Study design

120 faecal, 110 plasma and 24 PBMCs (Peripheral Blood Mononuclear Cells) samples were collected from a total of 172 subjects, consisting of 40 control, 32 RA, 32 RA_MD, 40 GA and 28 SLE, for analyses by 16 S rRNA, ITS (internal transcribed spacer) gene sequencing (n=120) and metabolomics (n=110), transcriptomics (n=18), and proteomics (n=6) and phosphorylated proteomics (n=4). The composition of RA_MD includes RA_T2D (RA with type 2 diabetes), RA_HLP (RA with hyperlipidaemia) and RA_AS (RA with atherosclerosis). Detailed samples and corresponding participant information can be found in the Supplementary Tables 1–3. Then, spearman’s correlation, protein-protein interaction (PPI) networks and shared pathway analyses were used to assess the multi-omics data correlations, and a machine learning model was built to identify disease-specific biomarkers for RA_MD diagnosis. Finally, the differences in the pathogenesis of MD among three rheumatic diseases, namely RA, GA and SLE, were explored (Fig. 1).

Open in a separate window

Fig. 1

The flowchart of our study design. RA, rheumatoid arthritis, RA_MD, RA patients with metabolic diseases, RA_T2D, RA patients with type 2 diabetes, RA_HLP, RA patients with hyperlipidaemia, RA_AS, RA patients with atherosclerosis, GA, gouty arthritis, GA_MD, GA patients with metabolic diseases, SLE, systemic lupus erythematosus, SLE_MD, SLE patients with metabolic diseases

Analysis of gut microbiome profiling (16 S/ITS)

Total DNA of the microbial community was extracted from fecal samples using E.Z.N.A.^® soil DNA kit (Omega Bio-tek, Norcross, GA, USA). The primers 338F (5’-ACTCCTACGGGAGGCAGCAG-3’) and 806R (5’-GGACTACHVGGGTWTCTAAT-3’) were used to amplify the bacteria 16 S rRNA gene fragments (V3–V4), and fungal internally transcribed spacer (ITS) fragments were amplified with primers ITS1F (5’-CTTGGTCATTTAGAGGAAGTAA-3’) and ITS2R (5’-GCTGCGTTCTTCATCGATGC-3’). Illumina MiSeq PE30 sequencing platform (Illumina, San Diego, CA, USA) was used to sequence the foregoing PCR amplicons. The Amplicon Sequence Variants (ASVs) were obtained through the QIIME2 (Quantitative Insights Into Microbial Ecology 2) process and analyzed taxonomically based on the SILVA 16 S rRNA (v 138) and ITS databases. Subsequently, the alpha and beta diversity, community composition, species variation, and Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt2) prediction analyses were performed using the online platform of Majorbio Cloud Platform (www.majorbio.com). Detailed sequencing and analyses methods are shown in Supplementary Methods 2.1.

Analysis of plasma metabolite profiling

The collected blood samples were processed with reference to the previous methods to obtain plasma samples [21, 22], and the metabolites extracted therefrom were subsequently sequenced by non-targeted metabolomics. The metabolite information was obtained by matched with the online Human Metabolome Database (HMDB) (https://hmdb.ca/) and Metlin (https: //metlin.scripps.edu/) database. Then, the above pre-processed data were uploaded onto the Majorbio Cloud Platform (www.majorbio.com) for multivariate statistical analysis such as Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA), student’s t-test, unpaired Wilcoxon test, and fold change analysis setting a Variable Importance in Projection (VIP) threshold≥1, | log2(fold change) | >1, and p<0.05. Metabolic pathway annotation was conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (http://www.kegg.jp/) to determine the pathways in which the different metabolites were involved. For detailed information, see Supplementary Methods 2.2, 2.3.

Gene expression data analysis

PBMCs were extracted from the collected blood subjects with reference to the methods of previous studies see Supplementary Methods 2.4 [21–23]. Details regarding RNA extraction and detection, PCR amplification and library construction, sequencing and quality control are described in the Supplementary Methods 2.5. Subsequently, bioinformatics analysis was performed on the expression matrices obtained from the above procedure. Gene Set Enrichment Analysis (GSEA) based on KEGG pathway database was carried out using an online platform for data analysis and visualization (https://www.bioinformatics.com.cn). DEseq2 package of the language R was employed to examine the differential expression of the transcriptome, and the Benjamini & Hochberg method was used to adjust p-values. A threshold of p<0.05 & | log2(fold change) | > 1 was set for significant differential expression, and a volcano plot was drawn using ggplot2.Using David (https://david.ncifcrf.gov/) for differential gene function annotation, Gene Ontology (GO) function enrichment analysis and KEGG pathway enrichment analysis.

DIA (data-independent acquisition) based proteomics and phosphoproteomics analysis

DIA based quantitative proteomic and phosphoproteomic analyses of PBMC proteins were done at a biological company (Norogene, Beijing, China). The data was searched against the human Uniprot fasta protein database using the library search software Spectronaut-PulsarX 14.0 (Biognosys, Zurich, Switzerland) in the DDA scanning mode. Meanwhile, to improve the quality of analysis results, Spectronaut-Pulsar software further filtered the search results. The DIA data was imported into the Spectronaut software, and the ion-pair chromatographic peaks were extracted according to the pulsar constructed DDA database, and the subion matching and peak area calculation were performed to realize the simultaneous characterization and quantification of the peptides. The retention time was corrected using the iRT kit (Biognosys, Zurich, Switzerland) added in the samples, and the Qvalue cutoff value of precursor ions was set at 0.01. The statistical analysis of the protein quantification results was performed by the t-test method, and those proteins with significant quantitative differences between RA and RA_MD groups (p<0.05, | log2(fold change) | > 1) were classified as significantly differentially expressed proteins. Detailed methods are described in the Supplementary Methods 2.6.

LASSO classifier

Datasets were divided 70%/30% into training and test sets and LASSO classifiers were constructed using the glmnet R package, which was then subjected to stratified 5-fold cross-validation to distinguish between RA_MD and RA. The accuracy of the generated classifiers was determined by calculating the Area Under the Curve (AUC) of receiver operating characteristic using test data.

Neocosmospora and Turicibacte r were positively correlated with TC in RA_MD patients

We evaluated the alpha diversity, and found that the RA_MD group was characterized by an increase in fungal Chao richness and Shannon diversity, in contrast to a decrease in bacterial Chao richness and Shannon diversity compared to the RA group (Fig. 2A, B and Supplementary Fig. 1A, B). Principal coordinate analysis (PCoA) revealed that bacterial community composition among the RA_MD, RA and control groups formed significantly different clusters (Fig. 2C) while, fungi community composition formed significant overlapping clusters (Supplementary Fig. 1C). Among the three groups, dominant fungal phyla were Ascomycota, Basidiomycota, and Mucoromycota (Fig. 2D). The principal bacterial phyla were embraced on Firmicutes, Actinobacteriota, Proteobacteria and Bacteroidota (Fig. 2F). Regarding the genus differences, the RA and RA_MD groups had higher abundance of Candida and Blautia (Fig. 2E), while the abundance of Aspergillus and Faecalibacterium lower than control (Fig. 2G). In addition, Venn diagrams illustrated more overlapping numbers of fungal and bacterial genus among the three groups (Supplementary Fig. 1D, E). Further, by analyzing differences between RA_MD and RA groups, we identified 4 and 12 significantly differential fungal and bacterial genus respectively (Fig. 2H, I and Supplementary Table 4). Specifically, Neocosmospora, Sebacina, Romboutsia and Turicibacter were significantly increased in the RA_MD group compared to the RA group, while Lycoperdon and Veillonella were significantly decreased. And what’s more, 12 functionally predicted metabolic pathways (Supplementary Fig. 1F) were obtained based on the PICRUSt2 tool and correlation clustering analysis was performed with the 12 significantly different bacteria in Fig. 2J. Subsequently, we found that Veillonella had the strongest positive correlation with both alpha-Linolenic acid metabolism (ko00592) and lysine degradation (ko00310) pathways. In addition, both Romboutsia and Turicibacter were significantly positively correlated with the primary bile acid biosynthesis (ko00120) pathway, while Turicibacter was also significantly positively correlated with galactose metabolism (ko00052) and starch and sucrose metabolism (ko00500) pathways. Meanwhile, Turicibacter and Neocosmospora were positively correlated with TC and LDL (Supplementary Fig. 1G). Overall, our findings suggested that the abundance of gut microbiota regulating host TC, cholesterol, and fatty acid metabolism was elevated in RA_MD patients.

Open in a separate window

Fig. 2

The alteration of the structural composition and diversity of microbial communities among control, RA and RA_MD. The differences of α-diversity by Kruskal-Wallis H test for Chao index in fungi (A) and Shannon index in bacteria (B). * p<0.05. (C) The Beta diversity of bacteria shown by principal co-ordinates analysis (PCOA) based on Bray Curtis distance analysis. Relative abundance of fungi at phylum (D) and genus(E) levels. Relative abundance of bacteria at phylum (F) and genus(G) levels. The Wilcoxon rank-sum test bar plot between RA and RA_MD at the fungal (H) and bacterial (I) genus levels. Values represented mean and standard error. (J) Spearman’s correlation heatmap of Ko pathways and significant bacteria at genus level

Lower levels of unsaturated fatty acids (UFAs) in RA_MD patients

With the nontargeted metabolomics analysis, the plasma metabolome showed a clear separation between RA_MD and RA in OPLS-DA score scatter plots (Fig. 3A, B). We detected 179, 176 and 175 annotated cationic metabolites in the control, RA, and RA_MD groups, respectively, with 224 anionic metabolites in all three groups. Then, we identified 42 significantly differentially expressed metabolites (DEMs) between RA_MD and RA (Supplementary Table 5), 28 upregulated and 14 downregulated (Fig. 3C). Afterwards, KEGG enrichment analysis of this 42 DEMs yielded 18 significant pathways, including 6 metabolic pathways such as synthesis and degradation of ketone bodies, linoleic acid metabolism, fatty acid biosynthesis, glycine, serine and threonine metabolism, glycerophospholipid metabolism and biosynthesis of unsaturated fatty acids (Fig. 3D). Interestingly, among the three-group comparative analyses of control, RA and RA_MD, we observed that there were 6 DEMs that are currently not annotated on any KEGG pathway, of which galactonic acid expression was sequentially elevated, and hexadecanedioic acid, 9-OxoODE, all cis-(6,9,12)-Linolenic acid, N6-Methyl-L-lysine and (4Z,7Z,10Z,13Z,16Z,19Z)-4,7,10, 13,1 6,19-Docosahexaenoic acid (DHA) expressions were sequentially decreased and significantly different (Fig. 3E-H and Supplementary Fig. 2A, B). Therefore, we combined this 6 unannotated DEMs with 11 DEMs from the 6 metabolic pathways for correlation cluster analysis with clinical indicators (Fig. 3I). Moreover, we also found that FFA (Free Fatty Acid) showed significantly positive correlations with UFAs, including oleic acid, 20-HETE, 9-OxoODE ci−9−Palmitoleic acid, erucic acid, all cis-(6,9,12)-Linolenic acid and DHA, as well as significantly negative correlations with L-tryptophan and L-threonine. It was important to note that only a small number of UFAs, such as linoleic acid and trans-2-Hydroxycinnamic acid, were found to be enriched in RA_MD patients. Our findings suggested that the depletion of UFAs (Supplementary Table 5) and N6-Methyl-L-lysine may have a significant impact on the development and regulation of MD in RA patients.

Open in a separate window

Fig. 3

Down-regulation of unsaturated fatty acids (UFAs) in RA_MD patients. Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) displayed the differential metabolites of anionic (A) and cationic (B) modes between RA_MD and RA. (C) Volcano plot showed that the differentially expressed metabolites between RA and RA_MD. (D) The KEGG enrichment analysis of 42 significant metabolites between RA_MD and RA. (E-H) The relative abundance of galactonic acid, 20-HETE, 9-OxoDE and all cis-(6, 9, 12)-linolenic acid among control, RA and RA_MD. (I) Spearman’s correlation heatmap of differential metabolites and clinical characteristics

Identification and functional analysis of differentially expressed genes profiles

GSEA revealed that lipid metabolic pathways, namely arachidonic acid metabolism, linoleic acid metabolism and primary bile acid biosynthesis, were significantly upregulated in the RA_MD group (Fig. 4A, B and Supplementary Fig. 3A). In total, 355 differential expressed genes (DEGs) were identified between RA_MD and RA, of which 68 DEGs were up-regulated and 287 DEGs were down-regulated (Supplementary Fig. 3B and Supplementary Table 6). Then, KEGG pathway analysis revealed a significant enrichment of genes involved in metabolic pathways, such as arachidonic acid metabolism, linoleic acid metabolism and cholesterol metabolism, which were highly relevant to MD (Fig. 4C). Based on the above metabolic pathway, we identified 8 significantly key DEGs between RA_MD and RA, of which APOC1, CETP, CYP2E1, and CYP2J2 were down-regulated and CYP1B1, CYP7A1, LPL, and GFPT2 were up-regulated in RA_MD (Fig. 4D-I and Supplementary Fig. 3D, E). Additionally, GO enrichment analysis displayed the top 20 functional terms, and we found that the biological functions of the 355 DEGs was also enriched in the metabolic process, namely, long−chain fatty acid biosynthetic process, linoleic acid metabolic process and cholesterol metabolic process (Supplementary Fig. 3C). Correlation analysis of these 8 DEGs revealed a significant correlation between CYP1B1 and the other genes (Fig. 4J). In addition, only APOC1 had a significant positive correlation with CRP (Supplementary Fig. 3F). These results revealed that dysregulation of metabolic pathways, especially lipid metabolic pathways and cholesterol metabolic pathway, play an indispensable role in RA_MD patients.

Open in a separate window

Fig. 4

Identification and functional analysis of differentially expressed genes profiles. (A, B) The Gene Set Enrichment Analysis. (C) Bubble plot depicted KEGG enrichment analysis of 355 differentially expressed genes. (D-I) The relative abundance of 6 significantly expressed genes. (J) The correlated relationship of 8 different genes that were associated with metabolic diseases

Combined multi-omics analysis revealed dysregulation of lipid metabolism

Given that all multi-omics data and clinical parameters demonstrated alterations in RA_MD compared to RA, we then assessed whether there were potential associations across multiple data types. We first combined gut microbiota, metabolites and clinical indicators in a correlation network analysis, which showed that linoleic acid (M2) and glycerophosphocholine (M4) were significantly positively correlated with Romboutsia (B5) and Turicibacter (B6), respectively. While, Lachnospira (B7) was negatively correlated with N6-Methyl-L-lysine and oleic acid (M10) that was substantially correlated with several UFAs as well as amino acids (Fig. 5A and Supplementary Table 7). Based on the shared metabolic pathways, we found that both transcriptomics and metabolomics profiles were mainly enriched in lipid metabolic pathways (Fig. 5B). In addition, our LASSO analyses combining microbial and metabolite data yielded less favorable results than analyses using metabolite data only, and we ultimately used 17 DEMs data and identified 4 metabolite biomarkers (M1: 1-Palmitoyl-sn-glycero-3-phosphocholine, M14: hexadecanedioic acid, M3: L−Tryptophan, M11: N6−Methyl−L−lysine), whose diagnostic performance was generally above 0.7 (AUC) and united AUC up to 0.958 (Fig. 5C and Supplementary Fig. 4). After that, we combined proteins, phosphoproteins and genes involved in cholesterol metabolism to perform the PPI network analysis, in which the phosphoprotein APOL1, where the phosphorylation occurs on serine 4 (S4), as well as the APOB protein, played a central role in the network analysis (Fig. 5D and Supplementary Fig. 5). Finally, the molecules obtained from multi-omics and RA_MD subgroups analyses were mainly focused on lipid metabolic pathways, followed by cholesterol metabolism pathways, amino acid metabolism pathways and carbohydrate metabolism pathways (Fig. 5E, Supplementary Fig. 6 and Supplementary Table 8). Our findings indicated that dysregulation of lipid metabolism was predominant in RA_MD and involved in gut microbiota, genes and proteins associated with TC, cholesterol and fatty acid metabolism.

Open in a separate window

Fig. 5

Combined multi-omics analysis revealed dysregulation of lipid metabolism. (A) Correlation network among differential metabolites, gut microbiota and clinical indicators. (B) The shared metabolic pathways between metabolomics and transcriptomics. (C) The Roc curve of significantly expressed gut microbiota and metabolites. (D) Protein-protein interaction (PPI) network map among differential genes, proteins and phosphoproteins involved in the cholesterol metabolism. (E) 3 types of metabolism pathways displayed the connections among metabolites, genes, proteins and phosphoproteins

We thank our colleagues in the Clinical Research Centre for their help, the volunteers who participated in this study for their active cooperation, and all the doctors and nurses in the Department of Rheumatology for their support. We also thank SMART and MetaboAnalyst for the figure preparation.