C. spinosa fruits (6/habitat) were collected from three districts, including Turpan, Karamay, and Ili in Xinjiang Uygur Autonomous Region of China, which were named as groups A, B, and C. The fruits were identified by Dr. Jiang He from the Xinjiang Institute of Materia Medica. The fruits from Turpan were green and medium in size ( Figure 1A ), the fruits from Karamay were the smallest and bottle-green or brown ( Figure 1B ), and the fruits from Ili were the largest and brown ( Figure 1C ). The GPS coordinates and climate information of the habitats can be found in Table 1 . Turpan had the least precipitation, the most arid climate, and the highest annual average temperature, while Ili had the most humid climate and the lowest annual average temperature, and Karamay had the middle climate index. The contents of N, P, and K in Ili soil were the highest.

C. spinosa fruits were extracted with 95% ethanol, then the extract was concentrated and lyophilized, named CSE. The yield of the extract was about 24%. CSEs prepared with C. spinosa fruits from Turpan, Karamay, and Ili were named CSEA, CSEB, and CSEC, respectively. To detect the anti-inflammatory effects of CSE in vitro, different concentrations (0.5, 1.0 and, 1.5 mg/mL) of CSEs were used to treat DCs in the presence of LPS. Results of flow cytometry showed that there was no significant change in cell proportion among all groups, suggesting that the selected doses of CSEs did not affect the viability of DCs ( Figure 2A ). Compared with the LPS group, CSEs did not inhibit the expression of surface molecule CD40 induced by LPS, whereas CSEC (1.5 mg/mL) significantly suppressed the expression of CD86 induced by LPS ( Figure 2B ). Three pro-inflammatory cytokines, including TNF-α, IL-6, and IL-12p40, were significantly decreased ( Figure 2C ).

In vivo, CSE treatment significantly suppressed the expression of pro-inflammatory cytokines TNF-α and IL-6 and relieved the degree of ear edema in the TPA-induced mouse ear edema model ( Figure 2D ).

These results showed that C. spinosa from all three different districts had anti-inflammatory effects, but the effect of fruits from Ili was better than the other two groups in inhibiting DC maturation.

Base peak ion chromatograms of the three groups were shown as fingerprints. There were significant differences among the samples of Ili and the other two districts ( Figure 3A ). A total of 82 compounds were identified, of which 47 were in positive ion mode ( Tables S1 , S3 ) and 43 were in negative ion mode ( Tables S2 , S4 ). Eight compounds were found in both modes. Principal component analysis (PCA) was performed on the 18 samples (6/habitat) ( Figure 3B ) or together with quality control (QC) ( Figure S1 ). All samples were within Hotelling’s T-squared ellipse, indicating that the detection system had good stability. In positive ion mode, PC1, PC2, and PC3 were 66.95%, 19.14%, and 8.98%, respectively, while in negative ion mode, PC1, PC2, and PC3 were 87.59%, 7.16% and, 1.29%, respectively ( Figure 3B ). The model of orthogonal partial least squares discriminant analysis (OPLS-DA) was stable and reliable ( Figure S2 ). The score plots of PCA and OPLS-DA exhibited an obvious separation among the samples.

The identified compounds were analyzed by hierarchical clustering analysis. The heatmap of hierarchical cluster analysis (HCA) showed that group C was separated from the other two groups ( Figure 3C ). Differential metabolites were screened by the Venn analysis in positive and negative modes, respectively ( Figure 3D ; Table S5 ).

Cosine similarity analysis showed a significant difference among the three groups in negative ion mode, and the similarity value in group C was 86% ( Figure 3E ). The clustering tree revealed that the distance between groups A and B was close, but group C was far from the other two groups ( Figure 3F ).

Stachydrine, ferulic acid, and chrysoeriol were selected to be quantified, and their contents in groups A, B, and C were consistent with the metabolomics ( Figure S3 ). These results suggested that the metabolic level of group C was considerably different from that of the other two groups. The metabolomics data were deposited in the OMIX repository, accession number OMIX004470 (https://ngdc.cncb.ac.cn/omix/preview/QoEYJ1oa).

By retrieving the database, 15 anti-inflammatory compounds were selected. They included one alkaloid, four lipids, three flavones, four phenols, one glycoside, and two other compounds ( Figure 4A ).

Considering the quantitative value ( Figure 4B ) and log2 (fold change) value of the compound ( Figure 4C ), the contents of almost compounds were similar between groups A and C except for oleoyl ethyl amid, 3,4-dihydroxybenzaldehyde, 2,3,5,6-tetramethylpyraz and vanillic acid, but the contents of almost compounds in group A and B were different with that in group C except for 6-gingerol and ferulic acid. The contents of three flavonoids including diosmetin, chrysoeriol, and astragalin in groups A and B were significantly higher than that in group C. The contents of linoleic acid, monoolein, and cappariloside A in group C were significantly higher than that in groups A and B, whereas the content of 2,3,5,6-tetramethylpyrazine in group B was highest. The contents of stachydrine, 6-gingerol, ferulic acid, and β-sitosterol in the three different districts showed no significant difference in the three groups.

The correlation analysis of the quantitative values of metabolites showed that three flavones had a significantly negative correlation with stachydrine, linoleic acid, monoolein, cappariloside A, and β-sitosterol. There was a significantly positive correlation between the three flavones and linoleoyl ethanolamide. Linoleoyl ethanolamide was negatively correlated with stachydrine, linoleic acid, monoolein, and cappariloside A. Stachydrine had a significantly high positive correlation with linoleic acid, monoolein, and cappariloside A. Stachydrine was significantly correlated with 3,4-dihydroxybenzaldehyde, ferulic acid and β-sitosterol. Among the four phenol compounds, ferulic acid, vanillic acid and, 3,4-dihydroxybenzaldehyde had a significant correlation, but the correlation among 6-gingerol and other compounds was weak. The contents of compounds in similar synthetic pathways existed a positive correlation ( Figure 4D ).

The contents of most anti-inflammatory compounds were strongly correlated with environmental factors. High evaporation, high temperature, and high Olsen-P contributed to the accumulation of the three flavones compounds. The trend of linoleoyl ethanolamide was similar to that of flavones. The trends of stachydrine, linoleic acid, monoolein, 6-gingerol, ferulic acid, cappariloside A, and β-sitosterol were opposite to that of flavone. On the other hand, 2,3,5,6-tetramethylpyrazine had a weak correlation with environmental factors ( Figure 4E ).

There was no difference in abscisic acid and tryptophan contents among the three groups, indicating that all the materials were in the same period of fruit development ( Figure S4 ). Compounds with consistent response trends to environmental factors also had a strong positive correlation in their content, indicating that the quality variation of C. spinosa was caused mainly by environmental factors.

To explore the anti-inflammatory mechanism of C. spinosa fruit, the above 15 anti-inflammatory compounds were used for network pharmacological analysis. 564 targets of compounds were collected, and 1,196 human targets of rheumatoid arthritis (RA) were obtained. Thus, after intersection analysis, 167 common targets were acquired ( Figure S5 ). The common targets were imputed into STRING to build the protein-protein interaction (PPI) network. Subsequently, the interaction network was constructed for illustrating the interactive relationship between compounds and targets ( Figure 5A ). Based on degree value, the top 20 were the core targets including MAPK1, STAT3, IL-6, TNF, VEGFA (vascular endothelial growth factor A), SRC, MMP9, and JUN ( Figure 5B ).

In addition, the common targets were subjected to GO and KEGG enrichment analysis, and the terms associated with RA of GO mainly focused on the regulation of interleukin-6 production, regulation of tumor necrosis factor production, regulation of interleukin-12 production, dendrite, immune receptor activity, cytokine activity, etc. ( Figure 5C ). KEGG pathways mainly focused on the TNF signaling pathway, NF-kappa B signaling pathway, Cytokine-cytokine receptor interaction, etc. ( Figure 5D ).

These results suggested that C. spinosa plays an anti-inflammatory role by affecting cytokine expression and lymphocyte differentiation.

To identify key active compounds in CSE that inhibit the maturation of LPS-induced DC, we examined the effect of stachydrine (ST), diosmetin (DIO), and 3,4-dihydroxybenzaldehyde (DHB) to treat DCs in the presence of LPS. Sachydrine and diosmetin did not inhibit the expression of costimulatory molecules (CD40 and CD86) and the secretion of pro-inflammatory cytokines IL-6 and TNF-α ( Figures 6A, B ). 3,4-dihydroxybenzaldehyde significantly suppressed the expression of CD40 and IL-6 induced by LPS at the concentration of 1 mM, and it significantly suppressed the expression of CD86 and TNF-α at the concentration of 0.5 mM ( Figure 6C ). The results showed that phenols were the main compounds that inhibited DCs maturation and secretion of pro-inflammatory cytokines in capers.

Transcriptome data of C. spinosa fruits were analyzed. PCA showed PC1, PC2, and PC3 were 49.31%, 17.50%, and 9.04%, respectively, and all samples were within the Hotelling’s T-squared ellipse. Group C was significantly separated from the other two groups ( Figure S6A ). Volcano Analysis showed that 2,432 genes were up-regulated, and 1,705 genes were down-regulated in group A and group B, 9,104 genes were up-regulated, and 4,209 genes were down-regulated in group A and group C, 4,872 genes were up-regulated, and 2.320 genes were down-regulated in group B and group C ( Figure S6B ). HCA and Venn’s analysis showed that group C was significantly different from the other two groups ( Figures S6C, D ). The enrichment differential pathways were concentrated in biosynthesis and metabolism pathways between groups A and C ( Figure S7 ). These data indicated that group C was significantly different from the other two groups at the transcriptional level. These transcript data were deposited into the NCBI database with identifier number PRJNA778809 (https://www.ncbi.nlm.nih.gov/sra/PRJNA778809).

The biosynthetic pathway was constructed based on three KEGG pathways: ‘Phenylpropanoid biosynthesis (Ko00940)’, ‘Stilbenoid, diarylheptanoid and gingerol biosynthesis (Ko00945)’, ‘Biosynthesis of alkaloids derived from shikimate pathway (Ko01063)’ ( Figure 7A ).

The genes encoding PAL (Cluster-19971.0), CYP73A (Cluster-22225.2180), COMT (Cluster-22225.2209, Cluster-23341.0, Cluster-30723.0), 4CL (Cluster-22225.12442, Cluster- 22225.4017), CCR (Cluster-22225.2355) were down-regulated in group C, but the genes encoding CYP73A (Cluster-22225.4290) and REF1 (Cluster-22225.2080, Cluster-22225.7764) were up-regulated in group C ( Figure 7B ). The qRT−PCR results showed good consistency with the RNA-Seq data. The qRT−PCR results indicated that the expression of CYP73A (Cluster-22225.2180) in groups A and B was significantly higher than that in group C, while the expression of CYP73A (Cluster-22225.4290) in group C was significantly higher than that in the other two groups. REF1 (Cluster-22225.7764) was highly expressed in all three groups ( Figure S8 ).

The expression level of REF1 (Cluster-22225.7764) indicated a strong positive correlation with ferulic acid content (r = 0.7667, P = 0.0214), which might be the key gene promoting ferulic acid biosynthesis in C. spinosa ( Figure 7C ). These results indicated that C. spinosa fruits from different districts affected the biosynthesis of secondary metabolites by regulating gene transcription.

The fruits of C. spinosa were collected from June to July 2019 from Turpan, Karamay, and Ili in Xinjiang Uygur Autonomous Region of China. The fruits were immediately frozen in liquid nitrogen and then stored at -80°C. The identification was carried out by Dr. Jiang He of Xinjiang Institute of Materia Medica.

BALB/c mice (20 ± 5 g) aged six weeks old were obtained from the experimental animal center of the Xinjiang Medical University (Urumqi, China). Mice were allowed free access to distilled water and standard food, mice room under conditions of maintained temperature (24~26°C) and humidity (69~71%) as well as a 12:12 h light-dark cycle. All experimental animal protocols were reviewed and approved by the Ethics Committee of Xinjiang University for the use of Laboratory Animals.

First, the dried fruits of C. spinosa were ground into a powder and passed through a 40-mesh sieve. Then, 50 g of the powder was mixed with 95% ethanol in a ratio of 1:6 (w/v). The mixture was ultra-sonicated for 20 minutes, followed by extraction for 2 hours in a water bath at 60°C. This extraction process was repeated three times, and the extracts were mixed. The extract was concentrated using rotary evaporation. Then, the extract was subjected to lyophilization and named ‘CSE’.

Immature DCs were induced from bone marrow cells of BALB/c mice by Granulocyte-macrophage colony-stimulating factor (GM-CSF, Peprotech, USA) (Aipire et al., 2017). The cells were cultured in RPMI-1640 medium containing 10% heat-inactivated fetal bovine serum (FBS), 100 units/ml penicillin-streptomycin, 50μM β-mercaptoethanol and 20 ng/ml GM-CSF. Subsequently, they were placed in a 5% CO₂, 37°C incubator. The cells were collected on the seventh day, cultured at a concentration of 1 × 10⁶/mL on a 24-well plate, and treated with different concentrations of CSEs or standards in the presence of 80 ng/mL LPS for 12h. The cells were collected and stained using the mAbs, including anti-mouse CD40, and anti-mouse APC-CD86. All samples were detected on FACSCalibur (BD Biosciences, USA), and the data were analyzed by the FlowJo software (v7.6). The cell supernatant was collected, and the cytokines (TNF-α, IL-6, and IL-12p40) were measured by ELISA kit. OD ₄₅₀ was measured using a microplate reader (Bio-Rad, USA).

TPA was used to induce the ear edema model in mice. The 45 mice were randomly divided into a control group (DMSO), model group (TPA, 75 μg/kg), positive control group (dexamethasone, DXM, 2.5 mg/kg), low-dose group (200 mg/kg) and high-dose group (500 mg/kg), with five mice in each group. The medicine was applied to the right ear for 6 h, ear edema degree of mice was calculated by right ear mass (mg)/left ear mass (mg), and the cytokines (TNF-α and IL-6) were measured by ELISA kit.

A 100 mg sample was extracted with 500 μL of 80% methanol. 2-Chloro-L-phenylalanine (1000:10) was used as the internal standard. After 30 s vortex, the samples were homogenized at 45 Hz for 4 min. Then the extracts were ultra-sonicated for 1 hour in the ice bath. The extracts were left at -20°C for 1 h and centrifuged at 12000 rpm for 15 min at 4°C. The supernatant was aspirated and filtered through a 0.22 μm filter membrane and set aside.

LC-MS/MS analysis was detected on an AB SCIEX Triple TOF 5600 mass spectrometer coupled with a Shimadzu Nexera UPLC LC-30A system with a Waters UPLC BEH C18 (1.7 μm × 2.1 × 100 mm) column. The flow rate was 0.4 mL/min, and the sample injection volume was 3 μL. The mobile phases were 0.1% formic acid aqueous solution (A) and 0.1% formic acid acetonitrile solution (B). The elution gradients were: 0-3.5 min, 95-85% A; 3.5-6 min, 85-70% A; 6-6.5, 70-70% A; 6.5-12 min, 70-30% A; 12-12.5 min, 30-30% A; 12.5-18 min, 30-0% A; 18-22 min, 0% A. Data were collected using Analyst TF 1.7 software to view and process.

The raw data were imported into Progenesis QI software for retention time correction, peak identification, peak extraction, peak integration, and peak alignment, and the corresponding TCM metabolic database was established. Substance identification of the unknown components was performed using the self-built mass spectrometry database. Multivariate statistical analysis, including PCA, and OPLS-DA, was performed using SIMCA software (V15.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden). Also, R software or GraphPad Prism software (V8.0, La Jolla, CA, USA) was used HCA, Venn analysis, cosine similarity analysis, radar plot, clustering tree, pie chart, histogram, and Pearson correlation analysis were performed in combination with univariate analysis for metabolite accumulation patterns between different samples.

Metabolites were extracted according to 2.2. Stachydrine, ferulic acid, and chrysinol (HPLC ≥ 98%, Yuanye, China) were dissolved in with methanolic solution, respectively. The prepared extracts were each taken 400 μL, mixed to prepare mixed standard solutions and a series of mixed standard solutions of different concentrations were prepared by stepwise dilution with methanol. The mass spectrometer apparatus and elution conditions were the same as in 3.4.

The compounds with anti-inflammatory functions were screened from metabolites for network pharmacological analysis. The chemical targets were screened by TCMSP databases (http://lsp.nwu.edu.cn/tcmsp.php), Swiss Target Prediction Database (http://www.swisstargetprediction.ch/), SEA databases (https://sea.bkslab.org/). The disease targets were screened by GeneCards (https://www.genecards.org), and OMIM databases (https://www.omim.org/). Furthermore, the keyword ‘rheumatoid arthritis’ was used to retrieve all targets under the condition of ‘Homo sapiens’. Venny software (v2.1.0) was used to intersect anti-inflammatory component targets and disease targets, which were named ‘the common targets’. Then, the common targets of RA and active ingredients were used to construct the PPI network by the STRING database (https://string-db.org/cgi/input.pl). The results were visualized using Cytoscape software (v3.6.1). The top 20 targets with degree value in the PPI network were named ‘core targets’. The R Project was used to visualize core targets. GO enrichment analysis of common targets was performed by Metascape (http://metascape.org).

Total RNA was extracted using TRIzol reagent, and genomic DNA was removed using DNase I (Takara) according to the manufacturer’s instructions. RNA quality and quantification were measured using a 2100 Bioanalyser (Agilent) and an ND-2000 (Nano-Drop Technologies) RNA-seq transcriptome libraries were then prepared from total RNA (1 μg) using the TruSeqTM RNA Sample Preparation Kit (Illumina, San Diego, CA, USA). The expression levels of each transcript were calculated according to the mapping reads per million exon kilobase fragment method to identify these DEGs in different samples. EdgeR (Empirical Analysis of Digital Gene Expression in R) was used for differential expression analysis and NR (NCBI non-redundant protein sequences) was used to annotate the assembled unigenes. In addition, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment analyses were performed for these DEGs using Goatools (https://github.com/tanghaibao/Goatools) and KOBAS (http://kobas.cbi.pku.edu.cn/home.do), respectively.

Based on the KEGG pathway analysis (https://www.kegg.jp/kegg/pathway.html), genes related to the synthesis of anti-inflammatory compounds were screened. The key genes were identified by Spearman correlation analysis using the GraphPad Prism 8 software.

Total RNA was reverse transcribed using the reverse transcription kit. Then, qPCR amplification was performed on the qTOWER3G qPCR instrument. The primer sequences are shown in Table S1 .

Statistical significance analysis was performed using one-way ANOVA or t-test and computed using GraphPad Prism software. A value of P < 0.05 was statistically significant.