We performed quantitative proteomics in plasma samples from 7 mild (M) and 10 severe (S) cases who were diagnosed with SFTS at the First Affiliated Hospital of Nanjing Medical University from April 2021 to July 2021. Five age- and gender-matched healthy (H) subjects were enrolled as controls (Cohort 1). To develop a better early severe prediction model, a total of 190 SFTS patients were enrolled in the same center from June 2015 to February 2021 (Cohort 2). Additionally, to validate its efficacy, 93 SFTS patients were prospectively recruited from March 2022 to November 2022 (Cohort 3). Plasma samples were obtained from all patients during the fever stage at the time of enrollment (Figure 1). Among the patients, plasma samples were collected from 86 cases at different stages of the disease for dynamic analysis of protein markers. In addition, 20 healthy subjects, whose serological tests were negative for DBV, were enrolled for comparison.

Diagnostic criteria of SFTS were as follows: (1) possibility of tick bite history; (2) acute fever, fatigue with thrombocytopenia; and (3) positive DBV ribonucleic acid (RNA) confirmed by real-time quantitative reverse-transcription polymerase chain reaction in plasma.

According to previous reports (Chen G. et al., 2022; Li and Wang, 2022), cases that met any one of the following criteria were classified as severe: (1) presenting with severe neurological symptoms, (2) multiple organ failure, (3) severe bleeding, (4) severe secondary infection, (5) platelet count <30 × 10⁹/L, (6) AST, LDH, CK >10 × ULN (upper limit of normal) or (7) death. Exclusive criteria were as follows: (1) laboratory-confirmed other pathogen infections, (2) underlying serious systemic diseases, or (3) severe immune deficiency. Demographic and laboratory data were obtained from all the patients. The survival status of all patients was followed for 28 days post-disease onset.

Written informed consent was obtained from each patient following the principles of the Declaration of Helsinki. The study was approved by the Research and Ethics Committee of the First Affiliated Hospital of Nanjing Medical University, Nanjing, China (Ethics approval number, 2022-SR-366).

Blood samples from all patients were collected in K2-EDTA tubes at the indicated time points. The blood was processed immediately to centrifuge at 670 g for 10 min at 20°C and the plasma was aliquoted and stored at −80°C until at the time of assays.

Viral RNA was extracted from plasma samples with a commercial kit (Daan Gene, Guangzhou, China) by a trained specialist. The extracted RNA was amplified using specific primers and probes by one-step real-time fluorescence qRT-PCR for the detection of a novel bunyavirus in all specimens according to the manufacturer’s instructions. Real-time qRT-PCR cycling was performed as follows: after reverse transcription at 50°C for 15 min, polymerase was activated at 95°C for 15 min, and amplification was undertaken for 45 cycles consisting of a denaturing step at 94°C for 15 s and an annealing-extension step at 55°C for 45 s. The fluorogenic signal emitted was collected during the extension step, and a cycle threshold was acquired for quantifying the viral load of the sample.

The samples were fractionated using a high pH reversed-phase fractionator, peptide fractions were subsequently analyzed by LC-MS/MS using the Thermo Scientific Orbitrap Exploris^™ 480 platform (Thermo Fisher Scientific) via data-independent acquisition (DIA) scan mode. Mass spectrometry data were used to generate a hybrid library in the Spectronaut software (Biognosys, version 15.7) for final protein identification and quantitation. All searches were performed against the human UniProt proteome sequences, LianChuan Biotechnology Company (Hangzhou, China) provides related technical support. Detailed procedures can be found in the Supplementary material.

Acquired data were analyzed by Excel 2021 and R (Version 4.2.0). Proteins with missing values in more than 60% of samples were excluded. The remaining missing values were interpolated by the median. Principal component analysis (PCA) and Clustering heatmaps of proteins using “ggplot2” and “pheatmap” packages, respectively. Differentially expressed proteins (DEPs) were defined as average ratio-fold change (FC) >1.5 as well as p-value <0.05. The DEPs were subjected to Gene Ontology (GO) functional clustering analyses.

Human protein ELISA kits were used to quantify plasma levels of proteins according to manufacturers’ instructions. Briefly, plasma samples were diluted according to the manufacturer’s dilution guidelines. Then, 100 μL of fixed dilution plasma sample was added to the plates and incubated for 90 min at 37°C. After washing, 100 μL biotinylated-specific antibody was immediately added to each well, and the plates were incubated for 60 min at 37°C. Followed by washing, 100 μL Avidin-horseradish peroxidase was added and incubated for 30 min at 37°C. Finally, the optical density (OD) value at 450 nm was determined after the addition of 90 μL tetramethyl-benzidine reagent and stop solution. The standard curve of each protein was generated by the determination of OD values from serial dilutions of the standard samples with known protein concentrations. Detailed information on ELISA kits can be found in Supplementary Table 1.

Statistical analyses were performed using SPSS 25.0 (IBM, Armonk, NY) or otherwise specified. Values are expressed as the mean ± standard deviation for normally distributed or as the median (p₂₅, p₇₅) for nonnormally distributed, whereas categorical variables are expressed as percentages. The independent Student t-test, nonparametric Mann–Whitney U test, or χ² test was used to determine differences among groups. Correlations were analyzed using Spearman’s correlation analysis. Logistic regression analysis was used to identify variables for predicting severe SFTS patients. The area under the receiver operating characteristic curve (auROC) was calculated and compared by the Z test (Delong’s method). Differences with a p < 0.05 were considered statistically significant.

To identify the potential protein biomarkers, we performed quantitative proteomics in plasma samples from Cohort 1 (5 H, 7 M, and 10 S). The clinical baseline characteristics of Cohort 1 were described in Table 1. A total of 583 proteins were identified by proteomics analysis. PCA showed that the protein profiles of the severe and mild patients were visibly distinct from the healthy control, and protein distribution in the severe group was segregated from the mild group (Figure 2A). Similarly, the heatmap clustering analysis also demonstrated differences in protein expression profiles among the three groups (Figure 2B), indicating unique proteomic features in severe cases for further exploration.

Next, we picked out several plasma proteins related to severe cases from proteomic results. The volcano plot showed that, Compared to healthy controls, a total of 120 significantly upregulated and 72 significantly downregulated DEPs were identified in mild patients (Figure 2C). GO enrichment analysis showed that DEPs were highly enriched in biological processes involved in humoral immune response, phagocytosis, hemostasis, and coagulation (Figure 2D). Compared to mild patients, a total of 83 significantly upregulated and 15 significantly downregulated DEPs were identified in severe patients (Figure 2E). The GO-terms pathways of the DEPs were highly enriched in processes related to proteasomal protein catabolic process, phagocytosis, and humoral immune response (Figure 2F). We further found that 46 DEPs exhibited an increasing trend, while three DEPs exhibited a decreasing trend among the H, M, and S groups (Figure 2G). Then, based on their abundance and biological functions primarily involved in processes such as inflammation, immunity response, metabolism, or coagulation which were related to pathological and physiological changes in the disease, 18 DEPs were finally selected for validation in plasma using ELISA (Table 2). Finally, consistent with proteomics results, four proteins, namely heat shock protein 90α (HSP90α), vascular cell adhesion protein 1 (VCAM1), proteasome subunit alpha type 1 (PSMA1), and nidogen 1 (NID1) were elevated in the early-stage plasma of severe SFTS patients (Figure 2H). Thus, we choose those four proteins to serve as candidate biomarkers for further research.

According to the typical clinical features, the course of SFTS has the following phases: the fever stage (1–7 days after onset, early-stage), the MOD stage (8–14 days), and the convalescent stage (>14 days). We analyzed the alternations of candidate protein expression in plasma from 86 cases in these three phases (20 H, 26 M, and 60 S). Results showed that the levels of candidate proteins including HSP90α, VCAM1, PSMA1, and NID1 were higher in the fever stage of SFTS patients compared with healthy controls, almost reaching plateaus in the MOD stage, and the concentrations were decreased in the convalescent stage (Figures 3A–D). Furthermore, the candidate protein levels were all higher in the severe cases compared to the mild cases both in the fever and MOD stages (Figures 3E–H). It was particularly noteworthy that the candidate protein levels were significantly higher in the non-survivors compared to the survivors in both the MOD and convalescence stages (Figures 3I–L).

Plasma samples in early-stage and clinical parameters were further obtained from 283 SFTS patients, then, we used ELISA to detect the candidate protein concentrations. A correlation analysis revealed that HSP90α, VCAM1, PSMA1, and NID1 were correlated with the levels of severity-related indicators such as PLT, APTT, AST, LDH, CK, and sCr (Figure 4A), suggesting these proteins can serve as markers reflecting the multi-organ damage in patients with SFTS.

These results indicated those four candidate proteins had a close association with the disease progression, and higher concentrations of those protein markers indicate a more severe state of illness.

In order to develop an early prediction model for severe illness, protein markers were detected in plasma by ELISA and clinical data were collected in the early stage from 190 patients with SFTS from June 2015 to February 2021 (Cohort 2). The clinical characteristics and the concentration of four protein markers were summarized in Table 3. We performed univariate logistic regression analysis to identify variables with p-values less than 0.001. Subsequently, we constructed an optimal model using the stepwise method, where a variable was included if its p-value was less than 0.05 and removed if its p-value exceeded 0.1. This process resulted in a model comprised of five predictor variables including LgDBV, CK, HSP90α, VCAM1, and PSMA1 (P5, all with auROC >0.75), named LCHVP. The model can be expressed as follows:

ROC curve analysis revealed that P5 had the highest auROC of 0.931 (95% CI, 0.885, 0.963) at the optimal cut-off value of 0.393. The sensitivity and specificity of P5 for predicting severe SFTS were 90.1 and 83.2%, respectively (Figure 4B and Table 4). Subsequently, we performed external validation in Cohort 3, which consisted of 93 patients from March 2022 to November 2022. There were no differences in age, death rate or any laboratory values between Cohort 2 and Cohort 3 (Table 5), and the results showed that the auROC in the validation group was 0.923 (Figure 4C). Moreover, we found that patients with a P5 value greater than 0.393 had a fatality rate of 29.9% (95% CI, 23.3–37.3%), which was significantly higher than the fatality rate in patients with lower P5 levels [1.9% (95% CI, 0.2–6.5%), p <0.001; Figure 4D].