined with the lowest common ancestor-based algorithm implemented with MEGAN (MEtaGenome Analyzer, Tübingen, Germany). The abundance of the taxonomy was defined as summing the abundance of genes matched.

The protein sequences of enzymes participating in LPS biosynthesis were acquired from the RefSeq database of the National Center for Biotechnology Information database (http://www.ncbi.nlm.nih.gov/) (Clementz et al., 1997; Raetz and Whitfield, 2002; Doerrler, 2006), including UDP-N-acetylglucosamine acyltransferase (LpxA), UDP-3-O-[3-hydroxymyristoyl] N-acetylglucosamine deacetylase (LpxC), UDP-3-O-[3-hydroxymyristoyl] glucosamine N-acyltransferase (LpxD), UDP-2,3-diacylglucosamine hydrolase (LpxH), lipid-A-disaccharide synthase (LpxB), tetraacyldisaccharide 4′-kinase (LpxK), 3-deoxy-D-manno-octulosonic-acid transferase (WaaA), Kdo2-lipid IVA lauroyltransferase/acyltransferase (LpxL), and lauroyl-Kdo2-lipid IVA myristoyltransferase (LpxM). BLASTP (version 2.6.0, Bethesda, Maryland, United States) was used with parameters of e-value <1e-5, sequence identity >50%, and coverage >50% as the cutoff in the reference genome resources to identify and align the protein sequences of LPS-biosynthesis-related enzymes within the non-redundant gene sequences of the study cohort. The relative abundance of genes encoding LPS-biosynthesis-related enzymes in each sample was calculated by summing the abundance of non-redundant genes annotated to the same enzyme. The taxonomic classification of gut microbiota harboring enzymatic genes were identified based on the taxonomic annotation of related genes.

DIAMOND (version 0.7.9.58, Tübingen, Germany) with default parameters, with the exception of–k50 –sensitive–e 0.00001, was applied to align all the genes in the catalog to the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (Release 73.1, Kyoto, Japan). KEGG orthologs (KOs) within the KEGG database platform (https://www.kegg.jp) mainly included target metabolism, genetic information processing, and so forth. Each protein was assigned to a corresponding KO with the highest-scoring annotated hits including ≥1 high-scoring segment pair scoring >60 hits. The query of LPS-biosynthesis-related KOs was performed in the KEGG database using the keywords “Kdo2-lipid A,” “CMP-Kdo,” “UDP-2,3-diacylglucosamine,” “UDP-GlcNAc,” “LPS (lipopolysaccharide),” and enzymes. The sum abundance of all the genes assigned to the identical property was considered as the abundance of a KO. To identify KOs associated with LPS biosynthesis in European women with T2D or NGT, the relative abundance of each KO within the group of samples obtained from patients with T2D was compared with its abundance in individuals with NGT. For each KO, for example, k, the odds ratio (OR) was calculated as follows: OR (k) = [Σs = T2D Ask/Σs = T2D (Σi≠k Asi)]/[Σs = NGT Ask/Σs = NGT (Σi≠k Asi)] as other investigators described previously (Greenblum et al., 2012). Ask denotes the abundance of KO k in sample s, T2D denotes the set of T2D samples, and NGT denotes the group of NGT samples. The structure of the formula could be summarized as [sum(s)/sum(a)], where “sum(s)” denotes the sum of KO1 in the T2D group, divide by sum(a) refers to the sum of non-KO1 in the T2D group. In the present study, the differential abundance score was defined as log2 (OR). Furthermore, the KOs related to LPS-biosynthesis-related enzymes were further classified as T2D-enriched [log2 (OR) > 0] or NGT-enriched [log2 (OR) < 0].

The quantitative variables with normal distribution were expressed as mean ± standard deviation, and the difference between groups was compared using a two-sided t-test. Data with non-normal distribution, including KOs, enzymatic genes, and harboring species, were expressed as median and quartiles, and the disparity between NGT and T2D was assessed using the Wilcoxon rank-sum test with Benjamin and Hochberg correction. A q value (corrected p < 0.05) represented a statistically significant difference. Spearman correlation analysis was performed with R (Version 3.3.3, Auckland, New Zealand), and Venn and Sankey plots were graphed with the packages UpSetR and fmsb in R, respectively.

The fecal shotgun metagenomic sequencing data of 70-year-old European women with normal and diabetic glucose control (Karlsson et al., 2013) were included in the present study to assess the gut microbial potentials for generating LPS and identify candidate LPS-producing bacteria in T2D. Age, sex, and geographical locations have been demonstrated to be confounding factors in investigations regarding the association of gut microbiota with T2D. Hence, the European cohort with elderly women, in whom the incidence of T2D was relatively high, was chosen to minimize the sources of variation. In addition, although Karlsson et al. (2013) suggested that medications in this cohort were unlikely to exert major confounding impacts in the mathematical model based on metagenomic profiles to discriminate patients with T2D, accumulating evidence indicated that the medicine use might directly affect the composition and functions of gut microbiota (Klünemann et al., 2021), especially for metformin (Ma et al., 2021; Mueller et al., 2021). Thus, patients with T2D under drug therapy were excluded from the present study, and ultimately 72 European women with normal (NGT, n = 43) or diabetic (T2D, n = 29) glucose control were examined.

Biometric and plasma measurements for the 72 participants are shown in Supplementary Table S1. The average age was 70.3 and 70.6 years for NGT individuals and patients with T2D, respectively. Glucose-related factors, such as fasting blood glucose, fasting insulin, and HbA1c, were significantly higher in patients with T2D (all p < 0.001). In addition, BMI, WC, and leptin and C-peptide levels were enriched among patients with T2D. TC and adiponectin levels significantly increased in the control group, while other baseline characteristics showed no difference between the two groups. The metagenomic data of genomic DNA from 72 fecal samples based on Illumina HiSeq 2000 were 217 Gb paired-end reads, with an average of 3.0 ± 1.2 Gb (mean ± SD) for each sample.

LPS is constituted of polysaccharide, O-antigen, and Kdo2-lipid A, and the crucial step of LPS biosynthesis is the process of Kdo2-lipid A biosynthesis. As the key component of LPS, Kdo2-lipid A is known to be catalyzed from UDP-GlcNAc, progressively to UDP-3-acyl N-acetylglucosamine, UDP-2,3-diacylglucosamine, Lipid X, Lipid IVA, and Kdo2-IVA, by using nine synthetic enzymes, including LpxA, LpxC, LpxD, LpxH, LpxB, LpxK, WaaA, LpxL, and LpxM (Figure 1A). These enzymes were considered to be responsible for LPS biosynthesis. We thus investigated their enrichment in the gut of individuals with NGT and patients with T2D based on the encoding gene sequence. LpxK and LpxC were observed to be the most abundant enzymes in the study cohort, while the level of LpxL and LpxM were relatively low (Figure 1B). According to the relative abundance of the total nine enzymes in each individual, PCA scatter plots roughly distinguished samples from individuals with NGT and patients with T2D, and a significant disparity between the groups was obtained in a single coordinate axis at PCA1 (Supplementary Figure S1).

The abundance of microbial genes for LPS-biosynthesis-related enzymes was mostly enhanced in the gut of patients with T2D, except for LpxA, compared with NGT controls. Intriguingly, the enzymatic gene abundance for LpxC, WaaA, and LpxM was strikingly augmented in patients with T2D, and LpxC coupled with WaaA displayed a higher fold change in T2D versus NGT (Figures 1B,C). The correlation and co-abundance relationship between LPS- biosynthesis-related enzymes in the study cohort were examined, and it was detected that LpxC was directly linked to LpxK, LpxH, and LpxM (Figure 1D). However, WaaA closely related to almost all the other enzymes, especially LpxD, LpxL, LpxB, LpxH, LpxM, and LpxK. The coordinated alterations of LPS-producing enzymes were confirmed, and this information suggested that the capacity and potential for LPS biosynthesis in the gut of patients subjected to T2D might be extremely elevated.

Other gut microbial metabolic products well established to participate in host physiology and pathology, including TMAO and SCFA derived from dietary choline, carnitine, phosphatidylcholines, and dietary fiber, were also focused. Although yeaX and CntA showed a reduced tendency, we found that the enzymatic genes for the biosynthesis of trimethylamine (TMA) were non-discriminatory between individuals with NGT and patients with T2D, as shown by CutD, CutC, TorA, CntB, GrdH, yeaX, and CntA in Supplementary Figures S2A,B. For enzymatic genes essential for SCFA biosynthesis, most enzymes increased in patients with T2D, such as propionyl CoA transferase, yciA and tesB, and CO dehydrogenase acetyl CoA synthase complex as well as tesA (Supplementary Figures S2C,D).

Considering the shifts of enzymatic genes for LPS formation in patients with T2D, the gut microbial functions relevant to the synthetic process and transformation pathways of LPS biosynthesis were assessed. Overall, 29 KOs were annotated with LPS, 2 KOs with UDP-GlcNAc, 2 KOs with UDP-2,3-diacylglucosamine, 1 KO with CMP-Kdo, 5 KOs with CMP-Kdo, and 9 KOs with Kdo2-lipid A-related enzymes through the KEGG database in the present study cohort, while unidentified KOs in NGT and T2D were not followed up (Figure 2A). We found 2 enzyme-associated KOs shared with Kdo2-lipidA, 1 KO coexisting with UDP-2,3-diacylglucosamine, and 45 KOs specific for LPS-biosynthesis-related compounds (Figure 2B). We clearly separated patients with T2D from individuals with NGT via PCA based on the relative abundance of the aforementioned 45 KOs in the gut (Figure 2C). In addition, an obvious distinction at component 1, which could explain 18.98% of the variation between NGTs and T2Ds, was detected. We thus performed Spearmen correlation analysis to investigate the linkage between 9 LPS-biosynthesis-related enzymes and 45 KOs (Figure 2D). Among the 45 annotated KOs, we identified 24 KOs prominently associated with all the enzymes. Nine KOs were positively correlated with LpxC, and one KO was negatively correlated with it. A total of 12 positive correlations were found between KOs and WaaA. KOs such as K02535 and K03269 were positively correlated with LpxM.

A comparison of KO enrichment between NGTs and T2Ds demonstrated nine differently distributed KOs, including K03643 (LPS-assembly lipoprotein), K05790 (lipopolysaccharide biosynthesis protein), K02535 (UDP-3-O-[3-hydroxymyristoyl] N-acetylglucosamine deacetylase), K02560 (lauroyl-Kdo2-lipid IVA myristoyltransferase), K03269 (UDP-2,3-diacylglucosamine hydrolase), K02527 (3-deoxy-D-manno-octulosonic-acid transferase), K00713 (LPS alpha-1,2-glucosyltransferase), K12983 (LPS beta-1,3-glucosyltransferase), and K19363 (lipopolysaccharide-induced tumor necrosis factor-alpha factor) (Figures 3A,B). Seven potential LPS-biosynthesis functional orthologs of K02535, K02560, K03269, K02527, K00713, K12983, and K19363 remarkably increased in the gut communities of patients with T2D compared with the control participants, while K03643 and K05790 were more abundant in NGTs. These findings further suggested aggravated microbial functions toward the formation of LPS in patients with T2D.

The enzymatic genes of LpxA, LpxC, LpxD, LpxH, LpxB, LpxK, WaaA, LpxL, and LpxM were aligned to the integrated nr database to identify the gut microbiota in which LPS-biosynthesis-related enzymes existed. The enzymatic distribution together with the taxonomic identification of enzyme-harboring bacteria was evaluated in individuals with NGT and patients with T2D (Supplementary Figure S3). Precisely, LpxC existed in 162 genera, and 536 species; LpxM distributed in 42 genera, and 89 species; 107 genera, and 316 species carried WaaA genes. In addition, the most abundant enzyme LpxK was detected in 229 genera, and 815 species. By removing the overlapped taxonomy between distinct enzymes, 339 unduplicated genera, and 1,192 non-repetitive species were identified in the study cohort (Figures 4A,B).

As a result, 335 genera, and 1,173 species harboring at least 1 LPS-biosynthesis-related enzymes were detectable in both individuals with NGT and patients with T2D. The top species related to LPS-producing enzymes were Faecalibacterium prausnitzii from Faecalibacterium, and Flavonifractor plautii from Flavonifractor (Figures 4C,D). Among these gut microbiota with LPS-biosynthesis-related enzymatic genes, we found 8 genera, and 33 species prominently enhanced in patients with T2D, while the abundance levels of 14 genera, and 56 species markedly reduced in the gut of patients with T2D (Figures 5A,B). For instance, Gemmiger, Papillibacter, Nitrospirillum, F. prausnitzii, Ruminococcus torques, and Gemmiger formicilis were remarkably enriched in patients with T2D, while those diminished in T2D mainly belonged to Shigella, Neglecta, Senegalimassilia, and Neglecta timonensis. (Figures 5C,D).

The interactions between the enzymatic genes of LpxA, LpxC, LpxD, LpxH, LpxB, LpxK, WaaA, LpxL, and LpxM and discriminative taxonomy in patients with T2D and individuals with NGT were examined to further explore the key LPS-producing microbes in patients with T2D. At the genus level, a positive correlation was detected between the abundance of Actinobacillus, Gemmiger, Papillibacter, and enzyme LpxC (Figure 6A). The abundance of Anaerobiospirillum showed a positive correlation simultaneously with LpxB, LpxD, LpxL, LpxM and WaaA. We also found that the abundance of intestinal species, including R. torques, Papillibacter cinnamivorans, Bacteroides pyogenes and Lanchnoclostridium sp., correlated with higher levels of LpxC, LpxM, and WaaA, respectively (Figure 6B).

Taken together, further analysis of the relationship among enzyme-harboring bacteria, LpxC, LpxM, WaaA, and LPS-biosynthesis-related KOs revealed 11 gut genera directly linked to the enzymes and subsequently associated with the 9 LPS-biosynthesis functional orthologs to varying degrees (Figure 7). The complicated correlation and network indicated that the overgrowth in potential LPS-producers, including Actinobacillus, Anaerobiospirillum, Gemmiger, and Papillibacter, might be responsible for the enhancement of LPS-biosynthesis-related enzymes in the gut of patients with T2D, and thus might potentially contribute to the excessive activation of bacterial functions implicated in LPS formation, such as LPS alpha-1,2-glucosyltransferase, 3-deoxy-D-manno-octulosonic-acid transferase, UDP-3-O-[3-hydroxymyristoyl] N-acetylglucosamine deacetylase, lauroyl-Kdo2-lipid IVA myristoyltransferase, UDP-2,3-diacylglucosamine hydrolase, and LPS beta-1,3-glucosyltransferase.