A total of 32 patients diagnosed with severe plaque psoriasis were recruited at the General Hospital of Ningxia Medical University (Ningxia Province, China) for this study from December 2018 to May 2019. At the same time, 29 healthy volunteers were recruited. The average age of the patients was 38.16 years, with a range of 17–74 years. For healthy controls, the average age was 35.53 years, with a range of 23–54 years. The severity of psoriasis was quantified by using the Psoriasis Area and Severity Index (PASI) score (38.96 ± 2.64, mean ± SE), the Psoriasis Global Assessment score (4.41 ± 0.13, mean ± SE), and the body surface area score (24.85 ± 2.82, mean ± SE). Patients who met the following criteria at the same time were included: patients with severe plaque psoriasis (PASI score ≥12) (Mrowietz et al., 2011), patients with at least half a year of disease duration, and patients who had previously received at least one course of systemic treatment without obvious improvement. The exclusion criteria were as follows: volunteers with severe liver or kidney damage, mental illness, hematopoietic dysfunction, or other serious organic disease; patients who received immunosuppressive treatment or high doses of glucocorticoids or retinoid treatment in the previous 2 months; and all participants, including healthy controls and psoriasis patients, who had used any skin care product or lotion in the previous week. None of the healthy volunteers had a history of any immune diseases, and none of them had any skin disorders. All samples and clinical information were obtained under the condition of informed consent. This study was conducted with the approval of the institutional review board of the General Hospital of Ningxia Medical University and in accordance with the Declaration of Helsinki.

Skin microbiome samples were collected according to a previous report (Oh et al., 2014). Briefly, a swab was rinsed with phosphate-buffered saline, and a defined skin area of approximately 2 × 2 cm² was swabbed at least 20 times to maximize the amount of microbiome DNA collected. All samples were stored at −80°C until extraction.

The plasma samples were collected on the same day after overnight fasting with a heparin sodium anticoagulant tube. The samples were then centrifuged at 3,000 rpm for 10 min at room temperature. The supernatants were collected and aliquoted into different tubes and stored at −80°C.

Genomic DNA from skin microbiome samples was extracted by using the Mobio Powersoil DNA Isolation Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. The V3 and V4 regions of the 16S rRNA genes were amplified by using Phusion^® High-Fidelity PCR Master Mix with GC Buffer (New England Biolabs, MA, United States) and the primers 341F and 806R. After purification of the polymerase chain reaction product by using AMPure XP magnetic beads (Beckman Coulter, IN, United States), the samples were analyzed by the Illumina NovaSeq 6000 platform (Illumina, CA, United States) through a paired-end sequencing strategy.

After Illumina sequencing, barcode and primer sequences were removed. Specific tags were generated by FLASH software¹ according to the overlap information of the reads. Then, we applied Trimmomatic software (v0.33) to remove the low-quality tags and obtained clean tags. Clean tags were further filtered to exclude the chimeric sequences by using UCHIME software (v4.2). Next, the remaining sequences with an identity >97% were classified as operational taxonomic units (OTUs) by using Uparse software². Taxonomic information was annotated by searching against the SSU rRNA database³. OTUs were then assigned to different phylogenetic levels (kingdom, phylum, class, order, family, genus, and species). Alpha diversity and beta diversity were analyzed by QIIME software (v1.9.1) based on the effective tags. The relative abundance and the difference in diversity were compared by Student t-test and the Wilcoxon rank-sum test. Furthermore, linear discriminant analysis coupled with effect size (LEfSe) was applied to identify microorganisms that can be used to discriminate psoriasis patients from people with no psoriasis.

A 100-μL plasma sample from each patient was mixed with 300 μL of methanol containing 1 μg/mL 2-chloro-L-phenylalanine (Hengbai Biotech, Shanghai, China) as the internal standard. After brief sonication in ice water for 10 min, all the samples were placed at −40°C for 1 h and centrifuged at 10,000 rpm for 15 min at 4°C. Then, the samples were resuspended in 100 μL of 50% acetonitrile. For quality control (QC) sample preparation, a mixture containing an equal volume (10 μL) of each plasma extract was prepared.

For liquid chromatography–mass spectrometry (LC-MS) metabolomic data collection, all plasma samples were analyzed by a 1290 UHPLC instrument (Agilent Technologies, CA, United States) coupled with a Thermo Q Exactive Focus (Thermo Fisher Scientific, MA, United States) by Biotree Ltd. (Shanghai, China), according to previously reported methods with minor modifications (He et al., 2020). Briefly, mobile phase A in positive ion mode was 0.1% formic acid in water, and in negative ion mode, it was 5 mmol/L ammonium acetate in water. Mobile phase B was acetonitrile. The elution gradient was set as follows: 1% B at 1 min, 99% B at 8 min, 99% B at 10 min, 1% B at 10.1 min, and 1% B at 12 min. The flow rate was set to 0.5 mL/min. The Q Exactive mass spectrometer was run at a spray voltage of 4.0 kV in positive mode and −3.6 kV in negative mode. Other ESI source conditions were as follows: sheath gas flow rate of 45 Arb, Aux gas flow rate of 15 Arb, and capillary temperature of 400°C. All MS1 and MS2 data were obtained under the control of Xcalibur (Thermo Fisher Scientific). A UPLC HSS T3 column (Waters, MA, United States) was used for all analyses. Organic reagents, including methanol, acetonitrile, and formic acid (HPLC grade), were purchased from CNW Technologies (Dusseldorf, Germany).

The extracted plasma samples were resuspended in 30 μL of methoxyamine hydrochloride (20 mg/mL in pyridine) and incubated at 80°C for 30 min. After derivatization with 40 mL of N,O-bis(trimethylsilyl)trifluoroacetamide (BSTFA) with 1% trimethylsilyldiethylamine (Sigma, Darmstadt, Germany) at 70°C for 1.5 h, the samples were cooled down gradually to room temperature. For QC sample preparation, a mixture containing an equal volume (10 μL) of each plasma extract was prepared. An additional 5 μL of saturated fatty acid methyl esters (Dr. Ehrenstorfer GmbH, Augsburg, Germany) dissolved in chloroform was added to the QC samples for gas chromatography (GC)–MS analysis.

Gas chromatography–time of flight (TOF)–MS analysis was carried out by using an Agilent 7890 gas chromatograph (Agilent Technologies) coupled with a Pegasus HT TOF mass spectrometer (LECO, Michigan, United States). In this analysis, a DB-5MS capillary column (30 m × 250 μm × 0.25 μm, Agilent Technologies) was used. The carrier gas was helium, the front inlet purge flow was set as 3 mL/min, and the gas flow rate was 1 mL/min. The temperature gradient was set as 50°C for 1 min, increased to 310°C at a rate of 20°C/min, and then maintained for 6 min. The front injection temperature, transfer line temperature, and ion source temperature were 280, 280, and 250°C, respectively. The energy was −70 eV in electron impact mode. The MS data were acquired in full-scan mode with an m/z range of 50–500 at a rate of 12.5 spectra per second after a solvent delay of 4.85 min. A 1-μL sample was injected for this analysis.

ProteoWizard software was used to transform the original LC-MS data to mzXML format. The data were processed by XCMS. GC-MS raw data were processed by Chroma TOF software. After peak identification, peak alignment, peak extraction, retention time (RT) correction, and peak integration, a three-dimensional data matrix was obtained. To make the metabolomics data reproducible and reliable, peaks with relative standard deviations greater than 30% in the QC samples were filtered out. The remaining peaks were identified by comparison of RT and mass to charge ratio (m/z) indexes in a library containing spectral information from the online database of HMDB⁴, Kyoto Encyclopedia of Genes and Genomes (KEGG)⁵, and the in-house library. The GC-MS data were matched with the LECO-Fiehn Rtx5 database. Peak intensity was quantified by using the area under the curve. The data matrix was further processed by removing the peaks with missing values in more than 50% of the samples and substituting the remaining missing values with half of the minimum value. Then, a new data matrix was generated by normalizing the data to the peak intensity of the internal standard.

Statistical analysis was performed by using Microsoft Excel (Microsoft Inc., Redmond, WA, United States) and R software version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria). The differential abundance of bacterial taxa at different levels (phylum, class, order, family, and genus) between psoriasis patients and healthy controls was calculated by the Wilcoxon rank-sum test and Metastat. The differences in alpha diversity indexes were determined by Student t-test. The beta diversity difference between psoriasis patients and the control group was analyzed by analysis of similarity (ANOSIM). To understand the difference in the metabolomic profile between psoriasis patients and healthy people, multivariate statistical analyses, including principal component analysis (PCA) and orthogonal projections to latent structure-discriminant analysis (OPLS-DA), were carried out. Small molecules with a VIP (variable importance in projection) >1 in OPLS-DA and p < 0.05 by Student t-test were considered significantly altered metabolites. Spearman correlation was carried out to determine the relationship between the skin microbiota and plasma metabolites.

We recruited 32 severe plaque psoriasis patients (PASI > 12) and 29 healthy controls to identify the psoriasis-related microbiota. After QC, the DNA sample amounts from only 26 patients and 10 controls were sufficient for 16S sequencing.

Overall, we obtained 83,998 effective tags and 7,887 OTUs according to 97% similarity. After taxonomic assignment against the Silva132 database, 7,606 OTUs were annotated at different phylogenetic levels (Supplementary Table 1). According to the species accumulation curve, the sequencing data and samples were sufficient for taxon identification. However, there were no significant differences between control individuals and psoriasis patients in terms of number of species on skin (Supplementary Figure 1A). In addition, the alpha diversity indexes, including the total observed species, Shannon index, ACE index, Simpson index, and Chao1 index, of the skin of psoriasis patients were not significantly different from those of the control group (Supplementary Figures 1B–F). To identify the microbes that were altered in psoriasis patients, we then conducted Student t test at the genus level (Figure 1A). The average abundance of Lactobacillus, which is widely distributed in the human gut and skin and plays a role as a lactic acid producer, was increased in psoriasis patients. This may suggest a potential positive role of Lactobacillus in regulating skin cell proliferation, which is consistent with a previous report that Lactobacillus was capable of enhancing skin repair after UV damage (Im et al., 2018). Moreover, the abundances of Thermomonas and Luteimonas, which are pathogenic members of Proteobacteria (phylum)_Gammaproteobacteria (class), were also increased, suggesting that the skin of psoriasis patients was a pathogenic environment. To further analyze the alterations in the microbiota in psoriasis patients, we applied Metastat, another widely used statistical analysis tool, to screen for significantly changed organisms (Figure 1B). Similarly, the change in the Lactobacillus abundance was also identified as an important alteration. In addition, another member of Gammaproteobacteria, Vibrio, was identified as being significantly elevated in psoriasis patients. Together, these data suggest an elevation in the abundance of pathogenic bacteria, especially Gammaproteobacteria, in psoriasis patients.

To further examine the alterations associated with psoriasis, we conducted LEfSe analysis. The main differences were the increase in abundance of undefined_Cyanobacteria (class unidentified_Cyanobacteria and order Cyanobacteria) in psoriasis patients (Figure 2A). Some differences were also observed at a lower taxonomic level. Psoriasis patients showed a loss in the abundance of the genus Citrobacter (Figure 2B). Taken together, these data indicate alterations in the commensal gut microbiome composition in psoriasis patients, suggesting dysregulation of the microbial community.

To further determine the functional impact of skin gut microbiota alterations in psoriasis, we predicted the KEGG pathways based on the 16S sequencing data by using PICRUSt software (Langille et al., 2013). Metabolic pathways ranked as the most abundant pathways predicted, accounting for approximately 50% of the pathways (Figure 3A). Among these pathways, carbohydrate metabolism and the metabolism of other amino acids were obviously decreased (Supplementary Figure 2). In contrast, pyrimidine and purine metabolism (nucleotide metabolism); glycolysis/gluconeogenesis, oxidative phosphorylation, and methane metabolism (energy metabolism); the metabolism of cofactors and vitamins; and the biosynthesis of other secondary metabolites showed an increase in psoriasis patients compared with healthy controls (Figure 3B and Supplementary Figure 2).

Microbiota alterations have been reported to be correlated with tissue metabolism in many studies (Liu et al., 2017; Olson et al., 2018). In addition, our data showed that skin microbiota–mediated small molecule metabolism was impacted by psoriasis (Figure 3). We thus investigated the metabolic alterations in psoriasis patients by applying GC and ultrahigh-pressure LC coupled with MS. In general, a total of 3,562 features and 716 metabolites were obtained (Supplementary Table 2). PCA showed very obvious separation of metabolic profiles between psoriasis patients and healthy controls, demonstrating different metabolic activities (Figure 4A). After statistical analysis, we obtained 117 significantly altered metabolites (VIP >1 and p < 0.05) (Figure 4B). Among them, we found several microbiome-generated metabolites that were also significantly changed. These included taurochenodeoxycholic acid (TCDCA), deoxycholic acid glycine conjugate (GDCA), chenodeoxycholic acid glycine conjugate, and L-kynurenine (Chávez-Talavera et al., 2017; Agus et al., 2018; He et al., 2020). To uncover the metabolic pathway alterations, we conducted KEGG pathway analysis of the differentially expressed metabolites by using Metaboanalyst⁶ (Figure 4C). Branched-chain amino acid metabolism (valine, leucine, and isoleucine biosynthesis), which was reported to be closely related to microbiota metabolic activity (Liu et al., 2017), was significantly altered. In addition, the metabolism of α-linolenic acid and linoleic acid, which reflect the inflammation status of tissues (Sergeant et al., 2016), was also significantly altered.

Many articles have reported the correlation of the gut microbiota and blood metabolism (Liu et al., 2017; He et al., 2020), whereas little is known about the relationship of the skin microbiota and blood metabolism. In this study, we carried out Spearman correlation analysis of the annotated skin microbiota at the genus level and the identified plasma metabolites. The association of the skin microbiota and plasma metabolites was different between healthy controls and psoriasis patients (Figures 5A,B), suggesting that the alteration of plasma metabolites was closely related to the skin microbiome. Interestingly, most of the associations between Lactobacillus and plasma metabolites and the associations between Enterococcus and plasma metabolites in healthy controls (Figure 5A) disappeared in psoriasis patients (Figure 5B). In addition, new correlations between Vibrio, Ferruginibacter, Romboutsia, and plasma metabolites were established in psoriasis patients (Figure 5B). The metabolites that showed a significant positive association with specific skin bacteria in both healthy controls and psoriasis patients were itaconic acid, crotonic acid, and heptadecanoic acid, which are involved in lipid metabolism (Figures 5A,B). Notably, some plasma metabolites were negatively associated with the skin microbiota in psoriasis patients. In addition to several lipids, xanthine, D-ribose 5-phosphate, and uric acid participate in nucleotide metabolism (Figure 5B). These results suggest a role of the skin microbiota in influencing lipid and nucleotide metabolism in psoriasis patients.

To determine a global relationship between the skin microbiota and plasma metabolism, we then conducted Spearman correlation analysis by using all the samples from healthy controls and systemic lupus erythematosus (SLE) patients. The correlations were shown in a Cytoscape network (Figure 5C), further suggesting a fundamental relationship of the skin microbiota and plasma metabolism. Interestingly, receiver operating characteristic (ROC) curve analysis revealed that many skin microbiota–associated plasma metabolites are potential biomarkers for SLE classification (Supplementary Figure 3).