This study used nine healthy, six-week-old male Sprague Dawley (SD) rats with body weight 200 ± 20 g obtained from the Beijing Weitong Lihua Laboratory Animal Technology Co., Ltd. The animals were reared in three separate cages (3 per cage) in the animal husbandry center with standard sterile feed and drinking water supply ad libitim, natural light, at room temperature 22-26°C, in humidity 45-65% and with bedding changed twice weekly to ensure the rats lived in a well-regulated, quiet, and clean environment. After 7 days of adaptive rearing, the formal experiment was carried out. Eating, drinking, activity level, and defecation were monitored daily. After all rats were confirmed to be healthy, they were assigned to either the control group (n = 3) 、LPS sepsis model group (n = 3)、Cecal ligation perforation model (CLP) sepsis model group (n = 3) into three groups using a random number table. Rats were intragastrically administered 2 mL of saline once a day for 7 consecutive days. One hour after the last administration, the control group was given intraperitoneal injection of 5ml/kg saline, LPS group was given intraperitoneal injection of endotoxin (LPS 2 mg/ml) 10 mg/kg (9), CLP (10)group was given cecal ligation and puncture. After 24 hours, samples were collected and weighed. The ileum tissue and serum were taken and stored at -80°C. The study was approved by the Ethics Committee of Hebei University of Chinese Medicine (Number:DWL2019023).

Ileum tissue or serum was processed to extract total RNA. Next, 2 ug of total RNA was denatured to use as template RNA for qRT-PCR (BIO-RAD, model CFX96) in a mixture containing Oligo dT Primer and random primers at 70°C for 5 min and then immediately cooled on ice. The processed RNA was then added to the reverse transcription reaction solution containing reaction buffer, MgCl2, PCR nucleotide mix, ribonuclease inhibitor, reverse transcriptase and nuclease-free water to bring the volume of the reaction solution to 20 µl. The reaction was mixed slowly, briefly centrifuged, and placed in the PCR machine using the following cycle: 25°C for 5 min (annealing), 42°C for 60 min (extension), 70°C for 15 min (inactivation), for cycles. Then the reaction was cooled on ice and stored at -20° C for later use. The Cq values of each target gene and the internal reference gene β-actin were obtained. The Q value of each target gene/the Q value of the first sample, that is, RQ==2-ΔΔCq, represents the relative quantitative value of the expression of each target gene, and the RQ value was used for statistical analysis. The primers were as follows: COMMD9:forward (5′- CATCAGAGCATTTCGTGGCG -3′) and reverse (5′- AAGGGCTGAACTGGAGAAGC -3′); CSF3R: for-ward (5′- TGAGGGAAACAGAAAGGCCC -3′) and reverse (5′- AGACCTAGGGGTGTAGCCTG -3′); NUB1: for-ward (5′- GAATGAAAACAAACGGCGGC -3′) and reverse (5′- TCTGCGCCATCCTTGAAAGT -3′); and beta-actin forward (5′- GCAGGAGTACGATGAGTCCG -3′) and reverse (5′- ACGCAGCTCAGTAACAGTCC -3′).

Western blot was used to detect the expression levels of COMMD9, CSF3R, and NUB1 proteins in rat ileum tissue. The ileum tissue was taken and placed in a centrifuge tube, and the protein in the ileum tissue was extracted using RIPA lysate. The supernatant was centrifuged and the protein concentration was measured using BCA method. We added the protein solution in a 4:1 ratio to 5×Loading buffer, denatured it in a boiling water bath for 25 minutes, performed 10% SDS-PAGE electrophoresis, and transferred the PVDF membrane for 30 minutes. The membrane was placed in a TBST incubation tank, shaken at room temperature with skim milk, and sealed for 2 hours. We added the diluted first antibody and incubated it at 4 °C on a shaker overnight. We used TBST to elute 3 times, each time for 5 minutes. The secondary antibody was diluted with TBST in a ratio of 1:5000, incubated at room temperature for 2 hours, and colored using ECL method. Finally, we used Image J software to analyze the grayscale values of the bands, and the relative expression level of the target protein=the grayscale value of the target protein/β- Action grayscale value.

Human microarrays were obtained from the NCBI Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/). GSE11755 including 31 specimens from patients with sepsis and 10 healthy control specimens, was on the foundation of the GPL570[HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array; GSE28750 including 21 specimens from patients with sepsis and 20 healthy controls, was on the foundation of the GPL570[HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array; GSE48080 including 20 specimens from patients with sepsis and 3 healthy controls, was on the foundation of the GPL4133 Agilent-014850 Whole Human Genome Microarray 4x44K G4112F (Feature Number version); and GSE12624 including 34 specimens from patients with sepsis and 36 healthy controls, was on the foundation of the GPL4204 GE Healthcare/Amersham Biosciences CodeLink UniSet Human I Bioarray. We combined the four datasets using the SVA package functionality of the R program, removing batch effects.

In this study, the limma package of R software was used to identify DEGs from the processed microarray data. The filter conditions were: |log2 Fold change (FC)| > 2, FDR < 0.05. Genes that met these criteria were identified as DEGs.

Gene Ontology (GO) analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were completed based on the classification of high- and low-risk patients using the “cluster Profiler” R package of R software in this study, and p < 0.05 was considered statistically significant.

We applied two machine learning algorithms, LASSO and Support vector machine (SVM), to predict sepsis status. LASSO is a compressed estimate that retains the advantages of subset shrinkage. It is a biased estimate for processing data with complex collinearity. LASSO determines whether the discrimination between sepsis samples and normal samples is significant through the “glmnet” package in R software.

SVM is a class of generalized linear classifiers that perform binary classification on data in a supervised learning manner. The RFE algorithm was used to screen the optimal genes from the metadata cohort, and SVM-RFE was used to screen suitable features to identify strong gene sets.

To determine the relationship between immune cell populations and sepsis, we determined the responses of 22 immune cell populations using the CIBERSORT method and assessed the association between these immune cells and the expression of key genes in normal samples and in samples from patients with sepsis.

Gene expression differences between specimens from patients with sepsis and normal specimens were compared using Student’s t-test. The ROC curve and AUC were calculated using the R package “proc” to test the classification effect of key genes in specimens from patients with sepsis and normal specimens. Statistical analysis was performed using R software (version 4.2.1) and GraphPad Prism (version 9.0.0) software, ∗ indicates p<0:05, ∗∗ indicates p<0:01.

In this study, four microarrays GSE11755, GSE12624, GSE28750, and GSE48080 were retrospectively analyzed that together included 106 specimens from patients with sepsis and 69 specimens from healthy subjects. After removing batch effects, the LIMMA software package was used to identify DEGS from metadata. A total of 50 DEGs were identified: 2 genes were significantly up-regulated and 48 genes were significantly down-regulated in samples from patients with sepsis compared to normal controls (Figure 1).

We used the ClusterProfile R software package to perform GO analysis and KEGG analysis on the 50 DEGs. The results showed that the DEGS were mainly involved in proteasomal protein catabolism process, proteasomal protein catabolic process, viral process, proteasome−mediated ubiquitin−dependent protein catabolic process, lymphocyte differentiation, focal adhesion, vesicle lumen, vesicle lumen, secretory granule lumen, transcription coregulator activity, transcription coregulator activity (Figure 2A). In addition, KEGG analysis showed some signaling pathways, including the Salmonella infection pathway, Human T−cell leukemia virus 1 infection, Epstein−Barr virus infection, were also significantly different between the two cohorts (Figure 2B).

We used the LASSO regression algorithm to interrogate the 50 DEGs for diagnostic markers of sepsis, which uncovered 44 potentially diagnostically relevant genes (Figure 3A). We also analyzed the 50 DEGs using the SVM-RFE algorithm, which identified eight feature subsets (Figure 3B). Only three genes were identified by both of the approaches: COMMD9, CSF3R, and NUB1 (Figure 3C), suggesting these genes may be involved in sepsis diagnose.

We found that, compared with normal samples, the expression levels of COMMD9 and CSF3R were all significantly down-regulated in samples from patients with sepsis compared to control samples (Figures 4A, B). NUB1 was significantly up-regulated in the samples (Figure 4C). We next performed ROC analysis of COMMD9, CSF3R, NUB1, and the results showed that COMMD9 (Figure 4D, AUC=0.841), CSF3R (Figure 4E, AUC=0.907), NUB1 (Figure 4F, AUC=0.719).

Infiltration of immunocytes in the tissue microenvironment is an independent prognostic indicator of overall survival. We therefore investigated the relationship between COMMD9, CSF3R, and NUB1 expression and the infiltration of immune cells in specimens from patients with sepsis and healthy controls. CIBERSORT was used to characterize the immune cell composition of the two cohorts (Figures 5A, B). And we also found some statistically significant differences in B cells memory, Plasma cells, T cells CD8, T cells CD4 native, T cells CD4 memory resting, T cells follicular helper, T cells regulatory, T cells gamma delta, NK cells activated, Monocytes, Macrophages M0, Dendritic cells resting, Dendritic cells activated, Mast cells resting, Mast cells activated, Eosinophils, Neutrophils (Figure 5C).This demonstrated that there is a correlation between lower expression of COMMD9, CSF3R and NUB1 and increased immune cell infiltration (Figures 6A–C). These data suggest that COMMD9, CSF3R, and NUB1 may regulate immune cells in patients with sepsis.

The PCR results in rat ileum tissue show that,compared with control group samples, the expression levels of COMMD9 and CSF3R were significantly decreased from rats with two sepsis model (LPS group=model1, CLP group=model2) (Figures 7A, B), and the expression level of NUB1 was significantly increased (Figure 7C).

The PCR results in rat serum show that, compared with control group samples, the expression levels of COMMD9 and CSF3R were significantly decreased from rats with two sepsis model (LPS group=model1, CLP group=model2) (Figures 7D, E), and the expression level of NUB1 was significantly increased (Figure 7F).

The WB results show in rat ileum tissue that (Figure 7G), compared with control group samples, the expression levels of COMMD9 and CSF3R were significantly decreased from rats with two sepsis model (LPS group=model1, CLP group=model2)(Figures 7H, I), and the expression level of NUB1 was significantly increased (Figure 7J).