Introduction

Sepsis is a major public health problem on a global scale and one of the main causes of death in intensive care units (ICU) [1, 2]. At least 5.3 million patients in the world are estimated to be diagnosed with sepsis each year, and the mortality of these patients remains approximately 30% [3,4,5]. The high mortality of sepsis patients is largely attributed to lack of accurate methods for early prediction of clinical outcome [6]. Increasing evidence [7, 8] indicates that the systemic immune response has a critical function to play in the pathogenesis and progression of sepsis. In the initial phase of sepsis, the immune response is dominated by pro-inflammatory processes and is favorable for the eradication of pathogens [9]. Progressive sepsis is mainly characterized by the suppression of the immune response, as seen by a decline in the function and number of immune cells [10]. The poor prognosis of sepsis may also be closely related to a compromised host immune system [11, 12], and more and more studies [13, 14] have suggested that novel immune biomarkers cannot only serve as potential predictors of sepsis prognosis but also can provide potential targets for immunotherapy of sepsis. Thus, it becomes necessary to explore immune biomarkers deeply to improve the clinical management of sepsis patients and their prognosis.

The transcriptomic research landscape has undergone a paradigm shift as a result of recent developments in high-throughput, next-generation sequencing and gene chips technology [15]. Numerous bioinformatics analyses and machine learning analyses have been conducted to explore the mRNA prognostic signatures and to direct clinical practice [16, 17]. Prognostic signatures based on immune-related genes (IRGs) have been described for a variety of types of cancer and have demonstrated high sensitivity and specificity [18,19,20]; however, these signatures have not been applied to predict the outcome of patients with sepsis. As a result, we sought to develop and validate an IRGs signature for predicting sepsis patient prognosis and to characterize the immune microenvironment in sepsis patients with varying prognostic risk.

In the present investigation, we conducted a systematic analysis of the immunogenomic landscape of sepsis using Gene Expression Omnibus (GEO) gene expression profiles, and we identified 7 IRGs. In both the training and validation cohorts, a unique IRGs prognostic signature was established and showed moderate predictive value for sepsis patient survival. Furthermore, the CIBERSORT algorithms and single-sample gene set enrichment analysis (ssGSEA) results showed that the high-risk group of sepsis patients were characterized by a decreased neutrophil count and by reduced inflammation-promoting function.

Methods

Acquisition of gene expression profiles and clinical information

GEO database (https://www.ncbi.nlm.nih.gov/geo/) [21] was the source of the level 3 RNA sequencing (RNA-seq) data and related clinical information of 802 sepsis patients (GSE65682 [22]). The clinical information encompassed age, gender, pneumonia, thrombocytopenia, ICU acquired infection, diabetes, abdominal sepsis, survival status and survival time. The healthy controls and sepsis patients with unavailable survival data were excluded. In the end 478 sepsis patients with integrated RNA-seq data and clinical information were screened. In the ArrayExpress database [23], the high-throughput sequencing data of E-MTAB-4451 dataset [24] and prognostic information were selected as external data sets to verify and analyze the model. The E-MTAB-4451 dataset contained a total of 114 samples of adult patients with sepsis. After excluding sepsis samples with incomplete clinical data, 106 sepsis samples were finally included. All the above samples were obtained from peripheral blood samples of septic patients within 24 h after admission to ICU. Subsequently, we performed principal components analysis (PCA) analysis on the expression values of the samples after batch correction. The workflow sketched in Fig. 1.

Fig. 1
figure 1

The study's flow chart. LASSO least absolute shrinkage and selection operator, ROC receiver operating characteristic curve, GO gene ontology, KEGG Kyoto encyclopedia of genes and genomes

Full size image

Identification of the immune-related genes

The Immunology Database and Analysis Portal (ImmPort) database (https://www.immport.org/) was employed to located IRGs. The ImmPort database had 2498 IRGs, which were listed in Additional file 1: Table S1.

Construction and validation of a prognostic immune-related genes signature

Four hundred and seventy-eight sepsis patients were randomly separated into two groups, one for training and one for validation, in a ratio of 1:1. Firstly, to assess the connections between IRGs and patient survival outcomes in the training cohort, univariate Cox regressions were performed. Secondly, the prognostic IRGs were recognized utilizing least absolute shrinkage and selection operator (LASSO) Cox regression [25], which reduced the number of IRGs with prognostic values. In addition, an interaction network of IGRs was created by the STRING database [26]. Thirdly, multivariable Cox regression analysis was employed to identify prognostic IRGs. Seven significant IRGs associated with survival were discovered according to the least Akaike information criterion (AIC) value. AIC was mainly used to measure the goodness of fit of the statistical model, and the model with the lowest AIC was the prediction model with the best fit. Fourth, according to the risk score, sepsis patients in the training and validation cohorts were separated into low-risk and high-risk groups. The following formula was used to determine the risk score: risk score = βgene (a) × EXPgene (a) + βgene (b) × EXPgene(b) + …  + βgene(n) × EXPgene(n), with EXPgene representing the expression level of the identified IRGs standardized by Z score and β representing the coefficient of that particular IRGs in multivariable Cox regression analysis. Kaplan–Meier analysis was performed with the R package “survival”, as well as “survminer”, to compare survival between low-risk and high-risk groups. For the purpose of assessing the accuracy of the constructed signature, the “pROC” R package was applied. Univariate and multivariable Cox regression analyses were recruited to evaluate the signature's capacity to serve as an independent prognostic yardstick in comparison to clinical features in the study population.

Enrichment analyses of the differentially expressed genes (DEGs)

The DEGs between low-risk and high-risk groups were obtained using the “limma” R package [27]. |log2FC|≥ 1 and P < 0.05 were regarded as the cutoff criterion for DEGs. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was performed using the "clusterProfiler" R package based on DEGs [28]. The gene set enrichment analysis (GSEA) (http://software.broadinstitute.org/gsea/index.js) was used to compare inflammatory pattern in the different risk groups.

Comprehensive analysis of immune status

The CIBERSORT algorithms [29] were employed to compare the fraction of immune cells between low-risk and high-risk groups. Furthermore, with ssGSEA, we estimated the infiltration score for the activity of 13 immune-related pathways [30]. Prospective immunological check-point genes were described in research articles.

Statistical analysis

Chi-squared tests were conducted to compare differences in proportions. Univariate and multivariable Cox regression analyses were implemented to distinguish IRGs linked with prognosis. The Kaplan–Meier analysis and the log-rank test were performed to compare the survival between low-risk and high-risk groups. The ssGSEA scores of immune cells or pathways were compared between the low-risk and high-risk groups using Mann–Whitney U test with P values adjusted by the Bonferroni-Holm (BH) method. The prognostic prediction signature's predictive accuracy was measured via receiver operating characteristic curve (ROC) analysis. All of these analyses entailed the use of SPSS software 23.0 and R software 4.0. P value less than 0.05 was considered to be statistically significant.

Results

The clinical information of the training and the validation cohort

Age, gender, the type of pneumonia, the proportion of thrombocytopenia, the proportion of ICU-acquired infection, the proportion of diabetes, the proportion of abdominal sepsis were not significantly different between the training and the validation cohort (P ≥ 0.05) (Table 1).

Table 1 The clinical characteristics of sepsis patients in the training and the validation cohort
Full size table

Construction of a prognostic immune-related genes signature

We conducted PCA analysis on the expression values of the samples after batch correction (Additional file 2: Fig. S1a, b). A total of 752 IGRs were expressed in sepsis patients. Firstly, following the univariate Cox regression analysis, we identified 126 IRGs that were linked with survival and had prognostic significance (Fig. 2, Additional file 1: Table 1, P ˂ 0.05). Secondly, to eliminate multicollinearity and to reduce the number of IRGs to 20, LASSO Cox regression analysis was applied. (Fig. 3a–c, Additional file 1: Table S2, P ˂ 0.05). Figure 3d and Fig. 3e depicted the network of interactions and the correlation between these IRGs. Thirdly, subsequent multivariable Cox regression analysis was used to construct a prognostic signature based on 7 IRGs (the C–C motif chemokine ligand 5 (CCL5), defensin alpha 4 (DEFA4), nuclear transcription factor Y subunit gamma (NFYC), estrogen receptor 1 (ESR1), tumor necrosis factor receptor superfamily member 8 (TNFRSF8), chemokine (C-X3-C motif) receptor 1 (CX3CR1), and serine protease inhibitor A3 (SERPINA3)) (Fig. 3f, Additional file 1: Table S3, P ˂ 0.05). As described earlier, the least AIC score aided in identifying of the IRGs signature (Table 2). The risk score was calculated as follows: − 0.465 × CCL5 + 0.215 × DEFA4 − 1 .487 × NFYC + 1.055 × ESR1 − 0.737 × TNFRSF8 − 0.228 × CX3CR1 + 1.003 × SERPINA3 (AIC = 522.36, Concordance index = 0.78). Patients were classified into a low-risk group (n = 131) and a high-risk group (n = 108) based on the median cut-off value of their risk score (Fig. 4a). Additionally, there was no significant difference in clinical data between the high-risk and low-risk groups in the training cohort (Table 3). Fourthly, The Kaplan–Meier curve indicated that the low-risk group had a considerably increased survival than the high-risk group (Fig. 4c). Finally, the ROC curves shown that the area under the curve (AUC) was 0.876 [95% confidence interval (CI) 0.793–0.946] (Fig. 4e). The results presented the sensitivity was 0.893, the specificity was 0.874, the precision was 0.685, the precision was 0.685, the negative predictive value was 0.964, and the Matthews correlation coefficient (MCC) value was 0.706.

Fig. 2
figure 2

Identification of the candidate top 50 immune-related genes associated with survival of sepsis patients

Full size image
Fig. 3
figure 3

An immune-related genes (IRGs) prognostic signature was constructed in sepsis patients. a 20 IRGs were screened by the LASSO regression model; b LASSO Cox regression analysis was used to compute the coefficients of IRGs; c forest plots illustrating the findings of a univariate Cox regression analysis examining the relationship between the 20 IRGs and 28-day survival in patients with sepsis; d the interactions of candidate IRGs are depicted in the protein–protein interaction network; e the correlation network of candidate IRGs; f forest plots illustrating the findings of a multivariable Cox regression analysis examining the relationship between the 20 IRGs and 28-day survival in patients with sepsis. LASSO Least absolute shrinkage and selection operator, AIC Akaike information criterion. *P < 0.05, †P < 0.01, ‡P < 0.001

Full size image
Table 2 Construction of prognostic signatures in patients with sepsis
Full size table
Fig. 4
figure 4

The 7 immune-related genes (IRGs) signature's prognostic performance in the training cohort and the validation cohort. a The median value and distribution of the risk scores calculated by this signature in the training cohort; b the median value and distribution of the risk scores calculated by this signature in the validation cohort; c Kaplan–Meier curves depicting the 28 days survival based on the risk score in sepsis patients in the training cohort; d Kaplan–Meier curves depicting the 28 days survival based on the risk score in sepsis patients in the Validation cohort; e the prognostic accuracy of the risk score is confirmed by the AUC of ROC curves Kaplan–Meier curves depicting the 28 days survival based on the risk score in sepsis patients in the training cohort; f the prognostic accuracy of the risk score is confirmed by the AUC of ROC curves Kaplan–Meier curves depicting the 28 days survival based on the risk score in sepsis patients in the validation cohort. AUC Area under curve, ROC Receiver operating characteristic curve

Full size image
Table 3 Association between signature and clinical characters in the training cohort
Full size table

Validation of the seven immune-related genes signature in the validation cohort

Similarly, patients were classified into a low-risk group (n = 119) and a high-risk group (n = 120) in the validation cohort (Fig. 4b). There was no significant difference in clinical data between the two risk groups in sepsis patients (Table 4). The Kaplan–Meier curve indicated that the low-risk group had a considerably increased survival than the high-risk group (Fig. 4d). The ROC curves indicate that the AUC was 0.855 (95% CI 0.781–0.932) (Fig. 4f). The results presented the sensitivity was 0.913, the specificity was 0.851, the precision was 0.663, the negative predictive value was 0.969, and the MCC value was 0.694 in the validation cohort. The expression levels of these 7 IRGs in the different outcomes groups were statistically significant (P < 0.05, Additional file 3: Fig. S2a–g). The verification of the external data set shows that the 28-day survival of the low-risk group is significantly higher than that of the high-risk group (Additional file 3: Fig. S2h) and the AUC of signature to predict 28-day survival in sepsis patients by risk score was 0.815 (P < 0.05, Additional file 3: Fig. S2i). The results presented the sensitivity was 0.889, the specificity was 0.789, the precision was 0.814, the negative predictive value was 0.872, and the MCC value was 0.682 in the external data set.

Table 4 Association between signature and clinical characters in the validation cohort
Full size table

Independent prognostic value of seven immune-related genes signature

Univariate Cox regression analysis revealed a correlation between the signatures of seven IRGs and the 28-day survival of sepsis patients (Fig. 5a and b). The 7 IRGs signature proved to be an independent prognostic factor in the multivariable Cox regression analysis (Fig. 5c, d).

Fig. 5
figure 5

The immune-related genes signature has independent prognostic significance in both the training and validation cohorts. a Univariate Cox regression analysis of 28 days survival with risk scores and clinical information in the training cohort; b univariate Cox regression analysis of 28 days survival with risk scores and clinical information.in the validation cohort; c multivariable Cox regression analysis of 28 days survival with risk scores and clinical information in the training cohort; d multivariable Cox regression analysis of 28 days survival with risk scores and clinical information in the validation cohort. ICU Intensive care unit

Full size image

Enrichment analyses of the differentially expressed genes

The DEGs between low-risk group and high-risk group were shown in Additional file 1: Table S4. We conducted GO enrichment and KEGG pathway analyses on the DEGs to better understand their biological functions and pathways. The GO enrichment analysis revealed that DEGs were clearly enriched in immune-related functions, such as neutrophil degranulation and activation, which are implicated in the immunological response in sepsis patients. (Fig. 6a). In addition, the markedly enriched pathways for DEGs were neutrophil extracellular trap formation, staphylococcus aureus infection, interleukin (IL)-17 signaling pathway and nucleotide-binding oligomerization domain (NOD)-like receptor signaling pathway in sepsis patients (Fig. 6b). Inflammatory response (NES = 2.02, P. adjust < 0.05) was considerably enriched in the low-risk group patients of sepsis (Additional file 4: Fig. S3).

Fig. 6
figure 6

The results of GO and KEGG enrichment analysis of differentially expressed genes between different risk groups in sepsis patients. The most significant GO enrichment analysis; b the most significant KEGG pathways enrichment analysis [28]. GO gene ontology, KEGG Kyoto encyclopedia of genes and genomes, IL interleukin, NOD nucleotide-binding oligomerization domain, FC fold-change, BP biological process, CC cellular component, MF molecular function

Full size image

The correlation between the risk score and immune status

We explored the immune cell infiltration landscape while using results of the CIBERSORT algorithm. Patients with sepsis in the high-risk group had higher ratios of neutrophils, monocyte, and T cells CD8 than those in the low-risk group (P < 0.05, Fig. 7a, b). Patients with sepsis in the high-risk group had lower ratios of plasma cell, T cells CD4 naive, T cells CD4 memory activated, natural killer (NK) cells resting, NK cells activated, macrophages M0, macrophages M2, eosinophils, monocytes mast cells activated and mast cells resting than those in the low-risk group (P < 0.05, Fig. 7a, b). Next, the enrichment scores of a variety of immune related functions or pathways were calculated by ssGSEA. The score of CC chemokine receptor, check point, inflammation promoting and T cell co-stimulation were lower in the high-risk group in sepsis patients (P < 0.05, Fig. 8a). Interestingly, the fraction of neutrophils was the largest statistically significant difference between the high-risk and low-risk groups, which was consistent with the findings in the GO and KEGG analysis. In addition, the sepsis patients in the high-risk group were characterized by upregulated expression of CD200R1 and leukocyte-associated immunoglobulin-like receptor 1, whereas the sepsis patients in the low-risk group were characterized by high expression of C10orf54, CD160, CD244, CD40, CD48, CD86, LAG3, TIGIT, TNFRSF14, TNFRSF25, TNFRSF8, TNFRSF9 and TNFSF14 (P < 0.05, Fig. 8b).

Fig. 7
figure 7

The immune cell infiltration landscape based on CIBERSORT algorithm between different risk groups in sepsis patients. Barplot (a) of the immune infiltrating cell proportions. Violin plot (b) demonstrated the differences in the proportions of immune infiltrating cell across different risk groups. NK natural killer

Full size image
Fig. 8
figure 8

The single-sample gene set enrichment analysis scores of sepsis patients in different risk groups are compared. a The scores of immune functions; b the expression levels of immune checkpoints. APC allophycocyanin, CCR CC chemokine receptor, HLA human leukocyte antigen, MHC major histocompatibility complex, IFN interferon. *P < 0.05, **P < 0.01, ***P < 0.001

Full size image

Discussion

Several traditional prognosis indicators are applied in clinical practice today, including the Sequential Organ Failure Assessment [31], quick Sequential Organ Failure Assessment [32], the Acute Physiology and Chronic Health Evaluation II [33], the Simplified Acute Physiology Score II [34], and C-reactive protein/albumin ratio [35]. Nevertheless, their performances are limited in specificity and sensitivity so that they have facilitated early diagnosis and prognosis prediction in patients with sepsis [36].

Molecular biomarkers are considered as non-invasive clinical methods that could objectively predict or evaluate status and progression of disease. Regulation of the immune response and function is known to play a critical role in the pathogenesis and progression of sepsis [37, 38]. Wong et al. [39, 40] found that some genes associated with innate immune response could be used to predict the prognosis of children with clinical sepsis and showed good clinical efficacy. However, these authors did not systematically integrate these genes into a signature, making it impossible to use them in clinical practice. A study [41] verified that sepsis risk scoring methods based on multi-gene biomarkers showed higher performance (sensitivity and specificity) in the ROC curves. Therefore, by using univariate Cox regression analysis, it was determined that 126 IRGs were substantially linked with sepsis patient survival. In the end, we managed to formulate a signature of 7 IRGs for the prognosis of sepsis patients: − 0.465 × CCL5 + 0.215 × DEFA4 − 1.487 × NFYC + 1.055 × ESR1 − 0.737 × TNFRSF8 − 0.228 × CX3CR1 + 1.003 × SERPINA3.

Among these 7 IRGs, which could predict the prognosis of sepsis patients, CCL5, DEFA4, ESR1 and CX3CR1 were broadly researched in previous studies. CCL5, a member of CC motif chemokines ligand, is recognized as an effective biomarker for the diagnosis of sepsis [42]. However, no study has found a relationship between CCL5 and the prognosis of sepsis. DEFA4, a member of the α-defensins family, has been shown to induce IL-6 release in macrophages in a toll-like receptor 4-independent manner [43]. Zhang et al. [44] elaborated that the expression of DEFA4 is strongly correlated with the severity of sepsis. Moreover, DEFA4 also may serve as a biomarker for clinical diagnosis and as a target for treatment of severe influenza infection [45]. ESR1 has been reported as a biomarker of septic syndrome in patients with coronavirus disease 2019 [46]. CX3CR1 is a G-protein coupled receptor, which is expressed on various cells, such as T lymphocytes, monocytes, natural killer cells, neurons and microglial cells [47]. Interestingly, not only is lower CX3CR1 expression associated with early and late mortality in critically sick patients, but also with septic shock progression [48, 49]. Although little is known about the roles of NFYC, TNFRSF8 and SERPINA3 in sepsis, NFYC is characterized as a new regulator of skeletal muscle immunometabolic signaling [50]. TNFRSF8/CD30 has been shown to inhibit the proliferation of autoreactive effector immune cells, hence assisting the body in resisting autoimmunity [51]. SERPINA3 is also an inflammatory cytokine gene, which could induce significant lung injury after influenza infection [52]. Underlying mechanisms of these three genes in sepsis still need further explorations.

In addition, we conducted GO and KEGG enrichment analysis based on the DEGs between different risk groups and discovered that immune-related biological processes and pathways, such as neutrophil degranulation, neutrophil activation involved immune response, neutrophil extracellular trap formation, staphylococcus aureus infection, IL-17 signaling pathway and NOD-like receptor signaling pathway, were unveiled. Furthermore, our immune cell infiltration landscape results indicated neutrophils had lower infiltration than those sepsis patients in the high-risk group. Not only are neutrophils the first line of defense, armed with the ability to recognize and respond to infection in the absence of normal receptor expression, but they are also activated in sepsis to produce reactive oxygen species, nitric oxide, cytokines, proteases, and kinins. Xini et al. [53] discovered that an early absolute CD64/CD15/CD45 neutrophils count lower than 2500/mm3 is independently associated with unfavorable outcome of sepsis. Another study found that reduced neutrophil CD16 expression predicted an increased risk of death in critically ill patients with sepsis [54]. As for immune functions, our ssGSEA results indicated that the high-risk groups had slighter inflammation-promoting in both the training and the validation cohort. Previous research [6] revealed pro-inflammatory processes aid in the clearance of pathogenic agents in the initial phase of sepsis.

Our study still included some limitations. First, we did not employ additional prospective real-world data to validate our prognostic signature's clinical efficacy. Second, owing to a paucity of relevant clinical data, the predictive model we created did not incorporate all relevant clinical data. Finally, it was determined that the relationships between the risk score and immunological status should be validated empirically.

In conclusion, our work established a unique prognostic signature of 7 IRGs. In both the training and validation cohorts, this signature was found to be independently linked with survival, providing insight into the prediction of sepsis prognosis. The differences in neutrophil infiltration were found to be correlated to the progression of sepsis. Thus, in the future, this marker may develop into a viable biomarker for predicting sepsis, allowing for more sensible ICU resource distribution.