Identification of prognostic biomarkers of sepsis and construction of ceRNA regulatory networks

Abstract

Sepsis is a life-threatening severe organ dysfunction, and the pathogenesis remains uncertain. Increasing evidence suggests that circRNAs, mRNAs, and microRNAs can interact to jointly regulate the development of sepsis. Identifying the interaction between ceRNA regulatory networks and sepsis may contribute to our deeper understanding of the pathogenesis of sepsis, bring new insights into early recognition and treatment of sepsis. Blood samples from sepsis patients in the Affiliated Hospital of Southwest Medical University were collected. RNA sequencing (mRNA/circRNA) was performed on Survivor group (n = 26) and Non-survivor group (n = 6), then quality control and differential expression analysis were performed. Subsequently, GO analysis was performed on the differential expression genes; Meta-analysis was used to screen for prognostic related genes; 10 × Single-cell RNA sequencing was used to annotate the cell distribution of core genes. Finally, combined with base complementary pairing and intergroup correlation analysis, a sepsis-associated circRNA-miRNA-mRNA regulatory network was constructed. Differential expression analysis screened 28 mRNAs and 16 circRNAs. GO results showed that differential expression genes were mainly involved in membrane raft, actin cytoskeleton, regulation of immune response, negative regulation of cAMP-dependent protein kinase activity, etc. Meta-analysis screened 2 core genes, GSPT1 and NPRL3, which are associated with sepsis prognosis. 10 × Single-cell RNA sequencing showed that GSPT1 and NPRL3 were widely localized in immune cells, mainly macrophages and T cells. A ceRNA network consisting of 4 circRNA, 26 miRNA, and 2 mRNA was constructed. GSPT1 and NPRL3 were lowly expressed in the sepsis Survivor group, compared with Non-survivor group, which may become novel prognostic biomarkers for sepsis. A sepsis-related ceRNA networks, which consists of 4 circRNA, 26 miRNA, and 2 core gene, may guide mechanistic studies.

Introduction

Sepsis is a life-threatening severe organ dysfunction caused by a dysregulated host response to infection that can lead to multiple organ dysfunction syndrome (MODS)¹. Sepsis is one of the important causes of patient death in intensive care units, and many programs have been applied in the treatment of sepsis, including fluid resuscitation, anti-infection, organ function support, blood purification and extracorporeal membrane oxygenation (ECMO), as well as prevention of complications^2,3. Despite the increasing number of septic patients receiving appropriate interventions, the clinical treatment outcomes are not satisfactory. This is attributed to the complex pathophysiological mechanisms of sepsis as well as the functional disturbances of multiple systems in the organism. On the other hand, the early identification of sepsis is the cornerstone of the rapid initiation of a sepsis regimen, which is critical for improving the prognosis of sepsis¹. However, the development of sepsis is a complex process, and its early diagnosis is rather difficult due to the lack of effective predictive means.

With the development of high-throughput sequencing technologies and big data analysis technologies, bioinformatics-based analysis technologies have enabled us to have a deeper understanding of the pathogenesis of sepsis, and many promising biomarkers of sepsis are gradually discovered. For example, Dai et al., using weighted gene co-expression network analysis (WGCNA), identified LPIN1 as a reliable biomarker for survival in sepsis patients⁴. In addition, the biomarkers of microRNAs (miRNAs), circRNAs, long non-coding RNAs (lncRNAs), and the human microbiome also provide important value in the diagnosis, treatment, and prognosis of sepsis⁵. The miRNAs are small non-coding RNA molecules that can hybridize the 3′untranslated regions (UTRs) of target mRNAs through seed sequences, which subsequently leads to the degradation or translation inhibition of the homologous target mRNAs, thereby achieving gene regulation at the post-transcriptional level⁶. In sepsis, miRNAs play an important role in processes such as body immune response, and in the meantime, the expression profile of miRNAs dysregulation has been identified, suggesting their potential value as diagnostic and prognostic biomarkers of sepsis^7,8. For example, miRNA-186 has been shown to ameliorate in sepsis-induced renal injury via the PTEN/PI3K/Akt/p53 pathway⁹. In addition, miRNAs can be encapsulated in extracellular vesicles or bind to proteins so that they are not degraded in the blood, and this stability also provides the conditions for miRNAs to be excellent biomarkers. The circRNAs are a class of RNAs with covalent bonds forming a closed ring structure, which is not susceptible to degradation by exonucleases and is more stable than linear RNAs. The vast majority of circRNAs are non-coding and play regulatory roles at the transcriptional or post-transcriptional level. CircRNAs can bind to and decrease the expression of their target miRNAs, thereby acting as sponges to inhibit transcription¹⁰. More and more studies have confirmed the potential of circRNAs in the diagnosis and treatment of sepsis, and studying the regulatory mechanism of circRNAs may help to find new targets for molecular intervention in sepsis^11,12. Based on the regulatory relationships of circRNAs, miRNAs and mRNAs, the researchers proposed the theory of competing endogenous RNA (ceRNA). Specifically, RNA molecules (such as mRNAs and circRNAs) targeted by a common miRNA can regulate each other indirectly by competing with the miRNA response elements¹³.

Identifying the interaction between ceRNA regulatory networks and sepsis may contribute to our deeper understanding of the pathogenesis of sepsis, bring new insights into early recognition and treatment of sepsis. Therefore, in this study, we collected peripheral blood samples from sepsis patients, performed mRNA and circRNA sequencing, and performed differential expression analysis of the sequencing results between sepsis survivor and non-survivor groups. Then, the comprehensive analysis including GO analysis and Meta-analysis were conducted to screen out the core targets affecting the prognosis of sepsis patients. Finally, its ceRNA regulatory network (circRNAs-miRNAs-mRNAs network) was constructed to provide a new understanding of the mechanism of sepsis. The specific process is shown in Fig. 1.

Fig. 1.

Flow chart. Firstly, the blood sample of sepsis survivors (n = 26) and non-survivors (n = 6) were collected for RNA sequencing(mRNAs/circRNAs) and differential expression analysis. Subsequently, GO analysis was performed on the differential genes and meta-analysis was used to screen for core genes; Single-cell RNA sequencing was used to clarify the cell localization of core genes. Based on the positive correlation between the gene and circRNA, combined with the base complementary pairing information in the CircBank database, the miRNA with binding sites with circRNA were predicted. On the other hand, miRNAs with binding sites to core genes were predicted through the miRWalk database. The miRNAs searched in the two databases were intersected to construct the circRNAs-miRNAs-mRNAs regulatory network.

Methods

Blood sample collection

Sepsis patients in Emergency Intensive Care Unit of the Affiliated Hospital of Southwest Medical University from January 2021 to December 2022 were continuously collected. The inclusion criteria were: (1) compliance with the Sepsis 3.0 criteria; (2) Age ≥ 18 years old; (3) The patient or legal representative consents to enter the experiment and signs the informed consent form. Pregnant or lactating women are excluded. All included participants signed an informed consent form. This study protocol passed the review of the Ethics Committee of the Affiliated Hospital of Southwest Medical University (ethics number: ky2018029). Peripheral blood samples were collected within 24 h of admission, and stored at -80 °C in the biobank of the Affiliated Hospital of Southwest Medical University. Samples were divided into Survivor group (SV, n = 26) and Non-survivor group (NS, n = 6). All methods were performed in accordance with the relevant guidelines. The clinical trial registration number is ChiCTR1900021261, and the trial registration date is 04/02/2019.

RNA sequencing

RNA sequencing was performed with the assistance of BGI (Shenzhen, China). The peripheral blood was centrifuged for 5 min (4℃, 12,000 rpm), and the supernatant was transferred into a new EP tube containing 0.3 mL of isoamyl alcohol (24:1). The mixture was centrifuged for 10 min (4℃, 12,000 rpm). The upper aqueous phase retaining RNA was transferred into a new tube, and centrifuged for 20 min (4℃, 13,600 rpm). After deserting the supernatant, air dry the pellet in a biosafety cabinet for 5–10 min. Approximately 1 µg total RNA per sample was treated with Ribo-Zero™ Magnetic Kit (Epicentre) to remove rRNA¹⁴. The RNA was fragmented by adding First Strand Master Mix (Invitrogen). The first-strand cDNA was generated using random primers reverse transcription, followed by second-strand cDNA synthesis. Several rounds of PCR amplification were performed using PCR Primer Cocktail and PCR Master Mix to enrich the cDNA fragments. The final library was examined by the Agilent 2100 bioanalyzer for fragment size distribution, and quantified using real-time quantitative PCR (QPCR) (TaqMan Probe). The Qualified libraries were sequenced on the Hiseq X-ten platform (BGI-Shenzhen, China). Relevant RNA sequencing data are stored in the China National GeneBank DataBase (CNGBdb, https://db.cngb.org/), and researchers can access our data at any time (registration number: CNP0002611).

Quality control and differential expression analysis

iDEP (Integrated Differential Expression and Pathway Analysis) (http://bioinformatics.sdstate.edu/idep/) is a common software for differential expression and pathway analysis of RNA sequencing data, which connects 63 R/Bioconductor packages, 2 web services, and comprehensive annotation and pathway databases for 220 plant and animal species^15,16. The iDEP software was used to quality control and logarithmic the data, density plot was used to demonstrate the homogeneity and comparability of normalized data. Subsequently, the DESeq2 method was used for statistical analysis, Fold Change (FC) ≥ 1.2 and False Discovery Rate (FDR) < 0.05 were used as criteria to screen for differentially expressed mRNAs and circRNAs between sepsis Survivor group and Non-survivor group.

GO analysis

Gene Ontology (GO) analysis is a common method for gene big data analysis, which contains three categories: Biological Process (BP) represents the process of a molecular activity event, including the functional of cells, tissues, organs and species, which is often the most relevant category for experimental research; Cellular Component (CC) represents cells or their external environment; Molecular Function (MF) is the active element that describes a gene product at the molecular level¹⁷. In order to view the functional enrichment of differentially expressed genes globally, we used OECloud (https://cloud.oebiotech.cn/task/category/enrich/?id=25) for GO functional enrichment analysis, the top 30 items with the most significant BP, CC, and MF were displayed, and the smaller the pvalue, the more significant the enrichment.

Meta-analysis

To further narrow down the core genes, and to verify the ability of core genes to determine prognosis, we downloaded multiple sepsis-related datasets from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) in anticipation of more objective results and reduced false-positive rates. The datasets GSE28750¹⁸, GSE54514¹⁹, GSE69528²⁰, GSE95233²¹, GSE67652²², GSE63042²³ were downloaded for meta-analysis, and the data were divided into Normal Control versus Sepsis, Survivor versus Non-survivor.

Single-cell RNA sequencing

Peripheral blood cells are a mixture of multi-cell lines, so we use 10 × Single-cell RNA sequencing for cell type annotation to clarify the specific cell localization of core genes and provide guidance for subsequent functional and mechanism studies. Library construction, sequencing and data analysis were completed under the guidance of OEbiotech (Shanghai, China). Cell Ranger software was used to demultiplex cellular barcodes, map reads to the genome using the STAR aligner, producing a matrix of gene counts versus cells. We processed the unique molecular identifier (UMI) count matrix using the R package Seurat²⁴. Library size normalization was performed with NormalizeData function in Seurat to obtain the normalized count²⁴. Top variable genes across single cells were identified using the method described in Macosko et al.²⁵. Graph-based clustering was performed to cluster cells according to their gene expression profile using the FindClusters function in Seurat²⁴. Cells were visualized using a 2-dimensional t-distributed stochastic neighbor embedding (t-SNE) algorithm with the RunTSNE function in Seurat²⁴. We used the FindAllMarkers function to identify marker genes of each cluster. It is generally considered that the marker gene of a cluster is highly expressed in this particular cluster and low expressed or not expressed in other clusters. Then, R package SingleR was used to infer the cell of origin of each of the single cells and identify cell types.

Construction of circRNAs-miRNAs-mRNAs network

The OmicShare platform (https://www.omicshare.com/) was used for intergroup correlation analysis to screen for differentially expressed circRNAs that were positively and strongly associated with core gene (correlation coefficient > 0.6 and pvalue < 0.05). Subsequently, miRNAs with binding sites to core circRNAs were predicted by base complementary pairing information in CircBank database (http://www.circbank.cn/index.html). On the other hand, miRNAs with binding sites to core genes were predicted by miRWalk database (http://mirwalk.umm.uni-heidelberg.de/). The miRNAs searched by the two databases were intersected to construct the circRNA-miRNA-mRNA regulatory network, which is involved in the prognosis regulation of sepsis.

Results

Demographic and clinical features

In this study, 26 sepsis survivors and 6 sepsis non-survivors were included. The average age of the Survivor group was 53.5, while the Non-survivor group was 58.83. The ALT in the Non-survivor group (212.2 ± 151.1) was significantly higher than that in the Survivor group (59.49 ± 12.61) and there was a statistically significant difference (P = 0.0462). The hs-TNT in the Non-survivor group (0.7868 ± 0.7253) was slightly lower and statistically significant than that in the Survivor group (0.07923 ± 0.0461) (P = 0.0476). Lymphocyte in the Non-survivor group (8.362 ± 3.798) was significantly higher and statistically significant than that in the Survivor group (1.653 ± 0.5527) (0.0030). The unpaired t-test was used for statistical analysis, and the results were expressed as mean ± standard deviation (Table 1).

Table 1.

Clinical characteristic of Sepsis Survivor group and Non-survivor group.

Clinical variable	Survivor	Non-survivor	P value
Age (years)	53.5 ± 2.204	58.83 ± 6.695	0.3442
ALT (U/L)	59.49 ± 12.61	212.2 ± 151.1	0.0462
TBIL (umol/L)	21.65 ± 3.284	16.93 ± 7.064	0.5406
Crea (umol/L)	149.3 ± 30.81	109.4 ± 14.47	0.5449
hs-TNT (ug/L)	0.07923 ± 0.0461	0.7868 ± 0.7253	0.0476
BNP (ng/L)	3142 ± 1198	6131 ± 5119	0.3903
WBC (109/L)	12.18 ± 1.644	12.02 ± 4.266	0.9680
Neutrophil (109/L)	10.59 ± 1.548	9.418 ± 3.066	0.7434
Macrophage (109/L)	0.7527 ± 0.2371	1.677 ± 1.335	0.2481
Lymphocyte (109/L)	1.653 ± 0.5527	8.362 ± 3.798	0.0030
PCT (ng/mL)	24.5 ± 6.744	41.3 ± 20.59	0.3501

ALT: alanine transaminase; TBIL: total bilirubin; Crea: creatinine; hs-TNT: hypersensivive troponin T; BNP: brain natriuretic peptide; WBC: white blood cell; PCT: procalcitonin.

Quality control and differential expression analysis

The mRNA and circRNA sequencing data were quality controlled and logarithmic, respectively. The density plot showed that the mRNA sequencing data were homogeneous and comparable (Fig. 2A). Subsequently, differential expression analysis was performed on the Survivor group and Non-survivor group, and 28 differentially expressed genes were screened for FC ≥ 1.2 and FDR < 0.05, of which 2 were upregulated and 26 were down-regulated (Fig. 2B) (Table 2). Bioinformatics analysis of circRNA data was performed in the same way, and the density plot showed that circRNA sequencing data were homogeneous and comparable (Fig. 2C). Differential expression analysis screened 16 differentially expressed circRNAs, of which 1 was up-regulated and 15 down-regulated (Fig. 2D).

Fig. 2.

Quality control and differential expression analysis. (A) Density plot of mRNA sequencing data, the abscissa is the expression value, the ordinate is the density, red represents Non-survivor group (NS), green represents Survivor group (SV), the two sets of data were homogeneous and comparable; (B) Differential expression analysis of mRNA sequencing data, red for 2 upregulated genes, blue for 26 downregulated genes; (C) Density plot of circRNA sequencing data, the two sets of data are homogeneous and comparable; (D) Differential expression analysis of circRNA sequencing data, red for 1 upregulated gene and blue for 15 downregulated genes.

Table 2.

Differentially expressed genes in the Sepsis Survivor group versus Non-survivor group.

Gene Symbol	Regulation	log2 Fold Change	FDR
SLFN12	Up	0.9904	0.0496
INTS6	Up	0.6686	0.0496
OR2W3	Down	-3.2779	0.0491
TUBB1	Down	-3.0759	0.0006
SELENBP1	Down	-2.9289	0.0440
SH3BGRL2	Down	-2.8013	0.0028
ITGB3	Down	-2.7668	0.0456
F13A1	Down	-2.6524	0.0016
CAVIN2	Down	-2.5840	0.0094
C9orf78	Down	-2.5597	0.0496
FAM210B	Down	-2.5454	0.0440
SPARC	Down	-2.5261	0.0233
TREML1	Down	-2.5047	0.0353
FCGR3A	Down	-2.4025	0.0004
MBNL3	Down	-2.3846	0.0496
PRSS1	Down	-2.3811	0.0076
PRKAR2B	Down	-2.2208	0.0447
GSPT1	Down	-2.1376	0.0496
NPRL3	Down	-2.0220	0.0496
PHOSPHO1	Down	-1.9369	0.0432
PGRMC1	Down	-1.9134	0.0098
MTURN	Down	-1.8931	0.0105
CD226	Down	-1.8104	0.0496
MMD	Down	-1.7825	0.0440
ZNF185	Down	-1.7572	0.0105
SPECC1	Down	-1.7350	0.0496
R3HDM4	Down	-1.3956	0.0491
PCBP2	Down	-1.0205	0.0496

GO analysis

GO function enrichment analysis based on 26 differentially expressed genes showed that these genes were mainly involved in translation release factor activity, cAMP-dependent protein kinase regulator activity, cAMP-dependent protein kinase inhibitor activity and other Molecular Functions. Mainly involved in membrane raft, actin cytoskeleton, cAMP-dependent protein kinase complex and other Cellular Components; Mainly involved in regulation of immune response, negative regulation of cAMP-dependent protein kinase activity, negative regulation of defense response to virus and other Biological Processes (Fig. 3A). The top 30 GO items with the most significant enrichment include: 5 Molecular Functions, 6 Cellular Components, and 19 Biological Processes (Fig. 3B) (Table 3).

Fig. 3.

GO analysis. A, MF, CC, BP enrichment of the top 10 items with the most significant, the abscissa is the specific entry, the ordinate is -log10 (p-value), p = 0.05 at the dotted line, the higher the bar graph, the smaller the p-value; B, The top 30 GO items with the most significant enrichment, the abscissa represents the enrichment score, the ordinate represents the enrichment entry, the size of the point represents the number of genes enriched, the different colors represent different p-value, the triangle represents molecular function, the square represents cellular component, the circle represents biological process, of which 5 are molecular functions, 6 are cellular components, and 19 are biological processes.

Table 3.

The top 30 GO items with the most significant enrichment.

Term	ID	Description	P value
CC	GO:0045121	membrane raft	0.0005
CC	GO:0015629	actin cytoskeleton	0.0012
BP	GO:0050776	regulation of immune response	0.0043
CC	GO:0005952	cAMP-dependent protein kinase complex	0.0067
MF	GO:0003747	translation release factor activity	0.0068
MF	GO:0008603	cAMP-dependent protein kinase regulator activity	0.0068
BP	GO:2000480	negative regulation of cAMP-dependent protein kinase activity	0.0076
BP	GO:0050687	negative regulation of defense response to virus	0.0076
BP	GO:0075522	IRES-dependent viral translational initiation	0.0076
BP	GO:0038202	TORC1 signaling	0.0076
CC	GO:0031091	platelet alpha granule	0.0081
BP	GO:0043249	erythrocyte maturation	0.0089
BP	GO:0050862	positive regulation of T cell receptor signaling pathway	0.0089
BP	GO:0039694	viral RNA genome replication	0.0089
MF	GO:0004862	cAMP-dependent protein kinase inhibitor activity	0.0096
BP	GO:0048738	cardiac muscle tissue development	0.0102
CC	GO:0030868	smooth endoplasmic reticulum membrane	0.0107
BP	GO:0035909	aorta morphogenesis	0.0114
CC	GO:0031224	intrinsic component of membrane	0.0121
MF	GO:0034236	protein kinase A catalytic subunit binding	0.0123
MF	GO:0004872	receptor activity	0.0128
BP	GO:0045662	negative regulation of myoblast differentiation	0.0139
BP	GO:0006646	phosphatidylethanolamine biosynthetic process	0.0139
BP	GO:0001816	cytokine production	0.0139
BP	GO:0001958	endochondral ossification	0.0139
BP	GO:0097320	plasma membrane tubulation	0.0152
BP	GO:0009235	cobalamin metabolic process	0.0152
BP	GO:0019835	cytolysis	0.0152
BP	GO:0045648	positive regulation of erythrocyte differentiation	0.0165
BP	GO:0045954	positive regulation of natural killer cell mediated cytotoxicity	0.0165

Meta-analysis

Meta-analysis of Normal Control group and Sepsis group based on GEO public database found that GSPT1 and NPRL3 were highly expressed in Normal Control group with statistically significant differences (P < 0.05) (Fig. 4A-B). Meta-analysis based on sepsis Survivor group and Non-survivor group showed that G1 to S phase transition 1 (GSPT1) was lowly expressed in Survivor group, while Nitrogen permease regulator-like 3 (NPRL3) was highly expressed in Survivor group, the difference was statistically significant (P < 0.05) (Fig. 4C-D).

Fig. 4.

Meta-analysis. (A, B) Based on meta-analysis of Normal Control group and Sepsis group, GSPT1 and NPRL3 were highly expressed in Normal Control group, and the difference was statistically significant; (C, D) Based on meta-analysis of Survivor group and Non-survivor group, GSPT1 was lowly expressed in Survivor group, while NPRL3 was highly expressed in Survivor group, with statistically significant differences. For testing for multi-data heterogeneity, select Random effects model for p < 0.05 and Fixed effect model for p ≥ 0.05.

Single-cell RNA sequencing

A total of 5 samples of Single-cell transcriptome sequencing were completed in this analysis. After quantitative quality control, the number of high-quality cells in each sample was distributed in 4050 ~ 10,191, and after removing two-cell, multicellular and apoptotic cells, the final number of cells obtained was distributed in 3108 ~ 8509, the average UMI in each cell was distribution in 519 ~ 8529, and the average gene number of each cell is distributed in 343 ~ 2337. After dimensionality reduction, there were 9 groups of cells, namely T cell (1, 2, 6 and 8 cluster), Monocyte (3 and 5 cluster), NK cell (4 cluster), B cell (7 cluster), Platelet (9 cluster) (Fig. 5A, B). Cell type annotation of the core genes showed that GSPT1 and NPRL3 were widely localized in immune cells, mainly macrophages and T cells (Fig. 5C, D).

Fig. 5.

Single-cell RNA sequencing. (A) Mixed sample sequencing, a total of 9 types of cell clusters with similar expression patterns were identified, the abscissa represents the first principal component after dimensionality reduction, the ordinate represents the second principal component, each dot represents a cell, and different groups of cells are distinguished by different colors; (B) Marker gene identification, each cell type was distinguished by different colors, green for T cells, orange for macrophages, blue for NK cells, black for B cells, brown for platelets; (C, D) Cell type annotation for GSPT1and NPRL3, GSPT1 and NPRL3 were widely localized in immune cells, mainly macrophages and T cells.

Construction of circRNAs-miRNAs-mRNAs network

Based on the intergroup correlation analysis and base complementary pairing principle, four core circRNAs and their targeted miRNAs were screened, including HSA_CIRC_0026658 with 34 targeted miRNA, HSA_CIRC_0026654 with 55 targeted miRNA, HSA_CIRC_0026660 with 27 targeted miRNA, and HSA_CIRC_0026657 with 27 targeted miRNA. On the other hand, 2428 targeted miRNAs of GSPT1 and 435 targeted miRNAs of NPRL3 were obtained through molecular interaction prediction. Finally, the forward and reverse predicted miRNAs were intersected to construct a ceRNA regulatory network, which consists of 4 circRNAs, 26 miRNAs, and 2 core genes (Fig. 6A-C) (Table 4).

Fig. 6.

circRNAs-miRNAs-mRNAs network. (A) Sankey diagram was composed of 4 circRNAs, 26 miRNAs and 2 mRNAs, which constitutes the regulatory relationship of circRNAs-miRNAs-mRNAs; (B, C) The ceRNA network of GSPT1and NPRL3, purple for circRNAs, pink for miRNAs, orange for mRNAs.

Table 4.

Composition of sepsis-related circRNA-miRNA-mRNA regulatory network.

Type	Core RNA
mRNA	GSPT1	NPRL3
miRNA	hsa-miR-873-3p	hsa-miR-4538
	hsa-miR-3064-5p	hsa-miR-7156-3p
	hsa-miR-4252	hsa-miR-5093
	hsa-miR-518c-5p	hsa-miR-937-5p
	hsa-miR-3929	hsa-miR-106b-3p
	hsa-miR-5187-3p	hsa-miR-99b-3p
	hsa-miR-6504-5p	hsa-miR-1295b-3p
	hsa-miR-612	hsa-miR-4725-3p
	hsa-miR-5192	hsa-miR-3153
	hsa-miR-5197-5p
	hsa-miR-937-5p
	hsa-miR-6811-3p
	hsa-miR-1293
	hsa-miR-491-5p
	hsa-miR-4725-3p
	hsa-miR-1304-5p
	hsa-miR-4651
	hsa-miR-520g-5p
	hsa-miR-6812-5p
circRNA	HSA_CIRC_0026658	HSA_CIRC_0026658
	HSA_CIRC_0026654	HSA_CIRC_0026654
	HSA_CIRC_0026660	HSA_CIRC_0026660
	HSA_CIRC_0026657	HSA_CIRC_0026657

Discussion

Sepsis is a complex clinical syndrome, which can be manifested as metabolic acidosis, coagulation disorder, shock and consciousness disorders, and can be accompanied by acute respiratory distress syndrome (ARDS), acute lung injury (ALI), acute heart failure and other complications^1,26. Therefore, the search for sepsis-specific biomarkers is critical. In this study, according to the two outcomes of 28-day survival and non-survival, we subclassified patients with sepsis into Survivor group and Non-survivor group, and 28 differentially expressed mRNAs and 16 differentially expressed circRNAs were screened by DEseq2 method. The Meta-analysis screened 2 core genes (GSPT1 and NPRL3) that affect the prognosis of sepsis, then Single-cell RNA sequencing showed that GSPT1 and NPRL3 were widely localized in immune cells. Combined with intergroup correlation analysis and molecular interaction prediction, we screened 4 core circRNAs (HSA_CIRC_0026658, HSA_CIRC_0026654, HSA_CIRC_0026660 and HSA_CIRC_0026657), and proposed a ceRNA regulatory network consisting of 4 circRNAs, 26 miRNAs and 2 core genes.

The UCSC genomic browser shows that GSPT1 is located in the long arm region 13 of human chromosome 16 (chr16 p13.13) and consists of 10 exons in the protein coding region. During transcription, GSPT1 forms two distinct transcripts by variable splicing, translating and encoding for 508 amino acids²⁷. GSPT1 is an essential gene for cell survival, and its protein products are involved in various biological processes, including the regulation of the cell cycle, cytoskeleton organization, and cell apoptosis²⁸. Eukaryotic peptide chain releasing factors (eRF) is a group of important proteins involved in intracellular protein synthesis released by nascent peptide chains in eukaryotes, including two types (eRF1 and eRF3). Among them, eRF3 includes two isoforms, eRF3a and eRF3b, encoded by GSPT1 and GSPT2, respectively. Eukaryotic translation termination factor 1 (ETF1/eRF1) has a transfer RNA-like structure and recognizes the termination codon in the A site of the translating ribosome^29,30. GSPT1/eRF3a is a GTPase that can bind to eRF1 to mediate stop codon recognition and ribosomal nascent protein release, thereby regulating cell cycle, proliferation and apoptosis³⁰. Through its C-terminus, GSPT1 interacts with ETF1 to form part of the translation termination complex that hydrolyzes the nascent protein and terminate its translation³¹. On the other hand, the N-terminal domain of GSPT1 contains 2 overlapping PAM2 motifs that are able to specifically recognize the MLLE domain of poly(A)-binding protein (PABP) or UPF1 to coordinate nonsense-mediated decay (NMD)^32,33. When the translation termination complex the GSPT1 structure encounters an appropriate stop codon near the tail of a poly (A), PABP is brought into proximity with and binds to GSPT1. In contrast, if the complex encounters a premature stop codon far from the poly (A) tail and PABP, GSPT1 is instead able to interact with UPF1 and the truncated protein is targeted for NMD³⁴. Therefore, GSPT1 is considered as a proto-oncogene that may be involved in tumorigenesis, and many previous studies have also clarified the regulatory function of GSPT1 in tumor development³⁵. For example, in gastrointestinal malignancies, GSPT1 is a direct functional target of miR-144, and miR-144 can inhibit the proliferation, invasion and migration of gastric cancer cells through targeted regulation of GSPT1^31,36. Meanwhile, GSPT1 may be related to gastric cancer subtyping, and some studies have observed increased GSPT1 mRNA levels in 70% of intestinal-type gastric tumors³⁶. GSPT1 can also negatively regulate the GSK-3β signaling pathway in colon cancer cells, thereby affecting the proliferation of colon cancer cells^27,37. Moreover, miR-27b-3p is regulated by aberrant DNA methylation and could also inhibit the malignant behavior of gastric cancer cells through targeted regulation of GSPT1³⁸. The miR-27b-3p/GSPT1 axis is also an important regulatory mechanism in non-small cell lung cancer (NSCLC), and the upregulation of GSPT1 expression will promote the proliferation, invasion and migration of NSCLC cells, which subsequently affects the prognosis³⁹. In leukemia, degradation of GSPT1 can cause TP53-independent cell death while sparing normal hematopoietic stem cells⁴⁰. Recent studies have found that thalidomide analogs degrade GSPT1 via CRL4^CRBN ubiquitin ligase, suggesting that GSPT1 is a new potential therapeutic target in the treatment of leukemia⁴¹. The N-terminus of GSPT1 also contains a polymorphic polyglycine repeat, and the structure can affect the interaction of GSPT1 with PABP and increase the risk of gastric cancer and breast cancer^28,42,43. Moreover, the N-terminus was also found to interact with the antiapoptotic protein survivin and the tumor suppressor p14^ARF, thus playing a key role in the tumor process, but the specific mechanism remains elusive^44,45. In conclusion, GSPT1 has been extensively studied as a promising antitumor therapeutic target, but this study provides the first finding that GSPT1 may also be involved in the development of sepsis. Through RNA sequencing, we found that GSPT1 was lowly expressed in Non-survivor group (FC = -4.2752, FDR = 0.0496). The GO analysis found that GSPT1 was involved in biological processes such as regulation of immune response and negative regulation of defense response to virus and other Biological Processes. Meta-analysis found that GSPT1 was lowly expressed in Sepsis group compared with Normal Control group. But compared with the Survivor group, GSPT1 was highly expressed in Non-survivor group. This difference in expression may be due to multiple factors such as individual differences in sequenced samples and differences in sepsis stages. Sepsis is a dysregulated immune response, and the Single-cell RNA sequencing found that GSPT1 was widely localized in immune cells, mainly macrophages and T cells. This may suggest that GSPT1 expressed in immune cells may play a role in the development of sepsis, but the specific mechanism will require subsequent in-depth study. Combined with molecular interaction prediction, we hypothesize that HSA_CIRC_0026658, HSA_CIRC_0026654, HSA_CIRC_0026660 and HSA_CIRC_0026657 can regulate the expression of GSPT1 by binding to downstream miRNA, and widely participate in macrophages and T cell-mediated immune responses, affecting the prognosis of patients with sepsis.

The NPRL3 gene (also known as C16orf35), located on chromosome 6p13.3 and linked to the NPRL2 gene, is a highly conserved gene widely expressed during body development, encoding a protein containing 569 amino acids. To date, 534 single-nucleotide variants (SNVs) in the NPRL3 have been included in the ClinVar database (https://www.ncbi.nlm.nih.Gov/clinvar), including 19 likely pathogenic variants and 25 pathogenic variants⁴⁶. Nprl3 (64 kDa), together with Dep Domain Containing 5 (DEPDC5; 177 kDa), and Nitrogen Permease Regulator-Like 2 (NPRL2; 44 kDa), constitute the GAP activity toward Rags 1 (GATOR1) complex. In mammalian non-neural cells, GATOR1 complex has GTPases active protein (GAP) activity, can modulates rapamycin (mTOR) pathway activity in response to intracellular amino acid levels by governing translocation of mammalian target of rapamycin C1 (mTORC1) to and from the lysosomal membrane, thus affect the metabolic pathway of ATP production, and finally realize the regulation of nutrient sensing and cellular metabolic state of the body^47,48. Under nutrient-sufficient conditions, the GATOR1 complex is inhibited, releasing mTORC1 to physically interact with the binding partner on the lysosome surface in an active conformation. Active mTORC1 can stimulate protein synthesis and cell growth through the phosphorylation of downstream effectors such as S6K and 4E-BP, while simultaneously inhibiting catabolic metabolism and autophagy⁴⁹. In contrast, when nutrient limitation or intracellular amino acid content is low, GATOR1 can inhibit the activity of mTOR pathway by preventing mTORC1 translocation to the lysosome surface (the remaining cytosol), thus promoting catabolism and negatively regulating cell growth^50,51. Thus, by regulating the activity of mTORC1, GATOR1 allows cells to rapidly adjust their metabolic status in response to extracellular and intracellular stimuli, thus affecting the development of diseases such as epilepsy, cancer, diabetes, stroke and neurodegenerative diseases^52–55. On the other hand, NPRL3 also has important regulatory functions in the hematological system. NPRL3 is a key regulator of erythroid metabolism, and the loss of NPRL3 will enhance mTORC1 signaling, suppresse autophagy, and disrupt glycolysis and redox control in erythroblasts. Meanwhile, in the absence of NPRL3, human CD34 + progenitors produce fewer enucleated cells and demonstrate dysregulated mTORC1 signalling in response to nutrient availability and erythropoietin⁵⁶. Furthermore, NPRL3 controls the autophagic flux, regulating mTORC1 responses in erythroid cells not only to amino acid availability, but also to iron and EPO. The proliferation of megakaryocytic progenitors is also regulated by NPRL3, and the mTOR target pathway plays an important role in megakaryocyte terminal differentiation, including platelet function⁵⁷.

In sepsis, the role of NPRL3 was also gradually revealed. It is well known that Purinergic signaling is an important factor influencing the inflammatory response in sepsis. The P2X₇ receptor (P2X₇R), as a ligand-gated ion channel, belongs to the purinergic type 2 receptor family (P2). In sepsis, extracellular adenosine triphosphate activates P2X₇R, then the NLRP3 inflammasome, and followed by caspase-1 activation, which releases the mature interleukin (IL)-1β, thereby promoting the inflammatory response⁵⁸. Zou et al. have found that circ_0001679 and circ_0001212 have potential ceRNAs regulatory relationships with NPRL3⁵⁹. P2X₇R antagonists exert a protective effect against sepsis-induced ALI in mice by regulating the expression of circ_0001679 and circ_0001212 as well as downstream NPRL3. This study found that NPRL3 was lowly expressed in Non-survivor group (FC = -4.044, FDR = 0.0496). Meta-analysis found that NPRL3 was lowly expressed in Sepsis group compared with Normal Control group, and compared with the Survivor group, NPRL3 was lowly expressed in Non-survivor group and was consistent with RNA sequencing results. Single-cell RNA sequencing found that NPRL3 was widely localized in immune cells, mainly macrophages and T cells. Combined with molecular interaction prediction, we hypothesize that core circRNA, such as HSA_CIRC_0026658, HSA_CIRC_0026654, and HSA_CIRC_0026660, HSA_CIRC_0026657 can bind to miRNA, thereby indirectly regulating the expression of downstream target gene NPRL3, mediating the immune response, and participating in the regulation of prognosis in patients with sepsis.

In summary, sepsis-related circRNAs, miRNAs, and mRNAs do not function through a single mechanism, while there are close interactions between different RNAs. Through RNA sequencing and bioinformatics analysis, we propose a ceRNA regulatory network consisting of 4 circRNAs, 26 miRNAs, and 2 core genes, which are closely related to the prognosis of sepsis and may serve as novel biomarkers to guide the study of sepsis mechanisms.

There are some shortcomings in this study. We identified two biomarkers (GSPT1 and NPRL3) affecting sepsis outcome by bioinformatics technology. However, the specific mechanisms by which they affect the progression of sepsis have not been elucidated, and whether these two target genes are synergistic or antagonistic, which requires further cytological experiments and animal model studies.

Author contributions

G.C. and W.Z. contributed to drafing of the manuscript and statistical analysis. C.W., M.C.and Y.H. contributed to data collection. Z.W. contributed to study design and guidance of the study. All authors reviewed the manuscript.

Data availability

The RNA sequencing data generated during the current study are stored in the China National GeneBank DataBase (CNGBdb, https://db.cngb.org/), and researchers can access our data at any time (registration number: CNP0002611).

Declarations

Competing interests

The authors declare no competing interests.