Ninety-nine patients with sepsis who were admitted to the Department of Critical Care Medicine of the First Affiliated Hospital of Anhui Medical University from June 2022 to September 2023 were divided to two groups: sepsis group (n ¼ 51) and septic shock group (n ¼ 58). The inclusion criteria were: (1) Age 18 years old; (2) Length of stay in ICU > 2 days; (3) The diagnosis of sepsis and septic shock conforms to the international consensus on the definition of Sepsis3.0 published in April 2016, septic shock is defined as hypotension after full fluid resuscitation (mean blood pressure 65 mmHg under pressor maintenance) and serum lactic acid > 2 mmol/L (Chiu and Legrand, 2021); (4) signed the informed consent form to be enrolled in the study. In addition, the following exclusion criteria were applied: pregnant patients, patients who have recently experienced an acute heart attack (Lu et al., 2012), and those with malignant tumors; as well as patients with conditions that may affect their baseline immune status, such as autoimmune diseases, chronic viral infections, and any history of long-term immunosuppressant therapy. ( Figure 1 ). The study was approved by the Ethics Committee for Clinical Research of the First Affiliated Hospital of Anhui Medical University (approval number PJ2024-02-38).

Vital signs (blood pressure, heart rate, respiratory rate, and body temperature), routine laboratory test results (creatinine, bilirubin, platelet count, C-reactive protein (CRP), Procalcitonin (PCT), hemoglobin, hematocrit, sodium, potassium, white blood cell count and blood culture), blood gas analysis (pH, lactate (Lac), PaO₂, PaCO₂, bicarbonate (HCO₃), and base excess (BE)), Sequential Organ Failure Assessment (SOFA) score, Acute Physiology and Chronic Health Evaluation (APACHE) II score and personal information (age, sex) were collected. A summary of participants characteristics are show in the Supplementary Tables 1 - 3 .

Six- to eight-weeks old C57BL/6 male mice were purchased from Anhui Medical University (Hefei, China). They were randomly assigned to following two groups: sham surgery and sepsis. Mice in the sepsis group underwent cecal ligation and puncture (CLP) to induce sepsis, a classical mode. Mice in the sham group only underwent laparotomy. Briefly, mice were anesthetized with 2% pentobarbital and maintained at 37°C using a heating pad. A 1-cm midline abdominal incision was made following skin disinfection. To induce moderate sepsis, the cecum was ligated at its midpoint and punctured with a 21-gauge needle. A small amount of stool was squeezed out through the puncture site. The cecum was placed back into the abdomen, followed by closure of the peritoneum, fasciae and abdominal musculature with a sterile 6–0 silk suture. For mice that underwent operation, 50 ml/kg normal saline was injected subcutaneously. All mice experiments were approved by the Animal Research Ethics Society of Anhui Medical University.

The immortalized bone marrow-derived macrophages (iBMDM) were purchased from the Cell Bank of Chinese Academy of Sciences (Shanghai, China), and passed in DMEM (Gibco, NY, USA) enriched with 10% fetal bovine serum at 37 °C in a 5% CO₂ humidified incubator. The cells were identified through short tandem repeats profiling and mycoplasma contamination was determined using the Cell Culture Contamination Detection Kit (Thermo Fisher). The iBMEM cells were seeded in 10-cm dishes into which 10 mL of the medium was added. Adherent cells were stimulated with 1 μg/mL LPS (Sigma-Aldrich, L6529) for 12 h following a medium change. Subsequently, five million cells per group were harvested for total RNA extraction. Each experiment was performed in triplicate. Three independent repeat experiments were conducted separately under the same experimental conditions to ensure the reproducibility and reliability of the results.

Peripheral blood mononuclear cells were isolated using the peripheral blood mononuclear cell kit (Solarbio, Beijing, China) following the manufacturer’s protocol. Mice liver tissues were obtained by literature format (Lu et al., 2012). Total RNA was extracted from the peripheral blood mononuclear cells and mouse liver using TRIzol reagent (Invitrogen Life Technologies, Carlsbad, CA, USA) to determine their transcript responses. Total RNA was quantified and purified using the NanoDrop ND-2000 spectrophotometer (Thermo Scientific, Waltham, MA). RT-qPCR was performed by using the 2× Universal SYBR Green Fast qPCR Mix (Nanjing Nuoweizan Biotechnology Co. Q111, LTD, China) and gene specific primers ( Supplementary Table 4 ) in a CFX96 Real-time PCR detection system (Bio-Rad Laboratories GmbH, Munich, Germany).

Data normality was assessed using the Kolmogorov-Smirnov test. For normally distributed data, descriptive statistics (x ̅ ± SD) were calculated, and group differences were analyzed using the t-test. Descriptive indicators Data that did not follow normal distribution were expressed as the median and interquartile range (Q1-Q3) and the Mann-Whitney U test was used to compare groups. Qualitative data were presented as the frequency and percentage (n (%)), and chi-square test was used to analyze differences between groups. The receiver operating characteristic ROC) was developed to explore sensitivity and specificity, and Spearman correlation analysis was employed to determine correlation between expression levels of miR-9-5P and the clinical data.

For multivariable analysis, variables that showed significant associations in the univariate analysis, as well as those considered clinically important, were included to ensure the clinical plausibility of the model. We used stepwise backward regression to construct the model and employed the Hosmer-Lemeshow goodness-of-fit test to evaluate the model’s fit, with p > 0.05 considered indicative of a good fit. All data were processed and analyzed using IBM SPSS Statistics 26.0 (IBM Corporation, Armonk, NY, USA), R software (4.3.1), and GraphPad Prism 9.0. All cell-based experiments were performed in triplicate, and statistical comparisons between groups were conducted using an independent sample t-test. P<0.05 was considered statistically significant.

The gene expression data were analyzed on the GEO database through the NCBI portal, which collects and stores high-throughput gene expression data. The GSE152371 dataset was retrieved and processed. A volcano map was constructed using the Prism software ( Figure 2 ) to display the differential microRNA, applying the selection criteria of |log₂FC| > 0.5 and FDR < 0.05). The analysis identified 18 up-regulated genes, including miR-9-5p, miR-125a-3p, miR-155-5p, and 15 down-regulated genes, such as miR-23b-3p, miR-149-5p, and miR-203a-3p. The fold change of miR-9-5p was 1.88, with p < 0.01.

A literature review identified six genes associated with sepsis, which were then further analyzed in a constructed lipopolysaccharide-treated mouse iBMDM macrophage inflammation model to investigate their corresponding microRNAs. Figure 3 shows that there was no significant difference in expression level of miR-125 and miR-23b. Considering the prevailing evidence for the role of miR-155 in tumors, inflammation, and other diseases, we selected miR-9-5p which exhibited elevated differential expression in human and mouse based on homologous sequencing analysis. Its expression was highly stable during cell validation. Therefore, miR-9-5p was selected for further experiments.

The expression level of miR-9-5p in peripheral blood mononuclear cells from the sepsis group, non-sepsis infection group, and control group was quantified and the results are shown in Table 1 The level of miR-9-5p in sepsis patients was significantly higher compared with that in the control group (10.13[3.21-23.94] vs. 0.86[0.36-1.88]). In addition, the expression level of miR-9-5p was significantly higher in sepsis patients relative to the non-sepsis infection group, with a median fold change of 2.08[1.68-3.18]. Moreover, the level of miR-9-5p expression showed significant differences among the three groups, with the highest expression in the sepsis group, followed by the infection without sepsis group, and the lowest in the healthy control group. This indicates the importance of infection in miR-9-5p expression.

The analysis, shown in Figures 4B, C , compares miR-9-5p expression levels across different infection etiologies and sites of infection. The data reveal no statistically significant differences in miR-9-5p levels between positive and negative bacterial cultures or between lung and abdominal infections.

The miR-9-5p level in peripheral blood mononuclear cells was significantly higher in the non-survival group compared to the survival group (14.06 [3.35-27.93] vs. 2.32 [1.75-3.61]) ( Table 2 , Figure 4D ). Subsequently, we developed a cecal ligation and puncture model using 6–8-week-old C57BL/6 mice, and blood samples were collected at 24 hours for q-PCR analysis. The mRNA levels of IL-1β ( Figure 4D ) and TNF-α ( Figure 4E ) in the CLP group were significantly elevated compared to the sham group, confirming the successful construction of the model. Additionally, the miR-9-5p mRNA level in the monocytes of the CLP group was significantly higher compared to levels in the sham group ( Figure 4F ), indicating that miR-9-5p mRNA levels in mouse sepsis models are upregulated compared to sham groups, which is consistent with clinical observations.

To investigate the relationship between miR-9-5p mRNA levels and clinical indicators during the sepsis progression, we analyzed the correlation between miR-9-5p levels in monocytes and various clinical indicators ( Figure 5 ). It was observed that miR-9-5p levels in peripheral blood mononuclear cells were positively correlated with several clinical indicators: SOFA score (r=0.655, P<0.001; Figure 6A ), APACHE II score (r=0.382, P<0.001; Figure 6B ), BNP ( Figure 6G ), CRP (r=0.318, P=0.0013; Figure 6C ), PCT (r=0.3583, P<0.001; Figure 6D ), Tbil (r=0.328, P<0.001; Figure 6H ), and BUN (r=0.335, P<0.001; Figure 6F ). Conversely, miR-9-5p levels were negatively correlated with PLT (r=-0.341, P<0.001; Figure 6E ) and eGFR (r=-0.358, P<0.001; Figure 6I ).

The SOFA score is often utilized to evaluate the degree of organ dysfunction in sepsis and directly reflects disease severity. A high SOFA score indicates poor prognosis, and a SOFA score greater than 2 reflects organ dysfunction. A scatter plot with error bars was developed for using the number of organ failures and the relative expression level of miR-9-5p ( Figure 7C ). Analysis of the plot demonstrated a positive association between the levels of miR-9-5p and the number of failed organs. Figure 7D further illustrates the relative expression levels of miR-9-5p across different stratified groups based on the median SOFA score of 8 with an interquartile range from 8 to 14. Notably, there were significant differences in the miR-9-5p level among the groups.

In further experiments, we explored whether miR-9-5p levels in peripheral blood mononuclear cells may have good diagnostic value in sepsis utilizing the ROC curve and AUC based on previous results. The ROC curve analysis for miR-9-5p levels in the ICU, presented in Table 3 , demonstrates its diagnostic value for sepsis. As shown in Figure 8 , the diagnostic accuracy of miR-9-5p significantly surpassed that of CRP and PCT, with AUC values of 0.672 and 0.678, respectively (P < 0.001). The ROC curve derived from the combination of all three markers yielded an AUC of 0.858, which was not statistically different from the diagnostic performance of miR-9-5p alone.

Results of the univariate ROC curve analysis demonstrated the predictive value of miR-9-5p for septic shock ( Table 4 ). Multi-index ROC curves were plotted to explore the predictive performance for septic shock ( Figure 9 ). The prediction accuracy of miR-9-5p in septic shock patients was statistically better than the SOFA score, CRP, PCT, and APACHE II scores(P < 0.05).

We generated ROC curves for the miR-9-5p levels to explore their prediction accuracy for in-hospital death among patients with sepsis in ICU ( Figure 10 ). The AUC was 0.794 at 95% confidence interval, 0.705-0.883. Further analysis was conducted to determine whether clinically feasible measures contributed to the prediction of mortality in patients with sepsis. The calculated AUC values for the SOFA score, APACHE II score, and CRP were 0.772 (95% CI: 0.679-0.866), 0.779 (95% CI: 0.688-0.872), and 0.723 (95% CI: 0.625-0.823). Predictive value of MiR-9-5p for mortality in patients with sepsis is equivalent to that of the SOFA score and APACHE II score. To develop a more accurate evaluation system, we constructed a combined predictive model incorporating all indicators using logistic regression analysis. The combined model yielded an AUC of 0.888, and its predictive accuracy was significantly improved compared to that of miR-9-5p alone (p < 0.05).