Patient information and sample collection. A total of 57 rosacea patients before and after treatment with doxycycline of sub-antimicrobial dose and 63 age-, sex-, and BMI-matched healthy controls were recruited from the Department of Dermatology in Xiangya Hospital, Central South University. As for the validation population, 30 patients diagnosed with rosacea and 20 age-, sex-, and BMI-matched healthy controls were recruited from the Department of Dermatology in the First Hospital of Changsha. Inclusion criteria included newly diagnosed and untreated rosacea patients according to the 2017 diagnostic criteria determined by the National Rosacea Society Expert Committee without any other metabolic comorbidities (52). The exclusion criteria for all the participants were systemic diseases and other skin disease, history of systemic immunomodulators or antibiotics, intake of prebiotics or probiotics, extreme diets in the previous 12 weeks, pregnancy, and lactation. Patients were clinically evaluated for IGA scores and CEA scores as previously described (53). Demographic and baseline clinical characteristics of 57 rosacea patients and 63 healthy controls are listed in Table 1. Serum samples of participants were collected at initial diagnosis and follow-up 8 weeks after oral doxycycline (40 mg/d) treatment. Peripheral blood samples were collected after overnight fasting for 10 hours and set aside for 30 minutes to obtain serum. All samples were stored at –80°C immediately after collection until analysis.

Targeted metabolomics sequencing and analysis. The metabonomics analysis was performed by Q300 Kit (Metabo-Profile) as previously described (11, 54). Specifically, metabolites were quantitated using a UPLC-MS/MS system (ACQUITY UPLC-Xevo TQ-S, Waters Corp.). All of the standards were obtained from MilliporeSigma, TRC Chemicals, and Steraloids Inc. To obtain individual stock solution, we weighed and prepared standards in water, methanol, hydrochloric acid solution, or sodium hydroxide solution. Each stock solution was appropriately mixed to create stock calibration solutions. A total of 25 μL of ice-bath serum was added to a 96-well plate. A total of 120 μL of ice-cold methanol with partial internal standards was added to each sample and then vortexed vigorously for 5 minutes. The plate was centrifuged at 4,000g for 30 minutes at 10°C, 30 μL of supernatant was transferred to a clean 96-well plate, and then 20 μL of derivative reagents was added to each well. The plate was sealed and the derivatization was carried out for 60 minutes at 30°C. Next, samples were diluted by 330 μL of ice-cold 50% methanol solution. Then the plate was stored at –20°C for 20 minutes and centrifuged at 4,000g at 4°C for 30 minutes. After transferring 135 μL of supernatant to a new 96-well plate containing 10 μL internal standards in each well, we added serial dilutions of derivatized stock standards to the left wells. Finally, the sealed plate was prepared for LC-MS analysis. The UPLC instrument settings were as follows: column: ACQUITY UPLC BEH C18 1.7 μM; column temperature: 40°C; sample manager temperature: 10°C; mobile phases: water with 0.1% formic acid (A) and acetonitrile (70:30, B); gradient conditions: 0–1 minutes (5% B), 1–11 minutes (5%–78% B), 11–13.5 minutes (78%–95% B), 13.5–14 minutes (95%–100% B), 14–16 minutes (100% B), 16–16.1 minutes (100%–5% B), 16.1–18 minutes (5% B); flow rate: 0.40 (mL/min); injection volume: 5.0 μL. The conditions of mass spectrometer were as follows: capillary: 1.5 kV and 2.0 kV; source temperature: 150°C; desolvation temperature: 550°C; desolvation gas flow: 1,000 L/h.

The acquired raw data were assessed by the MassLynx software (v4.1, Waters Corp.) to perform peak integration, quantitation, and calibration for each metabolite. Statistical analysis was processed by the powerful package from R studio. The concentration of substances was determined by making the comparison between the unknown sample and the calibration curve. Then, the iMAP software (version 1.0; Metabo-Profile) was operated for the targeted metabolites. Metabolites were identified based on the Human Metabolome Database and the Kyoto Encyclopedia of Genes and Genomes (KEGG). To understand the difference of metabolomics profiles between patients with rosacea and healthy people, multivariate statistical analyses, including PLS-DA and OPLS-DA, were carried out. Meanwhile, univariate statistical analyses, such as 2-tailed Student’s t test, Mann-Whitney-Wilcoxon (U test), 1-way ANOVA, Kruskal-Wallis, and correlation analysis, were used to identify the altered metabolites in patients with rosacea. KEGG pathway enrichment analysis was conducted using the KEGG database (version 89.1) with hypergeometric test comparing all the identified metabolites. The metabolites with variable importance on projection > 1 and P < 0.05 were considered significantly changed metabolites.

Mice and treatments. Age- and sex-matched BALB/c mice were purchased from Hunan SLAC Laboratory Animal Co., Ltd. All mice were bred and maintained under specific pathogen–free conditions with food and water ad libitum and were acclimatized to the new environment for 1 week before experiments. Eight-week-old BALB/c mice were treated with glutamic acid, aspartic acid, or leucine (MilliporeSigma) at a dosage of 2 mg/kg or 10 mg/kg per day from day 1 to day 5 by gastric perfusion. Mice from control group were treated with the same amount of normal saline. Capsaicin or vehicle was applied topically to the ears of the mice on day 5. The mice were imaged and euthanized within 30 minutes after topical application for ear lesion and subsequent analysis. Ear biopsy specimens were for histological analysis, immunohistochemistry, immunofluorescence, and quantitative PCR (qPCR); serum samples were for quantification of amino acids; DRG biopsy specimens were for qPCR and immunoblotting.

Quantification of amino acids of mouse serum samples. Peripheral blood samples of mice were centrifuged to obtain serum samples. We mixed 100 μL aliquots with 400 μL of cold methanol/acetonitrile (1:1, v/v) to remove the protein. The mixture was centrifuged for 20 minutes (14,000g, 4°C). The supernatant was dried in a vacuum centrifuge. For LC-MS analysis, the samples were redissolved in 100 μL acetonitrile/water (1:1, v/v) and adequately vortexed, then centrifuged (14,000g, 4°C, 15 minutes). The supernatants were collected for LC-MS/MS analysis. In electron spray ionization–positive modes, the conditions were set as follows: source temperature 500°C, ion source gas 1 (Gas1): 40, ion source gas 2 (Gas2): 40, curtain gas: 30, ion sapary voltage floating 5,500 V; adopt the multiple reaction monitoring mode detection ion pair. The Multiquant software was used to extract chromatographic peak area and retention time. The metabolites were identified by AA standards after retention time correction. All amino compound standards were purchased from MilliporeSigma.

Histological analysis. The histological analysis was carried out as per a previous study (55, 56). Formalin-fixed and paraffin-embedded ear skins of mice were cut into 5 μm skin sections and then stained with H&E. To determine the histological features, the number of infiltrating cells in the dermis was averaged in 6 randomly selected microscopic areas (original magnification, 200×) in each mouse.

Immunohistochemistry. Ear samples of mice were fixed in formalin and then embedded in paraffin and were cut into 5 μm skin sections. Immunohistochemistry was performed according to previous methods (57). Skin sections were incubated in primary antibody CD31 (1:100, Cell Signaling Technology, 77699s). The averaged circumference of primary capillaries stained by CD31 in 3 random hpfs for each mouse (n = 6) in each group was measured using ImageJ.

Immunofluorescence. Immunofluorescence of skin sections was conducted as previously described (58). Briefly, ear skins from mice were fixed in 4% paraformaldehyde and then frozen in OCT. Ear sections were cut into 8 μm, then washed with phosphate-buffered saline (PBS) 3 times. After being blocked for 60 minutes with blocking buffer (5% normal donkey serum, 1% BSA, 0.3% Triton X-100), sections were incubated in primary antibodies CD31 for immunofluorescence (1:100, BD Biosciences, 558736) and p-eNOS (1:100, Cell Signaling Technology, 9571s) overnight at 4°C. After wash, we added Alexa Fluor 488– or 594–conjugated secondary antibody (1:500, Thermo Fisher Scientific, A21208 and A21207, respectively) on sections for 60 minutes at room temperature. Next, we washed sections with PBS and counterstained them with DAPI. All pictures were acquired with an ECLIPSE Ni-U Microscope. The number of positive cells was counted and averaged in 5 randomly selected microscopic fields (original magnification, 200×) in each mouse.

Cell culture and treatment. HaCaT keratinocytes (immortalized human keratinocyte cell line), obtained from NTCC (Biovector Science Lab), were cultured in DMEM (Gibco, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin (Thermo Fisher Scientific), and 1% glutamine (Invitrogen) at 37°C in a humidified CO2 incubator (5% CO2). HDMEC line purchased from ATCC was cultured in MCB131 (Gibco, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin, 10 ng/mL human epidermal growth factor, 400 ng/mL hydrocortisone, and 1% glutamine at 37°C in a humidified CO2 incubator (5% CO2). For amino acid treatment, at a confluence of 50%, cells were starved overnight, then incubated with glutamic acid, aspartic acid, or leucine (at indicated doses) for the indicated time. All experiments were performed at least 3 times.

RT-qPCR. Total RNA was extracted from mouse ears, DRG neurons, HaCaT keratinocytes, and HDMECs by TRIzol reagent (Thermo Fisher Scientific), and a NanoDrop spectrophotometer (ND-2000, Thermo Fisher Scientific) was employed for RNA quality control. mRNA was reverse-transcribed into cDNA by the Maxima H Minus First Strand cDNA Synthesis Kit with dsDNase (Thermo Fisher Scientific) according to the manufacturer’s instructions. The real-time PCR was conducted with iTaq Universal SYBR Green Supermix (Bio-Rad) on a LightCycler 96 (Roche) thermocycler. The relative expression of each gene relative to GAPDH was analyzed by using the ΔCT method, and the fold change was normalized to the control group. The primer sequences of genes used in this study are listed in Supplemental Table 5.

Immunoblotting. The mouse DRG biopsies and cells, washed with cold PBS, were lysed with RIPA buffer including protease inhibitors (Thermo Fisher Scientific). We took the supernatant after centrifugation (12,000g, 15 minutes, 4°C). The proteins acquired were quantified through bicinchoninic acid assay (Thermo Fisher Scientific) and separated on SDS-PAGE and transferred to a PVDF membrane. The membrane was then blocked with 5% nonfat milk for 1 hour at room temperature and probed with primary antibodies overnight at 4°C. We washed the membrane 3 times in TBS containing 0.1% Tween 20 and incubated with secondary antibodies HRP-conjugated Goat anti-Mouse IgG (1:10,000, Santa Cruz Biotechnology, sc-2005) and HRP-conjugated Goat anti-Rabbit IgG secondary antibody (1:10,000, Santa Cruz Biotechnology, sc-2004) for 1 hour at room temperature. The immunoreactive bands were revealed by the HRP substrate (Luminata, MilliporeSigma) on ChemiDoc XRS+ system (Bio-Rad). GE Healthcare (now Cytiva) ImageQuant LAS 4000 Mini was used for data analysis. The primary antibodies in this study were as follows: Rabbit anti–p-eNOS (phospho S1177, 1:1,000, Cell Signaling Technology, 9571s), Rabbit anti-eNOS (1:1,000, Cell Signaling Technology, 9572s), Rabbit anti-VPAC2 (1:500, AiFang biological, AF06401), Rabbit anti-VIP (1:500, Santa Cruz Biotechnology, sc-25347), Rabbit anti–Lamin B (1:2,000, Abcam, ab16048), Mouse anti–α Tubulin (1:5,000, Abcam, ab7291), Mouse anti-GAPDH (1:10,000, Proteintech, 60004-1), and Mouse anti-β Actin (1:2,000, Santa Cruz Biotechnology, sc-47778).

Measurement of intracellular NO production. The production of NO level in HaCaT keratinocytes and HDMECs was measured using a fluorescent indicator, DAF-FM DA. At a confluence of 50%, we treated cells with amino acid for 1 hour and removed the supplement, then loaded with DAF-FM DA for 20 minutes at 37°C. Thereafter, cells were gently washed 3 times with PBS to remove extracellular DAF-FM DA. All pictures were acquired with an ECLIPSE Ni-U Microscope, and the average fluorescence intensity of 30 cells in each repeated experiment was measured with ImageJ (NIH).

Multiomics analysis of network pharmacology, transcriptomics, and metabolomics. The possible target genes of doxycycline were predicted based on BATMAN-TCM database, ChEMBL database, Comparative Toxicogenomics Database, DGIBD database, PharmMapper database, Swiss Target Prediction database, STITCH database, and Therapeutic Target Database, then by taking the intersection of all genes across databases. Next, the transcriptome of rat liver was analyzed by the DESeq package of R studio to obtain differentially expressed genes using liver sequencing data from rats with doxycycline or vehicle treatment deposited at the Gene Expression Omnibus database, under accession number GSE59923 (https://www.ncbi.nlm.nih.gov/geo/). After making genes’ names uniform across species, these molecular data were intersected with doxycycline-targeted genes. All the raw data of metabolites were standardized through normalization by sum, square root transformation, and Pareto scaling. The univariate statistical analyses were carried out to identify the altered metabolites in patients with rosacea before and after doxycycline treatment. Target genes by which doxycycline regulates glutamic acid and aspartic acid in rosacea patients were identified by metabolome-transcriptome combined analysis of varied metabolites and genes using MetaboAnalyst software (version 5.0).

Statistics. All statistical analyses were carried out using GraphPad Prism 6 (GraphPad Software). The normal distribution and similar variance of data from different groups were examined. The significance of differences (*P < 0.05, **P < 0.01) between groups was determined by 2-tailed unpaired Student’s t test for 2 groups’ comparisons or 1-way ANOVA with Bonferroni’s post hoc test for comparisons between multiple groups. When the data were not normally distributed or there existed heterogeneity variances between the 2 groups, statistical significance between different groups was determined by Whitney-Wilcoxon (U test) and Kruskal-Wallis. We performed Pearson’s r test or Spearman’s r test (for abnormally distributed data) for correlation analysis. All data represent the mean ± SEM.

Study approval. All human studies were approved by the ethical committee of the Xiangya Hospital, Central South University, and written informed consent was obtained from all participants. All mice were housed in specific pathogen–free conditions, and all procedures were conducted according to the instructions and permissions of the ethical committee of the Xiangya Hospital, Central South University.